commit 32282c0b71acf16f8871352383506d1544bdd16f Author: ModelHub XC Date: Sat Aug 15 17:06:00 2026 +0800 初始化项目，由ModelHub XC社区提供模型 Model: Neelectric/Llama-3.1-8B-Instruct_SFT_Math-220kv00.35 Source: Original Platform diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..52373fe --- /dev/null +++ b/.gitattributes @@ -0,0 +1,36 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000..6c09218 --- /dev/null +++ b/README.md @@ -0,0 +1,60 @@ +--- +base_model: meta-llama/Llama-3.1-8B-Instruct +datasets: Neelectric/Replay_0.04.OpenR1-Math-220k_extended.wildguardmix.Llama3_4096toks +library_name: transformers +model_name: Llama-3.1-8B-Instruct_SFT_Math-220kv00.35 +tags: +- generated_from_trainer +- sft +- open-r1 +- trl +licence: license +--- + +# Model Card for Llama-3.1-8B-Instruct_SFT_Math-220kv00.35 + +This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on the [Neelectric/Replay_0.04.OpenR1-Math-220k_extended.wildguardmix.Llama3_4096toks](https://huggingface.co/datasets/Neelectric/Replay_0.04.OpenR1-Math-220k_extended.wildguardmix.Llama3_4096toks) dataset. +It has been trained using [TRL](https://github.com/huggingface/trl). + +## Quick start + +```python +from transformers import pipeline + +question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?" +generator = pipeline("text-generation", model="Neelectric/Llama-3.1-8B-Instruct_SFT_Math-220kv00.35", device="cuda") +output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0] +print(output["generated_text"]) +``` + +## Training procedure + +[Visualize in Weights & Biases](https://wandb.ai/neelectric/open-r1_safety/runs/1ceyb6ir) + + +This model was trained with SFT. + +### Framework versions + +- TRL: 0.28.0.dev0 +- Transformers: 4.57.6 +- Pytorch: 2.9.0 +- Datasets: 4.5.0 +- Tokenizers: 0.22.2 + +## Citations + + + +Cite TRL as: + +```bibtex +@misc{vonwerra2022trl, + title = {{TRL: Transformer Reinforcement Learning}}, + author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallou{\'e}dec}, + year = 2020, + journal = {GitHub repository}, + publisher = {GitHub}, + howpublished = {\url{https://github.com/huggingface/trl}} +} +``` \ No newline at end of file diff --git a/all_results.json b/all_results.json new file mode 100644 index 0000000..96f55f2 --- /dev/null +++ b/all_results.json @@ -0,0 +1,8 @@ +{ + "total_flos": 2.554786723790219e+19, + "train_loss": 0.3743892161239599, + "train_runtime": 22902.1621, + "train_samples": 86158, + "train_samples_per_second": 11.286, + "train_steps_per_second": 0.705 +} \ No newline at end of file diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000..f0a4f20 --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,114 @@ +{{- bos_token }} +{%- if custom_tools is defined %} + {%- set tools = custom_tools %} +{%- endif %} +{%- if not tools_in_user_message is defined %} + {%- set tools_in_user_message = true %} +{%- endif %} +{%- if not date_string is defined %} + {%- set date_string = "26 Jul 2024" %} +{%- endif %} +{%- if not tools is defined %} + {%- set tools = none %} +{%- endif %} + +{#- This block extracts the system message, so we can slot it into the right place. #} +{%- if messages[0]['role'] == 'system' %} + {%- set system_message = messages[0]['content']|trim %} + {%- set messages = messages[1:] %} +{%- else %} + {%- set system_message = "You are a helpful AI Assistant that provides well-reasoned and detailed responses. You first think about the reasoning process as an internal monologue and then provide the user with the answer. Respond in the following format: +... + + +... +" %} +{%- endif %} + +{#- System message + builtin tools #} +{{- "<|start_header_id|>system<|end_header_id|>\n\n" }} +{%- if builtin_tools is defined or tools is not none %} + {{- "Environment: ipython\n" }} +{%- endif %} +{%- if builtin_tools is defined %} + {{- "Tools: " + builtin_tools | reject('equalto', 'code_interpreter') | join(", ") + "\n\n"}} +{%- endif %} +{{- "Cutting Knowledge Date: December 2023\n" }} +{{- "Today Date: " + date_string + "\n\n" }} +{%- if tools is not none and not tools_in_user_message %} + {{- "You have access to the following functions. To call a function, please respond with JSON for a function call." }} + {{- 'Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}.' }} + {{- "Do not use variables.\n\n" }} + {%- for t in tools %} + {{- t | tojson(indent=4) }} + {{- "\n\n" }} + {%- endfor %} +{%- endif %} +{{- system_message }} +{{- "<|eot_id|>" }} + +{#- Custom tools are passed in a user message with some extra guidance #} +{%- if tools_in_user_message and not tools is none %} + {#- Extract the first user message so we can plug it in here #} + {%- if messages | length != 0 %} + {%- set first_user_message = messages[0]['content']|trim %} + {%- set messages = messages[1:] %} + {%- else %} + {{- raise_exception("Cannot put tools in the first user message when there's no first user message!") }} +{%- endif %} + {{- '<|start_header_id|>user<|end_header_id|>\n\n' -}} + {{- "Given the following functions, please respond with a JSON for a function call " }} + {{- "with its proper arguments that best answers the given prompt.\n\n" }} + {{- 'Respond in the format {"name": function name, "parameters": dictionary of argument name and its value}.' }} + {{- "Do not use variables.\n\n" }} + {%- for t in tools %} + {{- t | tojson(indent=4) }} + {{- "\n\n" }} + {%- endfor %} + {{- first_user_message + "<|eot_id|>"}} +{%- endif %} + +{%- for message in messages %} + {%- if not (message.role == 'ipython' or message.role == 'tool' or 'tool_calls' in message) %} + {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' }} + {%- elif 'tool_calls' in message %} + {%- if not message.tool_calls|length == 1 %} + {{- raise_exception("This model only supports single tool-calls at once!") }} + {%- endif %} + {%- set tool_call = message.tool_calls[0].function %} + {%- if builtin_tools is defined and tool_call.name in builtin_tools %} + {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' -}} + {{- "<|python_tag|>" + tool_call.name + ".call(" }} + {%- for arg_name, arg_val in tool_call.arguments | items %} + {{- arg_name + '="' + arg_val + '"' }} + {%- if not loop.last %} + {{- ", " }} + {%- endif %} + {%- endfor %} + {{- ")" }} + {%- else %} + {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' -}} + {{- '{"name": "' + tool_call.name + '", ' }} + {{- '"parameters": ' }} + {{- tool_call.arguments | tojson }} + {{- "}" }} + {%- endif %} + {%- if builtin_tools is defined %} + {#- This means we're in ipython mode #} + {{- "<|eom_id|>" }} + {%- else %} + {{- "<|eot_id|>" }} + {%- endif %} + {%- elif message.role == "tool" or message.role == "ipython" %} + {{- "<|start_header_id|>ipython<|end_header_id|>\n\n" }} + {%- if message.content is mapping or message.content is iterable %} + {{- message.content | tojson }} + {%- else %} + {{- message.content }} + {%- endif %} + {{- "<|eot_id|>" }} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' }} +{%- endif %} diff --git a/config.json b/config.json new file mode 100644 index 0000000..e1d9068 --- /dev/null +++ b/config.json @@ -0,0 +1,35 @@ +{ + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "bos_token_id": 128000, + "dtype": "bfloat16", + "eos_token_id": 128009, + "head_dim": 128, + "hidden_act": "silu", + "hidden_size": 4096, + "initializer_range": 0.02, + "intermediate_size": 14336, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "num_attention_heads": 32, + "num_hidden_layers": 32, + "num_key_value_heads": 8, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "transformers_version": "4.57.6", + "use_cache": true, + "vocab_size": 128256 +} diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000..1996dc1 --- /dev/null +++ b/generation_config.json @@ -0,0 +1,8 @@ +{ + "bos_token_id": 128000, + "do_sample": true, + "eos_token_id": 128009, + "temperature": 0.6, + "top_p": 0.9, + "transformers_version": "4.57.6" +} diff --git a/model-00001-of-00004.safetensors b/model-00001-of-00004.safetensors new file mode 100644 index 0000000..e3f14ea --- /dev/null +++ b/model-00001-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa9670b103eeb0102d481deb43310f92675849b4a083f6dd3b06bda3f84e3bb4 +size 4976698672 diff --git a/model-00002-of-00004.safetensors b/model-00002-of-00004.safetensors new file mode 100644 index 0000000..4b32b0b --- /dev/null +++ b/model-00002-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f73fc56e2fd9ad99898a5550284452f060f9f64b631cde0ea89968c0a95ca3b0 +size 4999802720 diff --git a/model-00003-of-00004.safetensors b/model-00003-of-00004.safetensors new file mode 100644 index 0000000..7086ad4 --- /dev/null +++ b/model-00003-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:447bbd81ddcf60d38d946f2502c29dd160bfcb4b16719f96353a6633f22d9f91 +size 4915916176 diff --git a/model-00004-of-00004.safetensors b/model-00004-of-00004.safetensors new file mode 100644 index 0000000..e08c6e6 --- /dev/null +++ b/model-00004-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:549aa2c3611f58be268d5100430d4115fd27e7e8a0fd2c9d40eac5c650783b46 +size 1168138808 diff --git a/model.safetensors.index.json b/model.safetensors.index.json new file mode 100644 index 0000000..5c64f1e --- /dev/null +++ b/model.safetensors.index.json @@ -0,0 +1,299 @@ +{ + "metadata": { + "total_parameters": 8030261248, + "total_size": 16060522496 + }, + "weight_map": { + "lm_head.weight": "model-00004-of-00004.safetensors", + "model.embed_tokens.weight": "model-00001-of-00004.safetensors", + "model.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.0.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.1.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.10.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.10.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.11.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.12.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.13.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.14.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.15.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.15.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.15.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.16.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.16.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.16.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.17.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.17.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.17.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.18.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.18.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.18.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.19.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.19.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.19.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.2.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.20.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.20.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.20.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.20.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.20.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.20.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.20.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.20.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.21.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.21.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.22.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.23.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.23.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.23.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.24.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.24.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.24.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.25.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.25.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.25.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.26.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.26.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.26.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.27.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.27.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.27.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.28.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.28.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.28.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.29.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.29.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.29.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.3.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.30.input_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.30.mlp.down_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.post_attention_layernorm.weight": "model-00003-of-00004.safetensors", + "model.layers.30.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.30.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.input_layernorm.weight": "model-00004-of-00004.safetensors", + "model.layers.31.mlp.down_proj.weight": "model-00004-of-00004.safetensors", + "model.layers.31.mlp.gate_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.mlp.up_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.post_attention_layernorm.weight": "model-00004-of-00004.safetensors", + "model.layers.31.self_attn.k_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.self_attn.o_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.self_attn.q_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.31.self_attn.v_proj.weight": "model-00003-of-00004.safetensors", + "model.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.4.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.5.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.6.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.7.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.input_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.8.mlp.down_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.mlp.gate_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.mlp.up_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.post_attention_layernorm.weight": "model-00001-of-00004.safetensors", + "model.layers.8.self_attn.k_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.self_attn.o_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.self_attn.q_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.8.self_attn.v_proj.weight": "model-00001-of-00004.safetensors", + "model.layers.9.input_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.9.mlp.down_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.mlp.up_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00004.safetensors", + "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00004.safetensors", + "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00004.safetensors", + "model.norm.weight": "model-00004-of-00004.safetensors" + } +} diff --git a/special_tokens_map.json b/special_tokens_map.json new file mode 100644 index 0000000..e8f05fa --- /dev/null +++ b/special_tokens_map.json @@ -0,0 +1,10 @@ +{ + "bos_token": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false + }, + "eos_token": "<|eot_id|>" +} diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000..1c1d8d5 --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b +size 17209920 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000..8b0c7c1 --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,2062 @@ +{ + "added_tokens_decoder": { + "128000": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128001": { + "content": "<|end_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128002": { + "content": "<|reserved_special_token_0|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128003": { + "content": "<|reserved_special_token_1|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128004": { + "content": "<|finetune_right_pad_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128005": { + "content": "<|reserved_special_token_2|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128006": { + "content": "<|start_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128007": { + "content": "<|end_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128008": { + "content": "<|eom_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128009": { + "content": "<|eot_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128010": { + "content": "<|python_tag|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128011": { + "content": "<|reserved_special_token_3|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128012": { + "content": "<|reserved_special_token_4|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128013": { + "content": "<|reserved_special_token_5|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128014": { + "content": "<|reserved_special_token_6|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128015": { + "content": "<|reserved_special_token_7|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128016": { + "content": "<|reserved_special_token_8|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128017": { + "content": "<|reserved_special_token_9|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128018": { + "content": "<|reserved_special_token_10|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128019": { + "content": "<|reserved_special_token_11|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128020": { + "content": "<|reserved_special_token_12|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128021": { + "content": "<|reserved_special_token_13|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128022": { + "content": "<|reserved_special_token_14|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128023": { + "content": "<|reserved_special_token_15|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128024": { + "content": "<|reserved_special_token_16|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128025": { + "content": "<|reserved_special_token_17|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128026": { + "content": "<|reserved_special_token_18|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128027": { + "content": "<|reserved_special_token_19|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128028": { + "content": "<|reserved_special_token_20|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128029": { + "content": "<|reserved_special_token_21|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128030": { + "content": "<|reserved_special_token_22|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128031": { + "content": "<|reserved_special_token_23|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128032": { + "content": "<|reserved_special_token_24|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128033": { + "content": "<|reserved_special_token_25|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128034": { + "content": "<|reserved_special_token_26|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128035": { + "content": "<|reserved_special_token_27|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128036": { + "content": "<|reserved_special_token_28|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128037": { + "content": "<|reserved_special_token_29|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128038": { + "content": "<|reserved_special_token_30|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128039": { + "content": "<|reserved_special_token_31|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128040": { + "content": "<|reserved_special_token_32|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128041": { + "content": "<|reserved_special_token_33|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128042": { + "content": "<|reserved_special_token_34|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128043": { + "content": "<|reserved_special_token_35|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128044": { + "content": "<|reserved_special_token_36|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128045": { + "content": "<|reserved_special_token_37|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128046": { + "content": "<|reserved_special_token_38|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128047": { + "content": "<|reserved_special_token_39|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128048": { + "content": "<|reserved_special_token_40|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128049": { + "content": "<|reserved_special_token_41|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128050": { + "content": "<|reserved_special_token_42|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128051": { + "content": "<|reserved_special_token_43|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128052": { + "content": "<|reserved_special_token_44|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128053": { + "content": "<|reserved_special_token_45|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128054": { + "content": "<|reserved_special_token_46|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128055": { + "content": "<|reserved_special_token_47|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128056": { + "content": "<|reserved_special_token_48|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128057": { + "content": "<|reserved_special_token_49|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128058": { + "content": "<|reserved_special_token_50|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128059": { + "content": "<|reserved_special_token_51|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128060": { + "content": "<|reserved_special_token_52|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128061": { + "content": "<|reserved_special_token_53|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128062": { + "content": "<|reserved_special_token_54|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128063": { + "content": "<|reserved_special_token_55|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128064": { + "content": "<|reserved_special_token_56|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128065": { + "content": "<|reserved_special_token_57|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128066": { + "content": "<|reserved_special_token_58|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128067": { + "content": "<|reserved_special_token_59|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128068": { + "content": "<|reserved_special_token_60|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128069": { + "content": "<|reserved_special_token_61|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128070": { + "content": "<|reserved_special_token_62|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128071": { + "content": "<|reserved_special_token_63|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128072": { + "content": "<|reserved_special_token_64|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128073": { + "content": "<|reserved_special_token_65|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128074": { + "content": "<|reserved_special_token_66|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128075": { + "content": "<|reserved_special_token_67|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128076": { + "content": "<|reserved_special_token_68|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128077": { + "content": "<|reserved_special_token_69|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128078": { + "content": "<|reserved_special_token_70|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128079": { + "content": "<|reserved_special_token_71|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128080": { + "content": "<|reserved_special_token_72|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128081": { + "content": "<|reserved_special_token_73|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128082": { + "content": "<|reserved_special_token_74|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128083": { + "content": "<|reserved_special_token_75|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128084": { + "content": "<|reserved_special_token_76|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128085": { + "content": "<|reserved_special_token_77|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128086": { + "content": "<|reserved_special_token_78|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128087": { + "content": "<|reserved_special_token_79|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128088": { + "content": "<|reserved_special_token_80|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128089": { + "content": "<|reserved_special_token_81|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128090": { + "content": "<|reserved_special_token_82|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128091": { + "content": "<|reserved_special_token_83|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128092": { + "content": "<|reserved_special_token_84|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128093": { + "content": "<|reserved_special_token_85|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128094": { + "content": "<|reserved_special_token_86|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128095": { + "content": "<|reserved_special_token_87|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128096": { + "content": "<|reserved_special_token_88|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128097": { + "content": "<|reserved_special_token_89|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128098": { + "content": "<|reserved_special_token_90|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128099": { + "content": "<|reserved_special_token_91|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128100": { + "content": "<|reserved_special_token_92|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128101": { + "content": "<|reserved_special_token_93|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128102": { + "content": "<|reserved_special_token_94|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128103": { + "content": "<|reserved_special_token_95|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128104": { + "content": "<|reserved_special_token_96|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128105": { + "content": "<|reserved_special_token_97|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128106": { + "content": "<|reserved_special_token_98|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128107": { + "content": "<|reserved_special_token_99|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128108": { + "content": "<|reserved_special_token_100|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128109": { + "content": "<|reserved_special_token_101|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128110": { + "content": "<|reserved_special_token_102|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128111": { + "content": "<|reserved_special_token_103|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128112": { + "content": "<|reserved_special_token_104|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128113": { + "content": "<|reserved_special_token_105|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128114": { + "content": "<|reserved_special_token_106|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128115": { + "content": "<|reserved_special_token_107|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128116": { + "content": "<|reserved_special_token_108|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128117": { + "content": "<|reserved_special_token_109|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128118": { + "content": "<|reserved_special_token_110|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128119": { + "content": "<|reserved_special_token_111|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128120": { + "content": "<|reserved_special_token_112|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128121": { + "content": "<|reserved_special_token_113|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128122": { + "content": "<|reserved_special_token_114|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128123": { + "content": "<|reserved_special_token_115|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128124": { + "content": "<|reserved_special_token_116|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128125": { + "content": "<|reserved_special_token_117|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128126": { + "content": "<|reserved_special_token_118|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128127": { + "content": "<|reserved_special_token_119|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128128": { + "content": "<|reserved_special_token_120|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128129": { + "content": "<|reserved_special_token_121|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128130": { + "content": "<|reserved_special_token_122|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128131": { + "content": "<|reserved_special_token_123|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128132": { + "content": "<|reserved_special_token_124|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128133": { + "content": "<|reserved_special_token_125|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128134": { + "content": "<|reserved_special_token_126|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128135": { + "content": "<|reserved_special_token_127|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128136": { + "content": "<|reserved_special_token_128|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128137": { + "content": "<|reserved_special_token_129|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128138": { + "content": "<|reserved_special_token_130|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128139": { + "content": "<|reserved_special_token_131|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128140": { + "content": "<|reserved_special_token_132|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128141": { + "content": "<|reserved_special_token_133|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128142": { + "content": "<|reserved_special_token_134|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128143": { + "content": "<|reserved_special_token_135|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128144": { + "content": "<|reserved_special_token_136|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128145": { + "content": "<|reserved_special_token_137|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128146": { + "content": "<|reserved_special_token_138|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128147": { + "content": "<|reserved_special_token_139|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128148": { + "content": "<|reserved_special_token_140|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128149": { + "content": "<|reserved_special_token_141|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128150": { + "content": "<|reserved_special_token_142|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128151": { + "content": "<|reserved_special_token_143|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128152": { + "content": "<|reserved_special_token_144|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128153": { + "content": "<|reserved_special_token_145|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128154": { + "content": "<|reserved_special_token_146|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128155": { + "content": "<|reserved_special_token_147|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128156": { + "content": "<|reserved_special_token_148|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128157": { + "content": "<|reserved_special_token_149|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128158": { + "content": "<|reserved_special_token_150|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128159": { + "content": "<|reserved_special_token_151|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128160": { + "content": "<|reserved_special_token_152|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128161": { + "content": "<|reserved_special_token_153|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128162": { + "content": "<|reserved_special_token_154|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128163": { + "content": "<|reserved_special_token_155|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128164": { + "content": "<|reserved_special_token_156|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128165": { + "content": "<|reserved_special_token_157|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128166": { + "content": "<|reserved_special_token_158|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128167": { + "content": "<|reserved_special_token_159|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128168": { + "content": "<|reserved_special_token_160|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128169": { + "content": "<|reserved_special_token_161|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128170": { + "content": "<|reserved_special_token_162|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128171": { + "content": "<|reserved_special_token_163|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128172": { + "content": "<|reserved_special_token_164|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128173": { + "content": "<|reserved_special_token_165|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128174": { + "content": "<|reserved_special_token_166|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128175": { + "content": "<|reserved_special_token_167|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128176": { + "content": "<|reserved_special_token_168|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128177": { + "content": "<|reserved_special_token_169|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128178": { + "content": "<|reserved_special_token_170|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128179": { + "content": "<|reserved_special_token_171|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128180": { + "content": "<|reserved_special_token_172|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128181": { + "content": "<|reserved_special_token_173|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128182": { + "content": "<|reserved_special_token_174|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128183": { + "content": "<|reserved_special_token_175|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128184": { + "content": "<|reserved_special_token_176|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128185": { + "content": "<|reserved_special_token_177|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128186": { + "content": "<|reserved_special_token_178|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128187": { + "content": "<|reserved_special_token_179|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128188": { + "content": "<|reserved_special_token_180|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128189": { + "content": "<|reserved_special_token_181|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128190": { + "content": "<|reserved_special_token_182|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128191": { + "content": "<|reserved_special_token_183|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128192": { + "content": "<|reserved_special_token_184|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128193": { + "content": "<|reserved_special_token_185|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128194": { + "content": "<|reserved_special_token_186|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128195": { + "content": "<|reserved_special_token_187|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128196": { + "content": "<|reserved_special_token_188|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128197": { + "content": "<|reserved_special_token_189|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128198": { + "content": "<|reserved_special_token_190|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128199": { + "content": "<|reserved_special_token_191|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128200": { + "content": "<|reserved_special_token_192|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128201": { + "content": "<|reserved_special_token_193|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128202": { + "content": "<|reserved_special_token_194|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128203": { + "content": "<|reserved_special_token_195|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128204": { + "content": "<|reserved_special_token_196|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128205": { + "content": "<|reserved_special_token_197|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128206": { + "content": "<|reserved_special_token_198|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128207": { + "content": "<|reserved_special_token_199|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128208": { + "content": "<|reserved_special_token_200|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128209": { + "content": "<|reserved_special_token_201|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128210": { + "content": "<|reserved_special_token_202|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128211": { + "content": "<|reserved_special_token_203|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128212": { + "content": "<|reserved_special_token_204|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128213": { + "content": "<|reserved_special_token_205|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128214": { + "content": "<|reserved_special_token_206|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128215": { + "content": "<|reserved_special_token_207|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128216": { + "content": "<|reserved_special_token_208|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128217": { + "content": "<|reserved_special_token_209|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128218": { + "content": "<|reserved_special_token_210|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128219": { + "content": "<|reserved_special_token_211|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128220": { + "content": "<|reserved_special_token_212|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128221": { + "content": "<|reserved_special_token_213|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128222": { + "content": "<|reserved_special_token_214|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128223": { + "content": "<|reserved_special_token_215|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128224": { + "content": "<|reserved_special_token_216|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128225": { + "content": "<|reserved_special_token_217|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128226": { + "content": "<|reserved_special_token_218|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128227": { + "content": "<|reserved_special_token_219|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128228": { + "content": "<|reserved_special_token_220|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128229": { + "content": "<|reserved_special_token_221|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128230": { + "content": "<|reserved_special_token_222|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128231": { + "content": "<|reserved_special_token_223|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128232": { + "content": "<|reserved_special_token_224|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128233": { + "content": "<|reserved_special_token_225|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128234": { + "content": "<|reserved_special_token_226|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128235": { + "content": "<|reserved_special_token_227|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128236": { + "content": "<|reserved_special_token_228|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128237": { + "content": "<|reserved_special_token_229|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128238": { + "content": "<|reserved_special_token_230|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128239": { + "content": "<|reserved_special_token_231|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128240": { + "content": "<|reserved_special_token_232|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128241": { + "content": "<|reserved_special_token_233|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128242": { + "content": "<|reserved_special_token_234|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128243": { + "content": "<|reserved_special_token_235|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128244": { + "content": "<|reserved_special_token_236|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128245": { + "content": "<|reserved_special_token_237|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128246": { + "content": "<|reserved_special_token_238|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128247": { + "content": "<|reserved_special_token_239|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128248": { + "content": "<|reserved_special_token_240|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128249": { + "content": "<|reserved_special_token_241|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128250": { + "content": "<|reserved_special_token_242|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128251": { + "content": "<|reserved_special_token_243|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128252": { + "content": "<|reserved_special_token_244|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128253": { + "content": "<|reserved_special_token_245|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128254": { + "content": "<|reserved_special_token_246|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128255": { + "content": "<|reserved_special_token_247|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + } + }, + "bos_token": "<|begin_of_text|>", + "clean_up_tokenization_spaces": true, + "eos_token": "<|eot_id|>", + "extra_special_tokens": {}, + "model_input_names": [ + "input_ids", + "attention_mask" + ], + "model_max_length": 131072, + "tokenizer_class": "PreTrainedTokenizerFast" +} diff --git a/train_results.json b/train_results.json new file mode 100644 index 0000000..96f55f2 --- /dev/null +++ b/train_results.json @@ -0,0 +1,8 @@ +{ + "total_flos": 2.554786723790219e+19, + "train_loss": 0.3743892161239599, + "train_runtime": 22902.1621, + "train_samples": 86158, + "train_samples_per_second": 11.286, + "train_steps_per_second": 0.705 +} \ No newline at end of file diff --git a/trainer_state.json b/trainer_state.json new file mode 100644 index 0000000..bb690d6 --- /dev/null +++ b/trainer_state.json @@ -0,0 +1,145438 @@ +{ + "best_global_step": null, + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 3.0, + "eval_steps": 500, + "global_step": 16155, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.00018570102135561745, + "grad_norm": 12.444783210754395, + "learning_rate": 0.0, + "loss": 0.9336, + "mean_token_accuracy": 0.7542619109153748, + "num_tokens": 38904.0, + "step": 1 + }, + { + "epoch": 0.0003714020427112349, + "grad_norm": 15.545819282531738, + "learning_rate": 6.188118811881188e-10, + "loss": 0.9273, + "mean_token_accuracy": 0.7645187973976135, + "num_tokens": 69247.0, + "step": 2 + }, + { + "epoch": 0.0005571030640668524, + "grad_norm": 12.54469108581543, + "learning_rate": 1.2376237623762375e-09, + "loss": 0.9838, + "mean_token_accuracy": 0.7508255243301392, + "num_tokens": 107855.0, + "step": 3 + }, + { + "epoch": 0.0007428040854224698, + "grad_norm": 12.590400695800781, + "learning_rate": 1.8564356435643563e-09, + "loss": 1.0, + "mean_token_accuracy": 0.7442327737808228, + "num_tokens": 145563.0, + "step": 4 + }, + { + "epoch": 0.0009285051067780873, + "grad_norm": 16.1022891998291, + "learning_rate": 2.475247524752475e-09, + "loss": 0.9632, + "mean_token_accuracy": 0.7599184513092041, + "num_tokens": 175242.0, + "step": 5 + }, + { + "epoch": 0.0011142061281337048, + "grad_norm": 14.653234481811523, + "learning_rate": 3.0940594059405942e-09, + "loss": 0.9762, + "mean_token_accuracy": 0.7539225220680237, + "num_tokens": 207039.0, + "step": 6 + }, + { + "epoch": 0.0012999071494893223, + "grad_norm": 15.347660064697266, + "learning_rate": 3.7128712871287126e-09, + "loss": 0.9506, + "mean_token_accuracy": 0.7631068229675293, + "num_tokens": 238040.0, + "step": 7 + }, + { + "epoch": 0.0014856081708449396, + "grad_norm": 14.277806282043457, + "learning_rate": 4.331683168316832e-09, + "loss": 0.9528, + "mean_token_accuracy": 0.7606446743011475, + "num_tokens": 271554.0, + "step": 8 + }, + { + "epoch": 0.001671309192200557, + "grad_norm": 15.960329055786133, + "learning_rate": 4.95049504950495e-09, + "loss": 0.9441, + "mean_token_accuracy": 0.761651337146759, + "num_tokens": 301167.0, + "step": 9 + }, + { + "epoch": 0.0018570102135561746, + "grad_norm": 13.996705055236816, + "learning_rate": 5.569306930693069e-09, + "loss": 0.9125, + "mean_token_accuracy": 0.7706599235534668, + "num_tokens": 334553.0, + "step": 10 + }, + { + "epoch": 0.002042711234911792, + "grad_norm": 12.155610084533691, + "learning_rate": 6.1881188118811884e-09, + "loss": 0.9043, + "mean_token_accuracy": 0.7642819881439209, + "num_tokens": 373943.0, + "step": 11 + }, + { + "epoch": 0.0022284122562674096, + "grad_norm": 16.051088333129883, + "learning_rate": 6.806930693069306e-09, + "loss": 0.9777, + "mean_token_accuracy": 0.7579808831214905, + "num_tokens": 402798.0, + "step": 12 + }, + { + "epoch": 0.002414113277623027, + "grad_norm": 14.130428314208984, + "learning_rate": 7.425742574257425e-09, + "loss": 0.9225, + "mean_token_accuracy": 0.7617248296737671, + "num_tokens": 436304.0, + "step": 13 + }, + { + "epoch": 0.0025998142989786446, + "grad_norm": 13.452470779418945, + "learning_rate": 8.044554455445545e-09, + "loss": 0.9802, + "mean_token_accuracy": 0.7503772377967834, + "num_tokens": 471893.0, + "step": 14 + }, + { + "epoch": 0.002785515320334262, + "grad_norm": 14.324198722839355, + "learning_rate": 8.663366336633664e-09, + "loss": 0.9575, + "mean_token_accuracy": 0.7591619491577148, + "num_tokens": 504603.0, + "step": 15 + }, + { + "epoch": 0.002971216341689879, + "grad_norm": 13.692660331726074, + "learning_rate": 9.28217821782178e-09, + "loss": 0.9418, + "mean_token_accuracy": 0.7592573761940002, + "num_tokens": 539678.0, + "step": 16 + }, + { + "epoch": 0.003156917363045497, + "grad_norm": 15.104798316955566, + "learning_rate": 9.9009900990099e-09, + "loss": 0.9959, + "mean_token_accuracy": 0.7467846870422363, + "num_tokens": 570829.0, + "step": 17 + }, + { + "epoch": 0.003342618384401114, + "grad_norm": 12.035959243774414, + "learning_rate": 1.051980198019802e-08, + "loss": 0.9276, + "mean_token_accuracy": 0.7576239109039307, + "num_tokens": 610552.0, + "step": 18 + }, + { + "epoch": 0.003528319405756732, + "grad_norm": 15.51626205444336, + "learning_rate": 1.1138613861386138e-08, + "loss": 0.9964, + "mean_token_accuracy": 0.7444043159484863, + "num_tokens": 641154.0, + "step": 19 + }, + { + "epoch": 0.003714020427112349, + "grad_norm": 13.732726097106934, + "learning_rate": 1.1757425742574257e-08, + "loss": 0.9915, + "mean_token_accuracy": 0.7499734163284302, + "num_tokens": 675484.0, + "step": 20 + }, + { + "epoch": 0.0038997214484679664, + "grad_norm": 13.306537628173828, + "learning_rate": 1.2376237623762377e-08, + "loss": 0.9887, + "mean_token_accuracy": 0.7505103349685669, + "num_tokens": 711277.0, + "step": 21 + }, + { + "epoch": 0.004085422469823584, + "grad_norm": 11.255154609680176, + "learning_rate": 1.2995049504950493e-08, + "loss": 0.8696, + "mean_token_accuracy": 0.7725908756256104, + "num_tokens": 754298.0, + "step": 22 + }, + { + "epoch": 0.004271123491179201, + "grad_norm": 15.804388046264648, + "learning_rate": 1.3613861386138613e-08, + "loss": 1.0177, + "mean_token_accuracy": 0.7449629306793213, + "num_tokens": 784327.0, + "step": 23 + }, + { + "epoch": 0.004456824512534819, + "grad_norm": 12.218341827392578, + "learning_rate": 1.4232673267326732e-08, + "loss": 0.8952, + "mean_token_accuracy": 0.7690342664718628, + "num_tokens": 822562.0, + "step": 24 + }, + { + "epoch": 0.004642525533890436, + "grad_norm": 13.303646087646484, + "learning_rate": 1.485148514851485e-08, + "loss": 0.9999, + "mean_token_accuracy": 0.7471567392349243, + "num_tokens": 858213.0, + "step": 25 + }, + { + "epoch": 0.004828226555246054, + "grad_norm": 14.642655372619629, + "learning_rate": 1.5470297029702968e-08, + "loss": 1.0626, + "mean_token_accuracy": 0.7297772765159607, + "num_tokens": 890663.0, + "step": 26 + }, + { + "epoch": 0.005013927576601671, + "grad_norm": 13.094475746154785, + "learning_rate": 1.608910891089109e-08, + "loss": 0.9284, + "mean_token_accuracy": 0.7624815702438354, + "num_tokens": 927016.0, + "step": 27 + }, + { + "epoch": 0.005199628597957289, + "grad_norm": 14.410024642944336, + "learning_rate": 1.6707920792079207e-08, + "loss": 0.9572, + "mean_token_accuracy": 0.7575182914733887, + "num_tokens": 960219.0, + "step": 28 + }, + { + "epoch": 0.005385329619312906, + "grad_norm": 14.693161964416504, + "learning_rate": 1.732673267326733e-08, + "loss": 0.9588, + "mean_token_accuracy": 0.7581408023834229, + "num_tokens": 992194.0, + "step": 29 + }, + { + "epoch": 0.005571030640668524, + "grad_norm": 13.176468849182129, + "learning_rate": 1.7945544554455443e-08, + "loss": 0.9873, + "mean_token_accuracy": 0.7443438172340393, + "num_tokens": 1027482.0, + "step": 30 + }, + { + "epoch": 0.005756731662024141, + "grad_norm": 12.691295623779297, + "learning_rate": 1.856435643564356e-08, + "loss": 0.942, + "mean_token_accuracy": 0.761593222618103, + "num_tokens": 1064674.0, + "step": 31 + }, + { + "epoch": 0.005942432683379758, + "grad_norm": 16.066373825073242, + "learning_rate": 1.9183168316831682e-08, + "loss": 1.0014, + "mean_token_accuracy": 0.7457324862480164, + "num_tokens": 1093817.0, + "step": 32 + }, + { + "epoch": 0.006128133704735376, + "grad_norm": 11.703349113464355, + "learning_rate": 1.98019801980198e-08, + "loss": 0.9554, + "mean_token_accuracy": 0.757567286491394, + "num_tokens": 1134665.0, + "step": 33 + }, + { + "epoch": 0.006313834726090994, + "grad_norm": 12.973648071289062, + "learning_rate": 2.042079207920792e-08, + "loss": 0.9665, + "mean_token_accuracy": 0.7581090331077576, + "num_tokens": 1171162.0, + "step": 34 + }, + { + "epoch": 0.0064995357474466105, + "grad_norm": 12.914530754089355, + "learning_rate": 2.103960396039604e-08, + "loss": 0.997, + "mean_token_accuracy": 0.7444870471954346, + "num_tokens": 1206694.0, + "step": 35 + }, + { + "epoch": 0.006685236768802228, + "grad_norm": 12.297046661376953, + "learning_rate": 2.1658415841584157e-08, + "loss": 0.9351, + "mean_token_accuracy": 0.7628176212310791, + "num_tokens": 1244716.0, + "step": 36 + }, + { + "epoch": 0.006870937790157846, + "grad_norm": 14.532120704650879, + "learning_rate": 2.2277227722772275e-08, + "loss": 0.9464, + "mean_token_accuracy": 0.7637965679168701, + "num_tokens": 1276510.0, + "step": 37 + }, + { + "epoch": 0.007056638811513464, + "grad_norm": 13.555238723754883, + "learning_rate": 2.2896039603960393e-08, + "loss": 0.9571, + "mean_token_accuracy": 0.7556427717208862, + "num_tokens": 1311107.0, + "step": 38 + }, + { + "epoch": 0.0072423398328690805, + "grad_norm": 14.37629222869873, + "learning_rate": 2.3514851485148515e-08, + "loss": 1.016, + "mean_token_accuracy": 0.7459768056869507, + "num_tokens": 1343297.0, + "step": 39 + }, + { + "epoch": 0.007428040854224698, + "grad_norm": 13.859017372131348, + "learning_rate": 2.4133663366336632e-08, + "loss": 0.9728, + "mean_token_accuracy": 0.7555070519447327, + "num_tokens": 1376042.0, + "step": 40 + }, + { + "epoch": 0.007613741875580316, + "grad_norm": 15.43154239654541, + "learning_rate": 2.4752475247524754e-08, + "loss": 0.9656, + "mean_token_accuracy": 0.7566369771957397, + "num_tokens": 1406116.0, + "step": 41 + }, + { + "epoch": 0.007799442896935933, + "grad_norm": 14.5967378616333, + "learning_rate": 2.537128712871287e-08, + "loss": 0.9634, + "mean_token_accuracy": 0.7538175582885742, + "num_tokens": 1436680.0, + "step": 42 + }, + { + "epoch": 0.00798514391829155, + "grad_norm": 11.992701530456543, + "learning_rate": 2.5990099009900986e-08, + "loss": 0.9505, + "mean_token_accuracy": 0.7560255527496338, + "num_tokens": 1474213.0, + "step": 43 + }, + { + "epoch": 0.008170844939647167, + "grad_norm": 12.351685523986816, + "learning_rate": 2.6608910891089107e-08, + "loss": 0.9097, + "mean_token_accuracy": 0.768346905708313, + "num_tokens": 1510411.0, + "step": 44 + }, + { + "epoch": 0.008356545961002786, + "grad_norm": 11.671961784362793, + "learning_rate": 2.7227722772277225e-08, + "loss": 0.954, + "mean_token_accuracy": 0.7522638440132141, + "num_tokens": 1549277.0, + "step": 45 + }, + { + "epoch": 0.008542246982358403, + "grad_norm": 15.65117073059082, + "learning_rate": 2.7846534653465347e-08, + "loss": 1.0269, + "mean_token_accuracy": 0.7446528673171997, + "num_tokens": 1577060.0, + "step": 46 + }, + { + "epoch": 0.00872794800371402, + "grad_norm": 11.561203956604004, + "learning_rate": 2.8465346534653465e-08, + "loss": 0.9749, + "mean_token_accuracy": 0.7489598393440247, + "num_tokens": 1615629.0, + "step": 47 + }, + { + "epoch": 0.008913649025069638, + "grad_norm": 11.532244682312012, + "learning_rate": 2.9084158415841586e-08, + "loss": 0.9272, + "mean_token_accuracy": 0.7617202997207642, + "num_tokens": 1655077.0, + "step": 48 + }, + { + "epoch": 0.009099350046425255, + "grad_norm": 13.962568283081055, + "learning_rate": 2.97029702970297e-08, + "loss": 0.9381, + "mean_token_accuracy": 0.7614215612411499, + "num_tokens": 1687012.0, + "step": 49 + }, + { + "epoch": 0.009285051067780872, + "grad_norm": 10.939207077026367, + "learning_rate": 3.032178217821782e-08, + "loss": 0.8774, + "mean_token_accuracy": 0.775253415107727, + "num_tokens": 1727118.0, + "step": 50 + }, + { + "epoch": 0.00947075208913649, + "grad_norm": 12.20563793182373, + "learning_rate": 3.0940594059405936e-08, + "loss": 0.9302, + "mean_token_accuracy": 0.7608338594436646, + "num_tokens": 1762348.0, + "step": 51 + }, + { + "epoch": 0.009656453110492107, + "grad_norm": 11.045856475830078, + "learning_rate": 3.155940594059406e-08, + "loss": 0.8915, + "mean_token_accuracy": 0.768454909324646, + "num_tokens": 1802582.0, + "step": 52 + }, + { + "epoch": 0.009842154131847726, + "grad_norm": 11.3486967086792, + "learning_rate": 3.217821782178218e-08, + "loss": 0.98, + "mean_token_accuracy": 0.7467827200889587, + "num_tokens": 1842540.0, + "step": 53 + }, + { + "epoch": 0.010027855153203343, + "grad_norm": 12.458223342895508, + "learning_rate": 3.27970297029703e-08, + "loss": 0.9881, + "mean_token_accuracy": 0.7479250431060791, + "num_tokens": 1876902.0, + "step": 54 + }, + { + "epoch": 0.01021355617455896, + "grad_norm": 12.135812759399414, + "learning_rate": 3.3415841584158415e-08, + "loss": 0.9979, + "mean_token_accuracy": 0.7435544729232788, + "num_tokens": 1913471.0, + "step": 55 + }, + { + "epoch": 0.010399257195914578, + "grad_norm": 14.116676330566406, + "learning_rate": 3.403465346534653e-08, + "loss": 0.9866, + "mean_token_accuracy": 0.7488199472427368, + "num_tokens": 1944378.0, + "step": 56 + }, + { + "epoch": 0.010584958217270195, + "grad_norm": 13.079370498657227, + "learning_rate": 3.465346534653466e-08, + "loss": 1.0424, + "mean_token_accuracy": 0.7353452444076538, + "num_tokens": 1978441.0, + "step": 57 + }, + { + "epoch": 0.010770659238625812, + "grad_norm": 9.836256980895996, + "learning_rate": 3.527227722772277e-08, + "loss": 0.9687, + "mean_token_accuracy": 0.7476838827133179, + "num_tokens": 2017300.0, + "step": 58 + }, + { + "epoch": 0.01095636025998143, + "grad_norm": 9.491880416870117, + "learning_rate": 3.5891089108910886e-08, + "loss": 0.9626, + "mean_token_accuracy": 0.7477909326553345, + "num_tokens": 2057370.0, + "step": 59 + }, + { + "epoch": 0.011142061281337047, + "grad_norm": 9.533841133117676, + "learning_rate": 3.6509900990099004e-08, + "loss": 0.9368, + "mean_token_accuracy": 0.7545782327651978, + "num_tokens": 2096641.0, + "step": 60 + }, + { + "epoch": 0.011327762302692664, + "grad_norm": 10.448368072509766, + "learning_rate": 3.712871287128712e-08, + "loss": 0.9159, + "mean_token_accuracy": 0.7638678550720215, + "num_tokens": 2131466.0, + "step": 61 + }, + { + "epoch": 0.011513463324048283, + "grad_norm": 10.127710342407227, + "learning_rate": 3.774752475247525e-08, + "loss": 0.9266, + "mean_token_accuracy": 0.7613495588302612, + "num_tokens": 2167261.0, + "step": 62 + }, + { + "epoch": 0.0116991643454039, + "grad_norm": 10.141923904418945, + "learning_rate": 3.8366336633663365e-08, + "loss": 0.9351, + "mean_token_accuracy": 0.756965160369873, + "num_tokens": 2202636.0, + "step": 63 + }, + { + "epoch": 0.011884865366759517, + "grad_norm": 9.234869003295898, + "learning_rate": 3.898514851485148e-08, + "loss": 0.939, + "mean_token_accuracy": 0.7569757699966431, + "num_tokens": 2241934.0, + "step": 64 + }, + { + "epoch": 0.012070566388115135, + "grad_norm": 9.916848182678223, + "learning_rate": 3.96039603960396e-08, + "loss": 0.8626, + "mean_token_accuracy": 0.7784422636032104, + "num_tokens": 2277607.0, + "step": 65 + }, + { + "epoch": 0.012256267409470752, + "grad_norm": 8.825972557067871, + "learning_rate": 4.0222772277227725e-08, + "loss": 0.9641, + "mean_token_accuracy": 0.7508944272994995, + "num_tokens": 2319835.0, + "step": 66 + }, + { + "epoch": 0.012441968430826369, + "grad_norm": 11.527876853942871, + "learning_rate": 4.084158415841584e-08, + "loss": 1.0167, + "mean_token_accuracy": 0.7442053556442261, + "num_tokens": 2350998.0, + "step": 67 + }, + { + "epoch": 0.012627669452181987, + "grad_norm": 9.27322006225586, + "learning_rate": 4.146039603960396e-08, + "loss": 0.8639, + "mean_token_accuracy": 0.7738605737686157, + "num_tokens": 2389681.0, + "step": 68 + }, + { + "epoch": 0.012813370473537604, + "grad_norm": 9.779975891113281, + "learning_rate": 4.207920792079208e-08, + "loss": 0.9253, + "mean_token_accuracy": 0.7597540020942688, + "num_tokens": 2425749.0, + "step": 69 + }, + { + "epoch": 0.012999071494893221, + "grad_norm": 11.077693939208984, + "learning_rate": 4.26980198019802e-08, + "loss": 0.9438, + "mean_token_accuracy": 0.7574030756950378, + "num_tokens": 2457204.0, + "step": 70 + }, + { + "epoch": 0.01318477251624884, + "grad_norm": 12.247406959533691, + "learning_rate": 4.3316831683168315e-08, + "loss": 1.0538, + "mean_token_accuracy": 0.7312577962875366, + "num_tokens": 2484966.0, + "step": 71 + }, + { + "epoch": 0.013370473537604457, + "grad_norm": 8.728777885437012, + "learning_rate": 4.393564356435643e-08, + "loss": 0.9142, + "mean_token_accuracy": 0.7610812783241272, + "num_tokens": 2525286.0, + "step": 72 + }, + { + "epoch": 0.013556174558960075, + "grad_norm": 9.51921272277832, + "learning_rate": 4.455445544554455e-08, + "loss": 0.9028, + "mean_token_accuracy": 0.7652024030685425, + "num_tokens": 2562004.0, + "step": 73 + }, + { + "epoch": 0.013741875580315692, + "grad_norm": 11.14924430847168, + "learning_rate": 4.517326732673267e-08, + "loss": 1.007, + "mean_token_accuracy": 0.7411839962005615, + "num_tokens": 2593189.0, + "step": 74 + }, + { + "epoch": 0.013927576601671309, + "grad_norm": 10.184785842895508, + "learning_rate": 4.5792079207920787e-08, + "loss": 0.8987, + "mean_token_accuracy": 0.7662034034729004, + "num_tokens": 2628286.0, + "step": 75 + }, + { + "epoch": 0.014113277623026927, + "grad_norm": 9.912063598632812, + "learning_rate": 4.641089108910891e-08, + "loss": 0.8961, + "mean_token_accuracy": 0.7668171525001526, + "num_tokens": 2663408.0, + "step": 76 + }, + { + "epoch": 0.014298978644382544, + "grad_norm": 10.358297348022461, + "learning_rate": 4.702970297029703e-08, + "loss": 0.873, + "mean_token_accuracy": 0.774014413356781, + "num_tokens": 2700797.0, + "step": 77 + }, + { + "epoch": 0.014484679665738161, + "grad_norm": 10.596423149108887, + "learning_rate": 4.764851485148515e-08, + "loss": 0.9276, + "mean_token_accuracy": 0.7577248811721802, + "num_tokens": 2736996.0, + "step": 78 + }, + { + "epoch": 0.01467038068709378, + "grad_norm": 9.830811500549316, + "learning_rate": 4.8267326732673265e-08, + "loss": 0.8723, + "mean_token_accuracy": 0.7707773447036743, + "num_tokens": 2775850.0, + "step": 79 + }, + { + "epoch": 0.014856081708449397, + "grad_norm": 8.460518836975098, + "learning_rate": 4.888613861386138e-08, + "loss": 0.9351, + "mean_token_accuracy": 0.7525509595870972, + "num_tokens": 2814388.0, + "step": 80 + }, + { + "epoch": 0.015041782729805013, + "grad_norm": 7.915048599243164, + "learning_rate": 4.950495049504951e-08, + "loss": 0.8517, + "mean_token_accuracy": 0.7764618396759033, + "num_tokens": 2853190.0, + "step": 81 + }, + { + "epoch": 0.015227483751160632, + "grad_norm": 9.017470359802246, + "learning_rate": 5.0123762376237625e-08, + "loss": 0.9698, + "mean_token_accuracy": 0.7491247653961182, + "num_tokens": 2885520.0, + "step": 82 + }, + { + "epoch": 0.015413184772516249, + "grad_norm": 7.705904483795166, + "learning_rate": 5.074257425742574e-08, + "loss": 0.8844, + "mean_token_accuracy": 0.7630307078361511, + "num_tokens": 2920392.0, + "step": 83 + }, + { + "epoch": 0.015598885793871866, + "grad_norm": 8.2254056930542, + "learning_rate": 5.1361386138613855e-08, + "loss": 0.8888, + "mean_token_accuracy": 0.7679710388183594, + "num_tokens": 2951312.0, + "step": 84 + }, + { + "epoch": 0.015784586815227482, + "grad_norm": 7.3325066566467285, + "learning_rate": 5.198019801980197e-08, + "loss": 0.9242, + "mean_token_accuracy": 0.7554713487625122, + "num_tokens": 2987407.0, + "step": 85 + }, + { + "epoch": 0.0159702878365831, + "grad_norm": 6.548575401306152, + "learning_rate": 5.25990099009901e-08, + "loss": 0.9762, + "mean_token_accuracy": 0.7461804151535034, + "num_tokens": 3029284.0, + "step": 86 + }, + { + "epoch": 0.01615598885793872, + "grad_norm": 7.847725868225098, + "learning_rate": 5.3217821782178215e-08, + "loss": 0.9298, + "mean_token_accuracy": 0.7499889731407166, + "num_tokens": 3060173.0, + "step": 87 + }, + { + "epoch": 0.016341689879294335, + "grad_norm": 6.570877552032471, + "learning_rate": 5.383663366336633e-08, + "loss": 0.9117, + "mean_token_accuracy": 0.7606887817382812, + "num_tokens": 3098970.0, + "step": 88 + }, + { + "epoch": 0.016527390900649953, + "grad_norm": 6.760998725891113, + "learning_rate": 5.445544554455445e-08, + "loss": 0.913, + "mean_token_accuracy": 0.7557884454727173, + "num_tokens": 3136161.0, + "step": 89 + }, + { + "epoch": 0.016713091922005572, + "grad_norm": 6.5115556716918945, + "learning_rate": 5.5074257425742575e-08, + "loss": 0.8862, + "mean_token_accuracy": 0.7663400769233704, + "num_tokens": 3174137.0, + "step": 90 + }, + { + "epoch": 0.016898792943361187, + "grad_norm": 7.183831214904785, + "learning_rate": 5.569306930693069e-08, + "loss": 0.9673, + "mean_token_accuracy": 0.7451404333114624, + "num_tokens": 3207157.0, + "step": 91 + }, + { + "epoch": 0.017084493964716806, + "grad_norm": 6.172833442687988, + "learning_rate": 5.631188118811881e-08, + "loss": 0.8979, + "mean_token_accuracy": 0.7621138691902161, + "num_tokens": 3248242.0, + "step": 92 + }, + { + "epoch": 0.017270194986072424, + "grad_norm": 7.878584861755371, + "learning_rate": 5.693069306930693e-08, + "loss": 0.9371, + "mean_token_accuracy": 0.7524535059928894, + "num_tokens": 3277521.0, + "step": 93 + }, + { + "epoch": 0.01745589600742804, + "grad_norm": 6.92396879196167, + "learning_rate": 5.754950495049505e-08, + "loss": 0.9358, + "mean_token_accuracy": 0.7538275122642517, + "num_tokens": 3311908.0, + "step": 94 + }, + { + "epoch": 0.017641597028783658, + "grad_norm": 5.775569438934326, + "learning_rate": 5.816831683168317e-08, + "loss": 0.9166, + "mean_token_accuracy": 0.7556889653205872, + "num_tokens": 3352535.0, + "step": 95 + }, + { + "epoch": 0.017827298050139277, + "grad_norm": 7.018883228302002, + "learning_rate": 5.878712871287128e-08, + "loss": 0.9671, + "mean_token_accuracy": 0.7463970184326172, + "num_tokens": 3384867.0, + "step": 96 + }, + { + "epoch": 0.01801299907149489, + "grad_norm": 7.085948467254639, + "learning_rate": 5.94059405940594e-08, + "loss": 0.9164, + "mean_token_accuracy": 0.7624796628952026, + "num_tokens": 3414920.0, + "step": 97 + }, + { + "epoch": 0.01819870009285051, + "grad_norm": 6.6275715827941895, + "learning_rate": 6.002475247524753e-08, + "loss": 0.9541, + "mean_token_accuracy": 0.7514641284942627, + "num_tokens": 3448324.0, + "step": 98 + }, + { + "epoch": 0.01838440111420613, + "grad_norm": 6.088974952697754, + "learning_rate": 6.064356435643564e-08, + "loss": 0.8505, + "mean_token_accuracy": 0.7758772373199463, + "num_tokens": 3484005.0, + "step": 99 + }, + { + "epoch": 0.018570102135561744, + "grad_norm": 5.290814399719238, + "learning_rate": 6.126237623762376e-08, + "loss": 0.9044, + "mean_token_accuracy": 0.7614033222198486, + "num_tokens": 3527590.0, + "step": 100 + }, + { + "epoch": 0.018755803156917362, + "grad_norm": 5.843197822570801, + "learning_rate": 6.188118811881187e-08, + "loss": 0.8308, + "mean_token_accuracy": 0.7786649465560913, + "num_tokens": 3564091.0, + "step": 101 + }, + { + "epoch": 0.01894150417827298, + "grad_norm": 6.451364994049072, + "learning_rate": 6.25e-08, + "loss": 0.8421, + "mean_token_accuracy": 0.775981605052948, + "num_tokens": 3596705.0, + "step": 102 + }, + { + "epoch": 0.0191272051996286, + "grad_norm": 6.134764194488525, + "learning_rate": 6.311881188118812e-08, + "loss": 0.8695, + "mean_token_accuracy": 0.7682172060012817, + "num_tokens": 3630281.0, + "step": 103 + }, + { + "epoch": 0.019312906220984215, + "grad_norm": 5.666975975036621, + "learning_rate": 6.373762376237623e-08, + "loss": 0.8754, + "mean_token_accuracy": 0.7678150534629822, + "num_tokens": 3667327.0, + "step": 104 + }, + { + "epoch": 0.019498607242339833, + "grad_norm": 5.909034729003906, + "learning_rate": 6.435643564356436e-08, + "loss": 0.9086, + "mean_token_accuracy": 0.7566395998001099, + "num_tokens": 3702390.0, + "step": 105 + }, + { + "epoch": 0.019684308263695452, + "grad_norm": 5.517190456390381, + "learning_rate": 6.497524752475247e-08, + "loss": 0.8293, + "mean_token_accuracy": 0.7777027487754822, + "num_tokens": 3739772.0, + "step": 106 + }, + { + "epoch": 0.019870009285051067, + "grad_norm": 5.748786449432373, + "learning_rate": 6.55940594059406e-08, + "loss": 0.8226, + "mean_token_accuracy": 0.7819430828094482, + "num_tokens": 3775045.0, + "step": 107 + }, + { + "epoch": 0.020055710306406686, + "grad_norm": 5.864236354827881, + "learning_rate": 6.621287128712872e-08, + "loss": 0.8563, + "mean_token_accuracy": 0.7710198163986206, + "num_tokens": 3808431.0, + "step": 108 + }, + { + "epoch": 0.020241411327762304, + "grad_norm": 5.854403972625732, + "learning_rate": 6.683168316831683e-08, + "loss": 0.9446, + "mean_token_accuracy": 0.7458359599113464, + "num_tokens": 3843248.0, + "step": 109 + }, + { + "epoch": 0.02042711234911792, + "grad_norm": 5.133481502532959, + "learning_rate": 6.745049504950495e-08, + "loss": 0.8229, + "mean_token_accuracy": 0.7769081592559814, + "num_tokens": 3882584.0, + "step": 110 + }, + { + "epoch": 0.020612813370473538, + "grad_norm": 5.242915153503418, + "learning_rate": 6.806930693069307e-08, + "loss": 0.786, + "mean_token_accuracy": 0.7878711819648743, + "num_tokens": 3919680.0, + "step": 111 + }, + { + "epoch": 0.020798514391829157, + "grad_norm": 5.297712802886963, + "learning_rate": 6.868811881188119e-08, + "loss": 0.8781, + "mean_token_accuracy": 0.7646140456199646, + "num_tokens": 3957622.0, + "step": 112 + }, + { + "epoch": 0.02098421541318477, + "grad_norm": 5.9056010246276855, + "learning_rate": 6.930693069306931e-08, + "loss": 0.868, + "mean_token_accuracy": 0.7699576616287231, + "num_tokens": 3989434.0, + "step": 113 + }, + { + "epoch": 0.02116991643454039, + "grad_norm": 6.541084289550781, + "learning_rate": 6.992574257425743e-08, + "loss": 0.885, + "mean_token_accuracy": 0.756125271320343, + "num_tokens": 4016642.0, + "step": 114 + }, + { + "epoch": 0.02135561745589601, + "grad_norm": 6.142141342163086, + "learning_rate": 7.054455445544554e-08, + "loss": 0.9138, + "mean_token_accuracy": 0.7584670186042786, + "num_tokens": 4047030.0, + "step": 115 + }, + { + "epoch": 0.021541318477251624, + "grad_norm": 5.002140522003174, + "learning_rate": 7.116336633663365e-08, + "loss": 0.7976, + "mean_token_accuracy": 0.7819240093231201, + "num_tokens": 4085468.0, + "step": 116 + }, + { + "epoch": 0.021727019498607242, + "grad_norm": 6.434032917022705, + "learning_rate": 7.178217821782177e-08, + "loss": 0.8596, + "mean_token_accuracy": 0.7681440114974976, + "num_tokens": 4114195.0, + "step": 117 + }, + { + "epoch": 0.02191272051996286, + "grad_norm": 5.002919673919678, + "learning_rate": 7.24009900990099e-08, + "loss": 0.8731, + "mean_token_accuracy": 0.758535623550415, + "num_tokens": 4154377.0, + "step": 118 + }, + { + "epoch": 0.022098421541318476, + "grad_norm": 5.561726093292236, + "learning_rate": 7.301980198019801e-08, + "loss": 0.8332, + "mean_token_accuracy": 0.7730840444564819, + "num_tokens": 4188769.0, + "step": 119 + }, + { + "epoch": 0.022284122562674095, + "grad_norm": 5.663234233856201, + "learning_rate": 7.363861386138613e-08, + "loss": 0.8469, + "mean_token_accuracy": 0.7711501121520996, + "num_tokens": 4223131.0, + "step": 120 + }, + { + "epoch": 0.022469823584029713, + "grad_norm": 4.993075847625732, + "learning_rate": 7.425742574257424e-08, + "loss": 0.7999, + "mean_token_accuracy": 0.7795957326889038, + "num_tokens": 4263223.0, + "step": 121 + }, + { + "epoch": 0.02265552460538533, + "grad_norm": 5.545888423919678, + "learning_rate": 7.487623762376237e-08, + "loss": 0.8055, + "mean_token_accuracy": 0.7819740176200867, + "num_tokens": 4296647.0, + "step": 122 + }, + { + "epoch": 0.022841225626740947, + "grad_norm": 6.092235565185547, + "learning_rate": 7.54950495049505e-08, + "loss": 0.8428, + "mean_token_accuracy": 0.7672430276870728, + "num_tokens": 4327568.0, + "step": 123 + }, + { + "epoch": 0.023026926648096566, + "grad_norm": 6.7145209312438965, + "learning_rate": 7.61138613861386e-08, + "loss": 0.8448, + "mean_token_accuracy": 0.7722079753875732, + "num_tokens": 4354655.0, + "step": 124 + }, + { + "epoch": 0.02321262766945218, + "grad_norm": 5.369784832000732, + "learning_rate": 7.673267326732673e-08, + "loss": 0.7761, + "mean_token_accuracy": 0.7878938317298889, + "num_tokens": 4388362.0, + "step": 125 + }, + { + "epoch": 0.0233983286908078, + "grad_norm": 5.887394428253174, + "learning_rate": 7.735148514851484e-08, + "loss": 0.8524, + "mean_token_accuracy": 0.7648885250091553, + "num_tokens": 4420685.0, + "step": 126 + }, + { + "epoch": 0.023584029712163418, + "grad_norm": 4.94689416885376, + "learning_rate": 7.797029702970297e-08, + "loss": 0.7948, + "mean_token_accuracy": 0.7827721834182739, + "num_tokens": 4460100.0, + "step": 127 + }, + { + "epoch": 0.023769730733519033, + "grad_norm": 5.492007732391357, + "learning_rate": 7.858910891089109e-08, + "loss": 0.7711, + "mean_token_accuracy": 0.7832477688789368, + "num_tokens": 4493957.0, + "step": 128 + }, + { + "epoch": 0.02395543175487465, + "grad_norm": 5.483529090881348, + "learning_rate": 7.92079207920792e-08, + "loss": 0.8094, + "mean_token_accuracy": 0.7760226726531982, + "num_tokens": 4527595.0, + "step": 129 + }, + { + "epoch": 0.02414113277623027, + "grad_norm": 5.508817672729492, + "learning_rate": 7.982673267326733e-08, + "loss": 0.8138, + "mean_token_accuracy": 0.7728414535522461, + "num_tokens": 4562929.0, + "step": 130 + }, + { + "epoch": 0.024326833797585885, + "grad_norm": 5.609079837799072, + "learning_rate": 8.044554455445545e-08, + "loss": 0.8326, + "mean_token_accuracy": 0.7701128721237183, + "num_tokens": 4600869.0, + "step": 131 + }, + { + "epoch": 0.024512534818941504, + "grad_norm": 7.140466213226318, + "learning_rate": 8.106435643564356e-08, + "loss": 0.8399, + "mean_token_accuracy": 0.7689406275749207, + "num_tokens": 4628835.0, + "step": 132 + }, + { + "epoch": 0.024698235840297122, + "grad_norm": 6.286675930023193, + "learning_rate": 8.168316831683169e-08, + "loss": 0.778, + "mean_token_accuracy": 0.7834744453430176, + "num_tokens": 4662671.0, + "step": 133 + }, + { + "epoch": 0.024883936861652738, + "grad_norm": 6.818629264831543, + "learning_rate": 8.23019801980198e-08, + "loss": 0.7296, + "mean_token_accuracy": 0.7950568795204163, + "num_tokens": 4693980.0, + "step": 134 + }, + { + "epoch": 0.025069637883008356, + "grad_norm": 5.698207378387451, + "learning_rate": 8.292079207920792e-08, + "loss": 0.7856, + "mean_token_accuracy": 0.7788265943527222, + "num_tokens": 4732638.0, + "step": 135 + }, + { + "epoch": 0.025255338904363975, + "grad_norm": 5.46790885925293, + "learning_rate": 8.353960396039605e-08, + "loss": 0.7602, + "mean_token_accuracy": 0.7875475883483887, + "num_tokens": 4768844.0, + "step": 136 + }, + { + "epoch": 0.02544103992571959, + "grad_norm": 5.130856990814209, + "learning_rate": 8.415841584158416e-08, + "loss": 0.7799, + "mean_token_accuracy": 0.7810646891593933, + "num_tokens": 4809412.0, + "step": 137 + }, + { + "epoch": 0.02562674094707521, + "grad_norm": 5.601790904998779, + "learning_rate": 8.477722772277228e-08, + "loss": 0.7728, + "mean_token_accuracy": 0.7854694128036499, + "num_tokens": 4844366.0, + "step": 138 + }, + { + "epoch": 0.025812441968430827, + "grad_norm": 4.919003963470459, + "learning_rate": 8.53960396039604e-08, + "loss": 0.7472, + "mean_token_accuracy": 0.7879911065101624, + "num_tokens": 4883883.0, + "step": 139 + }, + { + "epoch": 0.025998142989786442, + "grad_norm": 4.830056190490723, + "learning_rate": 8.60148514851485e-08, + "loss": 0.8507, + "mean_token_accuracy": 0.7635051012039185, + "num_tokens": 4923634.0, + "step": 140 + }, + { + "epoch": 0.02618384401114206, + "grad_norm": 5.5861639976501465, + "learning_rate": 8.663366336633663e-08, + "loss": 0.7991, + "mean_token_accuracy": 0.7783869504928589, + "num_tokens": 4956314.0, + "step": 141 + }, + { + "epoch": 0.02636954503249768, + "grad_norm": 5.551412582397461, + "learning_rate": 8.725247524752474e-08, + "loss": 0.8448, + "mean_token_accuracy": 0.761555552482605, + "num_tokens": 4988887.0, + "step": 142 + }, + { + "epoch": 0.026555246053853298, + "grad_norm": 4.347281455993652, + "learning_rate": 8.787128712871287e-08, + "loss": 0.7434, + "mean_token_accuracy": 0.7924160957336426, + "num_tokens": 5032025.0, + "step": 143 + }, + { + "epoch": 0.026740947075208913, + "grad_norm": 5.050454616546631, + "learning_rate": 8.849009900990098e-08, + "loss": 0.7529, + "mean_token_accuracy": 0.7877185344696045, + "num_tokens": 5065641.0, + "step": 144 + }, + { + "epoch": 0.02692664809656453, + "grad_norm": 4.279937744140625, + "learning_rate": 8.91089108910891e-08, + "loss": 0.813, + "mean_token_accuracy": 0.7665966153144836, + "num_tokens": 5108114.0, + "step": 145 + }, + { + "epoch": 0.02711234911792015, + "grad_norm": 4.392427444458008, + "learning_rate": 8.972772277227723e-08, + "loss": 0.7931, + "mean_token_accuracy": 0.7763082981109619, + "num_tokens": 5144158.0, + "step": 146 + }, + { + "epoch": 0.027298050139275765, + "grad_norm": 4.067929267883301, + "learning_rate": 9.034653465346534e-08, + "loss": 0.7191, + "mean_token_accuracy": 0.7952808141708374, + "num_tokens": 5180857.0, + "step": 147 + }, + { + "epoch": 0.027483751160631384, + "grad_norm": 4.773965835571289, + "learning_rate": 9.096534653465346e-08, + "loss": 0.8019, + "mean_token_accuracy": 0.776703953742981, + "num_tokens": 5211650.0, + "step": 148 + }, + { + "epoch": 0.027669452181987002, + "grad_norm": 4.046422958374023, + "learning_rate": 9.158415841584157e-08, + "loss": 0.789, + "mean_token_accuracy": 0.7747517228126526, + "num_tokens": 5246376.0, + "step": 149 + }, + { + "epoch": 0.027855153203342618, + "grad_norm": 3.8854620456695557, + "learning_rate": 9.22029702970297e-08, + "loss": 0.7545, + "mean_token_accuracy": 0.7834163308143616, + "num_tokens": 5281069.0, + "step": 150 + }, + { + "epoch": 0.028040854224698236, + "grad_norm": 4.256216049194336, + "learning_rate": 9.282178217821782e-08, + "loss": 0.7449, + "mean_token_accuracy": 0.7825895547866821, + "num_tokens": 5312476.0, + "step": 151 + }, + { + "epoch": 0.028226555246053855, + "grad_norm": 3.8150997161865234, + "learning_rate": 9.344059405940593e-08, + "loss": 0.778, + "mean_token_accuracy": 0.7823340892791748, + "num_tokens": 5346779.0, + "step": 152 + }, + { + "epoch": 0.02841225626740947, + "grad_norm": 3.4176077842712402, + "learning_rate": 9.405940594059406e-08, + "loss": 0.7085, + "mean_token_accuracy": 0.7983806133270264, + "num_tokens": 5384925.0, + "step": 153 + }, + { + "epoch": 0.02859795728876509, + "grad_norm": 3.8595023155212402, + "learning_rate": 9.467821782178217e-08, + "loss": 0.7525, + "mean_token_accuracy": 0.7846523523330688, + "num_tokens": 5415990.0, + "step": 154 + }, + { + "epoch": 0.028783658310120707, + "grad_norm": 3.467108726501465, + "learning_rate": 9.52970297029703e-08, + "loss": 0.7205, + "mean_token_accuracy": 0.7983349561691284, + "num_tokens": 5451330.0, + "step": 155 + }, + { + "epoch": 0.028969359331476322, + "grad_norm": 3.3020498752593994, + "learning_rate": 9.591584158415842e-08, + "loss": 0.6817, + "mean_token_accuracy": 0.804574728012085, + "num_tokens": 5490144.0, + "step": 156 + }, + { + "epoch": 0.02915506035283194, + "grad_norm": 3.2188785076141357, + "learning_rate": 9.653465346534653e-08, + "loss": 0.6172, + "mean_token_accuracy": 0.8243669271469116, + "num_tokens": 5527259.0, + "step": 157 + }, + { + "epoch": 0.02934076137418756, + "grad_norm": 3.6319761276245117, + "learning_rate": 9.715346534653465e-08, + "loss": 0.7455, + "mean_token_accuracy": 0.7875147461891174, + "num_tokens": 5560384.0, + "step": 158 + }, + { + "epoch": 0.029526462395543174, + "grad_norm": 3.2796199321746826, + "learning_rate": 9.777227722772277e-08, + "loss": 0.7496, + "mean_token_accuracy": 0.7843752503395081, + "num_tokens": 5595374.0, + "step": 159 + }, + { + "epoch": 0.029712163416898793, + "grad_norm": 3.5124425888061523, + "learning_rate": 9.839108910891089e-08, + "loss": 0.7656, + "mean_token_accuracy": 0.7797623872756958, + "num_tokens": 5634705.0, + "step": 160 + }, + { + "epoch": 0.02989786443825441, + "grad_norm": 3.1758534908294678, + "learning_rate": 9.900990099009901e-08, + "loss": 0.7373, + "mean_token_accuracy": 0.7864295244216919, + "num_tokens": 5671322.0, + "step": 161 + }, + { + "epoch": 0.030083565459610027, + "grad_norm": 3.1099605560302734, + "learning_rate": 9.962871287128713e-08, + "loss": 0.7197, + "mean_token_accuracy": 0.7901744842529297, + "num_tokens": 5712453.0, + "step": 162 + }, + { + "epoch": 0.030269266480965645, + "grad_norm": 3.6041100025177, + "learning_rate": 1.0024752475247525e-07, + "loss": 0.7543, + "mean_token_accuracy": 0.7893853783607483, + "num_tokens": 5743494.0, + "step": 163 + }, + { + "epoch": 0.030454967502321264, + "grad_norm": 2.998434066772461, + "learning_rate": 1.0086633663366336e-07, + "loss": 0.6631, + "mean_token_accuracy": 0.8103863000869751, + "num_tokens": 5776981.0, + "step": 164 + }, + { + "epoch": 0.03064066852367688, + "grad_norm": 3.3240389823913574, + "learning_rate": 1.0148514851485149e-07, + "loss": 0.7217, + "mean_token_accuracy": 0.793874979019165, + "num_tokens": 5806808.0, + "step": 165 + }, + { + "epoch": 0.030826369545032498, + "grad_norm": 3.447122812271118, + "learning_rate": 1.021039603960396e-07, + "loss": 0.7072, + "mean_token_accuracy": 0.798749566078186, + "num_tokens": 5836875.0, + "step": 166 + }, + { + "epoch": 0.031012070566388116, + "grad_norm": 3.303818941116333, + "learning_rate": 1.0272277227722771e-07, + "loss": 0.7352, + "mean_token_accuracy": 0.7837268114089966, + "num_tokens": 5869471.0, + "step": 167 + }, + { + "epoch": 0.03119777158774373, + "grad_norm": 3.0724971294403076, + "learning_rate": 1.0334158415841583e-07, + "loss": 0.7622, + "mean_token_accuracy": 0.7796516418457031, + "num_tokens": 5904746.0, + "step": 168 + }, + { + "epoch": 0.03138347260909935, + "grad_norm": 2.9425907135009766, + "learning_rate": 1.0396039603960394e-07, + "loss": 0.6509, + "mean_token_accuracy": 0.8092695474624634, + "num_tokens": 5941678.0, + "step": 169 + }, + { + "epoch": 0.031569173630454965, + "grad_norm": 3.5998075008392334, + "learning_rate": 1.0457920792079207e-07, + "loss": 0.7008, + "mean_token_accuracy": 0.7922723293304443, + "num_tokens": 5972000.0, + "step": 170 + }, + { + "epoch": 0.03175487465181059, + "grad_norm": 3.4581315517425537, + "learning_rate": 1.051980198019802e-07, + "loss": 0.6553, + "mean_token_accuracy": 0.8087100982666016, + "num_tokens": 6000415.0, + "step": 171 + }, + { + "epoch": 0.0319405756731662, + "grad_norm": 3.2598915100097656, + "learning_rate": 1.058168316831683e-07, + "loss": 0.6743, + "mean_token_accuracy": 0.8029397130012512, + "num_tokens": 6036036.0, + "step": 172 + }, + { + "epoch": 0.03212627669452182, + "grad_norm": 2.953474283218384, + "learning_rate": 1.0643564356435643e-07, + "loss": 0.739, + "mean_token_accuracy": 0.7895759344100952, + "num_tokens": 6069130.0, + "step": 173 + }, + { + "epoch": 0.03231197771587744, + "grad_norm": 2.6922459602355957, + "learning_rate": 1.0705445544554454e-07, + "loss": 0.685, + "mean_token_accuracy": 0.7994952201843262, + "num_tokens": 6109129.0, + "step": 174 + }, + { + "epoch": 0.032497678737233054, + "grad_norm": 2.866082191467285, + "learning_rate": 1.0767326732673267e-07, + "loss": 0.7347, + "mean_token_accuracy": 0.7843265533447266, + "num_tokens": 6143461.0, + "step": 175 + }, + { + "epoch": 0.03268337975858867, + "grad_norm": 3.242677927017212, + "learning_rate": 1.0829207920792079e-07, + "loss": 0.6559, + "mean_token_accuracy": 0.806765079498291, + "num_tokens": 6172881.0, + "step": 176 + }, + { + "epoch": 0.03286908077994429, + "grad_norm": 2.778109312057495, + "learning_rate": 1.089108910891089e-07, + "loss": 0.6843, + "mean_token_accuracy": 0.8071509599685669, + "num_tokens": 6207330.0, + "step": 177 + }, + { + "epoch": 0.03305478180129991, + "grad_norm": 2.838510513305664, + "learning_rate": 1.0952970297029703e-07, + "loss": 0.6729, + "mean_token_accuracy": 0.8009219169616699, + "num_tokens": 6243315.0, + "step": 178 + }, + { + "epoch": 0.03324048282265552, + "grad_norm": 2.6666665077209473, + "learning_rate": 1.1014851485148515e-07, + "loss": 0.6591, + "mean_token_accuracy": 0.8073533773422241, + "num_tokens": 6282167.0, + "step": 179 + }, + { + "epoch": 0.033426183844011144, + "grad_norm": 3.157344102859497, + "learning_rate": 1.1076732673267326e-07, + "loss": 0.6854, + "mean_token_accuracy": 0.8017611503601074, + "num_tokens": 6319443.0, + "step": 180 + }, + { + "epoch": 0.03361188486536676, + "grad_norm": 2.653151273727417, + "learning_rate": 1.1138613861386139e-07, + "loss": 0.6346, + "mean_token_accuracy": 0.8126824498176575, + "num_tokens": 6359681.0, + "step": 181 + }, + { + "epoch": 0.033797585886722374, + "grad_norm": 2.797137498855591, + "learning_rate": 1.120049504950495e-07, + "loss": 0.6381, + "mean_token_accuracy": 0.8073254823684692, + "num_tokens": 6395834.0, + "step": 182 + }, + { + "epoch": 0.033983286908077996, + "grad_norm": 3.9239258766174316, + "learning_rate": 1.1262376237623762e-07, + "loss": 0.7047, + "mean_token_accuracy": 0.7946104407310486, + "num_tokens": 6433371.0, + "step": 183 + }, + { + "epoch": 0.03416898792943361, + "grad_norm": 4.384718418121338, + "learning_rate": 1.1324257425742575e-07, + "loss": 0.705, + "mean_token_accuracy": 0.7932132482528687, + "num_tokens": 6469962.0, + "step": 184 + }, + { + "epoch": 0.034354688950789226, + "grad_norm": 3.209401845932007, + "learning_rate": 1.1386138613861386e-07, + "loss": 0.6632, + "mean_token_accuracy": 0.8032293319702148, + "num_tokens": 6501469.0, + "step": 185 + }, + { + "epoch": 0.03454038997214485, + "grad_norm": 3.4410104751586914, + "learning_rate": 1.1448019801980198e-07, + "loss": 0.7321, + "mean_token_accuracy": 0.7862368822097778, + "num_tokens": 6534728.0, + "step": 186 + }, + { + "epoch": 0.034726090993500464, + "grad_norm": 3.2085020542144775, + "learning_rate": 1.150990099009901e-07, + "loss": 0.6728, + "mean_token_accuracy": 0.8036528825759888, + "num_tokens": 6568798.0, + "step": 187 + }, + { + "epoch": 0.03491179201485608, + "grad_norm": 2.9142847061157227, + "learning_rate": 1.1571782178217822e-07, + "loss": 0.6444, + "mean_token_accuracy": 0.8109070658683777, + "num_tokens": 6604704.0, + "step": 188 + }, + { + "epoch": 0.0350974930362117, + "grad_norm": 3.3579938411712646, + "learning_rate": 1.1633663366336634e-07, + "loss": 0.7208, + "mean_token_accuracy": 0.785463809967041, + "num_tokens": 6637160.0, + "step": 189 + }, + { + "epoch": 0.035283194057567316, + "grad_norm": 3.0370864868164062, + "learning_rate": 1.1695544554455445e-07, + "loss": 0.7193, + "mean_token_accuracy": 0.7908015251159668, + "num_tokens": 6674338.0, + "step": 190 + }, + { + "epoch": 0.03546889507892293, + "grad_norm": 3.082900285720825, + "learning_rate": 1.1757425742574257e-07, + "loss": 0.6713, + "mean_token_accuracy": 0.8032531142234802, + "num_tokens": 6710062.0, + "step": 191 + }, + { + "epoch": 0.03565459610027855, + "grad_norm": 2.7522170543670654, + "learning_rate": 1.1819306930693068e-07, + "loss": 0.6749, + "mean_token_accuracy": 0.8025619983673096, + "num_tokens": 6745734.0, + "step": 192 + }, + { + "epoch": 0.03584029712163417, + "grad_norm": 3.5607104301452637, + "learning_rate": 1.188118811881188e-07, + "loss": 0.5912, + "mean_token_accuracy": 0.8204594850540161, + "num_tokens": 6775109.0, + "step": 193 + }, + { + "epoch": 0.03602599814298978, + "grad_norm": 3.1023049354553223, + "learning_rate": 1.194306930693069e-07, + "loss": 0.7131, + "mean_token_accuracy": 0.7931435704231262, + "num_tokens": 6807340.0, + "step": 194 + }, + { + "epoch": 0.036211699164345405, + "grad_norm": 3.287001848220825, + "learning_rate": 1.2004950495049505e-07, + "loss": 0.6635, + "mean_token_accuracy": 0.8042355179786682, + "num_tokens": 6837749.0, + "step": 195 + }, + { + "epoch": 0.03639740018570102, + "grad_norm": 4.8302435874938965, + "learning_rate": 1.2066831683168316e-07, + "loss": 0.6192, + "mean_token_accuracy": 0.815776526927948, + "num_tokens": 6879616.0, + "step": 196 + }, + { + "epoch": 0.036583101207056636, + "grad_norm": 3.379643440246582, + "learning_rate": 1.2128712871287127e-07, + "loss": 0.6405, + "mean_token_accuracy": 0.8100081086158752, + "num_tokens": 6915033.0, + "step": 197 + }, + { + "epoch": 0.03676880222841226, + "grad_norm": 3.42671275138855, + "learning_rate": 1.2190594059405938e-07, + "loss": 0.6945, + "mean_token_accuracy": 0.7930051684379578, + "num_tokens": 6954252.0, + "step": 198 + }, + { + "epoch": 0.03695450324976787, + "grad_norm": 3.7194011211395264, + "learning_rate": 1.2252475247524752e-07, + "loss": 0.651, + "mean_token_accuracy": 0.8098219633102417, + "num_tokens": 6986961.0, + "step": 199 + }, + { + "epoch": 0.03714020427112349, + "grad_norm": 2.7487316131591797, + "learning_rate": 1.2314356435643563e-07, + "loss": 0.6507, + "mean_token_accuracy": 0.8050393462181091, + "num_tokens": 7019878.0, + "step": 200 + }, + { + "epoch": 0.03732590529247911, + "grad_norm": 3.6962525844573975, + "learning_rate": 1.2376237623762375e-07, + "loss": 0.6674, + "mean_token_accuracy": 0.8036860823631287, + "num_tokens": 7056740.0, + "step": 201 + }, + { + "epoch": 0.037511606313834725, + "grad_norm": 3.2000925540924072, + "learning_rate": 1.2438118811881188e-07, + "loss": 0.6676, + "mean_token_accuracy": 0.802141547203064, + "num_tokens": 7086320.0, + "step": 202 + }, + { + "epoch": 0.03769730733519034, + "grad_norm": 4.939861297607422, + "learning_rate": 1.25e-07, + "loss": 0.6796, + "mean_token_accuracy": 0.8012646436691284, + "num_tokens": 7120119.0, + "step": 203 + }, + { + "epoch": 0.03788300835654596, + "grad_norm": 3.217912197113037, + "learning_rate": 1.256188118811881e-07, + "loss": 0.6829, + "mean_token_accuracy": 0.7993265390396118, + "num_tokens": 7151490.0, + "step": 204 + }, + { + "epoch": 0.03806870937790158, + "grad_norm": 3.3996872901916504, + "learning_rate": 1.2623762376237624e-07, + "loss": 0.648, + "mean_token_accuracy": 0.8052247762680054, + "num_tokens": 7187927.0, + "step": 205 + }, + { + "epoch": 0.0382544103992572, + "grad_norm": 4.319382190704346, + "learning_rate": 1.2685643564356435e-07, + "loss": 0.6339, + "mean_token_accuracy": 0.8102648258209229, + "num_tokens": 7221432.0, + "step": 206 + }, + { + "epoch": 0.038440111420612814, + "grad_norm": 7.378921985626221, + "learning_rate": 1.2747524752475247e-07, + "loss": 0.6583, + "mean_token_accuracy": 0.8047616481781006, + "num_tokens": 7258169.0, + "step": 207 + }, + { + "epoch": 0.03862581244196843, + "grad_norm": 3.1033968925476074, + "learning_rate": 1.280940594059406e-07, + "loss": 0.6804, + "mean_token_accuracy": 0.8006247878074646, + "num_tokens": 7298453.0, + "step": 208 + }, + { + "epoch": 0.03881151346332405, + "grad_norm": 3.4689064025878906, + "learning_rate": 1.2871287128712872e-07, + "loss": 0.643, + "mean_token_accuracy": 0.8101931214332581, + "num_tokens": 7334443.0, + "step": 209 + }, + { + "epoch": 0.03899721448467967, + "grad_norm": 2.973416566848755, + "learning_rate": 1.2933168316831683e-07, + "loss": 0.6715, + "mean_token_accuracy": 0.8006471395492554, + "num_tokens": 7370812.0, + "step": 210 + }, + { + "epoch": 0.03918291550603528, + "grad_norm": 2.8593955039978027, + "learning_rate": 1.2995049504950494e-07, + "loss": 0.6408, + "mean_token_accuracy": 0.8068994879722595, + "num_tokens": 7409983.0, + "step": 211 + }, + { + "epoch": 0.039368616527390904, + "grad_norm": 3.0580804347991943, + "learning_rate": 1.3056930693069308e-07, + "loss": 0.6366, + "mean_token_accuracy": 0.8108865022659302, + "num_tokens": 7449739.0, + "step": 212 + }, + { + "epoch": 0.03955431754874652, + "grad_norm": 3.1187527179718018, + "learning_rate": 1.311881188118812e-07, + "loss": 0.6388, + "mean_token_accuracy": 0.8115429878234863, + "num_tokens": 7480464.0, + "step": 213 + }, + { + "epoch": 0.039740018570102134, + "grad_norm": 2.9031643867492676, + "learning_rate": 1.318069306930693e-07, + "loss": 0.6865, + "mean_token_accuracy": 0.7994449734687805, + "num_tokens": 7514182.0, + "step": 214 + }, + { + "epoch": 0.039925719591457756, + "grad_norm": 2.9524154663085938, + "learning_rate": 1.3242574257425744e-07, + "loss": 0.6487, + "mean_token_accuracy": 0.8086788654327393, + "num_tokens": 7552708.0, + "step": 215 + }, + { + "epoch": 0.04011142061281337, + "grad_norm": 3.3034138679504395, + "learning_rate": 1.3304455445544555e-07, + "loss": 0.6007, + "mean_token_accuracy": 0.818821132183075, + "num_tokens": 7585917.0, + "step": 216 + }, + { + "epoch": 0.040297121634168986, + "grad_norm": 3.0829663276672363, + "learning_rate": 1.3366336633663366e-07, + "loss": 0.657, + "mean_token_accuracy": 0.8041734099388123, + "num_tokens": 7626126.0, + "step": 217 + }, + { + "epoch": 0.04048282265552461, + "grad_norm": 3.272042751312256, + "learning_rate": 1.342821782178218e-07, + "loss": 0.6144, + "mean_token_accuracy": 0.8132901191711426, + "num_tokens": 7661118.0, + "step": 218 + }, + { + "epoch": 0.040668523676880224, + "grad_norm": 3.9012515544891357, + "learning_rate": 1.349009900990099e-07, + "loss": 0.7036, + "mean_token_accuracy": 0.7908287048339844, + "num_tokens": 7696058.0, + "step": 219 + }, + { + "epoch": 0.04085422469823584, + "grad_norm": 3.3553946018218994, + "learning_rate": 1.3551980198019802e-07, + "loss": 0.6562, + "mean_token_accuracy": 0.8064665794372559, + "num_tokens": 7728985.0, + "step": 220 + }, + { + "epoch": 0.04103992571959146, + "grad_norm": 3.053178310394287, + "learning_rate": 1.3613861386138613e-07, + "loss": 0.6562, + "mean_token_accuracy": 0.7990467548370361, + "num_tokens": 7762078.0, + "step": 221 + }, + { + "epoch": 0.041225626740947076, + "grad_norm": 3.2676098346710205, + "learning_rate": 1.3675742574257427e-07, + "loss": 0.6448, + "mean_token_accuracy": 0.805000901222229, + "num_tokens": 7794272.0, + "step": 222 + }, + { + "epoch": 0.04141132776230269, + "grad_norm": 3.652315139770508, + "learning_rate": 1.3737623762376238e-07, + "loss": 0.6682, + "mean_token_accuracy": 0.7999470233917236, + "num_tokens": 7829623.0, + "step": 223 + }, + { + "epoch": 0.04159702878365831, + "grad_norm": 2.600637435913086, + "learning_rate": 1.379950495049505e-07, + "loss": 0.676, + "mean_token_accuracy": 0.79767906665802, + "num_tokens": 7866550.0, + "step": 224 + }, + { + "epoch": 0.04178272980501393, + "grad_norm": 3.842426300048828, + "learning_rate": 1.3861386138613863e-07, + "loss": 0.6388, + "mean_token_accuracy": 0.8102327585220337, + "num_tokens": 7901619.0, + "step": 225 + }, + { + "epoch": 0.04196843082636954, + "grad_norm": 3.122432231903076, + "learning_rate": 1.3923267326732674e-07, + "loss": 0.6171, + "mean_token_accuracy": 0.8167226314544678, + "num_tokens": 7936287.0, + "step": 226 + }, + { + "epoch": 0.042154131847725165, + "grad_norm": 3.8244919776916504, + "learning_rate": 1.3985148514851485e-07, + "loss": 0.6436, + "mean_token_accuracy": 0.8107336759567261, + "num_tokens": 7963723.0, + "step": 227 + }, + { + "epoch": 0.04233983286908078, + "grad_norm": 3.6215474605560303, + "learning_rate": 1.40470297029703e-07, + "loss": 0.6079, + "mean_token_accuracy": 0.8186788558959961, + "num_tokens": 7997510.0, + "step": 228 + }, + { + "epoch": 0.042525533890436396, + "grad_norm": 5.247171401977539, + "learning_rate": 1.4108910891089107e-07, + "loss": 0.6213, + "mean_token_accuracy": 0.8157961368560791, + "num_tokens": 8031955.0, + "step": 229 + }, + { + "epoch": 0.04271123491179202, + "grad_norm": 3.5064468383789062, + "learning_rate": 1.4170792079207919e-07, + "loss": 0.6099, + "mean_token_accuracy": 0.8209146857261658, + "num_tokens": 8063563.0, + "step": 230 + }, + { + "epoch": 0.04289693593314763, + "grad_norm": 3.141932487487793, + "learning_rate": 1.423267326732673e-07, + "loss": 0.6624, + "mean_token_accuracy": 0.8013948798179626, + "num_tokens": 8100167.0, + "step": 231 + }, + { + "epoch": 0.04308263695450325, + "grad_norm": 3.3043036460876465, + "learning_rate": 1.4294554455445543e-07, + "loss": 0.6522, + "mean_token_accuracy": 0.8051804304122925, + "num_tokens": 8133440.0, + "step": 232 + }, + { + "epoch": 0.04326833797585887, + "grad_norm": 3.4928677082061768, + "learning_rate": 1.4356435643564355e-07, + "loss": 0.6235, + "mean_token_accuracy": 0.8141793012619019, + "num_tokens": 8167044.0, + "step": 233 + }, + { + "epoch": 0.043454038997214485, + "grad_norm": 3.177762031555176, + "learning_rate": 1.4418316831683166e-07, + "loss": 0.6506, + "mean_token_accuracy": 0.8119826316833496, + "num_tokens": 8197730.0, + "step": 234 + }, + { + "epoch": 0.0436397400185701, + "grad_norm": 3.5813777446746826, + "learning_rate": 1.448019801980198e-07, + "loss": 0.6124, + "mean_token_accuracy": 0.8174567818641663, + "num_tokens": 8233439.0, + "step": 235 + }, + { + "epoch": 0.04382544103992572, + "grad_norm": 3.5584263801574707, + "learning_rate": 1.454207920792079e-07, + "loss": 0.6884, + "mean_token_accuracy": 0.7938255071640015, + "num_tokens": 8268887.0, + "step": 236 + }, + { + "epoch": 0.04401114206128134, + "grad_norm": 3.3876335620880127, + "learning_rate": 1.4603960396039602e-07, + "loss": 0.6245, + "mean_token_accuracy": 0.8120059967041016, + "num_tokens": 8305519.0, + "step": 237 + }, + { + "epoch": 0.04419684308263695, + "grad_norm": 4.675374984741211, + "learning_rate": 1.4665841584158416e-07, + "loss": 0.6952, + "mean_token_accuracy": 0.7931209206581116, + "num_tokens": 8336389.0, + "step": 238 + }, + { + "epoch": 0.044382544103992574, + "grad_norm": 2.8562285900115967, + "learning_rate": 1.4727722772277227e-07, + "loss": 0.6606, + "mean_token_accuracy": 0.8037848472595215, + "num_tokens": 8377854.0, + "step": 239 + }, + { + "epoch": 0.04456824512534819, + "grad_norm": 2.9688334465026855, + "learning_rate": 1.4789603960396038e-07, + "loss": 0.5949, + "mean_token_accuracy": 0.8191940188407898, + "num_tokens": 8413994.0, + "step": 240 + }, + { + "epoch": 0.044753946146703805, + "grad_norm": 2.939880847930908, + "learning_rate": 1.485148514851485e-07, + "loss": 0.6857, + "mean_token_accuracy": 0.7939530611038208, + "num_tokens": 8449000.0, + "step": 241 + }, + { + "epoch": 0.04493964716805943, + "grad_norm": 3.5152132511138916, + "learning_rate": 1.4913366336633663e-07, + "loss": 0.5893, + "mean_token_accuracy": 0.8246385455131531, + "num_tokens": 8488609.0, + "step": 242 + }, + { + "epoch": 0.04512534818941504, + "grad_norm": 3.3308756351470947, + "learning_rate": 1.4975247524752474e-07, + "loss": 0.7001, + "mean_token_accuracy": 0.7939149141311646, + "num_tokens": 8521969.0, + "step": 243 + }, + { + "epoch": 0.04531104921077066, + "grad_norm": 2.9615519046783447, + "learning_rate": 1.5037128712871285e-07, + "loss": 0.6361, + "mean_token_accuracy": 0.8093383312225342, + "num_tokens": 8553935.0, + "step": 244 + }, + { + "epoch": 0.04549675023212628, + "grad_norm": 3.9913792610168457, + "learning_rate": 1.50990099009901e-07, + "loss": 0.6515, + "mean_token_accuracy": 0.8041991591453552, + "num_tokens": 8584203.0, + "step": 245 + }, + { + "epoch": 0.045682451253481894, + "grad_norm": 4.810832500457764, + "learning_rate": 1.516089108910891e-07, + "loss": 0.613, + "mean_token_accuracy": 0.819075345993042, + "num_tokens": 8613370.0, + "step": 246 + }, + { + "epoch": 0.04586815227483751, + "grad_norm": 3.4517061710357666, + "learning_rate": 1.522277227722772e-07, + "loss": 0.5981, + "mean_token_accuracy": 0.8196383118629456, + "num_tokens": 8647699.0, + "step": 247 + }, + { + "epoch": 0.04605385329619313, + "grad_norm": 4.190494060516357, + "learning_rate": 1.5284653465346535e-07, + "loss": 0.6532, + "mean_token_accuracy": 0.8050528168678284, + "num_tokens": 8686044.0, + "step": 248 + }, + { + "epoch": 0.046239554317548746, + "grad_norm": 3.779356002807617, + "learning_rate": 1.5346534653465346e-07, + "loss": 0.6584, + "mean_token_accuracy": 0.8021765351295471, + "num_tokens": 8716854.0, + "step": 249 + }, + { + "epoch": 0.04642525533890436, + "grad_norm": 2.5299487113952637, + "learning_rate": 1.5408415841584157e-07, + "loss": 0.6237, + "mean_token_accuracy": 0.8115344047546387, + "num_tokens": 8755297.0, + "step": 250 + }, + { + "epoch": 0.046610956360259984, + "grad_norm": 3.0765230655670166, + "learning_rate": 1.5470297029702968e-07, + "loss": 0.6125, + "mean_token_accuracy": 0.8155803680419922, + "num_tokens": 8794784.0, + "step": 251 + }, + { + "epoch": 0.0467966573816156, + "grad_norm": 3.036757230758667, + "learning_rate": 1.5532178217821782e-07, + "loss": 0.6077, + "mean_token_accuracy": 0.8137332797050476, + "num_tokens": 8830008.0, + "step": 252 + }, + { + "epoch": 0.046982358402971214, + "grad_norm": 3.0537264347076416, + "learning_rate": 1.5594059405940593e-07, + "loss": 0.6386, + "mean_token_accuracy": 0.809027373790741, + "num_tokens": 8863842.0, + "step": 253 + }, + { + "epoch": 0.047168059424326836, + "grad_norm": 3.1134917736053467, + "learning_rate": 1.5655940594059404e-07, + "loss": 0.5627, + "mean_token_accuracy": 0.8304451704025269, + "num_tokens": 8899508.0, + "step": 254 + }, + { + "epoch": 0.04735376044568245, + "grad_norm": 3.4824886322021484, + "learning_rate": 1.5717821782178218e-07, + "loss": 0.6653, + "mean_token_accuracy": 0.8034782409667969, + "num_tokens": 8928576.0, + "step": 255 + }, + { + "epoch": 0.047539461467038066, + "grad_norm": 3.1820626258850098, + "learning_rate": 1.577970297029703e-07, + "loss": 0.5615, + "mean_token_accuracy": 0.8281439542770386, + "num_tokens": 8965439.0, + "step": 256 + }, + { + "epoch": 0.04772516248839369, + "grad_norm": 3.519888162612915, + "learning_rate": 1.584158415841584e-07, + "loss": 0.6023, + "mean_token_accuracy": 0.8193531632423401, + "num_tokens": 9002148.0, + "step": 257 + }, + { + "epoch": 0.0479108635097493, + "grad_norm": 3.2247321605682373, + "learning_rate": 1.5903465346534654e-07, + "loss": 0.6228, + "mean_token_accuracy": 0.8106353282928467, + "num_tokens": 9035593.0, + "step": 258 + }, + { + "epoch": 0.04809656453110492, + "grad_norm": 3.825690746307373, + "learning_rate": 1.5965346534653465e-07, + "loss": 0.5856, + "mean_token_accuracy": 0.8199663162231445, + "num_tokens": 9070551.0, + "step": 259 + }, + { + "epoch": 0.04828226555246054, + "grad_norm": 3.326040267944336, + "learning_rate": 1.6027227722772276e-07, + "loss": 0.5931, + "mean_token_accuracy": 0.8208743333816528, + "num_tokens": 9105244.0, + "step": 260 + }, + { + "epoch": 0.048467966573816156, + "grad_norm": 2.978408098220825, + "learning_rate": 1.608910891089109e-07, + "loss": 0.6903, + "mean_token_accuracy": 0.7928614616394043, + "num_tokens": 9144726.0, + "step": 261 + }, + { + "epoch": 0.04865366759517177, + "grad_norm": 2.8701820373535156, + "learning_rate": 1.61509900990099e-07, + "loss": 0.5834, + "mean_token_accuracy": 0.8220723867416382, + "num_tokens": 9178188.0, + "step": 262 + }, + { + "epoch": 0.04883936861652739, + "grad_norm": 2.621427297592163, + "learning_rate": 1.6212871287128712e-07, + "loss": 0.6164, + "mean_token_accuracy": 0.817266047000885, + "num_tokens": 9222999.0, + "step": 263 + }, + { + "epoch": 0.04902506963788301, + "grad_norm": 2.8436756134033203, + "learning_rate": 1.6274752475247523e-07, + "loss": 0.6238, + "mean_token_accuracy": 0.810563325881958, + "num_tokens": 9264544.0, + "step": 264 + }, + { + "epoch": 0.04921077065923862, + "grad_norm": 3.2081360816955566, + "learning_rate": 1.6336633663366337e-07, + "loss": 0.5708, + "mean_token_accuracy": 0.8266994953155518, + "num_tokens": 9299663.0, + "step": 265 + }, + { + "epoch": 0.049396471680594245, + "grad_norm": 3.073690176010132, + "learning_rate": 1.6398514851485148e-07, + "loss": 0.6243, + "mean_token_accuracy": 0.8106973171234131, + "num_tokens": 9331571.0, + "step": 266 + }, + { + "epoch": 0.04958217270194986, + "grad_norm": 2.713021755218506, + "learning_rate": 1.646039603960396e-07, + "loss": 0.6179, + "mean_token_accuracy": 0.8137155771255493, + "num_tokens": 9366995.0, + "step": 267 + }, + { + "epoch": 0.049767873723305475, + "grad_norm": 3.711960792541504, + "learning_rate": 1.6522277227722773e-07, + "loss": 0.6505, + "mean_token_accuracy": 0.8058216571807861, + "num_tokens": 9402160.0, + "step": 268 + }, + { + "epoch": 0.0499535747446611, + "grad_norm": 3.6815810203552246, + "learning_rate": 1.6584158415841584e-07, + "loss": 0.5718, + "mean_token_accuracy": 0.8256638050079346, + "num_tokens": 9437495.0, + "step": 269 + }, + { + "epoch": 0.05013927576601671, + "grad_norm": 3.0866689682006836, + "learning_rate": 1.6646039603960396e-07, + "loss": 0.6863, + "mean_token_accuracy": 0.7873004674911499, + "num_tokens": 9471224.0, + "step": 270 + }, + { + "epoch": 0.05032497678737233, + "grad_norm": 3.1238837242126465, + "learning_rate": 1.670792079207921e-07, + "loss": 0.6181, + "mean_token_accuracy": 0.8102418780326843, + "num_tokens": 9508934.0, + "step": 271 + }, + { + "epoch": 0.05051067780872795, + "grad_norm": 2.9574286937713623, + "learning_rate": 1.676980198019802e-07, + "loss": 0.6291, + "mean_token_accuracy": 0.8087395429611206, + "num_tokens": 9548687.0, + "step": 272 + }, + { + "epoch": 0.050696378830083565, + "grad_norm": 2.792793035507202, + "learning_rate": 1.6831683168316832e-07, + "loss": 0.6373, + "mean_token_accuracy": 0.8068313598632812, + "num_tokens": 9588694.0, + "step": 273 + }, + { + "epoch": 0.05088207985143918, + "grad_norm": 3.103761911392212, + "learning_rate": 1.6893564356435643e-07, + "loss": 0.6146, + "mean_token_accuracy": 0.8116483688354492, + "num_tokens": 9621039.0, + "step": 274 + }, + { + "epoch": 0.0510677808727948, + "grad_norm": 4.675351619720459, + "learning_rate": 1.6955445544554456e-07, + "loss": 0.5219, + "mean_token_accuracy": 0.8396217823028564, + "num_tokens": 9651365.0, + "step": 275 + }, + { + "epoch": 0.05125348189415042, + "grad_norm": 2.9556798934936523, + "learning_rate": 1.7017326732673268e-07, + "loss": 0.6245, + "mean_token_accuracy": 0.8078462481498718, + "num_tokens": 9686524.0, + "step": 276 + }, + { + "epoch": 0.05143918291550603, + "grad_norm": 3.9609344005584717, + "learning_rate": 1.707920792079208e-07, + "loss": 0.6263, + "mean_token_accuracy": 0.8117012977600098, + "num_tokens": 9719009.0, + "step": 277 + }, + { + "epoch": 0.051624883936861654, + "grad_norm": 3.1611440181732178, + "learning_rate": 1.7141089108910893e-07, + "loss": 0.6093, + "mean_token_accuracy": 0.8138009309768677, + "num_tokens": 9749390.0, + "step": 278 + }, + { + "epoch": 0.05181058495821727, + "grad_norm": 3.009882926940918, + "learning_rate": 1.72029702970297e-07, + "loss": 0.5648, + "mean_token_accuracy": 0.831272542476654, + "num_tokens": 9793135.0, + "step": 279 + }, + { + "epoch": 0.051996285979572884, + "grad_norm": 4.100297451019287, + "learning_rate": 1.7264851485148512e-07, + "loss": 0.6692, + "mean_token_accuracy": 0.7995054721832275, + "num_tokens": 9819022.0, + "step": 280 + }, + { + "epoch": 0.052181987000928506, + "grad_norm": 2.5473577976226807, + "learning_rate": 1.7326732673267326e-07, + "loss": 0.552, + "mean_token_accuracy": 0.8304961919784546, + "num_tokens": 9858343.0, + "step": 281 + }, + { + "epoch": 0.05236768802228412, + "grad_norm": 3.2721238136291504, + "learning_rate": 1.7388613861386137e-07, + "loss": 0.6165, + "mean_token_accuracy": 0.8099657297134399, + "num_tokens": 9893429.0, + "step": 282 + }, + { + "epoch": 0.05255338904363974, + "grad_norm": 2.977297306060791, + "learning_rate": 1.7450495049504948e-07, + "loss": 0.6534, + "mean_token_accuracy": 0.8015214204788208, + "num_tokens": 9928085.0, + "step": 283 + }, + { + "epoch": 0.05273909006499536, + "grad_norm": 3.012636184692383, + "learning_rate": 1.751237623762376e-07, + "loss": 0.6338, + "mean_token_accuracy": 0.809065580368042, + "num_tokens": 9963982.0, + "step": 284 + }, + { + "epoch": 0.052924791086350974, + "grad_norm": 3.2748970985412598, + "learning_rate": 1.7574257425742573e-07, + "loss": 0.5669, + "mean_token_accuracy": 0.827668309211731, + "num_tokens": 9997599.0, + "step": 285 + }, + { + "epoch": 0.053110492107706596, + "grad_norm": 2.756164312362671, + "learning_rate": 1.7636138613861384e-07, + "loss": 0.6337, + "mean_token_accuracy": 0.8120453357696533, + "num_tokens": 10033982.0, + "step": 286 + }, + { + "epoch": 0.05329619312906221, + "grad_norm": 3.6357767581939697, + "learning_rate": 1.7698019801980195e-07, + "loss": 0.6682, + "mean_token_accuracy": 0.7946999073028564, + "num_tokens": 10062611.0, + "step": 287 + }, + { + "epoch": 0.053481894150417826, + "grad_norm": 3.3374502658843994, + "learning_rate": 1.775990099009901e-07, + "loss": 0.5892, + "mean_token_accuracy": 0.8190606832504272, + "num_tokens": 10092501.0, + "step": 288 + }, + { + "epoch": 0.05366759517177345, + "grad_norm": 4.11712646484375, + "learning_rate": 1.782178217821782e-07, + "loss": 0.6413, + "mean_token_accuracy": 0.8050824999809265, + "num_tokens": 10129995.0, + "step": 289 + }, + { + "epoch": 0.05385329619312906, + "grad_norm": 3.1246681213378906, + "learning_rate": 1.7883663366336631e-07, + "loss": 0.6133, + "mean_token_accuracy": 0.8130673766136169, + "num_tokens": 10163682.0, + "step": 290 + }, + { + "epoch": 0.05403899721448468, + "grad_norm": 3.606943130493164, + "learning_rate": 1.7945544554455445e-07, + "loss": 0.6289, + "mean_token_accuracy": 0.8103817701339722, + "num_tokens": 10191838.0, + "step": 291 + }, + { + "epoch": 0.0542246982358403, + "grad_norm": 2.9440996646881104, + "learning_rate": 1.8007425742574256e-07, + "loss": 0.5902, + "mean_token_accuracy": 0.8244055509567261, + "num_tokens": 10226772.0, + "step": 292 + }, + { + "epoch": 0.054410399257195916, + "grad_norm": 3.2272849082946777, + "learning_rate": 1.8069306930693067e-07, + "loss": 0.567, + "mean_token_accuracy": 0.8288002610206604, + "num_tokens": 10259524.0, + "step": 293 + }, + { + "epoch": 0.05459610027855153, + "grad_norm": 3.677459478378296, + "learning_rate": 1.8131188118811879e-07, + "loss": 0.6374, + "mean_token_accuracy": 0.8093141913414001, + "num_tokens": 10293233.0, + "step": 294 + }, + { + "epoch": 0.05478180129990715, + "grad_norm": 2.851046085357666, + "learning_rate": 1.8193069306930692e-07, + "loss": 0.6042, + "mean_token_accuracy": 0.8145656585693359, + "num_tokens": 10328010.0, + "step": 295 + }, + { + "epoch": 0.05496750232126277, + "grad_norm": 3.4766228199005127, + "learning_rate": 1.8254950495049503e-07, + "loss": 0.5791, + "mean_token_accuracy": 0.8201453685760498, + "num_tokens": 10360150.0, + "step": 296 + }, + { + "epoch": 0.05515320334261838, + "grad_norm": 2.7622992992401123, + "learning_rate": 1.8316831683168315e-07, + "loss": 0.584, + "mean_token_accuracy": 0.8191922903060913, + "num_tokens": 10395637.0, + "step": 297 + }, + { + "epoch": 0.055338904363974005, + "grad_norm": 2.7253713607788086, + "learning_rate": 1.8378712871287128e-07, + "loss": 0.5696, + "mean_token_accuracy": 0.8267407417297363, + "num_tokens": 10437181.0, + "step": 298 + }, + { + "epoch": 0.05552460538532962, + "grad_norm": 2.822077989578247, + "learning_rate": 1.844059405940594e-07, + "loss": 0.6447, + "mean_token_accuracy": 0.8073567152023315, + "num_tokens": 10476624.0, + "step": 299 + }, + { + "epoch": 0.055710306406685235, + "grad_norm": 3.17734432220459, + "learning_rate": 1.850247524752475e-07, + "loss": 0.5768, + "mean_token_accuracy": 0.8253787755966187, + "num_tokens": 10512624.0, + "step": 300 + }, + { + "epoch": 0.05589600742804086, + "grad_norm": 2.7789995670318604, + "learning_rate": 1.8564356435643564e-07, + "loss": 0.5517, + "mean_token_accuracy": 0.8297863602638245, + "num_tokens": 10550085.0, + "step": 301 + }, + { + "epoch": 0.05608170844939647, + "grad_norm": 2.442723274230957, + "learning_rate": 1.8626237623762376e-07, + "loss": 0.5305, + "mean_token_accuracy": 0.8351895809173584, + "num_tokens": 10588975.0, + "step": 302 + }, + { + "epoch": 0.05626740947075209, + "grad_norm": 2.738577127456665, + "learning_rate": 1.8688118811881187e-07, + "loss": 0.5897, + "mean_token_accuracy": 0.8172361850738525, + "num_tokens": 10629111.0, + "step": 303 + }, + { + "epoch": 0.05645311049210771, + "grad_norm": 3.0133056640625, + "learning_rate": 1.875e-07, + "loss": 0.6057, + "mean_token_accuracy": 0.8164430856704712, + "num_tokens": 10661517.0, + "step": 304 + }, + { + "epoch": 0.056638811513463325, + "grad_norm": 2.9028003215789795, + "learning_rate": 1.8811881188118812e-07, + "loss": 0.6417, + "mean_token_accuracy": 0.8035975098609924, + "num_tokens": 10701945.0, + "step": 305 + }, + { + "epoch": 0.05682451253481894, + "grad_norm": 2.666762113571167, + "learning_rate": 1.8873762376237623e-07, + "loss": 0.5706, + "mean_token_accuracy": 0.8271303772926331, + "num_tokens": 10743075.0, + "step": 306 + }, + { + "epoch": 0.05701021355617456, + "grad_norm": 2.941910982131958, + "learning_rate": 1.8935643564356434e-07, + "loss": 0.5276, + "mean_token_accuracy": 0.8347398042678833, + "num_tokens": 10780937.0, + "step": 307 + }, + { + "epoch": 0.05719591457753018, + "grad_norm": 3.100557565689087, + "learning_rate": 1.8997524752475248e-07, + "loss": 0.5667, + "mean_token_accuracy": 0.8239678144454956, + "num_tokens": 10816625.0, + "step": 308 + }, + { + "epoch": 0.05738161559888579, + "grad_norm": 3.242694139480591, + "learning_rate": 1.905940594059406e-07, + "loss": 0.5975, + "mean_token_accuracy": 0.8163655400276184, + "num_tokens": 10850319.0, + "step": 309 + }, + { + "epoch": 0.057567316620241414, + "grad_norm": 2.58435320854187, + "learning_rate": 1.912128712871287e-07, + "loss": 0.6664, + "mean_token_accuracy": 0.7962522506713867, + "num_tokens": 10887942.0, + "step": 310 + }, + { + "epoch": 0.05775301764159703, + "grad_norm": 2.932990550994873, + "learning_rate": 1.9183168316831684e-07, + "loss": 0.5845, + "mean_token_accuracy": 0.8230106830596924, + "num_tokens": 10922562.0, + "step": 311 + }, + { + "epoch": 0.057938718662952644, + "grad_norm": 3.0458033084869385, + "learning_rate": 1.9245049504950495e-07, + "loss": 0.6533, + "mean_token_accuracy": 0.8050445914268494, + "num_tokens": 10954693.0, + "step": 312 + }, + { + "epoch": 0.058124419684308266, + "grad_norm": 2.287778377532959, + "learning_rate": 1.9306930693069306e-07, + "loss": 0.6553, + "mean_token_accuracy": 0.8042346239089966, + "num_tokens": 10997376.0, + "step": 313 + }, + { + "epoch": 0.05831012070566388, + "grad_norm": 2.2840521335601807, + "learning_rate": 1.936881188118812e-07, + "loss": 0.5489, + "mean_token_accuracy": 0.8246674537658691, + "num_tokens": 11033774.0, + "step": 314 + }, + { + "epoch": 0.0584958217270195, + "grad_norm": 2.892524480819702, + "learning_rate": 1.943069306930693e-07, + "loss": 0.5771, + "mean_token_accuracy": 0.8248862028121948, + "num_tokens": 11066476.0, + "step": 315 + }, + { + "epoch": 0.05868152274837512, + "grad_norm": 2.7211899757385254, + "learning_rate": 1.9492574257425742e-07, + "loss": 0.5805, + "mean_token_accuracy": 0.820745587348938, + "num_tokens": 11101968.0, + "step": 316 + }, + { + "epoch": 0.058867223769730734, + "grad_norm": 3.1330504417419434, + "learning_rate": 1.9554455445544553e-07, + "loss": 0.5638, + "mean_token_accuracy": 0.8261843919754028, + "num_tokens": 11132484.0, + "step": 317 + }, + { + "epoch": 0.05905292479108635, + "grad_norm": 2.8268954753875732, + "learning_rate": 1.9616336633663367e-07, + "loss": 0.5556, + "mean_token_accuracy": 0.8312277793884277, + "num_tokens": 11168057.0, + "step": 318 + }, + { + "epoch": 0.05923862581244197, + "grad_norm": 3.1908466815948486, + "learning_rate": 1.9678217821782178e-07, + "loss": 0.6009, + "mean_token_accuracy": 0.8175234794616699, + "num_tokens": 11202648.0, + "step": 319 + }, + { + "epoch": 0.059424326833797586, + "grad_norm": 2.8709757328033447, + "learning_rate": 1.974009900990099e-07, + "loss": 0.554, + "mean_token_accuracy": 0.8293982744216919, + "num_tokens": 11232205.0, + "step": 320 + }, + { + "epoch": 0.0596100278551532, + "grad_norm": 2.7211599349975586, + "learning_rate": 1.9801980198019803e-07, + "loss": 0.6414, + "mean_token_accuracy": 0.8036176562309265, + "num_tokens": 11269738.0, + "step": 321 + }, + { + "epoch": 0.05979572887650882, + "grad_norm": 2.3829801082611084, + "learning_rate": 1.9863861386138614e-07, + "loss": 0.5529, + "mean_token_accuracy": 0.8280086517333984, + "num_tokens": 11307759.0, + "step": 322 + }, + { + "epoch": 0.05998142989786444, + "grad_norm": 2.4407639503479004, + "learning_rate": 1.9925742574257425e-07, + "loss": 0.6, + "mean_token_accuracy": 0.8118710517883301, + "num_tokens": 11345776.0, + "step": 323 + }, + { + "epoch": 0.06016713091922005, + "grad_norm": 2.542731523513794, + "learning_rate": 1.998762376237624e-07, + "loss": 0.6438, + "mean_token_accuracy": 0.8034313917160034, + "num_tokens": 11383937.0, + "step": 324 + }, + { + "epoch": 0.060352831940575676, + "grad_norm": 2.720210075378418, + "learning_rate": 2.004950495049505e-07, + "loss": 0.55, + "mean_token_accuracy": 0.8296675086021423, + "num_tokens": 11418390.0, + "step": 325 + }, + { + "epoch": 0.06053853296193129, + "grad_norm": 2.556610107421875, + "learning_rate": 2.011138613861386e-07, + "loss": 0.6063, + "mean_token_accuracy": 0.812538743019104, + "num_tokens": 11454208.0, + "step": 326 + }, + { + "epoch": 0.060724233983286906, + "grad_norm": 2.941575050354004, + "learning_rate": 2.0173267326732672e-07, + "loss": 0.5463, + "mean_token_accuracy": 0.8302255868911743, + "num_tokens": 11487674.0, + "step": 327 + }, + { + "epoch": 0.06090993500464253, + "grad_norm": 2.567660331726074, + "learning_rate": 2.0235148514851486e-07, + "loss": 0.5933, + "mean_token_accuracy": 0.814540445804596, + "num_tokens": 11526415.0, + "step": 328 + }, + { + "epoch": 0.06109563602599814, + "grad_norm": 2.750556230545044, + "learning_rate": 2.0297029702970297e-07, + "loss": 0.5655, + "mean_token_accuracy": 0.8239122033119202, + "num_tokens": 11560397.0, + "step": 329 + }, + { + "epoch": 0.06128133704735376, + "grad_norm": 2.3141605854034424, + "learning_rate": 2.0358910891089106e-07, + "loss": 0.5857, + "mean_token_accuracy": 0.8203139305114746, + "num_tokens": 11602758.0, + "step": 330 + }, + { + "epoch": 0.06146703806870938, + "grad_norm": 2.9669909477233887, + "learning_rate": 2.042079207920792e-07, + "loss": 0.5319, + "mean_token_accuracy": 0.8352600336074829, + "num_tokens": 11638728.0, + "step": 331 + }, + { + "epoch": 0.061652739090064995, + "grad_norm": 2.6393699645996094, + "learning_rate": 2.048267326732673e-07, + "loss": 0.621, + "mean_token_accuracy": 0.812468409538269, + "num_tokens": 11672241.0, + "step": 332 + }, + { + "epoch": 0.06183844011142061, + "grad_norm": 2.896944761276245, + "learning_rate": 2.0544554455445542e-07, + "loss": 0.6517, + "mean_token_accuracy": 0.802639901638031, + "num_tokens": 11710834.0, + "step": 333 + }, + { + "epoch": 0.06202414113277623, + "grad_norm": 2.720349073410034, + "learning_rate": 2.0606435643564356e-07, + "loss": 0.5601, + "mean_token_accuracy": 0.8287453651428223, + "num_tokens": 11744700.0, + "step": 334 + }, + { + "epoch": 0.06220984215413185, + "grad_norm": 2.8878321647644043, + "learning_rate": 2.0668316831683167e-07, + "loss": 0.581, + "mean_token_accuracy": 0.8179432153701782, + "num_tokens": 11779299.0, + "step": 335 + }, + { + "epoch": 0.06239554317548746, + "grad_norm": 2.5679409503936768, + "learning_rate": 2.0730198019801978e-07, + "loss": 0.5392, + "mean_token_accuracy": 0.8303713798522949, + "num_tokens": 11818169.0, + "step": 336 + }, + { + "epoch": 0.06258124419684308, + "grad_norm": 2.91257381439209, + "learning_rate": 2.079207920792079e-07, + "loss": 0.646, + "mean_token_accuracy": 0.8044091463088989, + "num_tokens": 11848564.0, + "step": 337 + }, + { + "epoch": 0.0627669452181987, + "grad_norm": 2.4053544998168945, + "learning_rate": 2.0853960396039603e-07, + "loss": 0.5687, + "mean_token_accuracy": 0.8248809576034546, + "num_tokens": 11886487.0, + "step": 338 + }, + { + "epoch": 0.06295264623955432, + "grad_norm": 2.7437562942504883, + "learning_rate": 2.0915841584158414e-07, + "loss": 0.6264, + "mean_token_accuracy": 0.8156580924987793, + "num_tokens": 11922157.0, + "step": 339 + }, + { + "epoch": 0.06313834726090993, + "grad_norm": 2.690556287765503, + "learning_rate": 2.0977722772277225e-07, + "loss": 0.5747, + "mean_token_accuracy": 0.8236458897590637, + "num_tokens": 11960190.0, + "step": 340 + }, + { + "epoch": 0.06332404828226555, + "grad_norm": 2.5125069618225098, + "learning_rate": 2.103960396039604e-07, + "loss": 0.5771, + "mean_token_accuracy": 0.8221901655197144, + "num_tokens": 11991269.0, + "step": 341 + }, + { + "epoch": 0.06350974930362117, + "grad_norm": 2.5982296466827393, + "learning_rate": 2.110148514851485e-07, + "loss": 0.5839, + "mean_token_accuracy": 0.8230233788490295, + "num_tokens": 12022664.0, + "step": 342 + }, + { + "epoch": 0.06369545032497678, + "grad_norm": 2.6107656955718994, + "learning_rate": 2.116336633663366e-07, + "loss": 0.5561, + "mean_token_accuracy": 0.8224133253097534, + "num_tokens": 12053766.0, + "step": 343 + }, + { + "epoch": 0.0638811513463324, + "grad_norm": 2.6496691703796387, + "learning_rate": 2.1225247524752475e-07, + "loss": 0.5788, + "mean_token_accuracy": 0.8296577334403992, + "num_tokens": 12080986.0, + "step": 344 + }, + { + "epoch": 0.06406685236768803, + "grad_norm": 2.7854361534118652, + "learning_rate": 2.1287128712871286e-07, + "loss": 0.555, + "mean_token_accuracy": 0.8273677229881287, + "num_tokens": 12116703.0, + "step": 345 + }, + { + "epoch": 0.06425255338904363, + "grad_norm": 2.2466022968292236, + "learning_rate": 2.1349009900990097e-07, + "loss": 0.5613, + "mean_token_accuracy": 0.824924111366272, + "num_tokens": 12154629.0, + "step": 346 + }, + { + "epoch": 0.06443825441039926, + "grad_norm": 2.907761573791504, + "learning_rate": 2.1410891089108908e-07, + "loss": 0.5586, + "mean_token_accuracy": 0.8258869647979736, + "num_tokens": 12184334.0, + "step": 347 + }, + { + "epoch": 0.06462395543175488, + "grad_norm": 3.0226938724517822, + "learning_rate": 2.1472772277227722e-07, + "loss": 0.5517, + "mean_token_accuracy": 0.8280020356178284, + "num_tokens": 12212743.0, + "step": 348 + }, + { + "epoch": 0.06480965645311049, + "grad_norm": 2.370757818222046, + "learning_rate": 2.1534653465346533e-07, + "loss": 0.6036, + "mean_token_accuracy": 0.8199981451034546, + "num_tokens": 12248434.0, + "step": 349 + }, + { + "epoch": 0.06499535747446611, + "grad_norm": 2.4226603507995605, + "learning_rate": 2.1596534653465344e-07, + "loss": 0.5307, + "mean_token_accuracy": 0.8336042165756226, + "num_tokens": 12283165.0, + "step": 350 + }, + { + "epoch": 0.06518105849582173, + "grad_norm": 2.4322164058685303, + "learning_rate": 2.1658415841584158e-07, + "loss": 0.5431, + "mean_token_accuracy": 0.8334487676620483, + "num_tokens": 12325454.0, + "step": 351 + }, + { + "epoch": 0.06536675951717734, + "grad_norm": 2.397152900695801, + "learning_rate": 2.172029702970297e-07, + "loss": 0.6161, + "mean_token_accuracy": 0.8131683468818665, + "num_tokens": 12362921.0, + "step": 352 + }, + { + "epoch": 0.06555246053853296, + "grad_norm": 2.3854305744171143, + "learning_rate": 2.178217821782178e-07, + "loss": 0.5529, + "mean_token_accuracy": 0.8288412094116211, + "num_tokens": 12397749.0, + "step": 353 + }, + { + "epoch": 0.06573816155988858, + "grad_norm": 2.2565574645996094, + "learning_rate": 2.1844059405940594e-07, + "loss": 0.5219, + "mean_token_accuracy": 0.8399975299835205, + "num_tokens": 12433366.0, + "step": 354 + }, + { + "epoch": 0.06592386258124419, + "grad_norm": 2.527730941772461, + "learning_rate": 2.1905940594059405e-07, + "loss": 0.5222, + "mean_token_accuracy": 0.8380529284477234, + "num_tokens": 12473982.0, + "step": 355 + }, + { + "epoch": 0.06610956360259981, + "grad_norm": 2.6789631843566895, + "learning_rate": 2.1967821782178216e-07, + "loss": 0.5492, + "mean_token_accuracy": 0.8303410410881042, + "num_tokens": 12507692.0, + "step": 356 + }, + { + "epoch": 0.06629526462395544, + "grad_norm": 2.6687965393066406, + "learning_rate": 2.202970297029703e-07, + "loss": 0.5618, + "mean_token_accuracy": 0.8257537484169006, + "num_tokens": 12538495.0, + "step": 357 + }, + { + "epoch": 0.06648096564531104, + "grad_norm": 2.178556203842163, + "learning_rate": 2.209158415841584e-07, + "loss": 0.5994, + "mean_token_accuracy": 0.8152117729187012, + "num_tokens": 12579154.0, + "step": 358 + }, + { + "epoch": 0.06666666666666667, + "grad_norm": 2.6066083908081055, + "learning_rate": 2.2153465346534652e-07, + "loss": 0.5916, + "mean_token_accuracy": 0.8194295763969421, + "num_tokens": 12609185.0, + "step": 359 + }, + { + "epoch": 0.06685236768802229, + "grad_norm": 2.6743850708007812, + "learning_rate": 2.2215346534653464e-07, + "loss": 0.5552, + "mean_token_accuracy": 0.8280736804008484, + "num_tokens": 12635395.0, + "step": 360 + }, + { + "epoch": 0.0670380687093779, + "grad_norm": 2.3187735080718994, + "learning_rate": 2.2277227722772277e-07, + "loss": 0.539, + "mean_token_accuracy": 0.8352199792861938, + "num_tokens": 12670517.0, + "step": 361 + }, + { + "epoch": 0.06722376973073352, + "grad_norm": 2.5096869468688965, + "learning_rate": 2.2339108910891088e-07, + "loss": 0.5371, + "mean_token_accuracy": 0.8324481248855591, + "num_tokens": 12706642.0, + "step": 362 + }, + { + "epoch": 0.06740947075208914, + "grad_norm": 2.5308382511138916, + "learning_rate": 2.24009900990099e-07, + "loss": 0.5532, + "mean_token_accuracy": 0.8267876505851746, + "num_tokens": 12739218.0, + "step": 363 + }, + { + "epoch": 0.06759517177344475, + "grad_norm": 2.1739563941955566, + "learning_rate": 2.2462871287128713e-07, + "loss": 0.5881, + "mean_token_accuracy": 0.8223356008529663, + "num_tokens": 12769967.0, + "step": 364 + }, + { + "epoch": 0.06778087279480037, + "grad_norm": 2.4076874256134033, + "learning_rate": 2.2524752475247524e-07, + "loss": 0.5557, + "mean_token_accuracy": 0.8273302316665649, + "num_tokens": 12804383.0, + "step": 365 + }, + { + "epoch": 0.06796657381615599, + "grad_norm": 2.965022563934326, + "learning_rate": 2.2586633663366336e-07, + "loss": 0.567, + "mean_token_accuracy": 0.8304897546768188, + "num_tokens": 12837788.0, + "step": 366 + }, + { + "epoch": 0.0681522748375116, + "grad_norm": 2.6384799480438232, + "learning_rate": 2.264851485148515e-07, + "loss": 0.5786, + "mean_token_accuracy": 0.8245049715042114, + "num_tokens": 12872107.0, + "step": 367 + }, + { + "epoch": 0.06833797585886722, + "grad_norm": 2.3075993061065674, + "learning_rate": 2.271039603960396e-07, + "loss": 0.6135, + "mean_token_accuracy": 0.8087102770805359, + "num_tokens": 12902889.0, + "step": 368 + }, + { + "epoch": 0.06852367688022284, + "grad_norm": 2.532459020614624, + "learning_rate": 2.2772277227722772e-07, + "loss": 0.59, + "mean_token_accuracy": 0.8161113262176514, + "num_tokens": 12940080.0, + "step": 369 + }, + { + "epoch": 0.06870937790157845, + "grad_norm": 2.3889002799987793, + "learning_rate": 2.2834158415841583e-07, + "loss": 0.5737, + "mean_token_accuracy": 0.8234008550643921, + "num_tokens": 12972813.0, + "step": 370 + }, + { + "epoch": 0.06889507892293407, + "grad_norm": 2.6219964027404785, + "learning_rate": 2.2896039603960397e-07, + "loss": 0.5968, + "mean_token_accuracy": 0.8206115961074829, + "num_tokens": 13005721.0, + "step": 371 + }, + { + "epoch": 0.0690807799442897, + "grad_norm": 2.5161826610565186, + "learning_rate": 2.2957920792079208e-07, + "loss": 0.5644, + "mean_token_accuracy": 0.8248375654220581, + "num_tokens": 13038874.0, + "step": 372 + }, + { + "epoch": 0.0692664809656453, + "grad_norm": 2.8676726818084717, + "learning_rate": 2.301980198019802e-07, + "loss": 0.5104, + "mean_token_accuracy": 0.8405030369758606, + "num_tokens": 13071309.0, + "step": 373 + }, + { + "epoch": 0.06945218198700093, + "grad_norm": 2.1719329357147217, + "learning_rate": 2.3081683168316833e-07, + "loss": 0.5381, + "mean_token_accuracy": 0.8325580954551697, + "num_tokens": 13113179.0, + "step": 374 + }, + { + "epoch": 0.06963788300835655, + "grad_norm": 2.3539440631866455, + "learning_rate": 2.3143564356435644e-07, + "loss": 0.5518, + "mean_token_accuracy": 0.8270180821418762, + "num_tokens": 13144151.0, + "step": 375 + }, + { + "epoch": 0.06982358402971216, + "grad_norm": 2.1696441173553467, + "learning_rate": 2.3205445544554455e-07, + "loss": 0.566, + "mean_token_accuracy": 0.8246710896492004, + "num_tokens": 13183136.0, + "step": 376 + }, + { + "epoch": 0.07000928505106778, + "grad_norm": 3.3961892127990723, + "learning_rate": 2.3267326732673269e-07, + "loss": 0.5072, + "mean_token_accuracy": 0.8382606506347656, + "num_tokens": 13208386.0, + "step": 377 + }, + { + "epoch": 0.0701949860724234, + "grad_norm": 2.6280739307403564, + "learning_rate": 2.332920792079208e-07, + "loss": 0.6289, + "mean_token_accuracy": 0.8032265901565552, + "num_tokens": 13246830.0, + "step": 378 + }, + { + "epoch": 0.07038068709377901, + "grad_norm": 2.2816498279571533, + "learning_rate": 2.339108910891089e-07, + "loss": 0.6195, + "mean_token_accuracy": 0.8110495805740356, + "num_tokens": 13286937.0, + "step": 379 + }, + { + "epoch": 0.07056638811513463, + "grad_norm": 2.392512321472168, + "learning_rate": 2.34529702970297e-07, + "loss": 0.6041, + "mean_token_accuracy": 0.8126296997070312, + "num_tokens": 13324850.0, + "step": 380 + }, + { + "epoch": 0.07075208913649025, + "grad_norm": 2.1712286472320557, + "learning_rate": 2.3514851485148513e-07, + "loss": 0.5568, + "mean_token_accuracy": 0.8243429660797119, + "num_tokens": 13360004.0, + "step": 381 + }, + { + "epoch": 0.07093779015784586, + "grad_norm": 2.593606472015381, + "learning_rate": 2.3576732673267324e-07, + "loss": 0.5603, + "mean_token_accuracy": 0.8230899572372437, + "num_tokens": 13392516.0, + "step": 382 + }, + { + "epoch": 0.07112349117920148, + "grad_norm": 2.634112596511841, + "learning_rate": 2.3638613861386135e-07, + "loss": 0.567, + "mean_token_accuracy": 0.8247793316841125, + "num_tokens": 13425989.0, + "step": 383 + }, + { + "epoch": 0.0713091922005571, + "grad_norm": 2.7148637771606445, + "learning_rate": 2.370049504950495e-07, + "loss": 0.6004, + "mean_token_accuracy": 0.8126302361488342, + "num_tokens": 13458137.0, + "step": 384 + }, + { + "epoch": 0.07149489322191271, + "grad_norm": 2.324324131011963, + "learning_rate": 2.376237623762376e-07, + "loss": 0.5491, + "mean_token_accuracy": 0.8288040161132812, + "num_tokens": 13491320.0, + "step": 385 + }, + { + "epoch": 0.07168059424326834, + "grad_norm": 2.5643298625946045, + "learning_rate": 2.3824257425742571e-07, + "loss": 0.6296, + "mean_token_accuracy": 0.811816930770874, + "num_tokens": 13520230.0, + "step": 386 + }, + { + "epoch": 0.07186629526462396, + "grad_norm": 2.090369462966919, + "learning_rate": 2.388613861386138e-07, + "loss": 0.5622, + "mean_token_accuracy": 0.8263182640075684, + "num_tokens": 13555792.0, + "step": 387 + }, + { + "epoch": 0.07205199628597957, + "grad_norm": 2.339742422103882, + "learning_rate": 2.3948019801980194e-07, + "loss": 0.6082, + "mean_token_accuracy": 0.813553512096405, + "num_tokens": 13590695.0, + "step": 388 + }, + { + "epoch": 0.07223769730733519, + "grad_norm": 2.2349462509155273, + "learning_rate": 2.400990099009901e-07, + "loss": 0.5262, + "mean_token_accuracy": 0.8367822170257568, + "num_tokens": 13622221.0, + "step": 389 + }, + { + "epoch": 0.07242339832869081, + "grad_norm": 2.2056586742401123, + "learning_rate": 2.407178217821782e-07, + "loss": 0.5561, + "mean_token_accuracy": 0.8230234384536743, + "num_tokens": 13658955.0, + "step": 390 + }, + { + "epoch": 0.07260909935004642, + "grad_norm": 2.0290589332580566, + "learning_rate": 2.413366336633663e-07, + "loss": 0.5791, + "mean_token_accuracy": 0.8212937116622925, + "num_tokens": 13696358.0, + "step": 391 + }, + { + "epoch": 0.07279480037140204, + "grad_norm": 1.8767321109771729, + "learning_rate": 2.4195544554455444e-07, + "loss": 0.5236, + "mean_token_accuracy": 0.8367720246315002, + "num_tokens": 13738305.0, + "step": 392 + }, + { + "epoch": 0.07298050139275766, + "grad_norm": 2.013927936553955, + "learning_rate": 2.4257425742574255e-07, + "loss": 0.5463, + "mean_token_accuracy": 0.8281389474868774, + "num_tokens": 13780463.0, + "step": 393 + }, + { + "epoch": 0.07316620241411327, + "grad_norm": 2.244053602218628, + "learning_rate": 2.4319306930693066e-07, + "loss": 0.5697, + "mean_token_accuracy": 0.8233824968338013, + "num_tokens": 13816460.0, + "step": 394 + }, + { + "epoch": 0.0733519034354689, + "grad_norm": 2.016507625579834, + "learning_rate": 2.4381188118811877e-07, + "loss": 0.5555, + "mean_token_accuracy": 0.8266714215278625, + "num_tokens": 13855540.0, + "step": 395 + }, + { + "epoch": 0.07353760445682452, + "grad_norm": 2.032062530517578, + "learning_rate": 2.4443069306930693e-07, + "loss": 0.5923, + "mean_token_accuracy": 0.8195043206214905, + "num_tokens": 13895620.0, + "step": 396 + }, + { + "epoch": 0.07372330547818012, + "grad_norm": 2.1489880084991455, + "learning_rate": 2.4504950495049505e-07, + "loss": 0.5961, + "mean_token_accuracy": 0.8197455406188965, + "num_tokens": 13931230.0, + "step": 397 + }, + { + "epoch": 0.07390900649953575, + "grad_norm": 2.3156235218048096, + "learning_rate": 2.4566831683168316e-07, + "loss": 0.5038, + "mean_token_accuracy": 0.842558741569519, + "num_tokens": 13966295.0, + "step": 398 + }, + { + "epoch": 0.07409470752089137, + "grad_norm": 2.7965874671936035, + "learning_rate": 2.4628712871287127e-07, + "loss": 0.5916, + "mean_token_accuracy": 0.8145913481712341, + "num_tokens": 13996871.0, + "step": 399 + }, + { + "epoch": 0.07428040854224698, + "grad_norm": 2.2016263008117676, + "learning_rate": 2.469059405940594e-07, + "loss": 0.5679, + "mean_token_accuracy": 0.822528600692749, + "num_tokens": 14032885.0, + "step": 400 + }, + { + "epoch": 0.0744661095636026, + "grad_norm": 1.9073041677474976, + "learning_rate": 2.475247524752475e-07, + "loss": 0.5405, + "mean_token_accuracy": 0.833532452583313, + "num_tokens": 14070588.0, + "step": 401 + }, + { + "epoch": 0.07465181058495822, + "grad_norm": 2.476433038711548, + "learning_rate": 2.4814356435643565e-07, + "loss": 0.5288, + "mean_token_accuracy": 0.8350086808204651, + "num_tokens": 14102594.0, + "step": 402 + }, + { + "epoch": 0.07483751160631383, + "grad_norm": 2.4013047218322754, + "learning_rate": 2.4876237623762377e-07, + "loss": 0.5201, + "mean_token_accuracy": 0.8383609056472778, + "num_tokens": 14130342.0, + "step": 403 + }, + { + "epoch": 0.07502321262766945, + "grad_norm": 2.0898029804229736, + "learning_rate": 2.493811881188119e-07, + "loss": 0.5529, + "mean_token_accuracy": 0.827180027961731, + "num_tokens": 14168359.0, + "step": 404 + }, + { + "epoch": 0.07520891364902507, + "grad_norm": 2.166151523590088, + "learning_rate": 2.5e-07, + "loss": 0.5784, + "mean_token_accuracy": 0.8176109790802002, + "num_tokens": 14202045.0, + "step": 405 + }, + { + "epoch": 0.07539461467038068, + "grad_norm": 2.056535243988037, + "learning_rate": 2.506188118811881e-07, + "loss": 0.5418, + "mean_token_accuracy": 0.8285390138626099, + "num_tokens": 14236184.0, + "step": 406 + }, + { + "epoch": 0.0755803156917363, + "grad_norm": 1.9077504873275757, + "learning_rate": 2.512376237623762e-07, + "loss": 0.5387, + "mean_token_accuracy": 0.8327031135559082, + "num_tokens": 14276623.0, + "step": 407 + }, + { + "epoch": 0.07576601671309192, + "grad_norm": 1.895838737487793, + "learning_rate": 2.518564356435643e-07, + "loss": 0.5545, + "mean_token_accuracy": 0.8212710618972778, + "num_tokens": 14313779.0, + "step": 408 + }, + { + "epoch": 0.07595171773444755, + "grad_norm": 1.786838412284851, + "learning_rate": 2.524752475247525e-07, + "loss": 0.5253, + "mean_token_accuracy": 0.8369919657707214, + "num_tokens": 14352039.0, + "step": 409 + }, + { + "epoch": 0.07613741875580315, + "grad_norm": 2.4797661304473877, + "learning_rate": 2.530940594059406e-07, + "loss": 0.5191, + "mean_token_accuracy": 0.8402203321456909, + "num_tokens": 14381325.0, + "step": 410 + }, + { + "epoch": 0.07632311977715878, + "grad_norm": 2.3064966201782227, + "learning_rate": 2.537128712871287e-07, + "loss": 0.6038, + "mean_token_accuracy": 0.8124874234199524, + "num_tokens": 14411505.0, + "step": 411 + }, + { + "epoch": 0.0765088207985144, + "grad_norm": 2.0030415058135986, + "learning_rate": 2.543316831683168e-07, + "loss": 0.5297, + "mean_token_accuracy": 0.8377548456192017, + "num_tokens": 14442282.0, + "step": 412 + }, + { + "epoch": 0.07669452181987, + "grad_norm": 2.390845775604248, + "learning_rate": 2.5495049504950493e-07, + "loss": 0.5286, + "mean_token_accuracy": 0.8335369825363159, + "num_tokens": 14469126.0, + "step": 413 + }, + { + "epoch": 0.07688022284122563, + "grad_norm": 2.0886924266815186, + "learning_rate": 2.5556930693069304e-07, + "loss": 0.5289, + "mean_token_accuracy": 0.836193323135376, + "num_tokens": 14504066.0, + "step": 414 + }, + { + "epoch": 0.07706592386258125, + "grad_norm": 2.005706548690796, + "learning_rate": 2.561881188118812e-07, + "loss": 0.4588, + "mean_token_accuracy": 0.8541619777679443, + "num_tokens": 14539928.0, + "step": 415 + }, + { + "epoch": 0.07725162488393686, + "grad_norm": 2.067627191543579, + "learning_rate": 2.568069306930693e-07, + "loss": 0.4888, + "mean_token_accuracy": 0.8442082405090332, + "num_tokens": 14574159.0, + "step": 416 + }, + { + "epoch": 0.07743732590529248, + "grad_norm": 2.6198155879974365, + "learning_rate": 2.5742574257425743e-07, + "loss": 0.6236, + "mean_token_accuracy": 0.8047987222671509, + "num_tokens": 14604923.0, + "step": 417 + }, + { + "epoch": 0.0776230269266481, + "grad_norm": 2.0661559104919434, + "learning_rate": 2.5804455445544554e-07, + "loss": 0.5142, + "mean_token_accuracy": 0.8364146947860718, + "num_tokens": 14638106.0, + "step": 418 + }, + { + "epoch": 0.07780872794800371, + "grad_norm": 1.8145461082458496, + "learning_rate": 2.5866336633663365e-07, + "loss": 0.495, + "mean_token_accuracy": 0.845639705657959, + "num_tokens": 14674068.0, + "step": 419 + }, + { + "epoch": 0.07799442896935933, + "grad_norm": 1.686231017112732, + "learning_rate": 2.5928217821782176e-07, + "loss": 0.5187, + "mean_token_accuracy": 0.8397143483161926, + "num_tokens": 14715788.0, + "step": 420 + }, + { + "epoch": 0.07818012999071496, + "grad_norm": 2.0945351123809814, + "learning_rate": 2.599009900990099e-07, + "loss": 0.6158, + "mean_token_accuracy": 0.8092693090438843, + "num_tokens": 14747334.0, + "step": 421 + }, + { + "epoch": 0.07836583101207056, + "grad_norm": 2.0048863887786865, + "learning_rate": 2.6051980198019804e-07, + "loss": 0.5755, + "mean_token_accuracy": 0.8265540599822998, + "num_tokens": 14780484.0, + "step": 422 + }, + { + "epoch": 0.07855153203342619, + "grad_norm": 1.8623619079589844, + "learning_rate": 2.6113861386138615e-07, + "loss": 0.4768, + "mean_token_accuracy": 0.8462073802947998, + "num_tokens": 14817657.0, + "step": 423 + }, + { + "epoch": 0.07873723305478181, + "grad_norm": 2.089392900466919, + "learning_rate": 2.6175742574257426e-07, + "loss": 0.565, + "mean_token_accuracy": 0.827667772769928, + "num_tokens": 14851146.0, + "step": 424 + }, + { + "epoch": 0.07892293407613742, + "grad_norm": 1.9494028091430664, + "learning_rate": 2.623762376237624e-07, + "loss": 0.5689, + "mean_token_accuracy": 0.8241245746612549, + "num_tokens": 14891234.0, + "step": 425 + }, + { + "epoch": 0.07910863509749304, + "grad_norm": 2.6755168437957764, + "learning_rate": 2.629950495049505e-07, + "loss": 0.5022, + "mean_token_accuracy": 0.843913197517395, + "num_tokens": 14921047.0, + "step": 426 + }, + { + "epoch": 0.07929433611884866, + "grad_norm": 1.940064549446106, + "learning_rate": 2.636138613861386e-07, + "loss": 0.5082, + "mean_token_accuracy": 0.8424521684646606, + "num_tokens": 14959358.0, + "step": 427 + }, + { + "epoch": 0.07948003714020427, + "grad_norm": 1.9022217988967896, + "learning_rate": 2.642326732673267e-07, + "loss": 0.5311, + "mean_token_accuracy": 0.8336639404296875, + "num_tokens": 15000708.0, + "step": 428 + }, + { + "epoch": 0.07966573816155989, + "grad_norm": 2.058509111404419, + "learning_rate": 2.6485148514851487e-07, + "loss": 0.5367, + "mean_token_accuracy": 0.8312013149261475, + "num_tokens": 15032878.0, + "step": 429 + }, + { + "epoch": 0.07985143918291551, + "grad_norm": 1.9216638803482056, + "learning_rate": 2.65470297029703e-07, + "loss": 0.5618, + "mean_token_accuracy": 0.8266297578811646, + "num_tokens": 15068293.0, + "step": 430 + }, + { + "epoch": 0.08003714020427112, + "grad_norm": 2.154123067855835, + "learning_rate": 2.660891089108911e-07, + "loss": 0.5121, + "mean_token_accuracy": 0.8439209461212158, + "num_tokens": 15100666.0, + "step": 431 + }, + { + "epoch": 0.08022284122562674, + "grad_norm": 2.0132291316986084, + "learning_rate": 2.667079207920792e-07, + "loss": 0.5221, + "mean_token_accuracy": 0.8368375897407532, + "num_tokens": 15134517.0, + "step": 432 + }, + { + "epoch": 0.08040854224698236, + "grad_norm": 1.8506748676300049, + "learning_rate": 2.673267326732673e-07, + "loss": 0.5253, + "mean_token_accuracy": 0.8387840986251831, + "num_tokens": 15168887.0, + "step": 433 + }, + { + "epoch": 0.08059424326833797, + "grad_norm": 1.7692924737930298, + "learning_rate": 2.6794554455445543e-07, + "loss": 0.4909, + "mean_token_accuracy": 0.8448084592819214, + "num_tokens": 15205714.0, + "step": 434 + }, + { + "epoch": 0.0807799442896936, + "grad_norm": 2.523268222808838, + "learning_rate": 2.685643564356436e-07, + "loss": 0.5319, + "mean_token_accuracy": 0.8345255851745605, + "num_tokens": 15239670.0, + "step": 435 + }, + { + "epoch": 0.08096564531104922, + "grad_norm": 2.0537190437316895, + "learning_rate": 2.691831683168317e-07, + "loss": 0.5222, + "mean_token_accuracy": 0.834814190864563, + "num_tokens": 15269436.0, + "step": 436 + }, + { + "epoch": 0.08115134633240483, + "grad_norm": 1.9077221155166626, + "learning_rate": 2.698019801980198e-07, + "loss": 0.5597, + "mean_token_accuracy": 0.823638916015625, + "num_tokens": 15301797.0, + "step": 437 + }, + { + "epoch": 0.08133704735376045, + "grad_norm": 2.296617031097412, + "learning_rate": 2.7042079207920793e-07, + "loss": 0.5118, + "mean_token_accuracy": 0.8322024941444397, + "num_tokens": 15327410.0, + "step": 438 + }, + { + "epoch": 0.08152274837511607, + "grad_norm": 2.153323173522949, + "learning_rate": 2.7103960396039604e-07, + "loss": 0.6199, + "mean_token_accuracy": 0.8092421293258667, + "num_tokens": 15358012.0, + "step": 439 + }, + { + "epoch": 0.08170844939647168, + "grad_norm": 1.7625986337661743, + "learning_rate": 2.7165841584158415e-07, + "loss": 0.469, + "mean_token_accuracy": 0.8525468707084656, + "num_tokens": 15395866.0, + "step": 440 + }, + { + "epoch": 0.0818941504178273, + "grad_norm": 2.0124599933624268, + "learning_rate": 2.7227722772277226e-07, + "loss": 0.5462, + "mean_token_accuracy": 0.8309603929519653, + "num_tokens": 15434444.0, + "step": 441 + }, + { + "epoch": 0.08207985143918292, + "grad_norm": 2.1331562995910645, + "learning_rate": 2.728960396039604e-07, + "loss": 0.5305, + "mean_token_accuracy": 0.8355808258056641, + "num_tokens": 15464895.0, + "step": 442 + }, + { + "epoch": 0.08226555246053853, + "grad_norm": 2.223982572555542, + "learning_rate": 2.7351485148514854e-07, + "loss": 0.5909, + "mean_token_accuracy": 0.8181522488594055, + "num_tokens": 15491587.0, + "step": 443 + }, + { + "epoch": 0.08245125348189415, + "grad_norm": 1.9911870956420898, + "learning_rate": 2.7413366336633665e-07, + "loss": 0.5263, + "mean_token_accuracy": 0.835385262966156, + "num_tokens": 15526093.0, + "step": 444 + }, + { + "epoch": 0.08263695450324977, + "grad_norm": 1.8870232105255127, + "learning_rate": 2.7475247524752476e-07, + "loss": 0.4896, + "mean_token_accuracy": 0.8459202647209167, + "num_tokens": 15561119.0, + "step": 445 + }, + { + "epoch": 0.08282265552460538, + "grad_norm": 1.8596055507659912, + "learning_rate": 2.7537128712871287e-07, + "loss": 0.5094, + "mean_token_accuracy": 0.8393924236297607, + "num_tokens": 15595906.0, + "step": 446 + }, + { + "epoch": 0.083008356545961, + "grad_norm": 1.8458961248397827, + "learning_rate": 2.75990099009901e-07, + "loss": 0.5341, + "mean_token_accuracy": 0.8325381875038147, + "num_tokens": 15634899.0, + "step": 447 + }, + { + "epoch": 0.08319405756731663, + "grad_norm": 1.8183977603912354, + "learning_rate": 2.7660891089108915e-07, + "loss": 0.5269, + "mean_token_accuracy": 0.8353437185287476, + "num_tokens": 15675605.0, + "step": 448 + }, + { + "epoch": 0.08337975858867223, + "grad_norm": 2.116732597351074, + "learning_rate": 2.7722772277227726e-07, + "loss": 0.472, + "mean_token_accuracy": 0.8499226570129395, + "num_tokens": 15712449.0, + "step": 449 + }, + { + "epoch": 0.08356545961002786, + "grad_norm": 1.7960814237594604, + "learning_rate": 2.7784653465346537e-07, + "loss": 0.553, + "mean_token_accuracy": 0.8276731371879578, + "num_tokens": 15753227.0, + "step": 450 + }, + { + "epoch": 0.08375116063138348, + "grad_norm": 1.884353756904602, + "learning_rate": 2.784653465346535e-07, + "loss": 0.5423, + "mean_token_accuracy": 0.8317908644676208, + "num_tokens": 15786026.0, + "step": 451 + }, + { + "epoch": 0.08393686165273909, + "grad_norm": 1.7354466915130615, + "learning_rate": 2.790841584158416e-07, + "loss": 0.5203, + "mean_token_accuracy": 0.8375537395477295, + "num_tokens": 15823182.0, + "step": 452 + }, + { + "epoch": 0.08412256267409471, + "grad_norm": 1.8454352617263794, + "learning_rate": 2.797029702970297e-07, + "loss": 0.5508, + "mean_token_accuracy": 0.8294247388839722, + "num_tokens": 15860194.0, + "step": 453 + }, + { + "epoch": 0.08430826369545033, + "grad_norm": 1.6487654447555542, + "learning_rate": 2.803217821782178e-07, + "loss": 0.5275, + "mean_token_accuracy": 0.836314857006073, + "num_tokens": 15899844.0, + "step": 454 + }, + { + "epoch": 0.08449396471680594, + "grad_norm": 2.146714448928833, + "learning_rate": 2.80940594059406e-07, + "loss": 0.5246, + "mean_token_accuracy": 0.834239661693573, + "num_tokens": 15927132.0, + "step": 455 + }, + { + "epoch": 0.08467966573816156, + "grad_norm": 1.5696858167648315, + "learning_rate": 2.8155940594059404e-07, + "loss": 0.485, + "mean_token_accuracy": 0.8452973365783691, + "num_tokens": 15967033.0, + "step": 456 + }, + { + "epoch": 0.08486536675951718, + "grad_norm": 1.6654512882232666, + "learning_rate": 2.8217821782178215e-07, + "loss": 0.5652, + "mean_token_accuracy": 0.8217026591300964, + "num_tokens": 16007936.0, + "step": 457 + }, + { + "epoch": 0.08505106778087279, + "grad_norm": 2.0108225345611572, + "learning_rate": 2.8279702970297026e-07, + "loss": 0.5623, + "mean_token_accuracy": 0.8293030858039856, + "num_tokens": 16041333.0, + "step": 458 + }, + { + "epoch": 0.08523676880222841, + "grad_norm": 1.8733339309692383, + "learning_rate": 2.8341584158415837e-07, + "loss": 0.5729, + "mean_token_accuracy": 0.8217974901199341, + "num_tokens": 16080508.0, + "step": 459 + }, + { + "epoch": 0.08542246982358404, + "grad_norm": 2.1130151748657227, + "learning_rate": 2.840346534653465e-07, + "loss": 0.5738, + "mean_token_accuracy": 0.8194166421890259, + "num_tokens": 16108706.0, + "step": 460 + }, + { + "epoch": 0.08560817084493964, + "grad_norm": 1.697487711906433, + "learning_rate": 2.846534653465346e-07, + "loss": 0.4959, + "mean_token_accuracy": 0.8398058414459229, + "num_tokens": 16143154.0, + "step": 461 + }, + { + "epoch": 0.08579387186629527, + "grad_norm": 1.9471834897994995, + "learning_rate": 2.8527227722772276e-07, + "loss": 0.5607, + "mean_token_accuracy": 0.825206995010376, + "num_tokens": 16179946.0, + "step": 462 + }, + { + "epoch": 0.08597957288765089, + "grad_norm": 1.9882665872573853, + "learning_rate": 2.8589108910891087e-07, + "loss": 0.4867, + "mean_token_accuracy": 0.8441665172576904, + "num_tokens": 16213324.0, + "step": 463 + }, + { + "epoch": 0.0861652739090065, + "grad_norm": 1.780660629272461, + "learning_rate": 2.86509900990099e-07, + "loss": 0.5405, + "mean_token_accuracy": 0.8322374820709229, + "num_tokens": 16251300.0, + "step": 464 + }, + { + "epoch": 0.08635097493036212, + "grad_norm": 1.8175368309020996, + "learning_rate": 2.871287128712871e-07, + "loss": 0.5193, + "mean_token_accuracy": 0.830790638923645, + "num_tokens": 16287298.0, + "step": 465 + }, + { + "epoch": 0.08653667595171774, + "grad_norm": 2.163339376449585, + "learning_rate": 2.877475247524752e-07, + "loss": 0.5451, + "mean_token_accuracy": 0.8253428936004639, + "num_tokens": 16319780.0, + "step": 466 + }, + { + "epoch": 0.08672237697307335, + "grad_norm": 1.7708479166030884, + "learning_rate": 2.883663366336633e-07, + "loss": 0.5067, + "mean_token_accuracy": 0.8394830822944641, + "num_tokens": 16354976.0, + "step": 467 + }, + { + "epoch": 0.08690807799442897, + "grad_norm": 1.6987744569778442, + "learning_rate": 2.889851485148514e-07, + "loss": 0.5126, + "mean_token_accuracy": 0.8383361101150513, + "num_tokens": 16394228.0, + "step": 468 + }, + { + "epoch": 0.08709377901578459, + "grad_norm": 1.9720706939697266, + "learning_rate": 2.896039603960396e-07, + "loss": 0.5159, + "mean_token_accuracy": 0.8335655927658081, + "num_tokens": 16426690.0, + "step": 469 + }, + { + "epoch": 0.0872794800371402, + "grad_norm": 1.8052607774734497, + "learning_rate": 2.902227722772277e-07, + "loss": 0.5371, + "mean_token_accuracy": 0.8292027115821838, + "num_tokens": 16459037.0, + "step": 470 + }, + { + "epoch": 0.08746518105849582, + "grad_norm": 1.8168179988861084, + "learning_rate": 2.908415841584158e-07, + "loss": 0.5759, + "mean_token_accuracy": 0.8153215050697327, + "num_tokens": 16497587.0, + "step": 471 + }, + { + "epoch": 0.08765088207985144, + "grad_norm": 1.6396875381469727, + "learning_rate": 2.914603960396039e-07, + "loss": 0.5052, + "mean_token_accuracy": 0.8404629230499268, + "num_tokens": 16535932.0, + "step": 472 + }, + { + "epoch": 0.08783658310120705, + "grad_norm": 2.0372257232666016, + "learning_rate": 2.9207920792079203e-07, + "loss": 0.5589, + "mean_token_accuracy": 0.8296735286712646, + "num_tokens": 16565615.0, + "step": 473 + }, + { + "epoch": 0.08802228412256267, + "grad_norm": 1.818084955215454, + "learning_rate": 2.9269801980198015e-07, + "loss": 0.4218, + "mean_token_accuracy": 0.8646214008331299, + "num_tokens": 16597731.0, + "step": 474 + }, + { + "epoch": 0.0882079851439183, + "grad_norm": 1.637273907661438, + "learning_rate": 2.933168316831683e-07, + "loss": 0.44, + "mean_token_accuracy": 0.8568915128707886, + "num_tokens": 16630603.0, + "step": 475 + }, + { + "epoch": 0.0883936861652739, + "grad_norm": 1.6685022115707397, + "learning_rate": 2.939356435643564e-07, + "loss": 0.4946, + "mean_token_accuracy": 0.843574583530426, + "num_tokens": 16664938.0, + "step": 476 + }, + { + "epoch": 0.08857938718662953, + "grad_norm": 1.7730249166488647, + "learning_rate": 2.9455445544554453e-07, + "loss": 0.583, + "mean_token_accuracy": 0.8183162808418274, + "num_tokens": 16702365.0, + "step": 477 + }, + { + "epoch": 0.08876508820798515, + "grad_norm": 1.8550258874893188, + "learning_rate": 2.9517326732673264e-07, + "loss": 0.5732, + "mean_token_accuracy": 0.8193913698196411, + "num_tokens": 16737847.0, + "step": 478 + }, + { + "epoch": 0.08895078922934076, + "grad_norm": 1.8552006483078003, + "learning_rate": 2.9579207920792076e-07, + "loss": 0.5387, + "mean_token_accuracy": 0.8313610553741455, + "num_tokens": 16767336.0, + "step": 479 + }, + { + "epoch": 0.08913649025069638, + "grad_norm": 1.8807014226913452, + "learning_rate": 2.9641089108910887e-07, + "loss": 0.5513, + "mean_token_accuracy": 0.8242890238761902, + "num_tokens": 16798911.0, + "step": 480 + }, + { + "epoch": 0.089322191272052, + "grad_norm": 2.0418269634246826, + "learning_rate": 2.97029702970297e-07, + "loss": 0.519, + "mean_token_accuracy": 0.8355849981307983, + "num_tokens": 16834139.0, + "step": 481 + }, + { + "epoch": 0.08950789229340761, + "grad_norm": 2.008049726486206, + "learning_rate": 2.9764851485148514e-07, + "loss": 0.4598, + "mean_token_accuracy": 0.849555492401123, + "num_tokens": 16864060.0, + "step": 482 + }, + { + "epoch": 0.08969359331476323, + "grad_norm": 1.7697172164916992, + "learning_rate": 2.9826732673267325e-07, + "loss": 0.5427, + "mean_token_accuracy": 0.8318970799446106, + "num_tokens": 16905216.0, + "step": 483 + }, + { + "epoch": 0.08987929433611885, + "grad_norm": 1.9187021255493164, + "learning_rate": 2.9888613861386136e-07, + "loss": 0.538, + "mean_token_accuracy": 0.8305767774581909, + "num_tokens": 16938023.0, + "step": 484 + }, + { + "epoch": 0.09006499535747446, + "grad_norm": 1.7977606058120728, + "learning_rate": 2.995049504950495e-07, + "loss": 0.5219, + "mean_token_accuracy": 0.8338809609413147, + "num_tokens": 16973218.0, + "step": 485 + }, + { + "epoch": 0.09025069637883008, + "grad_norm": 1.8227941989898682, + "learning_rate": 3.001237623762376e-07, + "loss": 0.5099, + "mean_token_accuracy": 0.8395107388496399, + "num_tokens": 17007367.0, + "step": 486 + }, + { + "epoch": 0.0904363974001857, + "grad_norm": 1.692259669303894, + "learning_rate": 3.007425742574257e-07, + "loss": 0.4965, + "mean_token_accuracy": 0.8419303297996521, + "num_tokens": 17045020.0, + "step": 487 + }, + { + "epoch": 0.09062209842154131, + "grad_norm": 2.020602226257324, + "learning_rate": 3.0136138613861386e-07, + "loss": 0.5868, + "mean_token_accuracy": 0.8154400587081909, + "num_tokens": 17075835.0, + "step": 488 + }, + { + "epoch": 0.09080779944289694, + "grad_norm": 1.8262180089950562, + "learning_rate": 3.01980198019802e-07, + "loss": 0.5287, + "mean_token_accuracy": 0.8315303921699524, + "num_tokens": 17108723.0, + "step": 489 + }, + { + "epoch": 0.09099350046425256, + "grad_norm": 1.8101849555969238, + "learning_rate": 3.025990099009901e-07, + "loss": 0.5163, + "mean_token_accuracy": 0.8389000296592712, + "num_tokens": 17141122.0, + "step": 490 + }, + { + "epoch": 0.09117920148560817, + "grad_norm": 1.9198344945907593, + "learning_rate": 3.032178217821782e-07, + "loss": 0.5236, + "mean_token_accuracy": 0.839504599571228, + "num_tokens": 17175350.0, + "step": 491 + }, + { + "epoch": 0.09136490250696379, + "grad_norm": 1.616322636604309, + "learning_rate": 3.038366336633663e-07, + "loss": 0.4773, + "mean_token_accuracy": 0.8493942022323608, + "num_tokens": 17211673.0, + "step": 492 + }, + { + "epoch": 0.09155060352831941, + "grad_norm": 1.9558771848678589, + "learning_rate": 3.044554455445544e-07, + "loss": 0.4818, + "mean_token_accuracy": 0.8423745632171631, + "num_tokens": 17239195.0, + "step": 493 + }, + { + "epoch": 0.09173630454967502, + "grad_norm": 1.8722807168960571, + "learning_rate": 3.0507425742574253e-07, + "loss": 0.5222, + "mean_token_accuracy": 0.8266782760620117, + "num_tokens": 17267550.0, + "step": 494 + }, + { + "epoch": 0.09192200557103064, + "grad_norm": 1.9182500839233398, + "learning_rate": 3.056930693069307e-07, + "loss": 0.6237, + "mean_token_accuracy": 0.8114364147186279, + "num_tokens": 17299233.0, + "step": 495 + }, + { + "epoch": 0.09210770659238626, + "grad_norm": 1.8739898204803467, + "learning_rate": 3.063118811881188e-07, + "loss": 0.5529, + "mean_token_accuracy": 0.820479154586792, + "num_tokens": 17334319.0, + "step": 496 + }, + { + "epoch": 0.09229340761374187, + "grad_norm": 1.6838759183883667, + "learning_rate": 3.069306930693069e-07, + "loss": 0.5635, + "mean_token_accuracy": 0.8221983909606934, + "num_tokens": 17374571.0, + "step": 497 + }, + { + "epoch": 0.09247910863509749, + "grad_norm": 1.9414350986480713, + "learning_rate": 3.0754950495049503e-07, + "loss": 0.6018, + "mean_token_accuracy": 0.8185757994651794, + "num_tokens": 17405216.0, + "step": 498 + }, + { + "epoch": 0.09266480965645311, + "grad_norm": 1.8391480445861816, + "learning_rate": 3.0816831683168314e-07, + "loss": 0.5069, + "mean_token_accuracy": 0.8420111536979675, + "num_tokens": 17436512.0, + "step": 499 + }, + { + "epoch": 0.09285051067780872, + "grad_norm": 1.8713886737823486, + "learning_rate": 3.0878712871287125e-07, + "loss": 0.5176, + "mean_token_accuracy": 0.833089292049408, + "num_tokens": 17467171.0, + "step": 500 + }, + { + "epoch": 0.09303621169916435, + "grad_norm": 1.9191545248031616, + "learning_rate": 3.0940594059405936e-07, + "loss": 0.4975, + "mean_token_accuracy": 0.8428254127502441, + "num_tokens": 17496714.0, + "step": 501 + }, + { + "epoch": 0.09322191272051997, + "grad_norm": 1.7737960815429688, + "learning_rate": 3.1002475247524753e-07, + "loss": 0.4925, + "mean_token_accuracy": 0.8449727892875671, + "num_tokens": 17531825.0, + "step": 502 + }, + { + "epoch": 0.09340761374187558, + "grad_norm": 1.8299193382263184, + "learning_rate": 3.1064356435643564e-07, + "loss": 0.5195, + "mean_token_accuracy": 0.837693452835083, + "num_tokens": 17568515.0, + "step": 503 + }, + { + "epoch": 0.0935933147632312, + "grad_norm": 1.8019686937332153, + "learning_rate": 3.1126237623762375e-07, + "loss": 0.5352, + "mean_token_accuracy": 0.829170286655426, + "num_tokens": 17605024.0, + "step": 504 + }, + { + "epoch": 0.09377901578458682, + "grad_norm": 1.733510136604309, + "learning_rate": 3.1188118811881186e-07, + "loss": 0.5448, + "mean_token_accuracy": 0.8276150226593018, + "num_tokens": 17641314.0, + "step": 505 + }, + { + "epoch": 0.09396471680594243, + "grad_norm": 1.6645065546035767, + "learning_rate": 3.1249999999999997e-07, + "loss": 0.5202, + "mean_token_accuracy": 0.8365569710731506, + "num_tokens": 17679636.0, + "step": 506 + }, + { + "epoch": 0.09415041782729805, + "grad_norm": 1.640458583831787, + "learning_rate": 3.131188118811881e-07, + "loss": 0.5116, + "mean_token_accuracy": 0.837882399559021, + "num_tokens": 17716694.0, + "step": 507 + }, + { + "epoch": 0.09433611884865367, + "grad_norm": 1.894782543182373, + "learning_rate": 3.1373762376237625e-07, + "loss": 0.5591, + "mean_token_accuracy": 0.8215577602386475, + "num_tokens": 17749263.0, + "step": 508 + }, + { + "epoch": 0.09452181987000928, + "grad_norm": 1.6590005159378052, + "learning_rate": 3.1435643564356436e-07, + "loss": 0.468, + "mean_token_accuracy": 0.8514742255210876, + "num_tokens": 17786672.0, + "step": 509 + }, + { + "epoch": 0.0947075208913649, + "grad_norm": 1.643105149269104, + "learning_rate": 3.1497524752475247e-07, + "loss": 0.5088, + "mean_token_accuracy": 0.8366231918334961, + "num_tokens": 17824328.0, + "step": 510 + }, + { + "epoch": 0.09489322191272052, + "grad_norm": 1.6786028146743774, + "learning_rate": 3.155940594059406e-07, + "loss": 0.4208, + "mean_token_accuracy": 0.8635756969451904, + "num_tokens": 17859890.0, + "step": 511 + }, + { + "epoch": 0.09507892293407613, + "grad_norm": 2.029130220413208, + "learning_rate": 3.162128712871287e-07, + "loss": 0.54, + "mean_token_accuracy": 0.8307360410690308, + "num_tokens": 17889614.0, + "step": 512 + }, + { + "epoch": 0.09526462395543175, + "grad_norm": 1.8802984952926636, + "learning_rate": 3.168316831683168e-07, + "loss": 0.4508, + "mean_token_accuracy": 0.855850338935852, + "num_tokens": 17917524.0, + "step": 513 + }, + { + "epoch": 0.09545032497678738, + "grad_norm": 1.5687652826309204, + "learning_rate": 3.174504950495049e-07, + "loss": 0.5319, + "mean_token_accuracy": 0.8344874382019043, + "num_tokens": 17956832.0, + "step": 514 + }, + { + "epoch": 0.09563602599814298, + "grad_norm": 1.6989080905914307, + "learning_rate": 3.180693069306931e-07, + "loss": 0.5087, + "mean_token_accuracy": 0.831308126449585, + "num_tokens": 17989928.0, + "step": 515 + }, + { + "epoch": 0.0958217270194986, + "grad_norm": 2.2267777919769287, + "learning_rate": 3.186881188118812e-07, + "loss": 0.523, + "mean_token_accuracy": 0.8391615748405457, + "num_tokens": 18015638.0, + "step": 516 + }, + { + "epoch": 0.09600742804085423, + "grad_norm": 2.010833501815796, + "learning_rate": 3.193069306930693e-07, + "loss": 0.5467, + "mean_token_accuracy": 0.8278308510780334, + "num_tokens": 18043550.0, + "step": 517 + }, + { + "epoch": 0.09619312906220984, + "grad_norm": 1.7127248048782349, + "learning_rate": 3.199257425742574e-07, + "loss": 0.4992, + "mean_token_accuracy": 0.8422673940658569, + "num_tokens": 18081343.0, + "step": 518 + }, + { + "epoch": 0.09637883008356546, + "grad_norm": 1.6851177215576172, + "learning_rate": 3.205445544554455e-07, + "loss": 0.506, + "mean_token_accuracy": 0.8377254009246826, + "num_tokens": 18120756.0, + "step": 519 + }, + { + "epoch": 0.09656453110492108, + "grad_norm": 1.8380271196365356, + "learning_rate": 3.2116336633663364e-07, + "loss": 0.5328, + "mean_token_accuracy": 0.8314522504806519, + "num_tokens": 18152832.0, + "step": 520 + }, + { + "epoch": 0.09675023212627669, + "grad_norm": 1.6565983295440674, + "learning_rate": 3.217821782178218e-07, + "loss": 0.5931, + "mean_token_accuracy": 0.8182106614112854, + "num_tokens": 18192453.0, + "step": 521 + }, + { + "epoch": 0.09693593314763231, + "grad_norm": 1.936331868171692, + "learning_rate": 3.224009900990099e-07, + "loss": 0.5215, + "mean_token_accuracy": 0.8341084122657776, + "num_tokens": 18222887.0, + "step": 522 + }, + { + "epoch": 0.09712163416898793, + "grad_norm": 1.6694364547729492, + "learning_rate": 3.23019801980198e-07, + "loss": 0.5218, + "mean_token_accuracy": 0.8424298167228699, + "num_tokens": 18260109.0, + "step": 523 + }, + { + "epoch": 0.09730733519034354, + "grad_norm": 1.7407671213150024, + "learning_rate": 3.2363861386138614e-07, + "loss": 0.5099, + "mean_token_accuracy": 0.8362129330635071, + "num_tokens": 18297426.0, + "step": 524 + }, + { + "epoch": 0.09749303621169916, + "grad_norm": 1.7095061540603638, + "learning_rate": 3.2425742574257425e-07, + "loss": 0.541, + "mean_token_accuracy": 0.8279873728752136, + "num_tokens": 18334332.0, + "step": 525 + }, + { + "epoch": 0.09767873723305479, + "grad_norm": 1.9492353200912476, + "learning_rate": 3.2487623762376236e-07, + "loss": 0.6307, + "mean_token_accuracy": 0.8079874515533447, + "num_tokens": 18372211.0, + "step": 526 + }, + { + "epoch": 0.0978644382544104, + "grad_norm": 1.7356730699539185, + "learning_rate": 3.2549504950495047e-07, + "loss": 0.5276, + "mean_token_accuracy": 0.831092894077301, + "num_tokens": 18408410.0, + "step": 527 + }, + { + "epoch": 0.09805013927576602, + "grad_norm": 1.7487376928329468, + "learning_rate": 3.2611386138613863e-07, + "loss": 0.5745, + "mean_token_accuracy": 0.8186458349227905, + "num_tokens": 18445538.0, + "step": 528 + }, + { + "epoch": 0.09823584029712164, + "grad_norm": 1.7247488498687744, + "learning_rate": 3.2673267326732674e-07, + "loss": 0.4846, + "mean_token_accuracy": 0.8401923179626465, + "num_tokens": 18477124.0, + "step": 529 + }, + { + "epoch": 0.09842154131847725, + "grad_norm": 1.7626416683197021, + "learning_rate": 3.2735148514851486e-07, + "loss": 0.5049, + "mean_token_accuracy": 0.837263822555542, + "num_tokens": 18517262.0, + "step": 530 + }, + { + "epoch": 0.09860724233983287, + "grad_norm": 1.8007792234420776, + "learning_rate": 3.2797029702970297e-07, + "loss": 0.4957, + "mean_token_accuracy": 0.8406906723976135, + "num_tokens": 18549428.0, + "step": 531 + }, + { + "epoch": 0.09879294336118849, + "grad_norm": 1.6855818033218384, + "learning_rate": 3.285891089108911e-07, + "loss": 0.6086, + "mean_token_accuracy": 0.8095021843910217, + "num_tokens": 18588167.0, + "step": 532 + }, + { + "epoch": 0.0989786443825441, + "grad_norm": 1.8815070390701294, + "learning_rate": 3.292079207920792e-07, + "loss": 0.5625, + "mean_token_accuracy": 0.825498640537262, + "num_tokens": 18620538.0, + "step": 533 + }, + { + "epoch": 0.09916434540389972, + "grad_norm": 1.7140294313430786, + "learning_rate": 3.298267326732673e-07, + "loss": 0.5227, + "mean_token_accuracy": 0.8372923135757446, + "num_tokens": 18656190.0, + "step": 534 + }, + { + "epoch": 0.09935004642525534, + "grad_norm": 1.8969377279281616, + "learning_rate": 3.3044554455445547e-07, + "loss": 0.5012, + "mean_token_accuracy": 0.8494536876678467, + "num_tokens": 18684564.0, + "step": 535 + }, + { + "epoch": 0.09953574744661095, + "grad_norm": 1.6831821203231812, + "learning_rate": 3.310643564356436e-07, + "loss": 0.518, + "mean_token_accuracy": 0.83775794506073, + "num_tokens": 18721617.0, + "step": 536 + }, + { + "epoch": 0.09972144846796657, + "grad_norm": 1.8032665252685547, + "learning_rate": 3.316831683168317e-07, + "loss": 0.509, + "mean_token_accuracy": 0.8361901044845581, + "num_tokens": 18755088.0, + "step": 537 + }, + { + "epoch": 0.0999071494893222, + "grad_norm": 1.949723720550537, + "learning_rate": 3.323019801980198e-07, + "loss": 0.5911, + "mean_token_accuracy": 0.8204377293586731, + "num_tokens": 18786151.0, + "step": 538 + }, + { + "epoch": 0.1000928505106778, + "grad_norm": 1.9157782793045044, + "learning_rate": 3.329207920792079e-07, + "loss": 0.5497, + "mean_token_accuracy": 0.828463077545166, + "num_tokens": 18818610.0, + "step": 539 + }, + { + "epoch": 0.10027855153203342, + "grad_norm": 1.8985443115234375, + "learning_rate": 3.33539603960396e-07, + "loss": 0.4989, + "mean_token_accuracy": 0.8392802476882935, + "num_tokens": 18847561.0, + "step": 540 + }, + { + "epoch": 0.10046425255338905, + "grad_norm": 1.6705195903778076, + "learning_rate": 3.341584158415842e-07, + "loss": 0.5017, + "mean_token_accuracy": 0.8408725261688232, + "num_tokens": 18885478.0, + "step": 541 + }, + { + "epoch": 0.10064995357474465, + "grad_norm": 1.8597270250320435, + "learning_rate": 3.347772277227723e-07, + "loss": 0.5279, + "mean_token_accuracy": 0.8314214944839478, + "num_tokens": 18917796.0, + "step": 542 + }, + { + "epoch": 0.10083565459610028, + "grad_norm": 1.9999680519104004, + "learning_rate": 3.353960396039604e-07, + "loss": 0.5583, + "mean_token_accuracy": 0.8258833885192871, + "num_tokens": 18945765.0, + "step": 543 + }, + { + "epoch": 0.1010213556174559, + "grad_norm": 1.7409385442733765, + "learning_rate": 3.360148514851485e-07, + "loss": 0.5086, + "mean_token_accuracy": 0.833605945110321, + "num_tokens": 18981284.0, + "step": 544 + }, + { + "epoch": 0.10120705663881151, + "grad_norm": 1.5525431632995605, + "learning_rate": 3.3663366336633663e-07, + "loss": 0.4647, + "mean_token_accuracy": 0.8478584289550781, + "num_tokens": 19020031.0, + "step": 545 + }, + { + "epoch": 0.10139275766016713, + "grad_norm": 1.6594605445861816, + "learning_rate": 3.3725247524752474e-07, + "loss": 0.5332, + "mean_token_accuracy": 0.8347307443618774, + "num_tokens": 19054886.0, + "step": 546 + }, + { + "epoch": 0.10157845868152275, + "grad_norm": 1.6461596488952637, + "learning_rate": 3.3787128712871285e-07, + "loss": 0.4845, + "mean_token_accuracy": 0.8493509292602539, + "num_tokens": 19089737.0, + "step": 547 + }, + { + "epoch": 0.10176415970287836, + "grad_norm": 1.6823958158493042, + "learning_rate": 3.38490099009901e-07, + "loss": 0.5305, + "mean_token_accuracy": 0.8338465690612793, + "num_tokens": 19129382.0, + "step": 548 + }, + { + "epoch": 0.10194986072423398, + "grad_norm": 1.5155205726623535, + "learning_rate": 3.3910891089108913e-07, + "loss": 0.4359, + "mean_token_accuracy": 0.8583313226699829, + "num_tokens": 19167261.0, + "step": 549 + }, + { + "epoch": 0.1021355617455896, + "grad_norm": 1.756888747215271, + "learning_rate": 3.3972772277227724e-07, + "loss": 0.4603, + "mean_token_accuracy": 0.8556375503540039, + "num_tokens": 19197885.0, + "step": 550 + }, + { + "epoch": 0.10232126276694521, + "grad_norm": 1.6422070264816284, + "learning_rate": 3.4034653465346535e-07, + "loss": 0.4887, + "mean_token_accuracy": 0.8396164178848267, + "num_tokens": 19233906.0, + "step": 551 + }, + { + "epoch": 0.10250696378830083, + "grad_norm": 1.6573132276535034, + "learning_rate": 3.4096534653465346e-07, + "loss": 0.5217, + "mean_token_accuracy": 0.835265576839447, + "num_tokens": 19270674.0, + "step": 552 + }, + { + "epoch": 0.10269266480965646, + "grad_norm": 1.6454635858535767, + "learning_rate": 3.415841584158416e-07, + "loss": 0.5033, + "mean_token_accuracy": 0.8357665538787842, + "num_tokens": 19307004.0, + "step": 553 + }, + { + "epoch": 0.10287836583101206, + "grad_norm": 1.7783631086349487, + "learning_rate": 3.4220297029702974e-07, + "loss": 0.5492, + "mean_token_accuracy": 0.8248679637908936, + "num_tokens": 19340916.0, + "step": 554 + }, + { + "epoch": 0.10306406685236769, + "grad_norm": 1.7299854755401611, + "learning_rate": 3.4282178217821785e-07, + "loss": 0.5011, + "mean_token_accuracy": 0.8368623852729797, + "num_tokens": 19375777.0, + "step": 555 + }, + { + "epoch": 0.10324976787372331, + "grad_norm": 1.6701256036758423, + "learning_rate": 3.4344059405940596e-07, + "loss": 0.5297, + "mean_token_accuracy": 0.8325037360191345, + "num_tokens": 19410829.0, + "step": 556 + }, + { + "epoch": 0.10343546889507892, + "grad_norm": 1.488848328590393, + "learning_rate": 3.44059405940594e-07, + "loss": 0.451, + "mean_token_accuracy": 0.8552123308181763, + "num_tokens": 19452579.0, + "step": 557 + }, + { + "epoch": 0.10362116991643454, + "grad_norm": 1.6362534761428833, + "learning_rate": 3.4467821782178213e-07, + "loss": 0.4931, + "mean_token_accuracy": 0.8436198830604553, + "num_tokens": 19488490.0, + "step": 558 + }, + { + "epoch": 0.10380687093779016, + "grad_norm": 1.8350750207901, + "learning_rate": 3.4529702970297024e-07, + "loss": 0.5068, + "mean_token_accuracy": 0.8392400741577148, + "num_tokens": 19517080.0, + "step": 559 + }, + { + "epoch": 0.10399257195914577, + "grad_norm": 1.735718011856079, + "learning_rate": 3.4591584158415835e-07, + "loss": 0.5267, + "mean_token_accuracy": 0.8344102501869202, + "num_tokens": 19549743.0, + "step": 560 + }, + { + "epoch": 0.10417827298050139, + "grad_norm": 1.8318581581115723, + "learning_rate": 3.465346534653465e-07, + "loss": 0.5014, + "mean_token_accuracy": 0.8424055576324463, + "num_tokens": 19579444.0, + "step": 561 + }, + { + "epoch": 0.10436397400185701, + "grad_norm": 1.6982282400131226, + "learning_rate": 3.4715346534653463e-07, + "loss": 0.5116, + "mean_token_accuracy": 0.834983229637146, + "num_tokens": 19618066.0, + "step": 562 + }, + { + "epoch": 0.10454967502321262, + "grad_norm": 1.5492479801177979, + "learning_rate": 3.4777227722772274e-07, + "loss": 0.4663, + "mean_token_accuracy": 0.8513998985290527, + "num_tokens": 19657969.0, + "step": 563 + }, + { + "epoch": 0.10473537604456824, + "grad_norm": 1.8261916637420654, + "learning_rate": 3.4839108910891085e-07, + "loss": 0.5099, + "mean_token_accuracy": 0.8377282619476318, + "num_tokens": 19692126.0, + "step": 564 + }, + { + "epoch": 0.10492107706592387, + "grad_norm": 1.69599187374115, + "learning_rate": 3.4900990099009896e-07, + "loss": 0.4902, + "mean_token_accuracy": 0.8423717021942139, + "num_tokens": 19729334.0, + "step": 565 + }, + { + "epoch": 0.10510677808727947, + "grad_norm": 1.6298195123672485, + "learning_rate": 3.496287128712871e-07, + "loss": 0.5429, + "mean_token_accuracy": 0.8272436261177063, + "num_tokens": 19768137.0, + "step": 566 + }, + { + "epoch": 0.1052924791086351, + "grad_norm": 1.8814820051193237, + "learning_rate": 3.502475247524752e-07, + "loss": 0.548, + "mean_token_accuracy": 0.827820897102356, + "num_tokens": 19797397.0, + "step": 567 + }, + { + "epoch": 0.10547818012999072, + "grad_norm": 1.9289182424545288, + "learning_rate": 3.5086633663366335e-07, + "loss": 0.5039, + "mean_token_accuracy": 0.8410490155220032, + "num_tokens": 19827800.0, + "step": 568 + }, + { + "epoch": 0.10566388115134633, + "grad_norm": 1.7111183404922485, + "learning_rate": 3.5148514851485146e-07, + "loss": 0.5667, + "mean_token_accuracy": 0.8238813877105713, + "num_tokens": 19864462.0, + "step": 569 + }, + { + "epoch": 0.10584958217270195, + "grad_norm": 1.6203938722610474, + "learning_rate": 3.5210396039603957e-07, + "loss": 0.5026, + "mean_token_accuracy": 0.8411579728126526, + "num_tokens": 19902065.0, + "step": 570 + }, + { + "epoch": 0.10603528319405757, + "grad_norm": 1.608491063117981, + "learning_rate": 3.527227722772277e-07, + "loss": 0.4826, + "mean_token_accuracy": 0.8415836691856384, + "num_tokens": 19937071.0, + "step": 571 + }, + { + "epoch": 0.10622098421541319, + "grad_norm": 1.6823087930679321, + "learning_rate": 3.533415841584158e-07, + "loss": 0.5104, + "mean_token_accuracy": 0.8369140625, + "num_tokens": 19969716.0, + "step": 572 + }, + { + "epoch": 0.1064066852367688, + "grad_norm": 1.7607920169830322, + "learning_rate": 3.539603960396039e-07, + "loss": 0.4876, + "mean_token_accuracy": 0.8450204730033875, + "num_tokens": 20003106.0, + "step": 573 + }, + { + "epoch": 0.10659238625812442, + "grad_norm": 1.7955044507980347, + "learning_rate": 3.5457920792079207e-07, + "loss": 0.43, + "mean_token_accuracy": 0.8621813058853149, + "num_tokens": 20030995.0, + "step": 574 + }, + { + "epoch": 0.10677808727948004, + "grad_norm": 1.8761379718780518, + "learning_rate": 3.551980198019802e-07, + "loss": 0.5839, + "mean_token_accuracy": 0.8156754970550537, + "num_tokens": 20067333.0, + "step": 575 + }, + { + "epoch": 0.10696378830083565, + "grad_norm": 1.6090093851089478, + "learning_rate": 3.558168316831683e-07, + "loss": 0.5353, + "mean_token_accuracy": 0.8317489624023438, + "num_tokens": 20106668.0, + "step": 576 + }, + { + "epoch": 0.10714948932219127, + "grad_norm": 1.6968432664871216, + "learning_rate": 3.564356435643564e-07, + "loss": 0.533, + "mean_token_accuracy": 0.832283616065979, + "num_tokens": 20139270.0, + "step": 577 + }, + { + "epoch": 0.1073351903435469, + "grad_norm": 1.6415810585021973, + "learning_rate": 3.570544554455445e-07, + "loss": 0.4692, + "mean_token_accuracy": 0.8483463525772095, + "num_tokens": 20177200.0, + "step": 578 + }, + { + "epoch": 0.1075208913649025, + "grad_norm": 1.609815239906311, + "learning_rate": 3.5767326732673263e-07, + "loss": 0.5314, + "mean_token_accuracy": 0.8274171352386475, + "num_tokens": 20218102.0, + "step": 579 + }, + { + "epoch": 0.10770659238625813, + "grad_norm": 1.6884757280349731, + "learning_rate": 3.5829207920792074e-07, + "loss": 0.5405, + "mean_token_accuracy": 0.8225094676017761, + "num_tokens": 20254910.0, + "step": 580 + }, + { + "epoch": 0.10789229340761375, + "grad_norm": 1.6051652431488037, + "learning_rate": 3.589108910891089e-07, + "loss": 0.5238, + "mean_token_accuracy": 0.8334180116653442, + "num_tokens": 20290915.0, + "step": 581 + }, + { + "epoch": 0.10807799442896936, + "grad_norm": 1.7093335390090942, + "learning_rate": 3.59529702970297e-07, + "loss": 0.5272, + "mean_token_accuracy": 0.8349688053131104, + "num_tokens": 20324664.0, + "step": 582 + }, + { + "epoch": 0.10826369545032498, + "grad_norm": 1.794880986213684, + "learning_rate": 3.601485148514851e-07, + "loss": 0.484, + "mean_token_accuracy": 0.8404098749160767, + "num_tokens": 20353824.0, + "step": 583 + }, + { + "epoch": 0.1084493964716806, + "grad_norm": 1.5689314603805542, + "learning_rate": 3.6076732673267324e-07, + "loss": 0.4641, + "mean_token_accuracy": 0.8489511609077454, + "num_tokens": 20391646.0, + "step": 584 + }, + { + "epoch": 0.10863509749303621, + "grad_norm": 1.6732203960418701, + "learning_rate": 3.6138613861386135e-07, + "loss": 0.4846, + "mean_token_accuracy": 0.8435413837432861, + "num_tokens": 20425319.0, + "step": 585 + }, + { + "epoch": 0.10882079851439183, + "grad_norm": 1.5869653224945068, + "learning_rate": 3.6200495049504946e-07, + "loss": 0.4726, + "mean_token_accuracy": 0.8454816937446594, + "num_tokens": 20460516.0, + "step": 586 + }, + { + "epoch": 0.10900649953574745, + "grad_norm": 2.0843703746795654, + "learning_rate": 3.6262376237623757e-07, + "loss": 0.4961, + "mean_token_accuracy": 0.8429202437400818, + "num_tokens": 20489962.0, + "step": 587 + }, + { + "epoch": 0.10919220055710306, + "grad_norm": 1.640283465385437, + "learning_rate": 3.6324257425742574e-07, + "loss": 0.5115, + "mean_token_accuracy": 0.842458188533783, + "num_tokens": 20526254.0, + "step": 588 + }, + { + "epoch": 0.10937790157845868, + "grad_norm": 1.712234616279602, + "learning_rate": 3.6386138613861385e-07, + "loss": 0.56, + "mean_token_accuracy": 0.8180286884307861, + "num_tokens": 20563906.0, + "step": 589 + }, + { + "epoch": 0.1095636025998143, + "grad_norm": 1.6264426708221436, + "learning_rate": 3.6448019801980196e-07, + "loss": 0.4941, + "mean_token_accuracy": 0.84257972240448, + "num_tokens": 20602258.0, + "step": 590 + }, + { + "epoch": 0.10974930362116991, + "grad_norm": 1.60197114944458, + "learning_rate": 3.6509900990099007e-07, + "loss": 0.5634, + "mean_token_accuracy": 0.8260484933853149, + "num_tokens": 20642044.0, + "step": 591 + }, + { + "epoch": 0.10993500464252554, + "grad_norm": 1.5913022756576538, + "learning_rate": 3.657178217821782e-07, + "loss": 0.5163, + "mean_token_accuracy": 0.8320798873901367, + "num_tokens": 20680443.0, + "step": 592 + }, + { + "epoch": 0.11012070566388116, + "grad_norm": 1.6833897829055786, + "learning_rate": 3.663366336633663e-07, + "loss": 0.4999, + "mean_token_accuracy": 0.840125322341919, + "num_tokens": 20717217.0, + "step": 593 + }, + { + "epoch": 0.11030640668523677, + "grad_norm": 1.6796013116836548, + "learning_rate": 3.6695544554455446e-07, + "loss": 0.4724, + "mean_token_accuracy": 0.8495485782623291, + "num_tokens": 20749442.0, + "step": 594 + }, + { + "epoch": 0.11049210770659239, + "grad_norm": 1.5891016721725464, + "learning_rate": 3.6757425742574257e-07, + "loss": 0.5081, + "mean_token_accuracy": 0.8362269997596741, + "num_tokens": 20785662.0, + "step": 595 + }, + { + "epoch": 0.11067780872794801, + "grad_norm": 1.5987659692764282, + "learning_rate": 3.681930693069307e-07, + "loss": 0.5151, + "mean_token_accuracy": 0.8366656303405762, + "num_tokens": 20823315.0, + "step": 596 + }, + { + "epoch": 0.11086350974930362, + "grad_norm": 1.631098985671997, + "learning_rate": 3.688118811881188e-07, + "loss": 0.4756, + "mean_token_accuracy": 0.8479884266853333, + "num_tokens": 20860768.0, + "step": 597 + }, + { + "epoch": 0.11104921077065924, + "grad_norm": 1.7056361436843872, + "learning_rate": 3.694306930693069e-07, + "loss": 0.5712, + "mean_token_accuracy": 0.8216796517372131, + "num_tokens": 20896393.0, + "step": 598 + }, + { + "epoch": 0.11123491179201486, + "grad_norm": 1.8487664461135864, + "learning_rate": 3.70049504950495e-07, + "loss": 0.5862, + "mean_token_accuracy": 0.8193373084068298, + "num_tokens": 20928147.0, + "step": 599 + }, + { + "epoch": 0.11142061281337047, + "grad_norm": 1.8244140148162842, + "learning_rate": 3.706683168316831e-07, + "loss": 0.5531, + "mean_token_accuracy": 0.8262332081794739, + "num_tokens": 20960783.0, + "step": 600 + }, + { + "epoch": 0.11160631383472609, + "grad_norm": 1.6308859586715698, + "learning_rate": 3.712871287128713e-07, + "loss": 0.4671, + "mean_token_accuracy": 0.8541148900985718, + "num_tokens": 20993918.0, + "step": 601 + }, + { + "epoch": 0.11179201485608171, + "grad_norm": 1.6137583255767822, + "learning_rate": 3.719059405940594e-07, + "loss": 0.4893, + "mean_token_accuracy": 0.8438692688941956, + "num_tokens": 21031312.0, + "step": 602 + }, + { + "epoch": 0.11197771587743732, + "grad_norm": 1.7470475435256958, + "learning_rate": 3.725247524752475e-07, + "loss": 0.5587, + "mean_token_accuracy": 0.8240047693252563, + "num_tokens": 21064516.0, + "step": 603 + }, + { + "epoch": 0.11216341689879294, + "grad_norm": 1.95088529586792, + "learning_rate": 3.731435643564356e-07, + "loss": 0.4851, + "mean_token_accuracy": 0.8452804088592529, + "num_tokens": 21089769.0, + "step": 604 + }, + { + "epoch": 0.11234911792014857, + "grad_norm": 1.5943224430084229, + "learning_rate": 3.7376237623762373e-07, + "loss": 0.4409, + "mean_token_accuracy": 0.8579726815223694, + "num_tokens": 21125649.0, + "step": 605 + }, + { + "epoch": 0.11253481894150417, + "grad_norm": 1.8617101907730103, + "learning_rate": 3.7438118811881185e-07, + "loss": 0.4534, + "mean_token_accuracy": 0.852006733417511, + "num_tokens": 21163355.0, + "step": 606 + }, + { + "epoch": 0.1127205199628598, + "grad_norm": 1.6122791767120361, + "learning_rate": 3.75e-07, + "loss": 0.5532, + "mean_token_accuracy": 0.8252540230751038, + "num_tokens": 21202727.0, + "step": 607 + }, + { + "epoch": 0.11290622098421542, + "grad_norm": 1.6458569765090942, + "learning_rate": 3.756188118811881e-07, + "loss": 0.5404, + "mean_token_accuracy": 0.8307160139083862, + "num_tokens": 21236678.0, + "step": 608 + }, + { + "epoch": 0.11309192200557103, + "grad_norm": 1.5440871715545654, + "learning_rate": 3.7623762376237623e-07, + "loss": 0.5002, + "mean_token_accuracy": 0.8407676815986633, + "num_tokens": 21273962.0, + "step": 609 + }, + { + "epoch": 0.11327762302692665, + "grad_norm": 1.5963959693908691, + "learning_rate": 3.7685643564356434e-07, + "loss": 0.4581, + "mean_token_accuracy": 0.8490836024284363, + "num_tokens": 21307822.0, + "step": 610 + }, + { + "epoch": 0.11346332404828227, + "grad_norm": 1.625346302986145, + "learning_rate": 3.7747524752475245e-07, + "loss": 0.4894, + "mean_token_accuracy": 0.8434799313545227, + "num_tokens": 21341297.0, + "step": 611 + }, + { + "epoch": 0.11364902506963788, + "grad_norm": 1.65431809425354, + "learning_rate": 3.7809405940594057e-07, + "loss": 0.5157, + "mean_token_accuracy": 0.8372693061828613, + "num_tokens": 21372832.0, + "step": 612 + }, + { + "epoch": 0.1138347260909935, + "grad_norm": 1.7563973665237427, + "learning_rate": 3.787128712871287e-07, + "loss": 0.5344, + "mean_token_accuracy": 0.8400204181671143, + "num_tokens": 21404012.0, + "step": 613 + }, + { + "epoch": 0.11402042711234912, + "grad_norm": 1.6989690065383911, + "learning_rate": 3.7933168316831684e-07, + "loss": 0.5376, + "mean_token_accuracy": 0.8354102373123169, + "num_tokens": 21439531.0, + "step": 614 + }, + { + "epoch": 0.11420612813370473, + "grad_norm": 1.884934663772583, + "learning_rate": 3.7995049504950495e-07, + "loss": 0.4985, + "mean_token_accuracy": 0.8389898538589478, + "num_tokens": 21467192.0, + "step": 615 + }, + { + "epoch": 0.11439182915506035, + "grad_norm": 1.733310580253601, + "learning_rate": 3.8056930693069306e-07, + "loss": 0.4893, + "mean_token_accuracy": 0.8454210758209229, + "num_tokens": 21500524.0, + "step": 616 + }, + { + "epoch": 0.11457753017641598, + "grad_norm": 1.5488287210464478, + "learning_rate": 3.811881188118812e-07, + "loss": 0.4456, + "mean_token_accuracy": 0.8551764488220215, + "num_tokens": 21536877.0, + "step": 617 + }, + { + "epoch": 0.11476323119777158, + "grad_norm": 1.6329642534255981, + "learning_rate": 3.818069306930693e-07, + "loss": 0.5596, + "mean_token_accuracy": 0.8267838954925537, + "num_tokens": 21572277.0, + "step": 618 + }, + { + "epoch": 0.1149489322191272, + "grad_norm": 1.6115750074386597, + "learning_rate": 3.824257425742574e-07, + "loss": 0.504, + "mean_token_accuracy": 0.8392457365989685, + "num_tokens": 21604927.0, + "step": 619 + }, + { + "epoch": 0.11513463324048283, + "grad_norm": 1.7403031587600708, + "learning_rate": 3.830445544554455e-07, + "loss": 0.474, + "mean_token_accuracy": 0.8473827838897705, + "num_tokens": 21635909.0, + "step": 620 + }, + { + "epoch": 0.11532033426183844, + "grad_norm": 1.5058507919311523, + "learning_rate": 3.836633663366337e-07, + "loss": 0.5186, + "mean_token_accuracy": 0.8377143144607544, + "num_tokens": 21676880.0, + "step": 621 + }, + { + "epoch": 0.11550603528319406, + "grad_norm": 1.5347193479537964, + "learning_rate": 3.842821782178218e-07, + "loss": 0.533, + "mean_token_accuracy": 0.8304312229156494, + "num_tokens": 21719591.0, + "step": 622 + }, + { + "epoch": 0.11569173630454968, + "grad_norm": 1.7826671600341797, + "learning_rate": 3.849009900990099e-07, + "loss": 0.4841, + "mean_token_accuracy": 0.8423103094100952, + "num_tokens": 21750187.0, + "step": 623 + }, + { + "epoch": 0.11587743732590529, + "grad_norm": 1.946004867553711, + "learning_rate": 3.85519801980198e-07, + "loss": 0.4774, + "mean_token_accuracy": 0.8456562757492065, + "num_tokens": 21776941.0, + "step": 624 + }, + { + "epoch": 0.11606313834726091, + "grad_norm": 1.641506314277649, + "learning_rate": 3.861386138613861e-07, + "loss": 0.5285, + "mean_token_accuracy": 0.8314101696014404, + "num_tokens": 21810608.0, + "step": 625 + }, + { + "epoch": 0.11624883936861653, + "grad_norm": 1.634078025817871, + "learning_rate": 3.8675742574257423e-07, + "loss": 0.483, + "mean_token_accuracy": 0.8449009656906128, + "num_tokens": 21846953.0, + "step": 626 + }, + { + "epoch": 0.11643454038997214, + "grad_norm": 1.456686019897461, + "learning_rate": 3.873762376237624e-07, + "loss": 0.5085, + "mean_token_accuracy": 0.836392879486084, + "num_tokens": 21887429.0, + "step": 627 + }, + { + "epoch": 0.11662024141132776, + "grad_norm": 1.8227591514587402, + "learning_rate": 3.879950495049505e-07, + "loss": 0.4681, + "mean_token_accuracy": 0.8516192436218262, + "num_tokens": 21912625.0, + "step": 628 + }, + { + "epoch": 0.11680594243268339, + "grad_norm": 1.5721920728683472, + "learning_rate": 3.886138613861386e-07, + "loss": 0.4697, + "mean_token_accuracy": 0.8512491583824158, + "num_tokens": 21947965.0, + "step": 629 + }, + { + "epoch": 0.116991643454039, + "grad_norm": 1.44223153591156, + "learning_rate": 3.8923267326732673e-07, + "loss": 0.4841, + "mean_token_accuracy": 0.8469125032424927, + "num_tokens": 21989839.0, + "step": 630 + }, + { + "epoch": 0.11717734447539462, + "grad_norm": 1.7128833532333374, + "learning_rate": 3.8985148514851484e-07, + "loss": 0.5466, + "mean_token_accuracy": 0.8331624865531921, + "num_tokens": 22021593.0, + "step": 631 + }, + { + "epoch": 0.11736304549675024, + "grad_norm": 1.75908362865448, + "learning_rate": 3.9047029702970295e-07, + "loss": 0.4761, + "mean_token_accuracy": 0.8495843410491943, + "num_tokens": 22049603.0, + "step": 632 + }, + { + "epoch": 0.11754874651810585, + "grad_norm": 1.591975450515747, + "learning_rate": 3.9108910891089106e-07, + "loss": 0.5213, + "mean_token_accuracy": 0.8342543840408325, + "num_tokens": 22086422.0, + "step": 633 + }, + { + "epoch": 0.11773444753946147, + "grad_norm": 1.6564503908157349, + "learning_rate": 3.9170792079207923e-07, + "loss": 0.5462, + "mean_token_accuracy": 0.8240696787834167, + "num_tokens": 22119909.0, + "step": 634 + }, + { + "epoch": 0.11792014856081709, + "grad_norm": 1.5976883172988892, + "learning_rate": 3.9232673267326734e-07, + "loss": 0.497, + "mean_token_accuracy": 0.8392539024353027, + "num_tokens": 22153434.0, + "step": 635 + }, + { + "epoch": 0.1181058495821727, + "grad_norm": 1.4601256847381592, + "learning_rate": 3.9294554455445545e-07, + "loss": 0.4488, + "mean_token_accuracy": 0.8560516834259033, + "num_tokens": 22191893.0, + "step": 636 + }, + { + "epoch": 0.11829155060352832, + "grad_norm": 1.6239207983016968, + "learning_rate": 3.9356435643564356e-07, + "loss": 0.5099, + "mean_token_accuracy": 0.836974024772644, + "num_tokens": 22225887.0, + "step": 637 + }, + { + "epoch": 0.11847725162488394, + "grad_norm": 1.9827145338058472, + "learning_rate": 3.9418316831683167e-07, + "loss": 0.5289, + "mean_token_accuracy": 0.8343889117240906, + "num_tokens": 22260862.0, + "step": 638 + }, + { + "epoch": 0.11866295264623955, + "grad_norm": 1.591559648513794, + "learning_rate": 3.948019801980198e-07, + "loss": 0.4897, + "mean_token_accuracy": 0.8453991413116455, + "num_tokens": 22296166.0, + "step": 639 + }, + { + "epoch": 0.11884865366759517, + "grad_norm": 1.4985312223434448, + "learning_rate": 3.954207920792079e-07, + "loss": 0.4646, + "mean_token_accuracy": 0.8486769199371338, + "num_tokens": 22333262.0, + "step": 640 + }, + { + "epoch": 0.1190343546889508, + "grad_norm": 1.6475110054016113, + "learning_rate": 3.9603960396039606e-07, + "loss": 0.4828, + "mean_token_accuracy": 0.8415999412536621, + "num_tokens": 22370621.0, + "step": 641 + }, + { + "epoch": 0.1192200557103064, + "grad_norm": 1.6011030673980713, + "learning_rate": 3.9665841584158417e-07, + "loss": 0.5135, + "mean_token_accuracy": 0.839846134185791, + "num_tokens": 22408478.0, + "step": 642 + }, + { + "epoch": 0.11940575673166202, + "grad_norm": 1.5811114311218262, + "learning_rate": 3.972772277227723e-07, + "loss": 0.5611, + "mean_token_accuracy": 0.8278487324714661, + "num_tokens": 22449449.0, + "step": 643 + }, + { + "epoch": 0.11959145775301765, + "grad_norm": 1.540831446647644, + "learning_rate": 3.978960396039604e-07, + "loss": 0.4658, + "mean_token_accuracy": 0.8482701778411865, + "num_tokens": 22486577.0, + "step": 644 + }, + { + "epoch": 0.11977715877437325, + "grad_norm": 1.6631540060043335, + "learning_rate": 3.985148514851485e-07, + "loss": 0.5437, + "mean_token_accuracy": 0.8261446952819824, + "num_tokens": 22523659.0, + "step": 645 + }, + { + "epoch": 0.11996285979572888, + "grad_norm": 1.5736151933670044, + "learning_rate": 3.991336633663366e-07, + "loss": 0.4934, + "mean_token_accuracy": 0.8422889709472656, + "num_tokens": 22560276.0, + "step": 646 + }, + { + "epoch": 0.1201485608170845, + "grad_norm": 1.587531328201294, + "learning_rate": 3.997524752475248e-07, + "loss": 0.5629, + "mean_token_accuracy": 0.8202238082885742, + "num_tokens": 22597514.0, + "step": 647 + }, + { + "epoch": 0.1203342618384401, + "grad_norm": 1.8609344959259033, + "learning_rate": 4.003712871287129e-07, + "loss": 0.5156, + "mean_token_accuracy": 0.837178111076355, + "num_tokens": 22623877.0, + "step": 648 + }, + { + "epoch": 0.12051996285979573, + "grad_norm": 1.6361204385757446, + "learning_rate": 4.00990099009901e-07, + "loss": 0.4417, + "mean_token_accuracy": 0.858230471611023, + "num_tokens": 22656192.0, + "step": 649 + }, + { + "epoch": 0.12070566388115135, + "grad_norm": 1.5906189680099487, + "learning_rate": 4.016089108910891e-07, + "loss": 0.4714, + "mean_token_accuracy": 0.8520704507827759, + "num_tokens": 22697747.0, + "step": 650 + }, + { + "epoch": 0.12089136490250696, + "grad_norm": 1.4743937253952026, + "learning_rate": 4.022277227722772e-07, + "loss": 0.4722, + "mean_token_accuracy": 0.8522481322288513, + "num_tokens": 22738564.0, + "step": 651 + }, + { + "epoch": 0.12107706592386258, + "grad_norm": 1.7243012189865112, + "learning_rate": 4.0284653465346534e-07, + "loss": 0.5432, + "mean_token_accuracy": 0.8283394575119019, + "num_tokens": 22773201.0, + "step": 652 + }, + { + "epoch": 0.1212627669452182, + "grad_norm": 1.446027398109436, + "learning_rate": 4.0346534653465345e-07, + "loss": 0.5077, + "mean_token_accuracy": 0.8370691537857056, + "num_tokens": 22816893.0, + "step": 653 + }, + { + "epoch": 0.12144846796657381, + "grad_norm": 1.7090229988098145, + "learning_rate": 4.040841584158416e-07, + "loss": 0.4908, + "mean_token_accuracy": 0.8432613611221313, + "num_tokens": 22850887.0, + "step": 654 + }, + { + "epoch": 0.12163416898792943, + "grad_norm": 1.739319920539856, + "learning_rate": 4.047029702970297e-07, + "loss": 0.4892, + "mean_token_accuracy": 0.8431053161621094, + "num_tokens": 22884171.0, + "step": 655 + }, + { + "epoch": 0.12181987000928506, + "grad_norm": 1.5954601764678955, + "learning_rate": 4.0532178217821783e-07, + "loss": 0.4648, + "mean_token_accuracy": 0.849423885345459, + "num_tokens": 22920929.0, + "step": 656 + }, + { + "epoch": 0.12200557103064066, + "grad_norm": 1.6251357793807983, + "learning_rate": 4.0594059405940595e-07, + "loss": 0.485, + "mean_token_accuracy": 0.8472598791122437, + "num_tokens": 22957449.0, + "step": 657 + }, + { + "epoch": 0.12219127205199629, + "grad_norm": 1.5947940349578857, + "learning_rate": 4.06559405940594e-07, + "loss": 0.475, + "mean_token_accuracy": 0.8452656269073486, + "num_tokens": 22992345.0, + "step": 658 + }, + { + "epoch": 0.12237697307335191, + "grad_norm": 1.689450740814209, + "learning_rate": 4.071782178217821e-07, + "loss": 0.4819, + "mean_token_accuracy": 0.8460685610771179, + "num_tokens": 23027117.0, + "step": 659 + }, + { + "epoch": 0.12256267409470752, + "grad_norm": 1.6989831924438477, + "learning_rate": 4.0779702970297023e-07, + "loss": 0.5541, + "mean_token_accuracy": 0.8238661885261536, + "num_tokens": 23061790.0, + "step": 660 + }, + { + "epoch": 0.12274837511606314, + "grad_norm": 1.7471518516540527, + "learning_rate": 4.084158415841584e-07, + "loss": 0.4751, + "mean_token_accuracy": 0.8452731370925903, + "num_tokens": 23088940.0, + "step": 661 + }, + { + "epoch": 0.12293407613741876, + "grad_norm": 1.54343581199646, + "learning_rate": 4.090346534653465e-07, + "loss": 0.4938, + "mean_token_accuracy": 0.8397390246391296, + "num_tokens": 23125559.0, + "step": 662 + }, + { + "epoch": 0.12311977715877437, + "grad_norm": 1.7521055936813354, + "learning_rate": 4.096534653465346e-07, + "loss": 0.5253, + "mean_token_accuracy": 0.8309329748153687, + "num_tokens": 23160074.0, + "step": 663 + }, + { + "epoch": 0.12330547818012999, + "grad_norm": 1.7814775705337524, + "learning_rate": 4.102722772277227e-07, + "loss": 0.5243, + "mean_token_accuracy": 0.8344423770904541, + "num_tokens": 23190233.0, + "step": 664 + }, + { + "epoch": 0.12349117920148561, + "grad_norm": 1.5174603462219238, + "learning_rate": 4.1089108910891084e-07, + "loss": 0.4487, + "mean_token_accuracy": 0.8575729727745056, + "num_tokens": 23225307.0, + "step": 665 + }, + { + "epoch": 0.12367688022284122, + "grad_norm": 1.4951279163360596, + "learning_rate": 4.1150990099009895e-07, + "loss": 0.575, + "mean_token_accuracy": 0.8190130591392517, + "num_tokens": 23269651.0, + "step": 666 + }, + { + "epoch": 0.12386258124419684, + "grad_norm": 1.6431885957717896, + "learning_rate": 4.121287128712871e-07, + "loss": 0.4584, + "mean_token_accuracy": 0.8527024388313293, + "num_tokens": 23304108.0, + "step": 667 + }, + { + "epoch": 0.12404828226555246, + "grad_norm": 1.7091227769851685, + "learning_rate": 4.127475247524752e-07, + "loss": 0.54, + "mean_token_accuracy": 0.8308014869689941, + "num_tokens": 23335271.0, + "step": 668 + }, + { + "epoch": 0.12423398328690807, + "grad_norm": 1.662562608718872, + "learning_rate": 4.1336633663366333e-07, + "loss": 0.4748, + "mean_token_accuracy": 0.8463528156280518, + "num_tokens": 23367830.0, + "step": 669 + }, + { + "epoch": 0.1244196843082637, + "grad_norm": 1.697886347770691, + "learning_rate": 4.1398514851485145e-07, + "loss": 0.5129, + "mean_token_accuracy": 0.8402224183082581, + "num_tokens": 23400212.0, + "step": 670 + }, + { + "epoch": 0.12460538532961932, + "grad_norm": 1.6780449151992798, + "learning_rate": 4.1460396039603956e-07, + "loss": 0.4872, + "mean_token_accuracy": 0.8411605358123779, + "num_tokens": 23432998.0, + "step": 671 + }, + { + "epoch": 0.12479108635097493, + "grad_norm": 1.4844837188720703, + "learning_rate": 4.1522277227722767e-07, + "loss": 0.4466, + "mean_token_accuracy": 0.8543069958686829, + "num_tokens": 23470937.0, + "step": 672 + }, + { + "epoch": 0.12497678737233055, + "grad_norm": 1.6528996229171753, + "learning_rate": 4.158415841584158e-07, + "loss": 0.5295, + "mean_token_accuracy": 0.8306185603141785, + "num_tokens": 23503671.0, + "step": 673 + }, + { + "epoch": 0.12516248839368616, + "grad_norm": 1.552987813949585, + "learning_rate": 4.1646039603960394e-07, + "loss": 0.5161, + "mean_token_accuracy": 0.8377339839935303, + "num_tokens": 23540554.0, + "step": 674 + }, + { + "epoch": 0.12534818941504178, + "grad_norm": 1.679099440574646, + "learning_rate": 4.1707920792079206e-07, + "loss": 0.502, + "mean_token_accuracy": 0.8354055881500244, + "num_tokens": 23572430.0, + "step": 675 + }, + { + "epoch": 0.1255338904363974, + "grad_norm": 1.6249885559082031, + "learning_rate": 4.1769801980198017e-07, + "loss": 0.4913, + "mean_token_accuracy": 0.8440750241279602, + "num_tokens": 23604759.0, + "step": 676 + }, + { + "epoch": 0.12571959145775302, + "grad_norm": 1.4850845336914062, + "learning_rate": 4.183168316831683e-07, + "loss": 0.4803, + "mean_token_accuracy": 0.8432856202125549, + "num_tokens": 23645870.0, + "step": 677 + }, + { + "epoch": 0.12590529247910864, + "grad_norm": 1.5043312311172485, + "learning_rate": 4.189356435643564e-07, + "loss": 0.5173, + "mean_token_accuracy": 0.8349855542182922, + "num_tokens": 23687164.0, + "step": 678 + }, + { + "epoch": 0.12609099350046427, + "grad_norm": 1.5098992586135864, + "learning_rate": 4.195544554455445e-07, + "loss": 0.4629, + "mean_token_accuracy": 0.8496581315994263, + "num_tokens": 23724715.0, + "step": 679 + }, + { + "epoch": 0.12627669452181986, + "grad_norm": 1.519008755683899, + "learning_rate": 4.2017326732673266e-07, + "loss": 0.4353, + "mean_token_accuracy": 0.855302095413208, + "num_tokens": 23761351.0, + "step": 680 + }, + { + "epoch": 0.12646239554317548, + "grad_norm": 1.6328212022781372, + "learning_rate": 4.207920792079208e-07, + "loss": 0.4632, + "mean_token_accuracy": 0.8523674607276917, + "num_tokens": 23793047.0, + "step": 681 + }, + { + "epoch": 0.1266480965645311, + "grad_norm": 1.6595330238342285, + "learning_rate": 4.214108910891089e-07, + "loss": 0.544, + "mean_token_accuracy": 0.8261988759040833, + "num_tokens": 23829862.0, + "step": 682 + }, + { + "epoch": 0.12683379758588673, + "grad_norm": 1.8978861570358276, + "learning_rate": 4.22029702970297e-07, + "loss": 0.5662, + "mean_token_accuracy": 0.8233460187911987, + "num_tokens": 23862306.0, + "step": 683 + }, + { + "epoch": 0.12701949860724235, + "grad_norm": 1.6356111764907837, + "learning_rate": 4.226485148514851e-07, + "loss": 0.5149, + "mean_token_accuracy": 0.8375217914581299, + "num_tokens": 23898745.0, + "step": 684 + }, + { + "epoch": 0.12720519962859797, + "grad_norm": 1.586569905281067, + "learning_rate": 4.232673267326732e-07, + "loss": 0.4465, + "mean_token_accuracy": 0.857149600982666, + "num_tokens": 23933838.0, + "step": 685 + }, + { + "epoch": 0.12739090064995356, + "grad_norm": 1.4552687406539917, + "learning_rate": 4.2388613861386133e-07, + "loss": 0.4474, + "mean_token_accuracy": 0.8569985628128052, + "num_tokens": 23974189.0, + "step": 686 + }, + { + "epoch": 0.1275766016713092, + "grad_norm": 1.5249155759811401, + "learning_rate": 4.245049504950495e-07, + "loss": 0.5437, + "mean_token_accuracy": 0.8273801803588867, + "num_tokens": 24016650.0, + "step": 687 + }, + { + "epoch": 0.1277623026926648, + "grad_norm": 1.6083168983459473, + "learning_rate": 4.251237623762376e-07, + "loss": 0.5044, + "mean_token_accuracy": 0.8392399549484253, + "num_tokens": 24051037.0, + "step": 688 + }, + { + "epoch": 0.12794800371402043, + "grad_norm": 1.579624056816101, + "learning_rate": 4.257425742574257e-07, + "loss": 0.4546, + "mean_token_accuracy": 0.852095901966095, + "num_tokens": 24083183.0, + "step": 689 + }, + { + "epoch": 0.12813370473537605, + "grad_norm": 1.6029794216156006, + "learning_rate": 4.2636138613861383e-07, + "loss": 0.5198, + "mean_token_accuracy": 0.8310338854789734, + "num_tokens": 24114789.0, + "step": 690 + }, + { + "epoch": 0.12831940575673167, + "grad_norm": 1.5096276998519897, + "learning_rate": 4.2698019801980194e-07, + "loss": 0.4779, + "mean_token_accuracy": 0.8471928834915161, + "num_tokens": 24151993.0, + "step": 691 + }, + { + "epoch": 0.12850510677808727, + "grad_norm": 1.407158374786377, + "learning_rate": 4.2759900990099005e-07, + "loss": 0.4534, + "mean_token_accuracy": 0.8555344343185425, + "num_tokens": 24193910.0, + "step": 692 + }, + { + "epoch": 0.1286908077994429, + "grad_norm": 1.5194460153579712, + "learning_rate": 4.2821782178217816e-07, + "loss": 0.4949, + "mean_token_accuracy": 0.8419919610023499, + "num_tokens": 24232225.0, + "step": 693 + }, + { + "epoch": 0.1288765088207985, + "grad_norm": 1.541111707687378, + "learning_rate": 4.2883663366336633e-07, + "loss": 0.4944, + "mean_token_accuracy": 0.839084267616272, + "num_tokens": 24268635.0, + "step": 694 + }, + { + "epoch": 0.12906220984215414, + "grad_norm": 1.7411558628082275, + "learning_rate": 4.2945544554455444e-07, + "loss": 0.5209, + "mean_token_accuracy": 0.8359770774841309, + "num_tokens": 24298587.0, + "step": 695 + }, + { + "epoch": 0.12924791086350976, + "grad_norm": 1.7298651933670044, + "learning_rate": 4.3007425742574255e-07, + "loss": 0.5324, + "mean_token_accuracy": 0.8307874202728271, + "num_tokens": 24330318.0, + "step": 696 + }, + { + "epoch": 0.12943361188486538, + "grad_norm": 1.6660915613174438, + "learning_rate": 4.3069306930693066e-07, + "loss": 0.459, + "mean_token_accuracy": 0.8522689342498779, + "num_tokens": 24361932.0, + "step": 697 + }, + { + "epoch": 0.12961931290622097, + "grad_norm": 1.7405248880386353, + "learning_rate": 4.313118811881188e-07, + "loss": 0.4884, + "mean_token_accuracy": 0.8477875590324402, + "num_tokens": 24393884.0, + "step": 698 + }, + { + "epoch": 0.1298050139275766, + "grad_norm": 1.6091721057891846, + "learning_rate": 4.319306930693069e-07, + "loss": 0.4954, + "mean_token_accuracy": 0.8425846099853516, + "num_tokens": 24431524.0, + "step": 699 + }, + { + "epoch": 0.12999071494893222, + "grad_norm": 1.5753612518310547, + "learning_rate": 4.3254950495049505e-07, + "loss": 0.4393, + "mean_token_accuracy": 0.855413556098938, + "num_tokens": 24463181.0, + "step": 700 + }, + { + "epoch": 0.13017641597028784, + "grad_norm": 1.4882272481918335, + "learning_rate": 4.3316831683168316e-07, + "loss": 0.4929, + "mean_token_accuracy": 0.8400996923446655, + "num_tokens": 24502792.0, + "step": 701 + }, + { + "epoch": 0.13036211699164346, + "grad_norm": 1.7235950231552124, + "learning_rate": 4.3378712871287127e-07, + "loss": 0.5001, + "mean_token_accuracy": 0.8377996683120728, + "num_tokens": 24535024.0, + "step": 702 + }, + { + "epoch": 0.13054781801299908, + "grad_norm": 1.5718026161193848, + "learning_rate": 4.344059405940594e-07, + "loss": 0.4325, + "mean_token_accuracy": 0.8608162999153137, + "num_tokens": 24567699.0, + "step": 703 + }, + { + "epoch": 0.13073351903435468, + "grad_norm": 1.5130095481872559, + "learning_rate": 4.350247524752475e-07, + "loss": 0.4792, + "mean_token_accuracy": 0.8473029136657715, + "num_tokens": 24606726.0, + "step": 704 + }, + { + "epoch": 0.1309192200557103, + "grad_norm": 1.528393268585205, + "learning_rate": 4.356435643564356e-07, + "loss": 0.4663, + "mean_token_accuracy": 0.8470776081085205, + "num_tokens": 24645099.0, + "step": 705 + }, + { + "epoch": 0.13110492107706592, + "grad_norm": 1.7140949964523315, + "learning_rate": 4.362623762376237e-07, + "loss": 0.5001, + "mean_token_accuracy": 0.8397121429443359, + "num_tokens": 24675791.0, + "step": 706 + }, + { + "epoch": 0.13129062209842154, + "grad_norm": 1.7720118761062622, + "learning_rate": 4.368811881188119e-07, + "loss": 0.5048, + "mean_token_accuracy": 0.8352887630462646, + "num_tokens": 24708562.0, + "step": 707 + }, + { + "epoch": 0.13147632311977717, + "grad_norm": 1.7593601942062378, + "learning_rate": 4.375e-07, + "loss": 0.5559, + "mean_token_accuracy": 0.8215998411178589, + "num_tokens": 24740866.0, + "step": 708 + }, + { + "epoch": 0.1316620241411328, + "grad_norm": 1.511965274810791, + "learning_rate": 4.381188118811881e-07, + "loss": 0.4722, + "mean_token_accuracy": 0.8469353914260864, + "num_tokens": 24777912.0, + "step": 709 + }, + { + "epoch": 0.13184772516248838, + "grad_norm": 1.671097755432129, + "learning_rate": 4.387376237623762e-07, + "loss": 0.4793, + "mean_token_accuracy": 0.8429821133613586, + "num_tokens": 24811658.0, + "step": 710 + }, + { + "epoch": 0.132033426183844, + "grad_norm": 1.5141048431396484, + "learning_rate": 4.3935643564356433e-07, + "loss": 0.4745, + "mean_token_accuracy": 0.8509767055511475, + "num_tokens": 24849246.0, + "step": 711 + }, + { + "epoch": 0.13221912720519963, + "grad_norm": 1.5677319765090942, + "learning_rate": 4.3997524752475244e-07, + "loss": 0.4754, + "mean_token_accuracy": 0.8481388688087463, + "num_tokens": 24882002.0, + "step": 712 + }, + { + "epoch": 0.13240482822655525, + "grad_norm": 1.8829594850540161, + "learning_rate": 4.405940594059406e-07, + "loss": 0.5307, + "mean_token_accuracy": 0.8312608003616333, + "num_tokens": 24909184.0, + "step": 713 + }, + { + "epoch": 0.13259052924791087, + "grad_norm": 1.4360761642456055, + "learning_rate": 4.412128712871287e-07, + "loss": 0.489, + "mean_token_accuracy": 0.8440259695053101, + "num_tokens": 24949751.0, + "step": 714 + }, + { + "epoch": 0.1327762302692665, + "grad_norm": 1.5244035720825195, + "learning_rate": 4.418316831683168e-07, + "loss": 0.4695, + "mean_token_accuracy": 0.8494842648506165, + "num_tokens": 24987647.0, + "step": 715 + }, + { + "epoch": 0.1329619312906221, + "grad_norm": 1.507218360900879, + "learning_rate": 4.4245049504950494e-07, + "loss": 0.5072, + "mean_token_accuracy": 0.8357590436935425, + "num_tokens": 25025951.0, + "step": 716 + }, + { + "epoch": 0.1331476323119777, + "grad_norm": 1.4306395053863525, + "learning_rate": 4.4306930693069305e-07, + "loss": 0.5348, + "mean_token_accuracy": 0.8296167850494385, + "num_tokens": 25068906.0, + "step": 717 + }, + { + "epoch": 0.13333333333333333, + "grad_norm": 1.6689122915267944, + "learning_rate": 4.4368811881188116e-07, + "loss": 0.4833, + "mean_token_accuracy": 0.8417131304740906, + "num_tokens": 25099749.0, + "step": 718 + }, + { + "epoch": 0.13351903435468895, + "grad_norm": 1.468581199645996, + "learning_rate": 4.4430693069306927e-07, + "loss": 0.4522, + "mean_token_accuracy": 0.8535372018814087, + "num_tokens": 25139210.0, + "step": 719 + }, + { + "epoch": 0.13370473537604458, + "grad_norm": 1.7525615692138672, + "learning_rate": 4.4492574257425744e-07, + "loss": 0.5265, + "mean_token_accuracy": 0.8315818309783936, + "num_tokens": 25170758.0, + "step": 720 + }, + { + "epoch": 0.1338904363974002, + "grad_norm": 1.5993191003799438, + "learning_rate": 4.4554455445544555e-07, + "loss": 0.507, + "mean_token_accuracy": 0.8375168442726135, + "num_tokens": 25203806.0, + "step": 721 + }, + { + "epoch": 0.1340761374187558, + "grad_norm": 1.6302974224090576, + "learning_rate": 4.4616336633663366e-07, + "loss": 0.5454, + "mean_token_accuracy": 0.8289918899536133, + "num_tokens": 25238893.0, + "step": 722 + }, + { + "epoch": 0.13426183844011141, + "grad_norm": 1.6976444721221924, + "learning_rate": 4.4678217821782177e-07, + "loss": 0.5277, + "mean_token_accuracy": 0.8320039510726929, + "num_tokens": 25273095.0, + "step": 723 + }, + { + "epoch": 0.13444753946146704, + "grad_norm": 1.536298155784607, + "learning_rate": 4.474009900990099e-07, + "loss": 0.4295, + "mean_token_accuracy": 0.8575966358184814, + "num_tokens": 25310308.0, + "step": 724 + }, + { + "epoch": 0.13463324048282266, + "grad_norm": 1.658776044845581, + "learning_rate": 4.48019801980198e-07, + "loss": 0.4622, + "mean_token_accuracy": 0.8509212136268616, + "num_tokens": 25345190.0, + "step": 725 + }, + { + "epoch": 0.13481894150417828, + "grad_norm": 1.53484308719635, + "learning_rate": 4.486386138613861e-07, + "loss": 0.5167, + "mean_token_accuracy": 0.834562361240387, + "num_tokens": 25384625.0, + "step": 726 + }, + { + "epoch": 0.1350046425255339, + "grad_norm": 1.7308988571166992, + "learning_rate": 4.4925742574257427e-07, + "loss": 0.5197, + "mean_token_accuracy": 0.8386857509613037, + "num_tokens": 25416983.0, + "step": 727 + }, + { + "epoch": 0.1351903435468895, + "grad_norm": 1.6932088136672974, + "learning_rate": 4.498762376237624e-07, + "loss": 0.4761, + "mean_token_accuracy": 0.8536689281463623, + "num_tokens": 25447090.0, + "step": 728 + }, + { + "epoch": 0.13537604456824512, + "grad_norm": 1.6452829837799072, + "learning_rate": 4.504950495049505e-07, + "loss": 0.486, + "mean_token_accuracy": 0.8427053689956665, + "num_tokens": 25481835.0, + "step": 729 + }, + { + "epoch": 0.13556174558960074, + "grad_norm": 1.4689900875091553, + "learning_rate": 4.511138613861386e-07, + "loss": 0.4958, + "mean_token_accuracy": 0.8417545557022095, + "num_tokens": 25525377.0, + "step": 730 + }, + { + "epoch": 0.13574744661095636, + "grad_norm": 1.5649467706680298, + "learning_rate": 4.517326732673267e-07, + "loss": 0.5218, + "mean_token_accuracy": 0.830729603767395, + "num_tokens": 25560642.0, + "step": 731 + }, + { + "epoch": 0.13593314763231198, + "grad_norm": 1.5933055877685547, + "learning_rate": 4.523514851485148e-07, + "loss": 0.504, + "mean_token_accuracy": 0.8405634164810181, + "num_tokens": 25599771.0, + "step": 732 + }, + { + "epoch": 0.1361188486536676, + "grad_norm": 1.668682336807251, + "learning_rate": 4.52970297029703e-07, + "loss": 0.4786, + "mean_token_accuracy": 0.847955584526062, + "num_tokens": 25632731.0, + "step": 733 + }, + { + "epoch": 0.1363045496750232, + "grad_norm": 1.6790298223495483, + "learning_rate": 4.535891089108911e-07, + "loss": 0.4894, + "mean_token_accuracy": 0.8407157063484192, + "num_tokens": 25664187.0, + "step": 734 + }, + { + "epoch": 0.13649025069637882, + "grad_norm": 1.5652387142181396, + "learning_rate": 4.542079207920792e-07, + "loss": 0.4925, + "mean_token_accuracy": 0.8451826572418213, + "num_tokens": 25699944.0, + "step": 735 + }, + { + "epoch": 0.13667595171773445, + "grad_norm": 1.5872045755386353, + "learning_rate": 4.548267326732673e-07, + "loss": 0.5228, + "mean_token_accuracy": 0.8373247385025024, + "num_tokens": 25738618.0, + "step": 736 + }, + { + "epoch": 0.13686165273909007, + "grad_norm": 1.5224766731262207, + "learning_rate": 4.5544554455445543e-07, + "loss": 0.5032, + "mean_token_accuracy": 0.840533435344696, + "num_tokens": 25777244.0, + "step": 737 + }, + { + "epoch": 0.1370473537604457, + "grad_norm": 1.539015769958496, + "learning_rate": 4.5606435643564354e-07, + "loss": 0.4762, + "mean_token_accuracy": 0.8491491675376892, + "num_tokens": 25812191.0, + "step": 738 + }, + { + "epoch": 0.1372330547818013, + "grad_norm": 1.54203200340271, + "learning_rate": 4.5668316831683166e-07, + "loss": 0.4489, + "mean_token_accuracy": 0.8551279902458191, + "num_tokens": 25845643.0, + "step": 739 + }, + { + "epoch": 0.1374187558031569, + "grad_norm": 1.503519058227539, + "learning_rate": 4.573019801980198e-07, + "loss": 0.4159, + "mean_token_accuracy": 0.8599215745925903, + "num_tokens": 25879118.0, + "step": 740 + }, + { + "epoch": 0.13760445682451253, + "grad_norm": 1.6392902135849, + "learning_rate": 4.5792079207920793e-07, + "loss": 0.4703, + "mean_token_accuracy": 0.84621262550354, + "num_tokens": 25911043.0, + "step": 741 + }, + { + "epoch": 0.13779015784586815, + "grad_norm": 1.608819603919983, + "learning_rate": 4.5853960396039604e-07, + "loss": 0.4987, + "mean_token_accuracy": 0.8388893008232117, + "num_tokens": 25947354.0, + "step": 742 + }, + { + "epoch": 0.13797585886722377, + "grad_norm": 1.648013710975647, + "learning_rate": 4.5915841584158415e-07, + "loss": 0.4397, + "mean_token_accuracy": 0.8571345806121826, + "num_tokens": 25979373.0, + "step": 743 + }, + { + "epoch": 0.1381615598885794, + "grad_norm": 1.5246095657348633, + "learning_rate": 4.5977722772277227e-07, + "loss": 0.4988, + "mean_token_accuracy": 0.8412118554115295, + "num_tokens": 26020443.0, + "step": 744 + }, + { + "epoch": 0.13834726090993502, + "grad_norm": 1.6800423860549927, + "learning_rate": 4.603960396039604e-07, + "loss": 0.4817, + "mean_token_accuracy": 0.8426128625869751, + "num_tokens": 26054005.0, + "step": 745 + }, + { + "epoch": 0.1385329619312906, + "grad_norm": 1.4738649129867554, + "learning_rate": 4.6101485148514854e-07, + "loss": 0.4929, + "mean_token_accuracy": 0.8407113552093506, + "num_tokens": 26092424.0, + "step": 746 + }, + { + "epoch": 0.13871866295264623, + "grad_norm": 1.5975757837295532, + "learning_rate": 4.6163366336633665e-07, + "loss": 0.4485, + "mean_token_accuracy": 0.8541855812072754, + "num_tokens": 26124553.0, + "step": 747 + }, + { + "epoch": 0.13890436397400185, + "grad_norm": 1.5041247606277466, + "learning_rate": 4.6225247524752476e-07, + "loss": 0.5276, + "mean_token_accuracy": 0.8330023288726807, + "num_tokens": 26163476.0, + "step": 748 + }, + { + "epoch": 0.13909006499535748, + "grad_norm": 1.5463732481002808, + "learning_rate": 4.628712871287129e-07, + "loss": 0.5081, + "mean_token_accuracy": 0.8362526893615723, + "num_tokens": 26199783.0, + "step": 749 + }, + { + "epoch": 0.1392757660167131, + "grad_norm": 1.6460078954696655, + "learning_rate": 4.63490099009901e-07, + "loss": 0.5153, + "mean_token_accuracy": 0.8365042209625244, + "num_tokens": 26234909.0, + "step": 750 + }, + { + "epoch": 0.13946146703806872, + "grad_norm": 1.5913317203521729, + "learning_rate": 4.641089108910891e-07, + "loss": 0.479, + "mean_token_accuracy": 0.846845269203186, + "num_tokens": 26272934.0, + "step": 751 + }, + { + "epoch": 0.13964716805942431, + "grad_norm": 1.7455875873565674, + "learning_rate": 4.647277227722772e-07, + "loss": 0.4764, + "mean_token_accuracy": 0.8442374467849731, + "num_tokens": 26300916.0, + "step": 752 + }, + { + "epoch": 0.13983286908077994, + "grad_norm": 1.6028952598571777, + "learning_rate": 4.6534653465346537e-07, + "loss": 0.4729, + "mean_token_accuracy": 0.8454976081848145, + "num_tokens": 26334968.0, + "step": 753 + }, + { + "epoch": 0.14001857010213556, + "grad_norm": 1.4050413370132446, + "learning_rate": 4.659653465346535e-07, + "loss": 0.4855, + "mean_token_accuracy": 0.8404747843742371, + "num_tokens": 26378057.0, + "step": 754 + }, + { + "epoch": 0.14020427112349118, + "grad_norm": 1.3998898267745972, + "learning_rate": 4.665841584158416e-07, + "loss": 0.4734, + "mean_token_accuracy": 0.8493123054504395, + "num_tokens": 26420887.0, + "step": 755 + }, + { + "epoch": 0.1403899721448468, + "grad_norm": 1.7887650728225708, + "learning_rate": 4.672029702970297e-07, + "loss": 0.5204, + "mean_token_accuracy": 0.8346466422080994, + "num_tokens": 26452759.0, + "step": 756 + }, + { + "epoch": 0.14057567316620243, + "grad_norm": 1.4245463609695435, + "learning_rate": 4.678217821782178e-07, + "loss": 0.4885, + "mean_token_accuracy": 0.8421844840049744, + "num_tokens": 26492319.0, + "step": 757 + }, + { + "epoch": 0.14076137418755802, + "grad_norm": 1.4993937015533447, + "learning_rate": 4.6844059405940593e-07, + "loss": 0.4691, + "mean_token_accuracy": 0.8482120037078857, + "num_tokens": 26530947.0, + "step": 758 + }, + { + "epoch": 0.14094707520891364, + "grad_norm": 1.7023545503616333, + "learning_rate": 4.69059405940594e-07, + "loss": 0.5225, + "mean_token_accuracy": 0.8343299031257629, + "num_tokens": 26564334.0, + "step": 759 + }, + { + "epoch": 0.14113277623026926, + "grad_norm": 1.4263851642608643, + "learning_rate": 4.6967821782178215e-07, + "loss": 0.436, + "mean_token_accuracy": 0.8546247482299805, + "num_tokens": 26603984.0, + "step": 760 + }, + { + "epoch": 0.14131847725162489, + "grad_norm": 1.469169020652771, + "learning_rate": 4.7029702970297026e-07, + "loss": 0.4493, + "mean_token_accuracy": 0.8542935252189636, + "num_tokens": 26642629.0, + "step": 761 + }, + { + "epoch": 0.1415041782729805, + "grad_norm": 1.6766211986541748, + "learning_rate": 4.709158415841584e-07, + "loss": 0.4841, + "mean_token_accuracy": 0.84818035364151, + "num_tokens": 26673227.0, + "step": 762 + }, + { + "epoch": 0.14168987929433613, + "grad_norm": 1.5975483655929565, + "learning_rate": 4.715346534653465e-07, + "loss": 0.5242, + "mean_token_accuracy": 0.8332997560501099, + "num_tokens": 26708555.0, + "step": 763 + }, + { + "epoch": 0.14187558031569172, + "grad_norm": 1.608867883682251, + "learning_rate": 4.721534653465346e-07, + "loss": 0.5351, + "mean_token_accuracy": 0.8352007269859314, + "num_tokens": 26745225.0, + "step": 764 + }, + { + "epoch": 0.14206128133704735, + "grad_norm": 1.5066673755645752, + "learning_rate": 4.727722772277227e-07, + "loss": 0.4836, + "mean_token_accuracy": 0.8423160910606384, + "num_tokens": 26782108.0, + "step": 765 + }, + { + "epoch": 0.14224698235840297, + "grad_norm": 1.5642155408859253, + "learning_rate": 4.733910891089108e-07, + "loss": 0.4699, + "mean_token_accuracy": 0.8498396277427673, + "num_tokens": 26817852.0, + "step": 766 + }, + { + "epoch": 0.1424326833797586, + "grad_norm": 1.5109705924987793, + "learning_rate": 4.74009900990099e-07, + "loss": 0.4576, + "mean_token_accuracy": 0.8516494631767273, + "num_tokens": 26855126.0, + "step": 767 + }, + { + "epoch": 0.1426183844011142, + "grad_norm": 1.6105948686599731, + "learning_rate": 4.746287128712871e-07, + "loss": 0.4772, + "mean_token_accuracy": 0.8462097644805908, + "num_tokens": 26888243.0, + "step": 768 + }, + { + "epoch": 0.14280408542246983, + "grad_norm": 1.5641427040100098, + "learning_rate": 4.752475247524752e-07, + "loss": 0.4861, + "mean_token_accuracy": 0.8422682285308838, + "num_tokens": 26926705.0, + "step": 769 + }, + { + "epoch": 0.14298978644382543, + "grad_norm": 1.674174189567566, + "learning_rate": 4.758663366336633e-07, + "loss": 0.5557, + "mean_token_accuracy": 0.823519229888916, + "num_tokens": 26960664.0, + "step": 770 + }, + { + "epoch": 0.14317548746518105, + "grad_norm": 1.5649446249008179, + "learning_rate": 4.7648514851485143e-07, + "loss": 0.4534, + "mean_token_accuracy": 0.8496809005737305, + "num_tokens": 26993944.0, + "step": 771 + }, + { + "epoch": 0.14336118848653667, + "grad_norm": 1.579086184501648, + "learning_rate": 4.771039603960396e-07, + "loss": 0.4232, + "mean_token_accuracy": 0.863535463809967, + "num_tokens": 27026529.0, + "step": 772 + }, + { + "epoch": 0.1435468895078923, + "grad_norm": 1.6137679815292358, + "learning_rate": 4.777227722772277e-07, + "loss": 0.4888, + "mean_token_accuracy": 0.8463207483291626, + "num_tokens": 27058872.0, + "step": 773 + }, + { + "epoch": 0.14373259052924792, + "grad_norm": 1.506001353263855, + "learning_rate": 4.783415841584158e-07, + "loss": 0.4588, + "mean_token_accuracy": 0.8524825572967529, + "num_tokens": 27095514.0, + "step": 774 + }, + { + "epoch": 0.14391829155060354, + "grad_norm": 1.6496598720550537, + "learning_rate": 4.789603960396039e-07, + "loss": 0.5246, + "mean_token_accuracy": 0.8269201517105103, + "num_tokens": 27130350.0, + "step": 775 + }, + { + "epoch": 0.14410399257195913, + "grad_norm": 1.7166026830673218, + "learning_rate": 4.79579207920792e-07, + "loss": 0.505, + "mean_token_accuracy": 0.8382904529571533, + "num_tokens": 27165646.0, + "step": 776 + }, + { + "epoch": 0.14428969359331476, + "grad_norm": 1.5456825494766235, + "learning_rate": 4.801980198019802e-07, + "loss": 0.4687, + "mean_token_accuracy": 0.8473352789878845, + "num_tokens": 27201174.0, + "step": 777 + }, + { + "epoch": 0.14447539461467038, + "grad_norm": 1.7866270542144775, + "learning_rate": 4.808168316831683e-07, + "loss": 0.4145, + "mean_token_accuracy": 0.8623378872871399, + "num_tokens": 27226713.0, + "step": 778 + }, + { + "epoch": 0.144661095636026, + "grad_norm": 1.769949197769165, + "learning_rate": 4.814356435643564e-07, + "loss": 0.4816, + "mean_token_accuracy": 0.8475251197814941, + "num_tokens": 27255463.0, + "step": 779 + }, + { + "epoch": 0.14484679665738162, + "grad_norm": 1.5703239440917969, + "learning_rate": 4.820544554455445e-07, + "loss": 0.4917, + "mean_token_accuracy": 0.8440717458724976, + "num_tokens": 27290613.0, + "step": 780 + }, + { + "epoch": 0.14503249767873724, + "grad_norm": 1.7722103595733643, + "learning_rate": 4.826732673267326e-07, + "loss": 0.4727, + "mean_token_accuracy": 0.8455804586410522, + "num_tokens": 27319350.0, + "step": 781 + }, + { + "epoch": 0.14521819870009284, + "grad_norm": 1.6204501390457153, + "learning_rate": 4.832920792079207e-07, + "loss": 0.4741, + "mean_token_accuracy": 0.8470124006271362, + "num_tokens": 27354972.0, + "step": 782 + }, + { + "epoch": 0.14540389972144846, + "grad_norm": 1.5379855632781982, + "learning_rate": 4.839108910891089e-07, + "loss": 0.458, + "mean_token_accuracy": 0.8528003692626953, + "num_tokens": 27389667.0, + "step": 783 + }, + { + "epoch": 0.14558960074280408, + "grad_norm": 1.5591789484024048, + "learning_rate": 4.84529702970297e-07, + "loss": 0.4743, + "mean_token_accuracy": 0.849867045879364, + "num_tokens": 27423651.0, + "step": 784 + }, + { + "epoch": 0.1457753017641597, + "grad_norm": 1.5585917234420776, + "learning_rate": 4.851485148514851e-07, + "loss": 0.5064, + "mean_token_accuracy": 0.8401622772216797, + "num_tokens": 27460736.0, + "step": 785 + }, + { + "epoch": 0.14596100278551533, + "grad_norm": 1.59889554977417, + "learning_rate": 4.857673267326733e-07, + "loss": 0.5196, + "mean_token_accuracy": 0.8324903249740601, + "num_tokens": 27496751.0, + "step": 786 + }, + { + "epoch": 0.14614670380687095, + "grad_norm": 1.5883113145828247, + "learning_rate": 4.863861386138613e-07, + "loss": 0.4651, + "mean_token_accuracy": 0.8515638113021851, + "num_tokens": 27531578.0, + "step": 787 + }, + { + "epoch": 0.14633240482822654, + "grad_norm": 1.5245078802108765, + "learning_rate": 4.870049504950495e-07, + "loss": 0.4938, + "mean_token_accuracy": 0.8424009680747986, + "num_tokens": 27567151.0, + "step": 788 + }, + { + "epoch": 0.14651810584958216, + "grad_norm": 1.592313289642334, + "learning_rate": 4.876237623762375e-07, + "loss": 0.5004, + "mean_token_accuracy": 0.8375993371009827, + "num_tokens": 27605881.0, + "step": 789 + }, + { + "epoch": 0.1467038068709378, + "grad_norm": 1.5540755987167358, + "learning_rate": 4.882425742574257e-07, + "loss": 0.4708, + "mean_token_accuracy": 0.850652277469635, + "num_tokens": 27639592.0, + "step": 790 + }, + { + "epoch": 0.1468895078922934, + "grad_norm": 1.6272708177566528, + "learning_rate": 4.888613861386139e-07, + "loss": 0.498, + "mean_token_accuracy": 0.8376474380493164, + "num_tokens": 27671445.0, + "step": 791 + }, + { + "epoch": 0.14707520891364903, + "grad_norm": 1.67556893825531, + "learning_rate": 4.894801980198019e-07, + "loss": 0.481, + "mean_token_accuracy": 0.845260500907898, + "num_tokens": 27703788.0, + "step": 792 + }, + { + "epoch": 0.14726090993500465, + "grad_norm": 1.6260119676589966, + "learning_rate": 4.900990099009901e-07, + "loss": 0.3992, + "mean_token_accuracy": 0.8663603663444519, + "num_tokens": 27734739.0, + "step": 793 + }, + { + "epoch": 0.14744661095636025, + "grad_norm": 1.7840667963027954, + "learning_rate": 4.907178217821781e-07, + "loss": 0.4903, + "mean_token_accuracy": 0.8402926921844482, + "num_tokens": 27765824.0, + "step": 794 + }, + { + "epoch": 0.14763231197771587, + "grad_norm": 1.6094152927398682, + "learning_rate": 4.913366336633663e-07, + "loss": 0.4782, + "mean_token_accuracy": 0.848449170589447, + "num_tokens": 27799949.0, + "step": 795 + }, + { + "epoch": 0.1478180129990715, + "grad_norm": 1.7541131973266602, + "learning_rate": 4.919554455445545e-07, + "loss": 0.4504, + "mean_token_accuracy": 0.8525524139404297, + "num_tokens": 27827508.0, + "step": 796 + }, + { + "epoch": 0.1480037140204271, + "grad_norm": 1.4388595819473267, + "learning_rate": 4.925742574257425e-07, + "loss": 0.5102, + "mean_token_accuracy": 0.8344461917877197, + "num_tokens": 27872967.0, + "step": 797 + }, + { + "epoch": 0.14818941504178273, + "grad_norm": 1.5677928924560547, + "learning_rate": 4.931930693069307e-07, + "loss": 0.4845, + "mean_token_accuracy": 0.8455353379249573, + "num_tokens": 27910327.0, + "step": 798 + }, + { + "epoch": 0.14837511606313836, + "grad_norm": 1.940756916999817, + "learning_rate": 4.938118811881188e-07, + "loss": 0.4659, + "mean_token_accuracy": 0.8478949069976807, + "num_tokens": 27931697.0, + "step": 799 + }, + { + "epoch": 0.14856081708449395, + "grad_norm": 1.5610709190368652, + "learning_rate": 4.944306930693069e-07, + "loss": 0.461, + "mean_token_accuracy": 0.8528463840484619, + "num_tokens": 27968331.0, + "step": 800 + }, + { + "epoch": 0.14874651810584957, + "grad_norm": 1.580933690071106, + "learning_rate": 4.95049504950495e-07, + "loss": 0.4857, + "mean_token_accuracy": 0.843619704246521, + "num_tokens": 28002881.0, + "step": 801 + }, + { + "epoch": 0.1489322191272052, + "grad_norm": 1.6016772985458374, + "learning_rate": 4.956683168316831e-07, + "loss": 0.4977, + "mean_token_accuracy": 0.8396314978599548, + "num_tokens": 28036967.0, + "step": 802 + }, + { + "epoch": 0.14911792014856082, + "grad_norm": 1.5636470317840576, + "learning_rate": 4.962871287128713e-07, + "loss": 0.539, + "mean_token_accuracy": 0.8253725171089172, + "num_tokens": 28076973.0, + "step": 803 + }, + { + "epoch": 0.14930362116991644, + "grad_norm": 1.695665955543518, + "learning_rate": 4.969059405940594e-07, + "loss": 0.5048, + "mean_token_accuracy": 0.8402115106582642, + "num_tokens": 28111216.0, + "step": 804 + }, + { + "epoch": 0.14948932219127206, + "grad_norm": 1.7667343616485596, + "learning_rate": 4.975247524752475e-07, + "loss": 0.5033, + "mean_token_accuracy": 0.8382422924041748, + "num_tokens": 28139327.0, + "step": 805 + }, + { + "epoch": 0.14967502321262766, + "grad_norm": 1.4026960134506226, + "learning_rate": 4.981435643564356e-07, + "loss": 0.4683, + "mean_token_accuracy": 0.8475581407546997, + "num_tokens": 28182560.0, + "step": 806 + }, + { + "epoch": 0.14986072423398328, + "grad_norm": 1.5748908519744873, + "learning_rate": 4.987623762376238e-07, + "loss": 0.4763, + "mean_token_accuracy": 0.8415488004684448, + "num_tokens": 28218051.0, + "step": 807 + }, + { + "epoch": 0.1500464252553389, + "grad_norm": 1.6175768375396729, + "learning_rate": 4.993811881188118e-07, + "loss": 0.4874, + "mean_token_accuracy": 0.8433800935745239, + "num_tokens": 28251217.0, + "step": 808 + }, + { + "epoch": 0.15023212627669452, + "grad_norm": 1.655200719833374, + "learning_rate": 5e-07, + "loss": 0.5019, + "mean_token_accuracy": 0.8327991366386414, + "num_tokens": 28282482.0, + "step": 809 + }, + { + "epoch": 0.15041782729805014, + "grad_norm": 1.6252599954605103, + "learning_rate": 5.00618811881188e-07, + "loss": 0.5005, + "mean_token_accuracy": 0.8435436487197876, + "num_tokens": 28313879.0, + "step": 810 + }, + { + "epoch": 0.15060352831940577, + "grad_norm": 1.7049094438552856, + "learning_rate": 5.012376237623762e-07, + "loss": 0.5109, + "mean_token_accuracy": 0.8395557403564453, + "num_tokens": 28348887.0, + "step": 811 + }, + { + "epoch": 0.15078922934076136, + "grad_norm": 1.5899314880371094, + "learning_rate": 5.018564356435643e-07, + "loss": 0.5574, + "mean_token_accuracy": 0.8223564624786377, + "num_tokens": 28385824.0, + "step": 812 + }, + { + "epoch": 0.15097493036211698, + "grad_norm": 1.7691572904586792, + "learning_rate": 5.024752475247524e-07, + "loss": 0.5208, + "mean_token_accuracy": 0.8341802358627319, + "num_tokens": 28415505.0, + "step": 813 + }, + { + "epoch": 0.1511606313834726, + "grad_norm": 1.5110489130020142, + "learning_rate": 5.030940594059405e-07, + "loss": 0.4798, + "mean_token_accuracy": 0.842044472694397, + "num_tokens": 28451690.0, + "step": 814 + }, + { + "epoch": 0.15134633240482823, + "grad_norm": 1.6161750555038452, + "learning_rate": 5.037128712871286e-07, + "loss": 0.4271, + "mean_token_accuracy": 0.8604742884635925, + "num_tokens": 28480835.0, + "step": 815 + }, + { + "epoch": 0.15153203342618385, + "grad_norm": 1.7257627248764038, + "learning_rate": 5.043316831683168e-07, + "loss": 0.4834, + "mean_token_accuracy": 0.8446006178855896, + "num_tokens": 28508956.0, + "step": 816 + }, + { + "epoch": 0.15171773444753947, + "grad_norm": 1.5443240404129028, + "learning_rate": 5.04950495049505e-07, + "loss": 0.5317, + "mean_token_accuracy": 0.8319887518882751, + "num_tokens": 28549777.0, + "step": 817 + }, + { + "epoch": 0.1519034354688951, + "grad_norm": 1.4595043659210205, + "learning_rate": 5.05569306930693e-07, + "loss": 0.4236, + "mean_token_accuracy": 0.8632307052612305, + "num_tokens": 28586960.0, + "step": 818 + }, + { + "epoch": 0.1520891364902507, + "grad_norm": 1.604403018951416, + "learning_rate": 5.061881188118812e-07, + "loss": 0.5622, + "mean_token_accuracy": 0.8238649368286133, + "num_tokens": 28622213.0, + "step": 819 + }, + { + "epoch": 0.1522748375116063, + "grad_norm": 1.6688228845596313, + "learning_rate": 5.068069306930693e-07, + "loss": 0.4281, + "mean_token_accuracy": 0.8615402579307556, + "num_tokens": 28652627.0, + "step": 820 + }, + { + "epoch": 0.15246053853296193, + "grad_norm": 1.5032716989517212, + "learning_rate": 5.074257425742574e-07, + "loss": 0.4601, + "mean_token_accuracy": 0.8510440587997437, + "num_tokens": 28689366.0, + "step": 821 + }, + { + "epoch": 0.15264623955431755, + "grad_norm": 1.454458475112915, + "learning_rate": 5.080445544554455e-07, + "loss": 0.4531, + "mean_token_accuracy": 0.8537479639053345, + "num_tokens": 28728030.0, + "step": 822 + }, + { + "epoch": 0.15283194057567318, + "grad_norm": 1.5590269565582275, + "learning_rate": 5.086633663366336e-07, + "loss": 0.492, + "mean_token_accuracy": 0.840910792350769, + "num_tokens": 28765885.0, + "step": 823 + }, + { + "epoch": 0.1530176415970288, + "grad_norm": 1.5641692876815796, + "learning_rate": 5.092821782178217e-07, + "loss": 0.4948, + "mean_token_accuracy": 0.8407539129257202, + "num_tokens": 28801616.0, + "step": 824 + }, + { + "epoch": 0.1532033426183844, + "grad_norm": 1.4761167764663696, + "learning_rate": 5.099009900990099e-07, + "loss": 0.4281, + "mean_token_accuracy": 0.8575116395950317, + "num_tokens": 28835252.0, + "step": 825 + }, + { + "epoch": 0.15338904363974, + "grad_norm": 1.6038954257965088, + "learning_rate": 5.105198019801979e-07, + "loss": 0.463, + "mean_token_accuracy": 0.8464599847793579, + "num_tokens": 28865083.0, + "step": 826 + }, + { + "epoch": 0.15357474466109564, + "grad_norm": 1.5677680969238281, + "learning_rate": 5.111386138613861e-07, + "loss": 0.5001, + "mean_token_accuracy": 0.837521493434906, + "num_tokens": 28900592.0, + "step": 827 + }, + { + "epoch": 0.15376044568245126, + "grad_norm": 1.4364687204360962, + "learning_rate": 5.117574257425741e-07, + "loss": 0.5081, + "mean_token_accuracy": 0.8391521573066711, + "num_tokens": 28941216.0, + "step": 828 + }, + { + "epoch": 0.15394614670380688, + "grad_norm": 1.6684316396713257, + "learning_rate": 5.123762376237624e-07, + "loss": 0.439, + "mean_token_accuracy": 0.8561709523200989, + "num_tokens": 28975300.0, + "step": 829 + }, + { + "epoch": 0.1541318477251625, + "grad_norm": 1.6277596950531006, + "learning_rate": 5.129950495049505e-07, + "loss": 0.5161, + "mean_token_accuracy": 0.8355869650840759, + "num_tokens": 29008510.0, + "step": 830 + }, + { + "epoch": 0.1543175487465181, + "grad_norm": 1.4874892234802246, + "learning_rate": 5.136138613861386e-07, + "loss": 0.5009, + "mean_token_accuracy": 0.8376832604408264, + "num_tokens": 29052957.0, + "step": 831 + }, + { + "epoch": 0.15450324976787372, + "grad_norm": 1.563879370689392, + "learning_rate": 5.142326732673267e-07, + "loss": 0.4608, + "mean_token_accuracy": 0.8500398993492126, + "num_tokens": 29088684.0, + "step": 832 + }, + { + "epoch": 0.15468895078922934, + "grad_norm": 1.474295735359192, + "learning_rate": 5.148514851485149e-07, + "loss": 0.4355, + "mean_token_accuracy": 0.8563698530197144, + "num_tokens": 29122102.0, + "step": 833 + }, + { + "epoch": 0.15487465181058496, + "grad_norm": 1.5027687549591064, + "learning_rate": 5.154702970297029e-07, + "loss": 0.444, + "mean_token_accuracy": 0.857669472694397, + "num_tokens": 29156117.0, + "step": 834 + }, + { + "epoch": 0.15506035283194058, + "grad_norm": 1.5236177444458008, + "learning_rate": 5.160891089108911e-07, + "loss": 0.5017, + "mean_token_accuracy": 0.8355900049209595, + "num_tokens": 29198384.0, + "step": 835 + }, + { + "epoch": 0.1552460538532962, + "grad_norm": 1.632037878036499, + "learning_rate": 5.167079207920791e-07, + "loss": 0.4857, + "mean_token_accuracy": 0.84271240234375, + "num_tokens": 29233316.0, + "step": 836 + }, + { + "epoch": 0.1554317548746518, + "grad_norm": 1.5702749490737915, + "learning_rate": 5.173267326732673e-07, + "loss": 0.4406, + "mean_token_accuracy": 0.8544520139694214, + "num_tokens": 29266350.0, + "step": 837 + }, + { + "epoch": 0.15561745589600742, + "grad_norm": 1.7142969369888306, + "learning_rate": 5.179455445544554e-07, + "loss": 0.4414, + "mean_token_accuracy": 0.8539483547210693, + "num_tokens": 29294143.0, + "step": 838 + }, + { + "epoch": 0.15580315691736304, + "grad_norm": 1.632889986038208, + "learning_rate": 5.185643564356435e-07, + "loss": 0.5001, + "mean_token_accuracy": 0.8373359441757202, + "num_tokens": 29331367.0, + "step": 839 + }, + { + "epoch": 0.15598885793871867, + "grad_norm": 1.6826059818267822, + "learning_rate": 5.191831683168316e-07, + "loss": 0.4906, + "mean_token_accuracy": 0.8364880681037903, + "num_tokens": 29360729.0, + "step": 840 + }, + { + "epoch": 0.1561745589600743, + "grad_norm": 1.5071558952331543, + "learning_rate": 5.198019801980198e-07, + "loss": 0.4741, + "mean_token_accuracy": 0.8470510244369507, + "num_tokens": 29399326.0, + "step": 841 + }, + { + "epoch": 0.1563602599814299, + "grad_norm": 1.4906725883483887, + "learning_rate": 5.204207920792078e-07, + "loss": 0.5126, + "mean_token_accuracy": 0.8364264369010925, + "num_tokens": 29438571.0, + "step": 842 + }, + { + "epoch": 0.1565459610027855, + "grad_norm": 1.4945049285888672, + "learning_rate": 5.210396039603961e-07, + "loss": 0.4121, + "mean_token_accuracy": 0.8625931739807129, + "num_tokens": 29470222.0, + "step": 843 + }, + { + "epoch": 0.15673166202414113, + "grad_norm": 1.6364635229110718, + "learning_rate": 5.216584158415841e-07, + "loss": 0.4936, + "mean_token_accuracy": 0.8396384716033936, + "num_tokens": 29503559.0, + "step": 844 + }, + { + "epoch": 0.15691736304549675, + "grad_norm": 1.6063969135284424, + "learning_rate": 5.222772277227723e-07, + "loss": 0.4726, + "mean_token_accuracy": 0.8489487767219543, + "num_tokens": 29535515.0, + "step": 845 + }, + { + "epoch": 0.15710306406685237, + "grad_norm": 1.7220953702926636, + "learning_rate": 5.228960396039604e-07, + "loss": 0.4955, + "mean_token_accuracy": 0.8364148139953613, + "num_tokens": 29566719.0, + "step": 846 + }, + { + "epoch": 0.157288765088208, + "grad_norm": 1.5457355976104736, + "learning_rate": 5.235148514851485e-07, + "loss": 0.4477, + "mean_token_accuracy": 0.8570770025253296, + "num_tokens": 29601293.0, + "step": 847 + }, + { + "epoch": 0.15747446610956362, + "grad_norm": 1.5344629287719727, + "learning_rate": 5.241336633663366e-07, + "loss": 0.5202, + "mean_token_accuracy": 0.8326659202575684, + "num_tokens": 29641925.0, + "step": 848 + }, + { + "epoch": 0.1576601671309192, + "grad_norm": 1.717697262763977, + "learning_rate": 5.247524752475247e-07, + "loss": 0.4716, + "mean_token_accuracy": 0.8433618545532227, + "num_tokens": 29670333.0, + "step": 849 + }, + { + "epoch": 0.15784586815227483, + "grad_norm": 1.5248788595199585, + "learning_rate": 5.253712871287128e-07, + "loss": 0.4973, + "mean_token_accuracy": 0.8407820463180542, + "num_tokens": 29707224.0, + "step": 850 + }, + { + "epoch": 0.15803156917363045, + "grad_norm": 1.4873743057250977, + "learning_rate": 5.25990099009901e-07, + "loss": 0.4662, + "mean_token_accuracy": 0.8461219668388367, + "num_tokens": 29747537.0, + "step": 851 + }, + { + "epoch": 0.15821727019498608, + "grad_norm": 1.5400516986846924, + "learning_rate": 5.26608910891089e-07, + "loss": 0.5366, + "mean_token_accuracy": 0.8307201862335205, + "num_tokens": 29787454.0, + "step": 852 + }, + { + "epoch": 0.1584029712163417, + "grad_norm": 1.3876968622207642, + "learning_rate": 5.272277227722772e-07, + "loss": 0.4723, + "mean_token_accuracy": 0.8488368988037109, + "num_tokens": 29828118.0, + "step": 853 + }, + { + "epoch": 0.15858867223769732, + "grad_norm": 1.3999327421188354, + "learning_rate": 5.278465346534653e-07, + "loss": 0.4541, + "mean_token_accuracy": 0.8536204695701599, + "num_tokens": 29867283.0, + "step": 854 + }, + { + "epoch": 0.15877437325905291, + "grad_norm": 1.5797796249389648, + "learning_rate": 5.284653465346534e-07, + "loss": 0.4376, + "mean_token_accuracy": 0.8558464646339417, + "num_tokens": 29898927.0, + "step": 855 + }, + { + "epoch": 0.15896007428040854, + "grad_norm": 1.4426690340042114, + "learning_rate": 5.290841584158416e-07, + "loss": 0.4587, + "mean_token_accuracy": 0.8504551649093628, + "num_tokens": 29935733.0, + "step": 856 + }, + { + "epoch": 0.15914577530176416, + "grad_norm": 1.503636121749878, + "learning_rate": 5.297029702970297e-07, + "loss": 0.4619, + "mean_token_accuracy": 0.8478189706802368, + "num_tokens": 29972181.0, + "step": 857 + }, + { + "epoch": 0.15933147632311978, + "grad_norm": 1.4859979152679443, + "learning_rate": 5.303217821782178e-07, + "loss": 0.4953, + "mean_token_accuracy": 0.8430246710777283, + "num_tokens": 30012554.0, + "step": 858 + }, + { + "epoch": 0.1595171773444754, + "grad_norm": 1.5797178745269775, + "learning_rate": 5.30940594059406e-07, + "loss": 0.5013, + "mean_token_accuracy": 0.8377883434295654, + "num_tokens": 30046547.0, + "step": 859 + }, + { + "epoch": 0.15970287836583102, + "grad_norm": 1.4501800537109375, + "learning_rate": 5.31559405940594e-07, + "loss": 0.4576, + "mean_token_accuracy": 0.8531315326690674, + "num_tokens": 30085941.0, + "step": 860 + }, + { + "epoch": 0.15988857938718662, + "grad_norm": 1.7931714057922363, + "learning_rate": 5.321782178217822e-07, + "loss": 0.5226, + "mean_token_accuracy": 0.8375471830368042, + "num_tokens": 30115289.0, + "step": 861 + }, + { + "epoch": 0.16007428040854224, + "grad_norm": 1.4619594812393188, + "learning_rate": 5.327970297029702e-07, + "loss": 0.4989, + "mean_token_accuracy": 0.8414062261581421, + "num_tokens": 30157993.0, + "step": 862 + }, + { + "epoch": 0.16025998142989786, + "grad_norm": 1.4268079996109009, + "learning_rate": 5.334158415841584e-07, + "loss": 0.4269, + "mean_token_accuracy": 0.8625410795211792, + "num_tokens": 30195076.0, + "step": 863 + }, + { + "epoch": 0.16044568245125349, + "grad_norm": 1.722464919090271, + "learning_rate": 5.340346534653465e-07, + "loss": 0.497, + "mean_token_accuracy": 0.8399566411972046, + "num_tokens": 30224548.0, + "step": 864 + }, + { + "epoch": 0.1606313834726091, + "grad_norm": 1.7233291864395142, + "learning_rate": 5.346534653465346e-07, + "loss": 0.5399, + "mean_token_accuracy": 0.8301750421524048, + "num_tokens": 30255228.0, + "step": 865 + }, + { + "epoch": 0.16081708449396473, + "grad_norm": 1.5603468418121338, + "learning_rate": 5.352722772277227e-07, + "loss": 0.496, + "mean_token_accuracy": 0.8403260111808777, + "num_tokens": 30292328.0, + "step": 866 + }, + { + "epoch": 0.16100278551532032, + "grad_norm": 1.741909146308899, + "learning_rate": 5.358910891089109e-07, + "loss": 0.4865, + "mean_token_accuracy": 0.8429231643676758, + "num_tokens": 30322952.0, + "step": 867 + }, + { + "epoch": 0.16118848653667595, + "grad_norm": 1.602752685546875, + "learning_rate": 5.365099009900989e-07, + "loss": 0.4599, + "mean_token_accuracy": 0.8500670194625854, + "num_tokens": 30354707.0, + "step": 868 + }, + { + "epoch": 0.16137418755803157, + "grad_norm": 1.6292188167572021, + "learning_rate": 5.371287128712872e-07, + "loss": 0.4655, + "mean_token_accuracy": 0.8461062908172607, + "num_tokens": 30387624.0, + "step": 869 + }, + { + "epoch": 0.1615598885793872, + "grad_norm": 1.441520094871521, + "learning_rate": 5.377475247524752e-07, + "loss": 0.456, + "mean_token_accuracy": 0.8540410399436951, + "num_tokens": 30425657.0, + "step": 870 + }, + { + "epoch": 0.1617455896007428, + "grad_norm": 1.5750640630722046, + "learning_rate": 5.383663366336634e-07, + "loss": 0.4562, + "mean_token_accuracy": 0.8518888354301453, + "num_tokens": 30463090.0, + "step": 871 + }, + { + "epoch": 0.16193129062209843, + "grad_norm": 1.7740108966827393, + "learning_rate": 5.389851485148515e-07, + "loss": 0.5106, + "mean_token_accuracy": 0.8340780735015869, + "num_tokens": 30495799.0, + "step": 872 + }, + { + "epoch": 0.16211699164345403, + "grad_norm": 1.6500680446624756, + "learning_rate": 5.396039603960396e-07, + "loss": 0.4987, + "mean_token_accuracy": 0.8419350385665894, + "num_tokens": 30528322.0, + "step": 873 + }, + { + "epoch": 0.16230269266480965, + "grad_norm": 1.3740317821502686, + "learning_rate": 5.402227722772277e-07, + "loss": 0.5007, + "mean_token_accuracy": 0.8367500305175781, + "num_tokens": 30572654.0, + "step": 874 + }, + { + "epoch": 0.16248839368616527, + "grad_norm": 1.718206524848938, + "learning_rate": 5.408415841584159e-07, + "loss": 0.4737, + "mean_token_accuracy": 0.8469717502593994, + "num_tokens": 30601666.0, + "step": 875 + }, + { + "epoch": 0.1626740947075209, + "grad_norm": 1.6712700128555298, + "learning_rate": 5.414603960396039e-07, + "loss": 0.4892, + "mean_token_accuracy": 0.8453923463821411, + "num_tokens": 30631087.0, + "step": 876 + }, + { + "epoch": 0.16285979572887652, + "grad_norm": 1.454543113708496, + "learning_rate": 5.420792079207921e-07, + "loss": 0.4981, + "mean_token_accuracy": 0.8419713973999023, + "num_tokens": 30672821.0, + "step": 877 + }, + { + "epoch": 0.16304549675023214, + "grad_norm": 1.5844913721084595, + "learning_rate": 5.426980198019801e-07, + "loss": 0.4671, + "mean_token_accuracy": 0.8485171794891357, + "num_tokens": 30705499.0, + "step": 878 + }, + { + "epoch": 0.16323119777158773, + "grad_norm": 1.6046355962753296, + "learning_rate": 5.433168316831683e-07, + "loss": 0.4345, + "mean_token_accuracy": 0.857948362827301, + "num_tokens": 30741451.0, + "step": 879 + }, + { + "epoch": 0.16341689879294335, + "grad_norm": 1.71597158908844, + "learning_rate": 5.439356435643564e-07, + "loss": 0.4437, + "mean_token_accuracy": 0.851965606212616, + "num_tokens": 30773515.0, + "step": 880 + }, + { + "epoch": 0.16360259981429898, + "grad_norm": 1.5747416019439697, + "learning_rate": 5.445544554455445e-07, + "loss": 0.4612, + "mean_token_accuracy": 0.8511077165603638, + "num_tokens": 30807794.0, + "step": 881 + }, + { + "epoch": 0.1637883008356546, + "grad_norm": 1.461687445640564, + "learning_rate": 5.451732673267327e-07, + "loss": 0.4796, + "mean_token_accuracy": 0.8426426649093628, + "num_tokens": 30846398.0, + "step": 882 + }, + { + "epoch": 0.16397400185701022, + "grad_norm": 1.7020009756088257, + "learning_rate": 5.457920792079208e-07, + "loss": 0.5233, + "mean_token_accuracy": 0.8342479467391968, + "num_tokens": 30877969.0, + "step": 883 + }, + { + "epoch": 0.16415970287836584, + "grad_norm": 1.806984543800354, + "learning_rate": 5.464108910891089e-07, + "loss": 0.4778, + "mean_token_accuracy": 0.8475086688995361, + "num_tokens": 30906001.0, + "step": 884 + }, + { + "epoch": 0.16434540389972144, + "grad_norm": 1.8196829557418823, + "learning_rate": 5.470297029702971e-07, + "loss": 0.4954, + "mean_token_accuracy": 0.8369537591934204, + "num_tokens": 30938301.0, + "step": 885 + }, + { + "epoch": 0.16453110492107706, + "grad_norm": 1.7519032955169678, + "learning_rate": 5.476485148514851e-07, + "loss": 0.5174, + "mean_token_accuracy": 0.8318129777908325, + "num_tokens": 30973277.0, + "step": 886 + }, + { + "epoch": 0.16471680594243268, + "grad_norm": 1.532757043838501, + "learning_rate": 5.482673267326733e-07, + "loss": 0.4344, + "mean_token_accuracy": 0.8600590229034424, + "num_tokens": 31009983.0, + "step": 887 + }, + { + "epoch": 0.1649025069637883, + "grad_norm": 1.4985839128494263, + "learning_rate": 5.488861386138614e-07, + "loss": 0.5012, + "mean_token_accuracy": 0.8407610058784485, + "num_tokens": 31049986.0, + "step": 888 + }, + { + "epoch": 0.16508820798514393, + "grad_norm": 1.5673892498016357, + "learning_rate": 5.495049504950495e-07, + "loss": 0.4512, + "mean_token_accuracy": 0.8518791198730469, + "num_tokens": 31085953.0, + "step": 889 + }, + { + "epoch": 0.16527390900649955, + "grad_norm": 1.61996591091156, + "learning_rate": 5.501237623762376e-07, + "loss": 0.4469, + "mean_token_accuracy": 0.8533370494842529, + "num_tokens": 31119068.0, + "step": 890 + }, + { + "epoch": 0.16545961002785514, + "grad_norm": 1.6052011251449585, + "learning_rate": 5.507425742574257e-07, + "loss": 0.4196, + "mean_token_accuracy": 0.8635424375534058, + "num_tokens": 31151368.0, + "step": 891 + }, + { + "epoch": 0.16564531104921076, + "grad_norm": 1.5313719511032104, + "learning_rate": 5.513613861386138e-07, + "loss": 0.4709, + "mean_token_accuracy": 0.8459845185279846, + "num_tokens": 31190084.0, + "step": 892 + }, + { + "epoch": 0.16583101207056639, + "grad_norm": 1.4847891330718994, + "learning_rate": 5.51980198019802e-07, + "loss": 0.4261, + "mean_token_accuracy": 0.8577818870544434, + "num_tokens": 31224516.0, + "step": 893 + }, + { + "epoch": 0.166016713091922, + "grad_norm": 1.5081522464752197, + "learning_rate": 5.5259900990099e-07, + "loss": 0.4524, + "mean_token_accuracy": 0.8515865802764893, + "num_tokens": 31259455.0, + "step": 894 + }, + { + "epoch": 0.16620241411327763, + "grad_norm": 1.5142757892608643, + "learning_rate": 5.532178217821783e-07, + "loss": 0.4414, + "mean_token_accuracy": 0.8542181849479675, + "num_tokens": 31295351.0, + "step": 895 + }, + { + "epoch": 0.16638811513463325, + "grad_norm": 1.4929407835006714, + "learning_rate": 5.538366336633663e-07, + "loss": 0.5037, + "mean_token_accuracy": 0.8363795280456543, + "num_tokens": 31334517.0, + "step": 896 + }, + { + "epoch": 0.16657381615598885, + "grad_norm": 1.4776026010513306, + "learning_rate": 5.544554455445545e-07, + "loss": 0.4435, + "mean_token_accuracy": 0.8535545468330383, + "num_tokens": 31374186.0, + "step": 897 + }, + { + "epoch": 0.16675951717734447, + "grad_norm": 1.557951807975769, + "learning_rate": 5.550742574257426e-07, + "loss": 0.4891, + "mean_token_accuracy": 0.8411794304847717, + "num_tokens": 31412037.0, + "step": 898 + }, + { + "epoch": 0.1669452181987001, + "grad_norm": 1.4451712369918823, + "learning_rate": 5.556930693069307e-07, + "loss": 0.4585, + "mean_token_accuracy": 0.8502227067947388, + "num_tokens": 31453816.0, + "step": 899 + }, + { + "epoch": 0.1671309192200557, + "grad_norm": 1.5923750400543213, + "learning_rate": 5.563118811881188e-07, + "loss": 0.5022, + "mean_token_accuracy": 0.8376522064208984, + "num_tokens": 31489019.0, + "step": 900 + }, + { + "epoch": 0.16731662024141133, + "grad_norm": 1.5639921426773071, + "learning_rate": 5.56930693069307e-07, + "loss": 0.4847, + "mean_token_accuracy": 0.8441964983940125, + "num_tokens": 31526061.0, + "step": 901 + }, + { + "epoch": 0.16750232126276696, + "grad_norm": 1.755102515220642, + "learning_rate": 5.57549504950495e-07, + "loss": 0.4635, + "mean_token_accuracy": 0.8460829257965088, + "num_tokens": 31556632.0, + "step": 902 + }, + { + "epoch": 0.16768802228412255, + "grad_norm": 1.4850378036499023, + "learning_rate": 5.581683168316832e-07, + "loss": 0.4824, + "mean_token_accuracy": 0.8409733176231384, + "num_tokens": 31596076.0, + "step": 903 + }, + { + "epoch": 0.16787372330547817, + "grad_norm": 1.433139681816101, + "learning_rate": 5.587871287128712e-07, + "loss": 0.4728, + "mean_token_accuracy": 0.8469420671463013, + "num_tokens": 31640073.0, + "step": 904 + }, + { + "epoch": 0.1680594243268338, + "grad_norm": 1.4510232210159302, + "learning_rate": 5.594059405940594e-07, + "loss": 0.4424, + "mean_token_accuracy": 0.852601945400238, + "num_tokens": 31676764.0, + "step": 905 + }, + { + "epoch": 0.16824512534818942, + "grad_norm": 1.5868135690689087, + "learning_rate": 5.600247524752475e-07, + "loss": 0.4392, + "mean_token_accuracy": 0.858235239982605, + "num_tokens": 31705898.0, + "step": 906 + }, + { + "epoch": 0.16843082636954504, + "grad_norm": 1.520656943321228, + "learning_rate": 5.606435643564356e-07, + "loss": 0.4834, + "mean_token_accuracy": 0.8415408134460449, + "num_tokens": 31740644.0, + "step": 907 + }, + { + "epoch": 0.16861652739090066, + "grad_norm": 1.5188664197921753, + "learning_rate": 5.612623762376237e-07, + "loss": 0.4561, + "mean_token_accuracy": 0.8533126711845398, + "num_tokens": 31776793.0, + "step": 908 + }, + { + "epoch": 0.16880222841225626, + "grad_norm": 1.5408841371536255, + "learning_rate": 5.61881188118812e-07, + "loss": 0.444, + "mean_token_accuracy": 0.8575922250747681, + "num_tokens": 31814167.0, + "step": 909 + }, + { + "epoch": 0.16898792943361188, + "grad_norm": 1.51926851272583, + "learning_rate": 5.625e-07, + "loss": 0.4727, + "mean_token_accuracy": 0.8503530025482178, + "num_tokens": 31849033.0, + "step": 910 + }, + { + "epoch": 0.1691736304549675, + "grad_norm": 1.6069588661193848, + "learning_rate": 5.631188118811881e-07, + "loss": 0.4928, + "mean_token_accuracy": 0.8400530815124512, + "num_tokens": 31882917.0, + "step": 911 + }, + { + "epoch": 0.16935933147632312, + "grad_norm": 1.7233785390853882, + "learning_rate": 5.637376237623762e-07, + "loss": 0.4882, + "mean_token_accuracy": 0.8418260216712952, + "num_tokens": 31912075.0, + "step": 912 + }, + { + "epoch": 0.16954503249767874, + "grad_norm": 1.6015000343322754, + "learning_rate": 5.643564356435643e-07, + "loss": 0.4578, + "mean_token_accuracy": 0.8487759828567505, + "num_tokens": 31946212.0, + "step": 913 + }, + { + "epoch": 0.16973073351903437, + "grad_norm": 1.702629804611206, + "learning_rate": 5.649752475247525e-07, + "loss": 0.4655, + "mean_token_accuracy": 0.8497940897941589, + "num_tokens": 31974730.0, + "step": 914 + }, + { + "epoch": 0.16991643454038996, + "grad_norm": 1.6740227937698364, + "learning_rate": 5.655940594059405e-07, + "loss": 0.4807, + "mean_token_accuracy": 0.8443004488945007, + "num_tokens": 32008277.0, + "step": 915 + }, + { + "epoch": 0.17010213556174558, + "grad_norm": 1.5044589042663574, + "learning_rate": 5.662128712871287e-07, + "loss": 0.5059, + "mean_token_accuracy": 0.8362588882446289, + "num_tokens": 32045331.0, + "step": 916 + }, + { + "epoch": 0.1702878365831012, + "grad_norm": 1.5828399658203125, + "learning_rate": 5.668316831683167e-07, + "loss": 0.4527, + "mean_token_accuracy": 0.8518726825714111, + "num_tokens": 32078357.0, + "step": 917 + }, + { + "epoch": 0.17047353760445683, + "grad_norm": 1.6658480167388916, + "learning_rate": 5.674504950495049e-07, + "loss": 0.4594, + "mean_token_accuracy": 0.8570519685745239, + "num_tokens": 32110211.0, + "step": 918 + }, + { + "epoch": 0.17065923862581245, + "grad_norm": 1.6181355714797974, + "learning_rate": 5.68069306930693e-07, + "loss": 0.5315, + "mean_token_accuracy": 0.8331624269485474, + "num_tokens": 32150602.0, + "step": 919 + }, + { + "epoch": 0.17084493964716807, + "grad_norm": 1.6800541877746582, + "learning_rate": 5.686881188118811e-07, + "loss": 0.4833, + "mean_token_accuracy": 0.8379623889923096, + "num_tokens": 32185820.0, + "step": 920 + }, + { + "epoch": 0.17103064066852366, + "grad_norm": 1.6102930307388306, + "learning_rate": 5.693069306930692e-07, + "loss": 0.476, + "mean_token_accuracy": 0.8496631383895874, + "num_tokens": 32218881.0, + "step": 921 + }, + { + "epoch": 0.1712163416898793, + "grad_norm": 1.593265414237976, + "learning_rate": 5.699257425742575e-07, + "loss": 0.4879, + "mean_token_accuracy": 0.8427813053131104, + "num_tokens": 32253998.0, + "step": 922 + }, + { + "epoch": 0.1714020427112349, + "grad_norm": 1.4365296363830566, + "learning_rate": 5.705445544554455e-07, + "loss": 0.4434, + "mean_token_accuracy": 0.8542829155921936, + "num_tokens": 32294008.0, + "step": 923 + }, + { + "epoch": 0.17158774373259053, + "grad_norm": 1.513513445854187, + "learning_rate": 5.711633663366337e-07, + "loss": 0.4304, + "mean_token_accuracy": 0.8563957214355469, + "num_tokens": 32327168.0, + "step": 924 + }, + { + "epoch": 0.17177344475394615, + "grad_norm": 1.3871961832046509, + "learning_rate": 5.717821782178217e-07, + "loss": 0.5189, + "mean_token_accuracy": 0.8328711986541748, + "num_tokens": 32371394.0, + "step": 925 + }, + { + "epoch": 0.17195914577530177, + "grad_norm": 1.644552230834961, + "learning_rate": 5.724009900990099e-07, + "loss": 0.4636, + "mean_token_accuracy": 0.8527946472167969, + "num_tokens": 32403125.0, + "step": 926 + }, + { + "epoch": 0.17214484679665737, + "grad_norm": 1.5336397886276245, + "learning_rate": 5.73019801980198e-07, + "loss": 0.4317, + "mean_token_accuracy": 0.8576558232307434, + "num_tokens": 32434151.0, + "step": 927 + }, + { + "epoch": 0.172330547818013, + "grad_norm": 1.480710744857788, + "learning_rate": 5.736386138613861e-07, + "loss": 0.4558, + "mean_token_accuracy": 0.856432318687439, + "num_tokens": 32473440.0, + "step": 928 + }, + { + "epoch": 0.1725162488393686, + "grad_norm": 1.5014722347259521, + "learning_rate": 5.742574257425742e-07, + "loss": 0.4453, + "mean_token_accuracy": 0.852313756942749, + "num_tokens": 32509839.0, + "step": 929 + }, + { + "epoch": 0.17270194986072424, + "grad_norm": 1.5404350757598877, + "learning_rate": 5.748762376237623e-07, + "loss": 0.4706, + "mean_token_accuracy": 0.8461785316467285, + "num_tokens": 32545261.0, + "step": 930 + }, + { + "epoch": 0.17288765088207986, + "grad_norm": 1.6055543422698975, + "learning_rate": 5.754950495049504e-07, + "loss": 0.4907, + "mean_token_accuracy": 0.8431409001350403, + "num_tokens": 32577560.0, + "step": 931 + }, + { + "epoch": 0.17307335190343548, + "grad_norm": 1.6678928136825562, + "learning_rate": 5.761138613861386e-07, + "loss": 0.4722, + "mean_token_accuracy": 0.8451206088066101, + "num_tokens": 32605885.0, + "step": 932 + }, + { + "epoch": 0.17325905292479107, + "grad_norm": 1.6863374710083008, + "learning_rate": 5.767326732673266e-07, + "loss": 0.4568, + "mean_token_accuracy": 0.8541457653045654, + "num_tokens": 32635123.0, + "step": 933 + }, + { + "epoch": 0.1734447539461467, + "grad_norm": 1.5018740892410278, + "learning_rate": 5.773514851485148e-07, + "loss": 0.4897, + "mean_token_accuracy": 0.8408173322677612, + "num_tokens": 32676474.0, + "step": 934 + }, + { + "epoch": 0.17363045496750232, + "grad_norm": 1.5443536043167114, + "learning_rate": 5.779702970297029e-07, + "loss": 0.4627, + "mean_token_accuracy": 0.8475151658058167, + "num_tokens": 32712596.0, + "step": 935 + }, + { + "epoch": 0.17381615598885794, + "grad_norm": 1.6354035139083862, + "learning_rate": 5.785891089108911e-07, + "loss": 0.5069, + "mean_token_accuracy": 0.8361011743545532, + "num_tokens": 32746308.0, + "step": 936 + }, + { + "epoch": 0.17400185701021356, + "grad_norm": 1.5172144174575806, + "learning_rate": 5.792079207920792e-07, + "loss": 0.4487, + "mean_token_accuracy": 0.8522474765777588, + "num_tokens": 32783352.0, + "step": 937 + }, + { + "epoch": 0.17418755803156918, + "grad_norm": 1.6058498620986938, + "learning_rate": 5.798267326732673e-07, + "loss": 0.448, + "mean_token_accuracy": 0.855571448802948, + "num_tokens": 32814673.0, + "step": 938 + }, + { + "epoch": 0.17437325905292478, + "grad_norm": 1.5221928358078003, + "learning_rate": 5.804455445544554e-07, + "loss": 0.4408, + "mean_token_accuracy": 0.8570124506950378, + "num_tokens": 32849631.0, + "step": 939 + }, + { + "epoch": 0.1745589600742804, + "grad_norm": 1.7127875089645386, + "learning_rate": 5.810643564356436e-07, + "loss": 0.4794, + "mean_token_accuracy": 0.8424862623214722, + "num_tokens": 32879281.0, + "step": 940 + }, + { + "epoch": 0.17474466109563602, + "grad_norm": 1.4210158586502075, + "learning_rate": 5.816831683168316e-07, + "loss": 0.4477, + "mean_token_accuracy": 0.8535293340682983, + "num_tokens": 32919830.0, + "step": 941 + }, + { + "epoch": 0.17493036211699164, + "grad_norm": 1.457305908203125, + "learning_rate": 5.823019801980198e-07, + "loss": 0.4553, + "mean_token_accuracy": 0.8506067991256714, + "num_tokens": 32957555.0, + "step": 942 + }, + { + "epoch": 0.17511606313834727, + "grad_norm": 1.730286955833435, + "learning_rate": 5.829207920792078e-07, + "loss": 0.4958, + "mean_token_accuracy": 0.8399770259857178, + "num_tokens": 32988501.0, + "step": 943 + }, + { + "epoch": 0.1753017641597029, + "grad_norm": 1.6282436847686768, + "learning_rate": 5.83539603960396e-07, + "loss": 0.5825, + "mean_token_accuracy": 0.81500643491745, + "num_tokens": 33022216.0, + "step": 944 + }, + { + "epoch": 0.17548746518105848, + "grad_norm": 1.6586447954177856, + "learning_rate": 5.841584158415841e-07, + "loss": 0.4492, + "mean_token_accuracy": 0.8581567406654358, + "num_tokens": 33056199.0, + "step": 945 + }, + { + "epoch": 0.1756731662024141, + "grad_norm": 1.4231637716293335, + "learning_rate": 5.847772277227722e-07, + "loss": 0.4482, + "mean_token_accuracy": 0.8539220094680786, + "num_tokens": 33093882.0, + "step": 946 + }, + { + "epoch": 0.17585886722376973, + "grad_norm": 1.4810374975204468, + "learning_rate": 5.853960396039603e-07, + "loss": 0.4455, + "mean_token_accuracy": 0.8565616607666016, + "num_tokens": 33132105.0, + "step": 947 + }, + { + "epoch": 0.17604456824512535, + "grad_norm": 1.554569959640503, + "learning_rate": 5.860148514851486e-07, + "loss": 0.5283, + "mean_token_accuracy": 0.8255206346511841, + "num_tokens": 33171955.0, + "step": 948 + }, + { + "epoch": 0.17623026926648097, + "grad_norm": 1.5321853160858154, + "learning_rate": 5.866336633663366e-07, + "loss": 0.4199, + "mean_token_accuracy": 0.861107587814331, + "num_tokens": 33207300.0, + "step": 949 + }, + { + "epoch": 0.1764159702878366, + "grad_norm": 1.4551554918289185, + "learning_rate": 5.872524752475248e-07, + "loss": 0.4344, + "mean_token_accuracy": 0.8569049835205078, + "num_tokens": 33249238.0, + "step": 950 + }, + { + "epoch": 0.1766016713091922, + "grad_norm": 1.421730399131775, + "learning_rate": 5.878712871287128e-07, + "loss": 0.4541, + "mean_token_accuracy": 0.8535851240158081, + "num_tokens": 33288636.0, + "step": 951 + }, + { + "epoch": 0.1767873723305478, + "grad_norm": 1.628795862197876, + "learning_rate": 5.88490099009901e-07, + "loss": 0.4673, + "mean_token_accuracy": 0.848853349685669, + "num_tokens": 33321751.0, + "step": 952 + }, + { + "epoch": 0.17697307335190343, + "grad_norm": 1.5722757577896118, + "learning_rate": 5.891089108910891e-07, + "loss": 0.4974, + "mean_token_accuracy": 0.8404688835144043, + "num_tokens": 33356531.0, + "step": 953 + }, + { + "epoch": 0.17715877437325905, + "grad_norm": 1.8195515871047974, + "learning_rate": 5.897277227722772e-07, + "loss": 0.4672, + "mean_token_accuracy": 0.8513277173042297, + "num_tokens": 33384283.0, + "step": 954 + }, + { + "epoch": 0.17734447539461468, + "grad_norm": 1.6485906839370728, + "learning_rate": 5.903465346534653e-07, + "loss": 0.4694, + "mean_token_accuracy": 0.8533116579055786, + "num_tokens": 33416890.0, + "step": 955 + }, + { + "epoch": 0.1775301764159703, + "grad_norm": 1.493458867073059, + "learning_rate": 5.909653465346535e-07, + "loss": 0.4263, + "mean_token_accuracy": 0.8602603673934937, + "num_tokens": 33455354.0, + "step": 956 + }, + { + "epoch": 0.1777158774373259, + "grad_norm": 1.4251010417938232, + "learning_rate": 5.915841584158415e-07, + "loss": 0.4695, + "mean_token_accuracy": 0.8466043472290039, + "num_tokens": 33494547.0, + "step": 957 + }, + { + "epoch": 0.17790157845868151, + "grad_norm": 1.6107829809188843, + "learning_rate": 5.922029702970297e-07, + "loss": 0.4941, + "mean_token_accuracy": 0.8395689725875854, + "num_tokens": 33530098.0, + "step": 958 + }, + { + "epoch": 0.17808727948003714, + "grad_norm": 1.4853886365890503, + "learning_rate": 5.928217821782177e-07, + "loss": 0.4815, + "mean_token_accuracy": 0.8435987234115601, + "num_tokens": 33568166.0, + "step": 959 + }, + { + "epoch": 0.17827298050139276, + "grad_norm": 1.6288955211639404, + "learning_rate": 5.934405940594059e-07, + "loss": 0.4266, + "mean_token_accuracy": 0.8644233345985413, + "num_tokens": 33601497.0, + "step": 960 + }, + { + "epoch": 0.17845868152274838, + "grad_norm": 1.4358787536621094, + "learning_rate": 5.94059405940594e-07, + "loss": 0.4861, + "mean_token_accuracy": 0.8381775617599487, + "num_tokens": 33643039.0, + "step": 961 + }, + { + "epoch": 0.178644382544104, + "grad_norm": 1.3587751388549805, + "learning_rate": 5.946782178217822e-07, + "loss": 0.435, + "mean_token_accuracy": 0.855964183807373, + "num_tokens": 33684743.0, + "step": 962 + }, + { + "epoch": 0.1788300835654596, + "grad_norm": 1.5789662599563599, + "learning_rate": 5.952970297029703e-07, + "loss": 0.4429, + "mean_token_accuracy": 0.8548628687858582, + "num_tokens": 33714692.0, + "step": 963 + }, + { + "epoch": 0.17901578458681522, + "grad_norm": 1.5568979978561401, + "learning_rate": 5.959158415841584e-07, + "loss": 0.4451, + "mean_token_accuracy": 0.8532006740570068, + "num_tokens": 33745807.0, + "step": 964 + }, + { + "epoch": 0.17920148560817084, + "grad_norm": 1.740329623222351, + "learning_rate": 5.965346534653465e-07, + "loss": 0.4637, + "mean_token_accuracy": 0.8499666452407837, + "num_tokens": 33773096.0, + "step": 965 + }, + { + "epoch": 0.17938718662952646, + "grad_norm": 1.6490665674209595, + "learning_rate": 5.971534653465347e-07, + "loss": 0.47, + "mean_token_accuracy": 0.8491395711898804, + "num_tokens": 33802898.0, + "step": 966 + }, + { + "epoch": 0.17957288765088208, + "grad_norm": 1.521094799041748, + "learning_rate": 5.977722772277227e-07, + "loss": 0.4783, + "mean_token_accuracy": 0.8473852872848511, + "num_tokens": 33836031.0, + "step": 967 + }, + { + "epoch": 0.1797585886722377, + "grad_norm": 1.7051059007644653, + "learning_rate": 5.983910891089109e-07, + "loss": 0.4134, + "mean_token_accuracy": 0.8674113750457764, + "num_tokens": 33867443.0, + "step": 968 + }, + { + "epoch": 0.1799442896935933, + "grad_norm": 1.6155893802642822, + "learning_rate": 5.99009900990099e-07, + "loss": 0.4743, + "mean_token_accuracy": 0.8462842702865601, + "num_tokens": 33898230.0, + "step": 969 + }, + { + "epoch": 0.18012999071494892, + "grad_norm": 1.6151524782180786, + "learning_rate": 5.996287128712871e-07, + "loss": 0.4715, + "mean_token_accuracy": 0.8471486568450928, + "num_tokens": 33931513.0, + "step": 970 + }, + { + "epoch": 0.18031569173630455, + "grad_norm": 1.5334842205047607, + "learning_rate": 6.002475247524752e-07, + "loss": 0.429, + "mean_token_accuracy": 0.8601546883583069, + "num_tokens": 33965675.0, + "step": 971 + }, + { + "epoch": 0.18050139275766017, + "grad_norm": 1.7083016633987427, + "learning_rate": 6.008663366336633e-07, + "loss": 0.4598, + "mean_token_accuracy": 0.8518246412277222, + "num_tokens": 33998877.0, + "step": 972 + }, + { + "epoch": 0.1806870937790158, + "grad_norm": 1.570328950881958, + "learning_rate": 6.014851485148514e-07, + "loss": 0.4474, + "mean_token_accuracy": 0.853256344795227, + "num_tokens": 34034074.0, + "step": 973 + }, + { + "epoch": 0.1808727948003714, + "grad_norm": 1.8059966564178467, + "learning_rate": 6.021039603960396e-07, + "loss": 0.4684, + "mean_token_accuracy": 0.8488630056381226, + "num_tokens": 34059832.0, + "step": 974 + }, + { + "epoch": 0.181058495821727, + "grad_norm": 1.5279606580734253, + "learning_rate": 6.027227722772277e-07, + "loss": 0.419, + "mean_token_accuracy": 0.8625336289405823, + "num_tokens": 34093620.0, + "step": 975 + }, + { + "epoch": 0.18124419684308263, + "grad_norm": 1.423856496810913, + "learning_rate": 6.033415841584159e-07, + "loss": 0.4593, + "mean_token_accuracy": 0.8513920307159424, + "num_tokens": 34132834.0, + "step": 976 + }, + { + "epoch": 0.18142989786443825, + "grad_norm": 1.5324586629867554, + "learning_rate": 6.03960396039604e-07, + "loss": 0.4931, + "mean_token_accuracy": 0.8429000377655029, + "num_tokens": 34169922.0, + "step": 977 + }, + { + "epoch": 0.18161559888579387, + "grad_norm": 1.5105315446853638, + "learning_rate": 6.045792079207921e-07, + "loss": 0.4251, + "mean_token_accuracy": 0.8594315052032471, + "num_tokens": 34205423.0, + "step": 978 + }, + { + "epoch": 0.1818012999071495, + "grad_norm": 1.5016509294509888, + "learning_rate": 6.051980198019802e-07, + "loss": 0.4666, + "mean_token_accuracy": 0.8450958728790283, + "num_tokens": 34240215.0, + "step": 979 + }, + { + "epoch": 0.18198700092850512, + "grad_norm": 1.4692977666854858, + "learning_rate": 6.058168316831683e-07, + "loss": 0.4499, + "mean_token_accuracy": 0.8528468608856201, + "num_tokens": 34278730.0, + "step": 980 + }, + { + "epoch": 0.18217270194986074, + "grad_norm": 1.6101914644241333, + "learning_rate": 6.064356435643564e-07, + "loss": 0.4799, + "mean_token_accuracy": 0.8427040576934814, + "num_tokens": 34312740.0, + "step": 981 + }, + { + "epoch": 0.18235840297121633, + "grad_norm": 1.5515029430389404, + "learning_rate": 6.070544554455446e-07, + "loss": 0.4587, + "mean_token_accuracy": 0.8499681353569031, + "num_tokens": 34347740.0, + "step": 982 + }, + { + "epoch": 0.18254410399257195, + "grad_norm": 1.6277036666870117, + "learning_rate": 6.076732673267326e-07, + "loss": 0.4418, + "mean_token_accuracy": 0.8554589152336121, + "num_tokens": 34376985.0, + "step": 983 + }, + { + "epoch": 0.18272980501392758, + "grad_norm": 1.5560928583145142, + "learning_rate": 6.082920792079208e-07, + "loss": 0.5067, + "mean_token_accuracy": 0.8365606665611267, + "num_tokens": 34414192.0, + "step": 984 + }, + { + "epoch": 0.1829155060352832, + "grad_norm": 1.5267292261123657, + "learning_rate": 6.089108910891088e-07, + "loss": 0.4936, + "mean_token_accuracy": 0.8388288021087646, + "num_tokens": 34448744.0, + "step": 985 + }, + { + "epoch": 0.18310120705663882, + "grad_norm": 1.62684166431427, + "learning_rate": 6.09529702970297e-07, + "loss": 0.5052, + "mean_token_accuracy": 0.8325832486152649, + "num_tokens": 34485240.0, + "step": 986 + }, + { + "epoch": 0.18328690807799444, + "grad_norm": 1.6881389617919922, + "learning_rate": 6.101485148514851e-07, + "loss": 0.4502, + "mean_token_accuracy": 0.8515995740890503, + "num_tokens": 34515930.0, + "step": 987 + }, + { + "epoch": 0.18347260909935004, + "grad_norm": 1.6027029752731323, + "learning_rate": 6.107673267326733e-07, + "loss": 0.5025, + "mean_token_accuracy": 0.8381980657577515, + "num_tokens": 34553518.0, + "step": 988 + }, + { + "epoch": 0.18365831012070566, + "grad_norm": 1.3657901287078857, + "learning_rate": 6.113861386138614e-07, + "loss": 0.448, + "mean_token_accuracy": 0.8527159690856934, + "num_tokens": 34597487.0, + "step": 989 + }, + { + "epoch": 0.18384401114206128, + "grad_norm": 1.7429687976837158, + "learning_rate": 6.120049504950496e-07, + "loss": 0.4617, + "mean_token_accuracy": 0.8500394225120544, + "num_tokens": 34626365.0, + "step": 990 + }, + { + "epoch": 0.1840297121634169, + "grad_norm": 1.5275764465332031, + "learning_rate": 6.126237623762376e-07, + "loss": 0.5339, + "mean_token_accuracy": 0.8254033327102661, + "num_tokens": 34666287.0, + "step": 991 + }, + { + "epoch": 0.18421541318477253, + "grad_norm": 1.5194909572601318, + "learning_rate": 6.132425742574258e-07, + "loss": 0.4958, + "mean_token_accuracy": 0.8387060165405273, + "num_tokens": 34701767.0, + "step": 992 + }, + { + "epoch": 0.18440111420612815, + "grad_norm": 1.4819986820220947, + "learning_rate": 6.138613861386138e-07, + "loss": 0.4267, + "mean_token_accuracy": 0.8619937896728516, + "num_tokens": 34738026.0, + "step": 993 + }, + { + "epoch": 0.18458681522748374, + "grad_norm": 1.4116616249084473, + "learning_rate": 6.14480198019802e-07, + "loss": 0.4663, + "mean_token_accuracy": 0.8479812145233154, + "num_tokens": 34783621.0, + "step": 994 + }, + { + "epoch": 0.18477251624883936, + "grad_norm": 1.4590340852737427, + "learning_rate": 6.150990099009901e-07, + "loss": 0.4154, + "mean_token_accuracy": 0.8652547597885132, + "num_tokens": 34819864.0, + "step": 995 + }, + { + "epoch": 0.18495821727019499, + "grad_norm": 1.5773911476135254, + "learning_rate": 6.157178217821782e-07, + "loss": 0.4584, + "mean_token_accuracy": 0.8531967401504517, + "num_tokens": 34855638.0, + "step": 996 + }, + { + "epoch": 0.1851439182915506, + "grad_norm": 1.5951586961746216, + "learning_rate": 6.163366336633663e-07, + "loss": 0.5221, + "mean_token_accuracy": 0.8318403363227844, + "num_tokens": 34892144.0, + "step": 997 + }, + { + "epoch": 0.18532961931290623, + "grad_norm": 1.6498111486434937, + "learning_rate": 6.169554455445544e-07, + "loss": 0.4831, + "mean_token_accuracy": 0.845610499382019, + "num_tokens": 34923286.0, + "step": 998 + }, + { + "epoch": 0.18551532033426185, + "grad_norm": 1.481733798980713, + "learning_rate": 6.175742574257425e-07, + "loss": 0.4266, + "mean_token_accuracy": 0.8601340055465698, + "num_tokens": 34960113.0, + "step": 999 + }, + { + "epoch": 0.18570102135561745, + "grad_norm": 1.3676879405975342, + "learning_rate": 6.181930693069307e-07, + "loss": 0.48, + "mean_token_accuracy": 0.8434886336326599, + "num_tokens": 35003582.0, + "step": 1000 + }, + { + "epoch": 0.18588672237697307, + "grad_norm": 1.5262980461120605, + "learning_rate": 6.188118811881187e-07, + "loss": 0.4865, + "mean_token_accuracy": 0.8419687151908875, + "num_tokens": 35040849.0, + "step": 1001 + }, + { + "epoch": 0.1860724233983287, + "grad_norm": 1.7411412000656128, + "learning_rate": 6.19430693069307e-07, + "loss": 0.4572, + "mean_token_accuracy": 0.8502882719039917, + "num_tokens": 35079990.0, + "step": 1002 + }, + { + "epoch": 0.1862581244196843, + "grad_norm": 1.5423033237457275, + "learning_rate": 6.200495049504951e-07, + "loss": 0.4944, + "mean_token_accuracy": 0.8412625789642334, + "num_tokens": 35115778.0, + "step": 1003 + }, + { + "epoch": 0.18644382544103993, + "grad_norm": 1.4590907096862793, + "learning_rate": 6.206683168316832e-07, + "loss": 0.4193, + "mean_token_accuracy": 0.8592380285263062, + "num_tokens": 35151252.0, + "step": 1004 + }, + { + "epoch": 0.18662952646239556, + "grad_norm": 1.5337902307510376, + "learning_rate": 6.212871287128713e-07, + "loss": 0.4124, + "mean_token_accuracy": 0.8689356446266174, + "num_tokens": 35182517.0, + "step": 1005 + }, + { + "epoch": 0.18681522748375115, + "grad_norm": 1.5841525793075562, + "learning_rate": 6.219059405940594e-07, + "loss": 0.4885, + "mean_token_accuracy": 0.8441529273986816, + "num_tokens": 35216227.0, + "step": 1006 + }, + { + "epoch": 0.18700092850510677, + "grad_norm": 1.5787426233291626, + "learning_rate": 6.225247524752475e-07, + "loss": 0.4547, + "mean_token_accuracy": 0.8532587289810181, + "num_tokens": 35252495.0, + "step": 1007 + }, + { + "epoch": 0.1871866295264624, + "grad_norm": 1.7400257587432861, + "learning_rate": 6.231435643564357e-07, + "loss": 0.4041, + "mean_token_accuracy": 0.8665786385536194, + "num_tokens": 35281734.0, + "step": 1008 + }, + { + "epoch": 0.18737233054781802, + "grad_norm": 1.6130889654159546, + "learning_rate": 6.237623762376237e-07, + "loss": 0.462, + "mean_token_accuracy": 0.8487545847892761, + "num_tokens": 35314319.0, + "step": 1009 + }, + { + "epoch": 0.18755803156917364, + "grad_norm": 1.6768362522125244, + "learning_rate": 6.243811881188119e-07, + "loss": 0.5665, + "mean_token_accuracy": 0.8190209269523621, + "num_tokens": 35347640.0, + "step": 1010 + }, + { + "epoch": 0.18774373259052926, + "grad_norm": 1.634291648864746, + "learning_rate": 6.249999999999999e-07, + "loss": 0.4754, + "mean_token_accuracy": 0.8422119617462158, + "num_tokens": 35378547.0, + "step": 1011 + }, + { + "epoch": 0.18792943361188486, + "grad_norm": 1.6376838684082031, + "learning_rate": 6.25618811881188e-07, + "loss": 0.4532, + "mean_token_accuracy": 0.854300856590271, + "num_tokens": 35410667.0, + "step": 1012 + }, + { + "epoch": 0.18811513463324048, + "grad_norm": 1.432205080986023, + "learning_rate": 6.262376237623762e-07, + "loss": 0.4273, + "mean_token_accuracy": 0.8596799969673157, + "num_tokens": 35449352.0, + "step": 1013 + }, + { + "epoch": 0.1883008356545961, + "grad_norm": 1.6312963962554932, + "learning_rate": 6.268564356435642e-07, + "loss": 0.475, + "mean_token_accuracy": 0.847540020942688, + "num_tokens": 35479857.0, + "step": 1014 + }, + { + "epoch": 0.18848653667595172, + "grad_norm": 1.56415855884552, + "learning_rate": 6.274752475247525e-07, + "loss": 0.4164, + "mean_token_accuracy": 0.8634781837463379, + "num_tokens": 35513071.0, + "step": 1015 + }, + { + "epoch": 0.18867223769730734, + "grad_norm": 1.706986427307129, + "learning_rate": 6.280940594059406e-07, + "loss": 0.4968, + "mean_token_accuracy": 0.8387096524238586, + "num_tokens": 35547170.0, + "step": 1016 + }, + { + "epoch": 0.18885793871866297, + "grad_norm": 1.6046862602233887, + "learning_rate": 6.287128712871287e-07, + "loss": 0.4804, + "mean_token_accuracy": 0.8418378829956055, + "num_tokens": 35579468.0, + "step": 1017 + }, + { + "epoch": 0.18904363974001856, + "grad_norm": 1.5008885860443115, + "learning_rate": 6.293316831683168e-07, + "loss": 0.4141, + "mean_token_accuracy": 0.859744668006897, + "num_tokens": 35615594.0, + "step": 1018 + }, + { + "epoch": 0.18922934076137418, + "grad_norm": 1.674046516418457, + "learning_rate": 6.299504950495049e-07, + "loss": 0.5089, + "mean_token_accuracy": 0.836916446685791, + "num_tokens": 35649464.0, + "step": 1019 + }, + { + "epoch": 0.1894150417827298, + "grad_norm": 1.6192600727081299, + "learning_rate": 6.30569306930693e-07, + "loss": 0.4978, + "mean_token_accuracy": 0.8385127782821655, + "num_tokens": 35685832.0, + "step": 1020 + }, + { + "epoch": 0.18960074280408543, + "grad_norm": 1.5116603374481201, + "learning_rate": 6.311881188118812e-07, + "loss": 0.496, + "mean_token_accuracy": 0.8415173292160034, + "num_tokens": 35723590.0, + "step": 1021 + }, + { + "epoch": 0.18978644382544105, + "grad_norm": 1.4693324565887451, + "learning_rate": 6.318069306930692e-07, + "loss": 0.4574, + "mean_token_accuracy": 0.8493682146072388, + "num_tokens": 35760406.0, + "step": 1022 + }, + { + "epoch": 0.18997214484679667, + "grad_norm": 1.8233959674835205, + "learning_rate": 6.324257425742574e-07, + "loss": 0.497, + "mean_token_accuracy": 0.8381164073944092, + "num_tokens": 35788205.0, + "step": 1023 + }, + { + "epoch": 0.19015784586815226, + "grad_norm": 1.562904953956604, + "learning_rate": 6.330445544554454e-07, + "loss": 0.4784, + "mean_token_accuracy": 0.8445022106170654, + "num_tokens": 35823802.0, + "step": 1024 + }, + { + "epoch": 0.1903435468895079, + "grad_norm": 1.645950198173523, + "learning_rate": 6.336633663366336e-07, + "loss": 0.5098, + "mean_token_accuracy": 0.8354158997535706, + "num_tokens": 35854146.0, + "step": 1025 + }, + { + "epoch": 0.1905292479108635, + "grad_norm": 1.5417104959487915, + "learning_rate": 6.342821782178217e-07, + "loss": 0.4368, + "mean_token_accuracy": 0.8571811318397522, + "num_tokens": 35887803.0, + "step": 1026 + }, + { + "epoch": 0.19071494893221913, + "grad_norm": 1.5678801536560059, + "learning_rate": 6.349009900990098e-07, + "loss": 0.4134, + "mean_token_accuracy": 0.8620186448097229, + "num_tokens": 35917732.0, + "step": 1027 + }, + { + "epoch": 0.19090064995357475, + "grad_norm": 1.5738519430160522, + "learning_rate": 6.35519801980198e-07, + "loss": 0.4746, + "mean_token_accuracy": 0.843426525592804, + "num_tokens": 35950564.0, + "step": 1028 + }, + { + "epoch": 0.19108635097493037, + "grad_norm": 1.7819502353668213, + "learning_rate": 6.361386138613862e-07, + "loss": 0.4555, + "mean_token_accuracy": 0.8506102561950684, + "num_tokens": 35978434.0, + "step": 1029 + }, + { + "epoch": 0.19127205199628597, + "grad_norm": 1.6660435199737549, + "learning_rate": 6.367574257425742e-07, + "loss": 0.4685, + "mean_token_accuracy": 0.8498207330703735, + "num_tokens": 36007214.0, + "step": 1030 + }, + { + "epoch": 0.1914577530176416, + "grad_norm": 1.4737880229949951, + "learning_rate": 6.373762376237624e-07, + "loss": 0.449, + "mean_token_accuracy": 0.8539069890975952, + "num_tokens": 36042943.0, + "step": 1031 + }, + { + "epoch": 0.1916434540389972, + "grad_norm": 1.4576165676116943, + "learning_rate": 6.379950495049504e-07, + "loss": 0.3804, + "mean_token_accuracy": 0.8739081621170044, + "num_tokens": 36077271.0, + "step": 1032 + }, + { + "epoch": 0.19182915506035284, + "grad_norm": 1.3899281024932861, + "learning_rate": 6.386138613861386e-07, + "loss": 0.4533, + "mean_token_accuracy": 0.8555138111114502, + "num_tokens": 36120101.0, + "step": 1033 + }, + { + "epoch": 0.19201485608170846, + "grad_norm": 1.6117069721221924, + "learning_rate": 6.392326732673267e-07, + "loss": 0.4671, + "mean_token_accuracy": 0.845642626285553, + "num_tokens": 36153840.0, + "step": 1034 + }, + { + "epoch": 0.19220055710306408, + "grad_norm": 1.6584511995315552, + "learning_rate": 6.398514851485148e-07, + "loss": 0.4537, + "mean_token_accuracy": 0.8540685772895813, + "num_tokens": 36182814.0, + "step": 1035 + }, + { + "epoch": 0.19238625812441967, + "grad_norm": 1.6541954278945923, + "learning_rate": 6.404702970297029e-07, + "loss": 0.4767, + "mean_token_accuracy": 0.8428916335105896, + "num_tokens": 36215206.0, + "step": 1036 + }, + { + "epoch": 0.1925719591457753, + "grad_norm": 1.425492525100708, + "learning_rate": 6.41089108910891e-07, + "loss": 0.4943, + "mean_token_accuracy": 0.8389164805412292, + "num_tokens": 36256635.0, + "step": 1037 + }, + { + "epoch": 0.19275766016713092, + "grad_norm": 1.594028115272522, + "learning_rate": 6.417079207920791e-07, + "loss": 0.4731, + "mean_token_accuracy": 0.8464908599853516, + "num_tokens": 36291550.0, + "step": 1038 + }, + { + "epoch": 0.19294336118848654, + "grad_norm": 1.6768651008605957, + "learning_rate": 6.423267326732673e-07, + "loss": 0.4776, + "mean_token_accuracy": 0.8460473418235779, + "num_tokens": 36323362.0, + "step": 1039 + }, + { + "epoch": 0.19312906220984216, + "grad_norm": 1.5433281660079956, + "learning_rate": 6.429455445544553e-07, + "loss": 0.4095, + "mean_token_accuracy": 0.8678934574127197, + "num_tokens": 36354440.0, + "step": 1040 + }, + { + "epoch": 0.19331476323119778, + "grad_norm": 1.6070492267608643, + "learning_rate": 6.435643564356436e-07, + "loss": 0.4503, + "mean_token_accuracy": 0.8533896207809448, + "num_tokens": 36389107.0, + "step": 1041 + }, + { + "epoch": 0.19350046425255338, + "grad_norm": 1.4521737098693848, + "learning_rate": 6.441831683168317e-07, + "loss": 0.4099, + "mean_token_accuracy": 0.8642911911010742, + "num_tokens": 36426653.0, + "step": 1042 + }, + { + "epoch": 0.193686165273909, + "grad_norm": 1.5604244470596313, + "learning_rate": 6.448019801980198e-07, + "loss": 0.4878, + "mean_token_accuracy": 0.8437103033065796, + "num_tokens": 36460887.0, + "step": 1043 + }, + { + "epoch": 0.19387186629526462, + "grad_norm": 1.5956929922103882, + "learning_rate": 6.454207920792079e-07, + "loss": 0.4265, + "mean_token_accuracy": 0.8606101870536804, + "num_tokens": 36495524.0, + "step": 1044 + }, + { + "epoch": 0.19405756731662024, + "grad_norm": 1.607574462890625, + "learning_rate": 6.46039603960396e-07, + "loss": 0.4383, + "mean_token_accuracy": 0.856639564037323, + "num_tokens": 36529102.0, + "step": 1045 + }, + { + "epoch": 0.19424326833797587, + "grad_norm": 1.5053415298461914, + "learning_rate": 6.466584158415841e-07, + "loss": 0.4751, + "mean_token_accuracy": 0.8462494015693665, + "num_tokens": 36567748.0, + "step": 1046 + }, + { + "epoch": 0.1944289693593315, + "grad_norm": 1.682963490486145, + "learning_rate": 6.472772277227723e-07, + "loss": 0.4585, + "mean_token_accuracy": 0.8527676463127136, + "num_tokens": 36599437.0, + "step": 1047 + }, + { + "epoch": 0.19461467038068708, + "grad_norm": 1.5928161144256592, + "learning_rate": 6.478960396039603e-07, + "loss": 0.497, + "mean_token_accuracy": 0.8412337303161621, + "num_tokens": 36631283.0, + "step": 1048 + }, + { + "epoch": 0.1948003714020427, + "grad_norm": 1.4908019304275513, + "learning_rate": 6.485148514851485e-07, + "loss": 0.4197, + "mean_token_accuracy": 0.8611693978309631, + "num_tokens": 36665283.0, + "step": 1049 + }, + { + "epoch": 0.19498607242339833, + "grad_norm": 1.4957945346832275, + "learning_rate": 6.491336633663366e-07, + "loss": 0.472, + "mean_token_accuracy": 0.8454830646514893, + "num_tokens": 36704763.0, + "step": 1050 + }, + { + "epoch": 0.19517177344475395, + "grad_norm": 1.599016785621643, + "learning_rate": 6.497524752475247e-07, + "loss": 0.4779, + "mean_token_accuracy": 0.8409935832023621, + "num_tokens": 36741649.0, + "step": 1051 + }, + { + "epoch": 0.19535747446610957, + "grad_norm": 1.5908052921295166, + "learning_rate": 6.503712871287128e-07, + "loss": 0.4521, + "mean_token_accuracy": 0.8535579442977905, + "num_tokens": 36776857.0, + "step": 1052 + }, + { + "epoch": 0.1955431754874652, + "grad_norm": 1.5552970170974731, + "learning_rate": 6.509900990099009e-07, + "loss": 0.4187, + "mean_token_accuracy": 0.8633887767791748, + "num_tokens": 36809547.0, + "step": 1053 + }, + { + "epoch": 0.1957288765088208, + "grad_norm": 1.3753327131271362, + "learning_rate": 6.51608910891089e-07, + "loss": 0.4532, + "mean_token_accuracy": 0.8531774282455444, + "num_tokens": 36848503.0, + "step": 1054 + }, + { + "epoch": 0.1959145775301764, + "grad_norm": 1.4918042421340942, + "learning_rate": 6.522277227722773e-07, + "loss": 0.3965, + "mean_token_accuracy": 0.8686127066612244, + "num_tokens": 36880394.0, + "step": 1055 + }, + { + "epoch": 0.19610027855153203, + "grad_norm": 1.7015947103500366, + "learning_rate": 6.528465346534653e-07, + "loss": 0.4083, + "mean_token_accuracy": 0.8646203875541687, + "num_tokens": 36904539.0, + "step": 1056 + }, + { + "epoch": 0.19628597957288765, + "grad_norm": 1.5977333784103394, + "learning_rate": 6.534653465346535e-07, + "loss": 0.4501, + "mean_token_accuracy": 0.8523575067520142, + "num_tokens": 36940274.0, + "step": 1057 + }, + { + "epoch": 0.19647168059424328, + "grad_norm": 1.7566486597061157, + "learning_rate": 6.540841584158415e-07, + "loss": 0.4726, + "mean_token_accuracy": 0.8527195453643799, + "num_tokens": 36968499.0, + "step": 1058 + }, + { + "epoch": 0.1966573816155989, + "grad_norm": 1.907609462738037, + "learning_rate": 6.547029702970297e-07, + "loss": 0.4928, + "mean_token_accuracy": 0.8444831371307373, + "num_tokens": 36996454.0, + "step": 1059 + }, + { + "epoch": 0.1968430826369545, + "grad_norm": 1.602123737335205, + "learning_rate": 6.553217821782178e-07, + "loss": 0.4531, + "mean_token_accuracy": 0.8558104038238525, + "num_tokens": 37029225.0, + "step": 1060 + }, + { + "epoch": 0.1970287836583101, + "grad_norm": 1.4328404664993286, + "learning_rate": 6.559405940594059e-07, + "loss": 0.3881, + "mean_token_accuracy": 0.8732984066009521, + "num_tokens": 37064975.0, + "step": 1061 + }, + { + "epoch": 0.19721448467966574, + "grad_norm": 1.5483906269073486, + "learning_rate": 6.56559405940594e-07, + "loss": 0.4724, + "mean_token_accuracy": 0.8483233451843262, + "num_tokens": 37098650.0, + "step": 1062 + }, + { + "epoch": 0.19740018570102136, + "grad_norm": 1.5798721313476562, + "learning_rate": 6.571782178217822e-07, + "loss": 0.4418, + "mean_token_accuracy": 0.8533951044082642, + "num_tokens": 37129882.0, + "step": 1063 + }, + { + "epoch": 0.19758588672237698, + "grad_norm": 1.5181933641433716, + "learning_rate": 6.577970297029702e-07, + "loss": 0.4569, + "mean_token_accuracy": 0.8497394323348999, + "num_tokens": 37168055.0, + "step": 1064 + }, + { + "epoch": 0.1977715877437326, + "grad_norm": 1.6394799947738647, + "learning_rate": 6.584158415841584e-07, + "loss": 0.5187, + "mean_token_accuracy": 0.8342341184616089, + "num_tokens": 37200332.0, + "step": 1065 + }, + { + "epoch": 0.1979572887650882, + "grad_norm": 1.563173770904541, + "learning_rate": 6.590346534653464e-07, + "loss": 0.446, + "mean_token_accuracy": 0.8536227345466614, + "num_tokens": 37236698.0, + "step": 1066 + }, + { + "epoch": 0.19814298978644382, + "grad_norm": 1.3908958435058594, + "learning_rate": 6.596534653465346e-07, + "loss": 0.4193, + "mean_token_accuracy": 0.8612645268440247, + "num_tokens": 37275295.0, + "step": 1067 + }, + { + "epoch": 0.19832869080779944, + "grad_norm": 1.4470516443252563, + "learning_rate": 6.602722772277228e-07, + "loss": 0.4093, + "mean_token_accuracy": 0.8657152652740479, + "num_tokens": 37312194.0, + "step": 1068 + }, + { + "epoch": 0.19851439182915506, + "grad_norm": 1.4726054668426514, + "learning_rate": 6.608910891089109e-07, + "loss": 0.4116, + "mean_token_accuracy": 0.8600234985351562, + "num_tokens": 37348488.0, + "step": 1069 + }, + { + "epoch": 0.19870009285051068, + "grad_norm": 1.4548107385635376, + "learning_rate": 6.61509900990099e-07, + "loss": 0.4645, + "mean_token_accuracy": 0.8481894731521606, + "num_tokens": 37390179.0, + "step": 1070 + }, + { + "epoch": 0.1988857938718663, + "grad_norm": 1.4825160503387451, + "learning_rate": 6.621287128712872e-07, + "loss": 0.4956, + "mean_token_accuracy": 0.8394612073898315, + "num_tokens": 37429409.0, + "step": 1071 + }, + { + "epoch": 0.1990714948932219, + "grad_norm": 1.4498724937438965, + "learning_rate": 6.627475247524752e-07, + "loss": 0.413, + "mean_token_accuracy": 0.8605217933654785, + "num_tokens": 37465734.0, + "step": 1072 + }, + { + "epoch": 0.19925719591457752, + "grad_norm": 1.5420514345169067, + "learning_rate": 6.633663366336634e-07, + "loss": 0.4678, + "mean_token_accuracy": 0.8465415835380554, + "num_tokens": 37504039.0, + "step": 1073 + }, + { + "epoch": 0.19944289693593314, + "grad_norm": 1.4913285970687866, + "learning_rate": 6.639851485148514e-07, + "loss": 0.4921, + "mean_token_accuracy": 0.8405884504318237, + "num_tokens": 37543181.0, + "step": 1074 + }, + { + "epoch": 0.19962859795728877, + "grad_norm": 1.815704345703125, + "learning_rate": 6.646039603960396e-07, + "loss": 0.5336, + "mean_token_accuracy": 0.8297287821769714, + "num_tokens": 37574434.0, + "step": 1075 + }, + { + "epoch": 0.1998142989786444, + "grad_norm": 1.3873200416564941, + "learning_rate": 6.652227722772277e-07, + "loss": 0.4411, + "mean_token_accuracy": 0.8537024259567261, + "num_tokens": 37616122.0, + "step": 1076 + }, + { + "epoch": 0.2, + "grad_norm": 1.5997439622879028, + "learning_rate": 6.658415841584158e-07, + "loss": 0.4732, + "mean_token_accuracy": 0.8494065999984741, + "num_tokens": 37649358.0, + "step": 1077 + }, + { + "epoch": 0.2001857010213556, + "grad_norm": 1.6272141933441162, + "learning_rate": 6.664603960396039e-07, + "loss": 0.4622, + "mean_token_accuracy": 0.8481531739234924, + "num_tokens": 37682772.0, + "step": 1078 + }, + { + "epoch": 0.20037140204271123, + "grad_norm": 1.6526198387145996, + "learning_rate": 6.67079207920792e-07, + "loss": 0.4865, + "mean_token_accuracy": 0.8433482050895691, + "num_tokens": 37716506.0, + "step": 1079 + }, + { + "epoch": 0.20055710306406685, + "grad_norm": 1.4985765218734741, + "learning_rate": 6.676980198019801e-07, + "loss": 0.4191, + "mean_token_accuracy": 0.8635498285293579, + "num_tokens": 37749094.0, + "step": 1080 + }, + { + "epoch": 0.20074280408542247, + "grad_norm": 1.619638204574585, + "learning_rate": 6.683168316831684e-07, + "loss": 0.4744, + "mean_token_accuracy": 0.8457293510437012, + "num_tokens": 37785480.0, + "step": 1081 + }, + { + "epoch": 0.2009285051067781, + "grad_norm": 1.4949405193328857, + "learning_rate": 6.689356435643564e-07, + "loss": 0.4981, + "mean_token_accuracy": 0.8363882303237915, + "num_tokens": 37826791.0, + "step": 1082 + }, + { + "epoch": 0.20111420612813372, + "grad_norm": 1.6618151664733887, + "learning_rate": 6.695544554455446e-07, + "loss": 0.5139, + "mean_token_accuracy": 0.8332310914993286, + "num_tokens": 37859832.0, + "step": 1083 + }, + { + "epoch": 0.2012999071494893, + "grad_norm": 1.7233420610427856, + "learning_rate": 6.701732673267327e-07, + "loss": 0.5274, + "mean_token_accuracy": 0.8295243382453918, + "num_tokens": 37891668.0, + "step": 1084 + }, + { + "epoch": 0.20148560817084493, + "grad_norm": 1.631632685661316, + "learning_rate": 6.707920792079208e-07, + "loss": 0.4852, + "mean_token_accuracy": 0.8475005626678467, + "num_tokens": 37923169.0, + "step": 1085 + }, + { + "epoch": 0.20167130919220055, + "grad_norm": 1.571901559829712, + "learning_rate": 6.714108910891089e-07, + "loss": 0.4921, + "mean_token_accuracy": 0.8442637920379639, + "num_tokens": 37958287.0, + "step": 1086 + }, + { + "epoch": 0.20185701021355618, + "grad_norm": 1.6555556058883667, + "learning_rate": 6.72029702970297e-07, + "loss": 0.4943, + "mean_token_accuracy": 0.8408446907997131, + "num_tokens": 37992120.0, + "step": 1087 + }, + { + "epoch": 0.2020427112349118, + "grad_norm": 1.4463049173355103, + "learning_rate": 6.726485148514851e-07, + "loss": 0.4617, + "mean_token_accuracy": 0.8467822670936584, + "num_tokens": 38031166.0, + "step": 1088 + }, + { + "epoch": 0.20222841225626742, + "grad_norm": 1.5436522960662842, + "learning_rate": 6.732673267326733e-07, + "loss": 0.4235, + "mean_token_accuracy": 0.8596829175949097, + "num_tokens": 38063632.0, + "step": 1089 + }, + { + "epoch": 0.20241411327762301, + "grad_norm": 1.5041804313659668, + "learning_rate": 6.738861386138613e-07, + "loss": 0.4698, + "mean_token_accuracy": 0.8467593789100647, + "num_tokens": 38102809.0, + "step": 1090 + }, + { + "epoch": 0.20259981429897864, + "grad_norm": 1.5638625621795654, + "learning_rate": 6.745049504950495e-07, + "loss": 0.4817, + "mean_token_accuracy": 0.8438512682914734, + "num_tokens": 38136264.0, + "step": 1091 + }, + { + "epoch": 0.20278551532033426, + "grad_norm": 1.4176543951034546, + "learning_rate": 6.751237623762375e-07, + "loss": 0.4717, + "mean_token_accuracy": 0.8426836729049683, + "num_tokens": 38176109.0, + "step": 1092 + }, + { + "epoch": 0.20297121634168988, + "grad_norm": 1.5049761533737183, + "learning_rate": 6.757425742574257e-07, + "loss": 0.4315, + "mean_token_accuracy": 0.8593673706054688, + "num_tokens": 38212418.0, + "step": 1093 + }, + { + "epoch": 0.2031569173630455, + "grad_norm": 1.6116994619369507, + "learning_rate": 6.763613861386139e-07, + "loss": 0.4859, + "mean_token_accuracy": 0.8440889716148376, + "num_tokens": 38244161.0, + "step": 1094 + }, + { + "epoch": 0.20334261838440112, + "grad_norm": 1.3873881101608276, + "learning_rate": 6.76980198019802e-07, + "loss": 0.3832, + "mean_token_accuracy": 0.8702813982963562, + "num_tokens": 38284516.0, + "step": 1095 + }, + { + "epoch": 0.20352831940575672, + "grad_norm": 1.570275902748108, + "learning_rate": 6.775990099009901e-07, + "loss": 0.4992, + "mean_token_accuracy": 0.8382570743560791, + "num_tokens": 38320394.0, + "step": 1096 + }, + { + "epoch": 0.20371402042711234, + "grad_norm": 1.4338064193725586, + "learning_rate": 6.782178217821783e-07, + "loss": 0.4634, + "mean_token_accuracy": 0.8500097990036011, + "num_tokens": 38359624.0, + "step": 1097 + }, + { + "epoch": 0.20389972144846796, + "grad_norm": 1.495401382446289, + "learning_rate": 6.788366336633663e-07, + "loss": 0.4196, + "mean_token_accuracy": 0.8612534403800964, + "num_tokens": 38395386.0, + "step": 1098 + }, + { + "epoch": 0.20408542246982359, + "grad_norm": 1.5063695907592773, + "learning_rate": 6.794554455445545e-07, + "loss": 0.4583, + "mean_token_accuracy": 0.8534071445465088, + "num_tokens": 38430232.0, + "step": 1099 + }, + { + "epoch": 0.2042711234911792, + "grad_norm": 1.5486724376678467, + "learning_rate": 6.800742574257425e-07, + "loss": 0.4528, + "mean_token_accuracy": 0.8520316481590271, + "num_tokens": 38460866.0, + "step": 1100 + }, + { + "epoch": 0.20445682451253483, + "grad_norm": 1.5773266553878784, + "learning_rate": 6.806930693069307e-07, + "loss": 0.4822, + "mean_token_accuracy": 0.84807288646698, + "num_tokens": 38495168.0, + "step": 1101 + }, + { + "epoch": 0.20464252553389042, + "grad_norm": 1.704846978187561, + "learning_rate": 6.813118811881188e-07, + "loss": 0.4346, + "mean_token_accuracy": 0.8579217195510864, + "num_tokens": 38522978.0, + "step": 1102 + }, + { + "epoch": 0.20482822655524605, + "grad_norm": 1.3483177423477173, + "learning_rate": 6.819306930693069e-07, + "loss": 0.3919, + "mean_token_accuracy": 0.8719755411148071, + "num_tokens": 38566767.0, + "step": 1103 + }, + { + "epoch": 0.20501392757660167, + "grad_norm": 1.7191760540008545, + "learning_rate": 6.82549504950495e-07, + "loss": 0.5453, + "mean_token_accuracy": 0.8304745554924011, + "num_tokens": 38600728.0, + "step": 1104 + }, + { + "epoch": 0.2051996285979573, + "grad_norm": 1.5470739603042603, + "learning_rate": 6.831683168316831e-07, + "loss": 0.4645, + "mean_token_accuracy": 0.8472757339477539, + "num_tokens": 38634212.0, + "step": 1105 + }, + { + "epoch": 0.2053853296193129, + "grad_norm": 1.5248674154281616, + "learning_rate": 6.837871287128712e-07, + "loss": 0.4397, + "mean_token_accuracy": 0.8526740074157715, + "num_tokens": 38667236.0, + "step": 1106 + }, + { + "epoch": 0.20557103064066853, + "grad_norm": 1.5880519151687622, + "learning_rate": 6.844059405940595e-07, + "loss": 0.441, + "mean_token_accuracy": 0.8572697639465332, + "num_tokens": 38698036.0, + "step": 1107 + }, + { + "epoch": 0.20575673166202413, + "grad_norm": 1.4766626358032227, + "learning_rate": 6.850247524752475e-07, + "loss": 0.4592, + "mean_token_accuracy": 0.849970817565918, + "num_tokens": 38736033.0, + "step": 1108 + }, + { + "epoch": 0.20594243268337975, + "grad_norm": 1.5281565189361572, + "learning_rate": 6.856435643564357e-07, + "loss": 0.4549, + "mean_token_accuracy": 0.8523398637771606, + "num_tokens": 38770142.0, + "step": 1109 + }, + { + "epoch": 0.20612813370473537, + "grad_norm": 1.6320265531539917, + "learning_rate": 6.862623762376238e-07, + "loss": 0.5423, + "mean_token_accuracy": 0.828122615814209, + "num_tokens": 38805410.0, + "step": 1110 + }, + { + "epoch": 0.206313834726091, + "grad_norm": 1.6179717779159546, + "learning_rate": 6.868811881188119e-07, + "loss": 0.4648, + "mean_token_accuracy": 0.8515916466712952, + "num_tokens": 38838300.0, + "step": 1111 + }, + { + "epoch": 0.20649953574744662, + "grad_norm": 1.53328537940979, + "learning_rate": 6.875e-07, + "loss": 0.4666, + "mean_token_accuracy": 0.8492602705955505, + "num_tokens": 38872734.0, + "step": 1112 + }, + { + "epoch": 0.20668523676880224, + "grad_norm": 1.4558610916137695, + "learning_rate": 6.88118811881188e-07, + "loss": 0.4877, + "mean_token_accuracy": 0.8419700860977173, + "num_tokens": 38911971.0, + "step": 1113 + }, + { + "epoch": 0.20687093779015783, + "grad_norm": 1.511066198348999, + "learning_rate": 6.887376237623762e-07, + "loss": 0.4541, + "mean_token_accuracy": 0.849535346031189, + "num_tokens": 38945424.0, + "step": 1114 + }, + { + "epoch": 0.20705663881151345, + "grad_norm": 1.588633418083191, + "learning_rate": 6.893564356435643e-07, + "loss": 0.4628, + "mean_token_accuracy": 0.8492035865783691, + "num_tokens": 38977422.0, + "step": 1115 + }, + { + "epoch": 0.20724233983286908, + "grad_norm": 1.5221989154815674, + "learning_rate": 6.899752475247524e-07, + "loss": 0.4227, + "mean_token_accuracy": 0.8595235347747803, + "num_tokens": 39009590.0, + "step": 1116 + }, + { + "epoch": 0.2074280408542247, + "grad_norm": 1.584003210067749, + "learning_rate": 6.905940594059405e-07, + "loss": 0.5052, + "mean_token_accuracy": 0.835084080696106, + "num_tokens": 39048068.0, + "step": 1117 + }, + { + "epoch": 0.20761374187558032, + "grad_norm": 1.5731462240219116, + "learning_rate": 6.912128712871287e-07, + "loss": 0.4624, + "mean_token_accuracy": 0.851472020149231, + "num_tokens": 39083222.0, + "step": 1118 + }, + { + "epoch": 0.20779944289693594, + "grad_norm": 1.5078703165054321, + "learning_rate": 6.918316831683167e-07, + "loss": 0.453, + "mean_token_accuracy": 0.853527307510376, + "num_tokens": 39123147.0, + "step": 1119 + }, + { + "epoch": 0.20798514391829154, + "grad_norm": 1.5544111728668213, + "learning_rate": 6.924504950495049e-07, + "loss": 0.4526, + "mean_token_accuracy": 0.8526049852371216, + "num_tokens": 39159945.0, + "step": 1120 + }, + { + "epoch": 0.20817084493964716, + "grad_norm": 1.4678242206573486, + "learning_rate": 6.93069306930693e-07, + "loss": 0.4653, + "mean_token_accuracy": 0.849774956703186, + "num_tokens": 39198878.0, + "step": 1121 + }, + { + "epoch": 0.20835654596100278, + "grad_norm": 1.6616065502166748, + "learning_rate": 6.936881188118812e-07, + "loss": 0.4826, + "mean_token_accuracy": 0.8419184684753418, + "num_tokens": 39228911.0, + "step": 1122 + }, + { + "epoch": 0.2085422469823584, + "grad_norm": 1.535508394241333, + "learning_rate": 6.943069306930693e-07, + "loss": 0.4753, + "mean_token_accuracy": 0.8472795486450195, + "num_tokens": 39265795.0, + "step": 1123 + }, + { + "epoch": 0.20872794800371403, + "grad_norm": 1.5924707651138306, + "learning_rate": 6.949257425742574e-07, + "loss": 0.4557, + "mean_token_accuracy": 0.8511142730712891, + "num_tokens": 39299277.0, + "step": 1124 + }, + { + "epoch": 0.20891364902506965, + "grad_norm": 1.5208399295806885, + "learning_rate": 6.955445544554455e-07, + "loss": 0.4392, + "mean_token_accuracy": 0.85471111536026, + "num_tokens": 39335489.0, + "step": 1125 + }, + { + "epoch": 0.20909935004642524, + "grad_norm": 1.6387044191360474, + "learning_rate": 6.961633663366336e-07, + "loss": 0.4905, + "mean_token_accuracy": 0.8406044244766235, + "num_tokens": 39367933.0, + "step": 1126 + }, + { + "epoch": 0.20928505106778086, + "grad_norm": 1.571984887123108, + "learning_rate": 6.967821782178217e-07, + "loss": 0.5011, + "mean_token_accuracy": 0.8362583518028259, + "num_tokens": 39402119.0, + "step": 1127 + }, + { + "epoch": 0.20947075208913649, + "grad_norm": 1.4647079706192017, + "learning_rate": 6.974009900990099e-07, + "loss": 0.4489, + "mean_token_accuracy": 0.8527936935424805, + "num_tokens": 39436775.0, + "step": 1128 + }, + { + "epoch": 0.2096564531104921, + "grad_norm": 1.5780577659606934, + "learning_rate": 6.980198019801979e-07, + "loss": 0.5336, + "mean_token_accuracy": 0.8311004638671875, + "num_tokens": 39474147.0, + "step": 1129 + }, + { + "epoch": 0.20984215413184773, + "grad_norm": 1.4247492551803589, + "learning_rate": 6.986386138613861e-07, + "loss": 0.4596, + "mean_token_accuracy": 0.8486394286155701, + "num_tokens": 39514122.0, + "step": 1130 + }, + { + "epoch": 0.21002785515320335, + "grad_norm": 1.4937063455581665, + "learning_rate": 6.992574257425742e-07, + "loss": 0.4242, + "mean_token_accuracy": 0.8595213294029236, + "num_tokens": 39549811.0, + "step": 1131 + }, + { + "epoch": 0.21021355617455895, + "grad_norm": 1.4682190418243408, + "learning_rate": 6.998762376237623e-07, + "loss": 0.4619, + "mean_token_accuracy": 0.8493261933326721, + "num_tokens": 39586341.0, + "step": 1132 + }, + { + "epoch": 0.21039925719591457, + "grad_norm": 1.6055172681808472, + "learning_rate": 7.004950495049504e-07, + "loss": 0.4561, + "mean_token_accuracy": 0.8517153859138489, + "num_tokens": 39620147.0, + "step": 1133 + }, + { + "epoch": 0.2105849582172702, + "grad_norm": 1.584733009338379, + "learning_rate": 7.011138613861386e-07, + "loss": 0.478, + "mean_token_accuracy": 0.8436816930770874, + "num_tokens": 39653816.0, + "step": 1134 + }, + { + "epoch": 0.2107706592386258, + "grad_norm": 1.5357565879821777, + "learning_rate": 7.017326732673267e-07, + "loss": 0.4271, + "mean_token_accuracy": 0.8634210824966431, + "num_tokens": 39687108.0, + "step": 1135 + }, + { + "epoch": 0.21095636025998143, + "grad_norm": 1.4118906259536743, + "learning_rate": 7.023514851485149e-07, + "loss": 0.4712, + "mean_token_accuracy": 0.848807156085968, + "num_tokens": 39730606.0, + "step": 1136 + }, + { + "epoch": 0.21114206128133706, + "grad_norm": 1.6895644664764404, + "learning_rate": 7.029702970297029e-07, + "loss": 0.4753, + "mean_token_accuracy": 0.8452766537666321, + "num_tokens": 39765258.0, + "step": 1137 + }, + { + "epoch": 0.21132776230269265, + "grad_norm": 1.3897324800491333, + "learning_rate": 7.035891089108911e-07, + "loss": 0.443, + "mean_token_accuracy": 0.8555893898010254, + "num_tokens": 39811056.0, + "step": 1138 + }, + { + "epoch": 0.21151346332404827, + "grad_norm": 1.533629059791565, + "learning_rate": 7.042079207920791e-07, + "loss": 0.4823, + "mean_token_accuracy": 0.8422044515609741, + "num_tokens": 39851619.0, + "step": 1139 + }, + { + "epoch": 0.2116991643454039, + "grad_norm": 1.5091952085494995, + "learning_rate": 7.048267326732673e-07, + "loss": 0.44, + "mean_token_accuracy": 0.8552661538124084, + "num_tokens": 39888428.0, + "step": 1140 + }, + { + "epoch": 0.21188486536675952, + "grad_norm": 1.7783372402191162, + "learning_rate": 7.054455445544554e-07, + "loss": 0.5051, + "mean_token_accuracy": 0.8374416828155518, + "num_tokens": 39917657.0, + "step": 1141 + }, + { + "epoch": 0.21207056638811514, + "grad_norm": 1.5481700897216797, + "learning_rate": 7.060643564356435e-07, + "loss": 0.4413, + "mean_token_accuracy": 0.8560487031936646, + "num_tokens": 39953547.0, + "step": 1142 + }, + { + "epoch": 0.21225626740947076, + "grad_norm": 1.781058669090271, + "learning_rate": 7.066831683168316e-07, + "loss": 0.4854, + "mean_token_accuracy": 0.8433303236961365, + "num_tokens": 39982484.0, + "step": 1143 + }, + { + "epoch": 0.21244196843082638, + "grad_norm": 1.6881399154663086, + "learning_rate": 7.073019801980198e-07, + "loss": 0.4176, + "mean_token_accuracy": 0.863178014755249, + "num_tokens": 40009386.0, + "step": 1144 + }, + { + "epoch": 0.21262766945218198, + "grad_norm": 1.5459980964660645, + "learning_rate": 7.079207920792078e-07, + "loss": 0.4436, + "mean_token_accuracy": 0.8552030324935913, + "num_tokens": 40044425.0, + "step": 1145 + }, + { + "epoch": 0.2128133704735376, + "grad_norm": 1.6224570274353027, + "learning_rate": 7.08539603960396e-07, + "loss": 0.4254, + "mean_token_accuracy": 0.8571311235427856, + "num_tokens": 40077366.0, + "step": 1146 + }, + { + "epoch": 0.21299907149489322, + "grad_norm": 1.4047973155975342, + "learning_rate": 7.091584158415841e-07, + "loss": 0.4135, + "mean_token_accuracy": 0.8639869689941406, + "num_tokens": 40116316.0, + "step": 1147 + }, + { + "epoch": 0.21318477251624884, + "grad_norm": 1.6572012901306152, + "learning_rate": 7.097772277227723e-07, + "loss": 0.4439, + "mean_token_accuracy": 0.8543880581855774, + "num_tokens": 40149765.0, + "step": 1148 + }, + { + "epoch": 0.21337047353760447, + "grad_norm": 1.3933677673339844, + "learning_rate": 7.103960396039604e-07, + "loss": 0.4437, + "mean_token_accuracy": 0.85743248462677, + "num_tokens": 40191300.0, + "step": 1149 + }, + { + "epoch": 0.2135561745589601, + "grad_norm": 1.482879400253296, + "learning_rate": 7.110148514851485e-07, + "loss": 0.4395, + "mean_token_accuracy": 0.8527455925941467, + "num_tokens": 40230878.0, + "step": 1150 + }, + { + "epoch": 0.21374187558031568, + "grad_norm": 1.6358965635299683, + "learning_rate": 7.116336633663366e-07, + "loss": 0.4306, + "mean_token_accuracy": 0.8605477809906006, + "num_tokens": 40264409.0, + "step": 1151 + }, + { + "epoch": 0.2139275766016713, + "grad_norm": 1.5897819995880127, + "learning_rate": 7.122524752475248e-07, + "loss": 0.5048, + "mean_token_accuracy": 0.8390353918075562, + "num_tokens": 40302968.0, + "step": 1152 + }, + { + "epoch": 0.21411327762302693, + "grad_norm": 1.4544954299926758, + "learning_rate": 7.128712871287128e-07, + "loss": 0.4447, + "mean_token_accuracy": 0.8549149036407471, + "num_tokens": 40340102.0, + "step": 1153 + }, + { + "epoch": 0.21429897864438255, + "grad_norm": 1.4892985820770264, + "learning_rate": 7.13490099009901e-07, + "loss": 0.4675, + "mean_token_accuracy": 0.8456056118011475, + "num_tokens": 40378007.0, + "step": 1154 + }, + { + "epoch": 0.21448467966573817, + "grad_norm": 1.5925674438476562, + "learning_rate": 7.14108910891089e-07, + "loss": 0.4148, + "mean_token_accuracy": 0.8646817207336426, + "num_tokens": 40411948.0, + "step": 1155 + }, + { + "epoch": 0.2146703806870938, + "grad_norm": 1.4599883556365967, + "learning_rate": 7.147277227722772e-07, + "loss": 0.3975, + "mean_token_accuracy": 0.8679069876670837, + "num_tokens": 40447784.0, + "step": 1156 + }, + { + "epoch": 0.2148560817084494, + "grad_norm": 1.5938372611999512, + "learning_rate": 7.153465346534653e-07, + "loss": 0.4554, + "mean_token_accuracy": 0.8531363606452942, + "num_tokens": 40478381.0, + "step": 1157 + }, + { + "epoch": 0.215041782729805, + "grad_norm": 1.5446078777313232, + "learning_rate": 7.159653465346534e-07, + "loss": 0.4552, + "mean_token_accuracy": 0.8489019870758057, + "num_tokens": 40513132.0, + "step": 1158 + }, + { + "epoch": 0.21522748375116063, + "grad_norm": 1.6750433444976807, + "learning_rate": 7.165841584158415e-07, + "loss": 0.4649, + "mean_token_accuracy": 0.8501793146133423, + "num_tokens": 40543797.0, + "step": 1159 + }, + { + "epoch": 0.21541318477251625, + "grad_norm": 1.5513733625411987, + "learning_rate": 7.172029702970297e-07, + "loss": 0.4856, + "mean_token_accuracy": 0.8428598642349243, + "num_tokens": 40578508.0, + "step": 1160 + }, + { + "epoch": 0.21559888579387188, + "grad_norm": 1.466648817062378, + "learning_rate": 7.178217821782178e-07, + "loss": 0.439, + "mean_token_accuracy": 0.8584212064743042, + "num_tokens": 40612131.0, + "step": 1161 + }, + { + "epoch": 0.2157845868152275, + "grad_norm": 1.5757054090499878, + "learning_rate": 7.18440594059406e-07, + "loss": 0.5085, + "mean_token_accuracy": 0.8367133736610413, + "num_tokens": 40648916.0, + "step": 1162 + }, + { + "epoch": 0.2159702878365831, + "grad_norm": 1.6642451286315918, + "learning_rate": 7.19059405940594e-07, + "loss": 0.4502, + "mean_token_accuracy": 0.8521846532821655, + "num_tokens": 40683371.0, + "step": 1163 + }, + { + "epoch": 0.2161559888579387, + "grad_norm": 1.6186586618423462, + "learning_rate": 7.196782178217822e-07, + "loss": 0.538, + "mean_token_accuracy": 0.8277232646942139, + "num_tokens": 40721096.0, + "step": 1164 + }, + { + "epoch": 0.21634168987929434, + "grad_norm": 1.4583386182785034, + "learning_rate": 7.202970297029703e-07, + "loss": 0.504, + "mean_token_accuracy": 0.8361032009124756, + "num_tokens": 40760758.0, + "step": 1165 + }, + { + "epoch": 0.21652739090064996, + "grad_norm": 1.6281952857971191, + "learning_rate": 7.209158415841584e-07, + "loss": 0.4483, + "mean_token_accuracy": 0.8498798608779907, + "num_tokens": 40793881.0, + "step": 1166 + }, + { + "epoch": 0.21671309192200558, + "grad_norm": 1.5764274597167969, + "learning_rate": 7.215346534653465e-07, + "loss": 0.4636, + "mean_token_accuracy": 0.8521010875701904, + "num_tokens": 40828609.0, + "step": 1167 + }, + { + "epoch": 0.2168987929433612, + "grad_norm": 1.6511107683181763, + "learning_rate": 7.221534653465346e-07, + "loss": 0.474, + "mean_token_accuracy": 0.8495153784751892, + "num_tokens": 40859650.0, + "step": 1168 + }, + { + "epoch": 0.2170844939647168, + "grad_norm": 1.545771837234497, + "learning_rate": 7.227722772277227e-07, + "loss": 0.4175, + "mean_token_accuracy": 0.8616595268249512, + "num_tokens": 40895391.0, + "step": 1169 + }, + { + "epoch": 0.21727019498607242, + "grad_norm": 1.712541103363037, + "learning_rate": 7.233910891089109e-07, + "loss": 0.4909, + "mean_token_accuracy": 0.8424320220947266, + "num_tokens": 40927811.0, + "step": 1170 + }, + { + "epoch": 0.21745589600742804, + "grad_norm": 1.772177815437317, + "learning_rate": 7.240099009900989e-07, + "loss": 0.4266, + "mean_token_accuracy": 0.8595831990242004, + "num_tokens": 40957493.0, + "step": 1171 + }, + { + "epoch": 0.21764159702878366, + "grad_norm": 1.6268877983093262, + "learning_rate": 7.246287128712871e-07, + "loss": 0.4355, + "mean_token_accuracy": 0.8575856685638428, + "num_tokens": 40990342.0, + "step": 1172 + }, + { + "epoch": 0.21782729805013928, + "grad_norm": 1.6315324306488037, + "learning_rate": 7.252475247524751e-07, + "loss": 0.4878, + "mean_token_accuracy": 0.8421289920806885, + "num_tokens": 41023359.0, + "step": 1173 + }, + { + "epoch": 0.2180129990714949, + "grad_norm": 1.6840238571166992, + "learning_rate": 7.258663366336634e-07, + "loss": 0.4312, + "mean_token_accuracy": 0.8592489361763, + "num_tokens": 41051890.0, + "step": 1174 + }, + { + "epoch": 0.2181987000928505, + "grad_norm": 1.6080312728881836, + "learning_rate": 7.264851485148515e-07, + "loss": 0.4396, + "mean_token_accuracy": 0.8552461862564087, + "num_tokens": 41085499.0, + "step": 1175 + }, + { + "epoch": 0.21838440111420612, + "grad_norm": 1.473501443862915, + "learning_rate": 7.271039603960396e-07, + "loss": 0.4005, + "mean_token_accuracy": 0.864985466003418, + "num_tokens": 41119436.0, + "step": 1176 + }, + { + "epoch": 0.21857010213556174, + "grad_norm": 1.6824321746826172, + "learning_rate": 7.277227722772277e-07, + "loss": 0.5042, + "mean_token_accuracy": 0.8371383547782898, + "num_tokens": 41152789.0, + "step": 1177 + }, + { + "epoch": 0.21875580315691737, + "grad_norm": 1.659924030303955, + "learning_rate": 7.283415841584159e-07, + "loss": 0.4938, + "mean_token_accuracy": 0.8438553214073181, + "num_tokens": 41187141.0, + "step": 1178 + }, + { + "epoch": 0.218941504178273, + "grad_norm": 1.659791111946106, + "learning_rate": 7.289603960396039e-07, + "loss": 0.4779, + "mean_token_accuracy": 0.8443917036056519, + "num_tokens": 41220167.0, + "step": 1179 + }, + { + "epoch": 0.2191272051996286, + "grad_norm": 1.7183812856674194, + "learning_rate": 7.295792079207921e-07, + "loss": 0.5123, + "mean_token_accuracy": 0.8358460664749146, + "num_tokens": 41249073.0, + "step": 1180 + }, + { + "epoch": 0.2193129062209842, + "grad_norm": 1.4896446466445923, + "learning_rate": 7.301980198019801e-07, + "loss": 0.447, + "mean_token_accuracy": 0.8528468608856201, + "num_tokens": 41288789.0, + "step": 1181 + }, + { + "epoch": 0.21949860724233983, + "grad_norm": 1.4133692979812622, + "learning_rate": 7.308168316831683e-07, + "loss": 0.4618, + "mean_token_accuracy": 0.8499352931976318, + "num_tokens": 41328840.0, + "step": 1182 + }, + { + "epoch": 0.21968430826369545, + "grad_norm": 1.6145316362380981, + "learning_rate": 7.314356435643564e-07, + "loss": 0.4344, + "mean_token_accuracy": 0.8601670265197754, + "num_tokens": 41360179.0, + "step": 1183 + }, + { + "epoch": 0.21987000928505107, + "grad_norm": 1.593434453010559, + "learning_rate": 7.320544554455445e-07, + "loss": 0.414, + "mean_token_accuracy": 0.8671658039093018, + "num_tokens": 41390056.0, + "step": 1184 + }, + { + "epoch": 0.2200557103064067, + "grad_norm": 1.4062772989273071, + "learning_rate": 7.326732673267326e-07, + "loss": 0.4613, + "mean_token_accuracy": 0.8509339690208435, + "num_tokens": 41429412.0, + "step": 1185 + }, + { + "epoch": 0.22024141132776232, + "grad_norm": 1.4886677265167236, + "learning_rate": 7.332920792079207e-07, + "loss": 0.4587, + "mean_token_accuracy": 0.8500378131866455, + "num_tokens": 41467807.0, + "step": 1186 + }, + { + "epoch": 0.2204271123491179, + "grad_norm": 1.7437236309051514, + "learning_rate": 7.339108910891089e-07, + "loss": 0.5057, + "mean_token_accuracy": 0.8370476961135864, + "num_tokens": 41499774.0, + "step": 1187 + }, + { + "epoch": 0.22061281337047353, + "grad_norm": 1.615572452545166, + "learning_rate": 7.345297029702971e-07, + "loss": 0.4615, + "mean_token_accuracy": 0.8501166105270386, + "num_tokens": 41529593.0, + "step": 1188 + }, + { + "epoch": 0.22079851439182915, + "grad_norm": 1.4633607864379883, + "learning_rate": 7.351485148514851e-07, + "loss": 0.4671, + "mean_token_accuracy": 0.8442049622535706, + "num_tokens": 41570528.0, + "step": 1189 + }, + { + "epoch": 0.22098421541318478, + "grad_norm": 1.5039689540863037, + "learning_rate": 7.357673267326733e-07, + "loss": 0.5046, + "mean_token_accuracy": 0.8370694518089294, + "num_tokens": 41611435.0, + "step": 1190 + }, + { + "epoch": 0.2211699164345404, + "grad_norm": 1.5413516759872437, + "learning_rate": 7.363861386138614e-07, + "loss": 0.464, + "mean_token_accuracy": 0.8495321273803711, + "num_tokens": 41646290.0, + "step": 1191 + }, + { + "epoch": 0.22135561745589602, + "grad_norm": 1.6349103450775146, + "learning_rate": 7.370049504950495e-07, + "loss": 0.4419, + "mean_token_accuracy": 0.8500088453292847, + "num_tokens": 41677040.0, + "step": 1192 + }, + { + "epoch": 0.22154131847725161, + "grad_norm": 1.6323388814926147, + "learning_rate": 7.376237623762376e-07, + "loss": 0.4631, + "mean_token_accuracy": 0.8491129875183105, + "num_tokens": 41708957.0, + "step": 1193 + }, + { + "epoch": 0.22172701949860724, + "grad_norm": 1.7865383625030518, + "learning_rate": 7.382425742574257e-07, + "loss": 0.4641, + "mean_token_accuracy": 0.8455409407615662, + "num_tokens": 41735947.0, + "step": 1194 + }, + { + "epoch": 0.22191272051996286, + "grad_norm": 1.4583977460861206, + "learning_rate": 7.388613861386138e-07, + "loss": 0.3951, + "mean_token_accuracy": 0.8681349754333496, + "num_tokens": 41770427.0, + "step": 1195 + }, + { + "epoch": 0.22209842154131848, + "grad_norm": 1.5558910369873047, + "learning_rate": 7.39480198019802e-07, + "loss": 0.435, + "mean_token_accuracy": 0.8555497527122498, + "num_tokens": 41805953.0, + "step": 1196 + }, + { + "epoch": 0.2222841225626741, + "grad_norm": 1.6229467391967773, + "learning_rate": 7.4009900990099e-07, + "loss": 0.4391, + "mean_token_accuracy": 0.8548873066902161, + "num_tokens": 41839444.0, + "step": 1197 + }, + { + "epoch": 0.22246982358402972, + "grad_norm": 1.5658625364303589, + "learning_rate": 7.407178217821782e-07, + "loss": 0.4401, + "mean_token_accuracy": 0.8539747595787048, + "num_tokens": 41872488.0, + "step": 1198 + }, + { + "epoch": 0.22265552460538532, + "grad_norm": 1.5027172565460205, + "learning_rate": 7.413366336633662e-07, + "loss": 0.4359, + "mean_token_accuracy": 0.8571944236755371, + "num_tokens": 41910064.0, + "step": 1199 + }, + { + "epoch": 0.22284122562674094, + "grad_norm": 1.5289262533187866, + "learning_rate": 7.419554455445545e-07, + "loss": 0.4041, + "mean_token_accuracy": 0.8632766604423523, + "num_tokens": 41941750.0, + "step": 1200 + }, + { + "epoch": 0.22302692664809656, + "grad_norm": 1.584272861480713, + "learning_rate": 7.425742574257426e-07, + "loss": 0.4513, + "mean_token_accuracy": 0.8545924425125122, + "num_tokens": 41973980.0, + "step": 1201 + }, + { + "epoch": 0.22321262766945218, + "grad_norm": 1.5396822690963745, + "learning_rate": 7.431930693069307e-07, + "loss": 0.4572, + "mean_token_accuracy": 0.852363109588623, + "num_tokens": 42009052.0, + "step": 1202 + }, + { + "epoch": 0.2233983286908078, + "grad_norm": 1.7013139724731445, + "learning_rate": 7.438118811881188e-07, + "loss": 0.5087, + "mean_token_accuracy": 0.8383480310440063, + "num_tokens": 42040641.0, + "step": 1203 + }, + { + "epoch": 0.22358402971216343, + "grad_norm": 1.5840765237808228, + "learning_rate": 7.44430693069307e-07, + "loss": 0.4155, + "mean_token_accuracy": 0.8618103265762329, + "num_tokens": 42072745.0, + "step": 1204 + }, + { + "epoch": 0.22376973073351902, + "grad_norm": 1.5376715660095215, + "learning_rate": 7.45049504950495e-07, + "loss": 0.4643, + "mean_token_accuracy": 0.8430824875831604, + "num_tokens": 42108334.0, + "step": 1205 + }, + { + "epoch": 0.22395543175487465, + "grad_norm": 1.4907869100570679, + "learning_rate": 7.456683168316832e-07, + "loss": 0.4469, + "mean_token_accuracy": 0.8543407320976257, + "num_tokens": 42149223.0, + "step": 1206 + }, + { + "epoch": 0.22414113277623027, + "grad_norm": 1.6362640857696533, + "learning_rate": 7.462871287128712e-07, + "loss": 0.488, + "mean_token_accuracy": 0.8387395739555359, + "num_tokens": 42182081.0, + "step": 1207 + }, + { + "epoch": 0.2243268337975859, + "grad_norm": 1.5627877712249756, + "learning_rate": 7.469059405940594e-07, + "loss": 0.4417, + "mean_token_accuracy": 0.8572646379470825, + "num_tokens": 42215174.0, + "step": 1208 + }, + { + "epoch": 0.2245125348189415, + "grad_norm": 1.607716679573059, + "learning_rate": 7.475247524752475e-07, + "loss": 0.5239, + "mean_token_accuracy": 0.8318374156951904, + "num_tokens": 42251160.0, + "step": 1209 + }, + { + "epoch": 0.22469823584029713, + "grad_norm": 1.4422587156295776, + "learning_rate": 7.481435643564356e-07, + "loss": 0.4319, + "mean_token_accuracy": 0.8588768243789673, + "num_tokens": 42290698.0, + "step": 1210 + }, + { + "epoch": 0.22488393686165273, + "grad_norm": 1.5889050960540771, + "learning_rate": 7.487623762376237e-07, + "loss": 0.4633, + "mean_token_accuracy": 0.8475885391235352, + "num_tokens": 42328115.0, + "step": 1211 + }, + { + "epoch": 0.22506963788300835, + "grad_norm": 1.5379856824874878, + "learning_rate": 7.493811881188119e-07, + "loss": 0.418, + "mean_token_accuracy": 0.8607211112976074, + "num_tokens": 42362089.0, + "step": 1212 + }, + { + "epoch": 0.22525533890436397, + "grad_norm": 1.5072904825210571, + "learning_rate": 7.5e-07, + "loss": 0.4647, + "mean_token_accuracy": 0.850859522819519, + "num_tokens": 42400618.0, + "step": 1213 + }, + { + "epoch": 0.2254410399257196, + "grad_norm": 1.5587711334228516, + "learning_rate": 7.506188118811881e-07, + "loss": 0.447, + "mean_token_accuracy": 0.8537989258766174, + "num_tokens": 42434774.0, + "step": 1214 + }, + { + "epoch": 0.22562674094707522, + "grad_norm": 1.4279643297195435, + "learning_rate": 7.512376237623762e-07, + "loss": 0.4358, + "mean_token_accuracy": 0.8568121790885925, + "num_tokens": 42475245.0, + "step": 1215 + }, + { + "epoch": 0.22581244196843084, + "grad_norm": 1.5077108144760132, + "learning_rate": 7.518564356435643e-07, + "loss": 0.4241, + "mean_token_accuracy": 0.8603102564811707, + "num_tokens": 42511246.0, + "step": 1216 + }, + { + "epoch": 0.22599814298978643, + "grad_norm": 1.5122359991073608, + "learning_rate": 7.524752475247525e-07, + "loss": 0.4786, + "mean_token_accuracy": 0.846035361289978, + "num_tokens": 42548830.0, + "step": 1217 + }, + { + "epoch": 0.22618384401114205, + "grad_norm": 1.3925737142562866, + "learning_rate": 7.530940594059405e-07, + "loss": 0.4176, + "mean_token_accuracy": 0.8617126941680908, + "num_tokens": 42590447.0, + "step": 1218 + }, + { + "epoch": 0.22636954503249768, + "grad_norm": 1.6500908136367798, + "learning_rate": 7.537128712871287e-07, + "loss": 0.476, + "mean_token_accuracy": 0.8448625802993774, + "num_tokens": 42622151.0, + "step": 1219 + }, + { + "epoch": 0.2265552460538533, + "grad_norm": 1.5757486820220947, + "learning_rate": 7.543316831683167e-07, + "loss": 0.5184, + "mean_token_accuracy": 0.833704948425293, + "num_tokens": 42657048.0, + "step": 1220 + }, + { + "epoch": 0.22674094707520892, + "grad_norm": 1.4849835634231567, + "learning_rate": 7.549504950495049e-07, + "loss": 0.456, + "mean_token_accuracy": 0.8505909442901611, + "num_tokens": 42693576.0, + "step": 1221 + }, + { + "epoch": 0.22692664809656454, + "grad_norm": 1.5361301898956299, + "learning_rate": 7.55569306930693e-07, + "loss": 0.4318, + "mean_token_accuracy": 0.8598217964172363, + "num_tokens": 42728905.0, + "step": 1222 + }, + { + "epoch": 0.22711234911792014, + "grad_norm": 1.75863516330719, + "learning_rate": 7.561881188118811e-07, + "loss": 0.4723, + "mean_token_accuracy": 0.8477343320846558, + "num_tokens": 42756485.0, + "step": 1223 + }, + { + "epoch": 0.22729805013927576, + "grad_norm": 1.674654245376587, + "learning_rate": 7.568069306930692e-07, + "loss": 0.478, + "mean_token_accuracy": 0.8452330827713013, + "num_tokens": 42788424.0, + "step": 1224 + }, + { + "epoch": 0.22748375116063138, + "grad_norm": 1.489061713218689, + "learning_rate": 7.574257425742574e-07, + "loss": 0.4565, + "mean_token_accuracy": 0.8528265953063965, + "num_tokens": 42826565.0, + "step": 1225 + }, + { + "epoch": 0.227669452181987, + "grad_norm": 1.3693829774856567, + "learning_rate": 7.580445544554454e-07, + "loss": 0.3989, + "mean_token_accuracy": 0.8677758574485779, + "num_tokens": 42866321.0, + "step": 1226 + }, + { + "epoch": 0.22785515320334263, + "grad_norm": 1.5899354219436646, + "learning_rate": 7.586633663366337e-07, + "loss": 0.4154, + "mean_token_accuracy": 0.8611201643943787, + "num_tokens": 42898475.0, + "step": 1227 + }, + { + "epoch": 0.22804085422469825, + "grad_norm": 1.5869072675704956, + "learning_rate": 7.592821782178217e-07, + "loss": 0.4145, + "mean_token_accuracy": 0.8630833625793457, + "num_tokens": 42927961.0, + "step": 1228 + }, + { + "epoch": 0.22822655524605384, + "grad_norm": 1.4021514654159546, + "learning_rate": 7.599009900990099e-07, + "loss": 0.4697, + "mean_token_accuracy": 0.8455132246017456, + "num_tokens": 42972384.0, + "step": 1229 + }, + { + "epoch": 0.22841225626740946, + "grad_norm": 1.4563896656036377, + "learning_rate": 7.60519801980198e-07, + "loss": 0.4487, + "mean_token_accuracy": 0.8517312407493591, + "num_tokens": 43011694.0, + "step": 1230 + }, + { + "epoch": 0.22859795728876509, + "grad_norm": 1.9144316911697388, + "learning_rate": 7.611386138613861e-07, + "loss": 0.4699, + "mean_token_accuracy": 0.8486813306808472, + "num_tokens": 43046268.0, + "step": 1231 + }, + { + "epoch": 0.2287836583101207, + "grad_norm": 1.4897873401641846, + "learning_rate": 7.617574257425742e-07, + "loss": 0.4404, + "mean_token_accuracy": 0.8552415370941162, + "num_tokens": 43083471.0, + "step": 1232 + }, + { + "epoch": 0.22896935933147633, + "grad_norm": 1.6358568668365479, + "learning_rate": 7.623762376237624e-07, + "loss": 0.5071, + "mean_token_accuracy": 0.8403970003128052, + "num_tokens": 43118684.0, + "step": 1233 + }, + { + "epoch": 0.22915506035283195, + "grad_norm": 1.5312243700027466, + "learning_rate": 7.629950495049504e-07, + "loss": 0.4372, + "mean_token_accuracy": 0.8600543737411499, + "num_tokens": 43154350.0, + "step": 1234 + }, + { + "epoch": 0.22934076137418755, + "grad_norm": 1.3821169137954712, + "learning_rate": 7.636138613861386e-07, + "loss": 0.4072, + "mean_token_accuracy": 0.86419278383255, + "num_tokens": 43193187.0, + "step": 1235 + }, + { + "epoch": 0.22952646239554317, + "grad_norm": 1.480642318725586, + "learning_rate": 7.642326732673266e-07, + "loss": 0.4646, + "mean_token_accuracy": 0.8475643396377563, + "num_tokens": 43233770.0, + "step": 1236 + }, + { + "epoch": 0.2297121634168988, + "grad_norm": 1.5021047592163086, + "learning_rate": 7.648514851485148e-07, + "loss": 0.4513, + "mean_token_accuracy": 0.8515834212303162, + "num_tokens": 43268825.0, + "step": 1237 + }, + { + "epoch": 0.2298978644382544, + "grad_norm": 1.558079719543457, + "learning_rate": 7.654702970297029e-07, + "loss": 0.4495, + "mean_token_accuracy": 0.8546339273452759, + "num_tokens": 43305154.0, + "step": 1238 + }, + { + "epoch": 0.23008356545961003, + "grad_norm": 1.6087857484817505, + "learning_rate": 7.66089108910891e-07, + "loss": 0.413, + "mean_token_accuracy": 0.8651865720748901, + "num_tokens": 43335093.0, + "step": 1239 + }, + { + "epoch": 0.23026926648096566, + "grad_norm": 1.566331148147583, + "learning_rate": 7.667079207920792e-07, + "loss": 0.4512, + "mean_token_accuracy": 0.8522684574127197, + "num_tokens": 43366530.0, + "step": 1240 + }, + { + "epoch": 0.23045496750232125, + "grad_norm": 1.3699676990509033, + "learning_rate": 7.673267326732673e-07, + "loss": 0.4114, + "mean_token_accuracy": 0.8634394407272339, + "num_tokens": 43407032.0, + "step": 1241 + }, + { + "epoch": 0.23064066852367687, + "grad_norm": 1.5982376337051392, + "learning_rate": 7.679455445544554e-07, + "loss": 0.539, + "mean_token_accuracy": 0.8274896144866943, + "num_tokens": 43443475.0, + "step": 1242 + }, + { + "epoch": 0.2308263695450325, + "grad_norm": 1.5604732036590576, + "learning_rate": 7.685643564356436e-07, + "loss": 0.4947, + "mean_token_accuracy": 0.8421031832695007, + "num_tokens": 43481946.0, + "step": 1243 + }, + { + "epoch": 0.23101207056638812, + "grad_norm": 1.402350902557373, + "learning_rate": 7.691831683168316e-07, + "loss": 0.4137, + "mean_token_accuracy": 0.863640546798706, + "num_tokens": 43522275.0, + "step": 1244 + }, + { + "epoch": 0.23119777158774374, + "grad_norm": 1.4388238191604614, + "learning_rate": 7.698019801980198e-07, + "loss": 0.4061, + "mean_token_accuracy": 0.8640608787536621, + "num_tokens": 43554865.0, + "step": 1245 + }, + { + "epoch": 0.23138347260909936, + "grad_norm": 1.4681462049484253, + "learning_rate": 7.704207920792079e-07, + "loss": 0.4215, + "mean_token_accuracy": 0.8625714778900146, + "num_tokens": 43593839.0, + "step": 1246 + }, + { + "epoch": 0.23156917363045496, + "grad_norm": 1.534790277481079, + "learning_rate": 7.71039603960396e-07, + "loss": 0.4896, + "mean_token_accuracy": 0.8428617119789124, + "num_tokens": 43638004.0, + "step": 1247 + }, + { + "epoch": 0.23175487465181058, + "grad_norm": 1.6186977624893188, + "learning_rate": 7.716584158415841e-07, + "loss": 0.4553, + "mean_token_accuracy": 0.8517822027206421, + "num_tokens": 43672610.0, + "step": 1248 + }, + { + "epoch": 0.2319405756731662, + "grad_norm": 1.4602463245391846, + "learning_rate": 7.722772277227722e-07, + "loss": 0.4605, + "mean_token_accuracy": 0.8486963510513306, + "num_tokens": 43715967.0, + "step": 1249 + }, + { + "epoch": 0.23212627669452182, + "grad_norm": 1.5513283014297485, + "learning_rate": 7.728960396039603e-07, + "loss": 0.4675, + "mean_token_accuracy": 0.8481011390686035, + "num_tokens": 43751004.0, + "step": 1250 + }, + { + "epoch": 0.23231197771587744, + "grad_norm": 1.4422229528427124, + "learning_rate": 7.735148514851485e-07, + "loss": 0.486, + "mean_token_accuracy": 0.8428928256034851, + "num_tokens": 43790127.0, + "step": 1251 + }, + { + "epoch": 0.23249767873723307, + "grad_norm": 1.565048098564148, + "learning_rate": 7.741336633663365e-07, + "loss": 0.4755, + "mean_token_accuracy": 0.8435755372047424, + "num_tokens": 43828231.0, + "step": 1252 + }, + { + "epoch": 0.23268337975858866, + "grad_norm": 1.6303343772888184, + "learning_rate": 7.747524752475248e-07, + "loss": 0.4735, + "mean_token_accuracy": 0.8423050045967102, + "num_tokens": 43863410.0, + "step": 1253 + }, + { + "epoch": 0.23286908077994428, + "grad_norm": 1.55360746383667, + "learning_rate": 7.753712871287128e-07, + "loss": 0.434, + "mean_token_accuracy": 0.8590545654296875, + "num_tokens": 43898964.0, + "step": 1254 + }, + { + "epoch": 0.2330547818012999, + "grad_norm": 1.5033200979232788, + "learning_rate": 7.75990099009901e-07, + "loss": 0.421, + "mean_token_accuracy": 0.8610086441040039, + "num_tokens": 43930499.0, + "step": 1255 + }, + { + "epoch": 0.23324048282265553, + "grad_norm": 1.5334839820861816, + "learning_rate": 7.766089108910891e-07, + "loss": 0.4921, + "mean_token_accuracy": 0.841152548789978, + "num_tokens": 43966557.0, + "step": 1256 + }, + { + "epoch": 0.23342618384401115, + "grad_norm": 1.5731157064437866, + "learning_rate": 7.772277227722772e-07, + "loss": 0.4647, + "mean_token_accuracy": 0.8518496751785278, + "num_tokens": 44001755.0, + "step": 1257 + }, + { + "epoch": 0.23361188486536677, + "grad_norm": 1.4922224283218384, + "learning_rate": 7.778465346534653e-07, + "loss": 0.4322, + "mean_token_accuracy": 0.8568310141563416, + "num_tokens": 44040768.0, + "step": 1258 + }, + { + "epoch": 0.23379758588672236, + "grad_norm": 1.4740142822265625, + "learning_rate": 7.784653465346535e-07, + "loss": 0.4293, + "mean_token_accuracy": 0.8577536344528198, + "num_tokens": 44078883.0, + "step": 1259 + }, + { + "epoch": 0.233983286908078, + "grad_norm": 1.487377405166626, + "learning_rate": 7.790841584158415e-07, + "loss": 0.544, + "mean_token_accuracy": 0.8277024030685425, + "num_tokens": 44117034.0, + "step": 1260 + }, + { + "epoch": 0.2341689879294336, + "grad_norm": 1.5697124004364014, + "learning_rate": 7.797029702970297e-07, + "loss": 0.4606, + "mean_token_accuracy": 0.8517211675643921, + "num_tokens": 44151003.0, + "step": 1261 + }, + { + "epoch": 0.23435468895078923, + "grad_norm": 1.4732869863510132, + "learning_rate": 7.803217821782177e-07, + "loss": 0.4171, + "mean_token_accuracy": 0.8637513518333435, + "num_tokens": 44187068.0, + "step": 1262 + }, + { + "epoch": 0.23454038997214485, + "grad_norm": 1.5932550430297852, + "learning_rate": 7.809405940594059e-07, + "loss": 0.5349, + "mean_token_accuracy": 0.8279194831848145, + "num_tokens": 44227966.0, + "step": 1263 + }, + { + "epoch": 0.23472609099350047, + "grad_norm": 1.6834546327590942, + "learning_rate": 7.81559405940594e-07, + "loss": 0.4406, + "mean_token_accuracy": 0.8530500531196594, + "num_tokens": 44257348.0, + "step": 1264 + }, + { + "epoch": 0.23491179201485607, + "grad_norm": 1.5671309232711792, + "learning_rate": 7.821782178217821e-07, + "loss": 0.5016, + "mean_token_accuracy": 0.8344933390617371, + "num_tokens": 44293072.0, + "step": 1265 + }, + { + "epoch": 0.2350974930362117, + "grad_norm": 1.4934707880020142, + "learning_rate": 7.827970297029702e-07, + "loss": 0.41, + "mean_token_accuracy": 0.8649945259094238, + "num_tokens": 44329882.0, + "step": 1266 + }, + { + "epoch": 0.2352831940575673, + "grad_norm": 1.5836609601974487, + "learning_rate": 7.834158415841585e-07, + "loss": 0.4781, + "mean_token_accuracy": 0.8448036313056946, + "num_tokens": 44364983.0, + "step": 1267 + }, + { + "epoch": 0.23546889507892294, + "grad_norm": 1.6754525899887085, + "learning_rate": 7.840346534653465e-07, + "loss": 0.4529, + "mean_token_accuracy": 0.8601035475730896, + "num_tokens": 44394090.0, + "step": 1268 + }, + { + "epoch": 0.23565459610027856, + "grad_norm": 1.5686665773391724, + "learning_rate": 7.846534653465347e-07, + "loss": 0.4427, + "mean_token_accuracy": 0.8553849458694458, + "num_tokens": 44429817.0, + "step": 1269 + }, + { + "epoch": 0.23584029712163418, + "grad_norm": 1.596701979637146, + "learning_rate": 7.852722772277227e-07, + "loss": 0.4852, + "mean_token_accuracy": 0.8446241617202759, + "num_tokens": 44466256.0, + "step": 1270 + }, + { + "epoch": 0.23602599814298977, + "grad_norm": 1.7270569801330566, + "learning_rate": 7.858910891089109e-07, + "loss": 0.3985, + "mean_token_accuracy": 0.8702479600906372, + "num_tokens": 44492774.0, + "step": 1271 + }, + { + "epoch": 0.2362116991643454, + "grad_norm": 1.6672645807266235, + "learning_rate": 7.86509900990099e-07, + "loss": 0.4743, + "mean_token_accuracy": 0.8471047878265381, + "num_tokens": 44522747.0, + "step": 1272 + }, + { + "epoch": 0.23639740018570102, + "grad_norm": 1.392680287361145, + "learning_rate": 7.871287128712871e-07, + "loss": 0.3968, + "mean_token_accuracy": 0.8662760257720947, + "num_tokens": 44561263.0, + "step": 1273 + }, + { + "epoch": 0.23658310120705664, + "grad_norm": 1.56077241897583, + "learning_rate": 7.877475247524752e-07, + "loss": 0.4412, + "mean_token_accuracy": 0.8531150817871094, + "num_tokens": 44595229.0, + "step": 1274 + }, + { + "epoch": 0.23676880222841226, + "grad_norm": 1.6463212966918945, + "learning_rate": 7.883663366336633e-07, + "loss": 0.4287, + "mean_token_accuracy": 0.8593493700027466, + "num_tokens": 44625549.0, + "step": 1275 + }, + { + "epoch": 0.23695450324976788, + "grad_norm": 1.6177208423614502, + "learning_rate": 7.889851485148514e-07, + "loss": 0.5024, + "mean_token_accuracy": 0.8397656083106995, + "num_tokens": 44661786.0, + "step": 1276 + }, + { + "epoch": 0.23714020427112348, + "grad_norm": 1.4270811080932617, + "learning_rate": 7.896039603960396e-07, + "loss": 0.4112, + "mean_token_accuracy": 0.8657333850860596, + "num_tokens": 44699027.0, + "step": 1277 + }, + { + "epoch": 0.2373259052924791, + "grad_norm": 1.5459648370742798, + "learning_rate": 7.902227722772276e-07, + "loss": 0.3911, + "mean_token_accuracy": 0.8685911297798157, + "num_tokens": 44727326.0, + "step": 1278 + }, + { + "epoch": 0.23751160631383472, + "grad_norm": 1.5577179193496704, + "learning_rate": 7.908415841584158e-07, + "loss": 0.42, + "mean_token_accuracy": 0.8607267141342163, + "num_tokens": 44759648.0, + "step": 1279 + }, + { + "epoch": 0.23769730733519034, + "grad_norm": 1.4520708322525024, + "learning_rate": 7.91460396039604e-07, + "loss": 0.4824, + "mean_token_accuracy": 0.8441563844680786, + "num_tokens": 44802673.0, + "step": 1280 + }, + { + "epoch": 0.23788300835654597, + "grad_norm": 1.6063169240951538, + "learning_rate": 7.920792079207921e-07, + "loss": 0.4843, + "mean_token_accuracy": 0.8414177894592285, + "num_tokens": 44835928.0, + "step": 1281 + }, + { + "epoch": 0.2380687093779016, + "grad_norm": 1.4624202251434326, + "learning_rate": 7.926980198019802e-07, + "loss": 0.441, + "mean_token_accuracy": 0.8543336987495422, + "num_tokens": 44874607.0, + "step": 1282 + }, + { + "epoch": 0.23825441039925718, + "grad_norm": 1.4613076448440552, + "learning_rate": 7.933168316831683e-07, + "loss": 0.4994, + "mean_token_accuracy": 0.8363093137741089, + "num_tokens": 44914449.0, + "step": 1283 + }, + { + "epoch": 0.2384401114206128, + "grad_norm": 1.5127795934677124, + "learning_rate": 7.939356435643564e-07, + "loss": 0.4425, + "mean_token_accuracy": 0.8529750108718872, + "num_tokens": 44949461.0, + "step": 1284 + }, + { + "epoch": 0.23862581244196843, + "grad_norm": 1.5754141807556152, + "learning_rate": 7.945544554455446e-07, + "loss": 0.4836, + "mean_token_accuracy": 0.8454117774963379, + "num_tokens": 44983222.0, + "step": 1285 + }, + { + "epoch": 0.23881151346332405, + "grad_norm": 1.593916893005371, + "learning_rate": 7.951732673267326e-07, + "loss": 0.4216, + "mean_token_accuracy": 0.8639448881149292, + "num_tokens": 45015257.0, + "step": 1286 + }, + { + "epoch": 0.23899721448467967, + "grad_norm": 1.7213512659072876, + "learning_rate": 7.957920792079208e-07, + "loss": 0.4368, + "mean_token_accuracy": 0.8542894124984741, + "num_tokens": 45044458.0, + "step": 1287 + }, + { + "epoch": 0.2391829155060353, + "grad_norm": 1.5743945837020874, + "learning_rate": 7.964108910891088e-07, + "loss": 0.4642, + "mean_token_accuracy": 0.8472671508789062, + "num_tokens": 45082939.0, + "step": 1288 + }, + { + "epoch": 0.2393686165273909, + "grad_norm": 1.5224170684814453, + "learning_rate": 7.97029702970297e-07, + "loss": 0.4551, + "mean_token_accuracy": 0.8505495190620422, + "num_tokens": 45118156.0, + "step": 1289 + }, + { + "epoch": 0.2395543175487465, + "grad_norm": 1.6953049898147583, + "learning_rate": 7.976485148514851e-07, + "loss": 0.4671, + "mean_token_accuracy": 0.8494342565536499, + "num_tokens": 45152788.0, + "step": 1290 + }, + { + "epoch": 0.23974001857010213, + "grad_norm": 1.4936819076538086, + "learning_rate": 7.982673267326732e-07, + "loss": 0.3873, + "mean_token_accuracy": 0.8720020055770874, + "num_tokens": 45187126.0, + "step": 1291 + }, + { + "epoch": 0.23992571959145775, + "grad_norm": 1.45512056350708, + "learning_rate": 7.988861386138613e-07, + "loss": 0.3862, + "mean_token_accuracy": 0.8726439476013184, + "num_tokens": 45219212.0, + "step": 1292 + }, + { + "epoch": 0.24011142061281338, + "grad_norm": 1.7388708591461182, + "learning_rate": 7.995049504950496e-07, + "loss": 0.4526, + "mean_token_accuracy": 0.852599024772644, + "num_tokens": 45246371.0, + "step": 1293 + }, + { + "epoch": 0.240297121634169, + "grad_norm": 1.6813011169433594, + "learning_rate": 8.001237623762376e-07, + "loss": 0.4712, + "mean_token_accuracy": 0.846221923828125, + "num_tokens": 45278894.0, + "step": 1294 + }, + { + "epoch": 0.2404828226555246, + "grad_norm": 1.461052417755127, + "learning_rate": 8.007425742574258e-07, + "loss": 0.4235, + "mean_token_accuracy": 0.8579013347625732, + "num_tokens": 45314263.0, + "step": 1295 + }, + { + "epoch": 0.2406685236768802, + "grad_norm": 1.5562947988510132, + "learning_rate": 8.013613861386138e-07, + "loss": 0.4601, + "mean_token_accuracy": 0.8504585027694702, + "num_tokens": 45347222.0, + "step": 1296 + }, + { + "epoch": 0.24085422469823584, + "grad_norm": 1.5501072406768799, + "learning_rate": 8.01980198019802e-07, + "loss": 0.4274, + "mean_token_accuracy": 0.8569250106811523, + "num_tokens": 45382999.0, + "step": 1297 + }, + { + "epoch": 0.24103992571959146, + "grad_norm": 1.5803505182266235, + "learning_rate": 8.025990099009901e-07, + "loss": 0.4469, + "mean_token_accuracy": 0.8518024682998657, + "num_tokens": 45413238.0, + "step": 1298 + }, + { + "epoch": 0.24122562674094708, + "grad_norm": 1.6413494348526, + "learning_rate": 8.032178217821782e-07, + "loss": 0.4225, + "mean_token_accuracy": 0.8628045916557312, + "num_tokens": 45444298.0, + "step": 1299 + }, + { + "epoch": 0.2414113277623027, + "grad_norm": 1.63889741897583, + "learning_rate": 8.038366336633663e-07, + "loss": 0.4499, + "mean_token_accuracy": 0.8536858558654785, + "num_tokens": 45475474.0, + "step": 1300 + }, + { + "epoch": 0.2415970287836583, + "grad_norm": 1.676720380783081, + "learning_rate": 8.044554455445544e-07, + "loss": 0.4322, + "mean_token_accuracy": 0.8597753047943115, + "num_tokens": 45504095.0, + "step": 1301 + }, + { + "epoch": 0.24178272980501392, + "grad_norm": 1.563698172569275, + "learning_rate": 8.050742574257425e-07, + "loss": 0.4993, + "mean_token_accuracy": 0.8374680876731873, + "num_tokens": 45539427.0, + "step": 1302 + }, + { + "epoch": 0.24196843082636954, + "grad_norm": 1.5250937938690186, + "learning_rate": 8.056930693069307e-07, + "loss": 0.4167, + "mean_token_accuracy": 0.861163854598999, + "num_tokens": 45573654.0, + "step": 1303 + }, + { + "epoch": 0.24215413184772516, + "grad_norm": 1.5871188640594482, + "learning_rate": 8.063118811881187e-07, + "loss": 0.452, + "mean_token_accuracy": 0.8482578992843628, + "num_tokens": 45606692.0, + "step": 1304 + }, + { + "epoch": 0.24233983286908078, + "grad_norm": 1.6369822025299072, + "learning_rate": 8.069306930693069e-07, + "loss": 0.4423, + "mean_token_accuracy": 0.8545703291893005, + "num_tokens": 45636259.0, + "step": 1305 + }, + { + "epoch": 0.2425255338904364, + "grad_norm": 1.570095181465149, + "learning_rate": 8.075495049504951e-07, + "loss": 0.499, + "mean_token_accuracy": 0.8354225158691406, + "num_tokens": 45675460.0, + "step": 1306 + }, + { + "epoch": 0.24271123491179203, + "grad_norm": 1.4395654201507568, + "learning_rate": 8.081683168316832e-07, + "loss": 0.4371, + "mean_token_accuracy": 0.8573066592216492, + "num_tokens": 45711996.0, + "step": 1307 + }, + { + "epoch": 0.24289693593314762, + "grad_norm": 1.4619207382202148, + "learning_rate": 8.087871287128713e-07, + "loss": 0.3997, + "mean_token_accuracy": 0.8698283433914185, + "num_tokens": 45748993.0, + "step": 1308 + }, + { + "epoch": 0.24308263695450325, + "grad_norm": 1.6419230699539185, + "learning_rate": 8.094059405940594e-07, + "loss": 0.4562, + "mean_token_accuracy": 0.8483133316040039, + "num_tokens": 45779255.0, + "step": 1309 + }, + { + "epoch": 0.24326833797585887, + "grad_norm": 1.68636155128479, + "learning_rate": 8.100247524752475e-07, + "loss": 0.3736, + "mean_token_accuracy": 0.8738021850585938, + "num_tokens": 45806137.0, + "step": 1310 + }, + { + "epoch": 0.2434540389972145, + "grad_norm": 1.5067870616912842, + "learning_rate": 8.106435643564357e-07, + "loss": 0.4212, + "mean_token_accuracy": 0.8610032200813293, + "num_tokens": 45841215.0, + "step": 1311 + }, + { + "epoch": 0.2436397400185701, + "grad_norm": 1.2717363834381104, + "learning_rate": 8.112623762376237e-07, + "loss": 0.402, + "mean_token_accuracy": 0.8687849044799805, + "num_tokens": 45885763.0, + "step": 1312 + }, + { + "epoch": 0.24382544103992573, + "grad_norm": 1.5936086177825928, + "learning_rate": 8.118811881188119e-07, + "loss": 0.4443, + "mean_token_accuracy": 0.855157732963562, + "num_tokens": 45918341.0, + "step": 1313 + }, + { + "epoch": 0.24401114206128133, + "grad_norm": 1.4886248111724854, + "learning_rate": 8.125e-07, + "loss": 0.4352, + "mean_token_accuracy": 0.8552592992782593, + "num_tokens": 45952445.0, + "step": 1314 + }, + { + "epoch": 0.24419684308263695, + "grad_norm": 1.708380937576294, + "learning_rate": 8.13118811881188e-07, + "loss": 0.4361, + "mean_token_accuracy": 0.8580891489982605, + "num_tokens": 45980414.0, + "step": 1315 + }, + { + "epoch": 0.24438254410399257, + "grad_norm": 1.5287585258483887, + "learning_rate": 8.137376237623762e-07, + "loss": 0.4237, + "mean_token_accuracy": 0.860247015953064, + "num_tokens": 46012956.0, + "step": 1316 + }, + { + "epoch": 0.2445682451253482, + "grad_norm": 1.4732983112335205, + "learning_rate": 8.143564356435642e-07, + "loss": 0.4453, + "mean_token_accuracy": 0.852357029914856, + "num_tokens": 46050002.0, + "step": 1317 + }, + { + "epoch": 0.24475394614670382, + "grad_norm": 1.4367311000823975, + "learning_rate": 8.149752475247524e-07, + "loss": 0.4262, + "mean_token_accuracy": 0.8595421314239502, + "num_tokens": 46087295.0, + "step": 1318 + }, + { + "epoch": 0.24493964716805944, + "grad_norm": 1.4418565034866333, + "learning_rate": 8.155940594059405e-07, + "loss": 0.4236, + "mean_token_accuracy": 0.8618999719619751, + "num_tokens": 46127014.0, + "step": 1319 + }, + { + "epoch": 0.24512534818941503, + "grad_norm": 1.4841254949569702, + "learning_rate": 8.162128712871287e-07, + "loss": 0.4186, + "mean_token_accuracy": 0.8592259883880615, + "num_tokens": 46164373.0, + "step": 1320 + }, + { + "epoch": 0.24531104921077065, + "grad_norm": 1.5779815912246704, + "learning_rate": 8.168316831683168e-07, + "loss": 0.4519, + "mean_token_accuracy": 0.8519753217697144, + "num_tokens": 46195711.0, + "step": 1321 + }, + { + "epoch": 0.24549675023212628, + "grad_norm": 1.5444676876068115, + "learning_rate": 8.174504950495049e-07, + "loss": 0.4676, + "mean_token_accuracy": 0.8476699590682983, + "num_tokens": 46235475.0, + "step": 1322 + }, + { + "epoch": 0.2456824512534819, + "grad_norm": 1.6887174844741821, + "learning_rate": 8.18069306930693e-07, + "loss": 0.4317, + "mean_token_accuracy": 0.8540675640106201, + "num_tokens": 46262990.0, + "step": 1323 + }, + { + "epoch": 0.24586815227483752, + "grad_norm": 1.6910669803619385, + "learning_rate": 8.186881188118812e-07, + "loss": 0.4749, + "mean_token_accuracy": 0.8497043251991272, + "num_tokens": 46291339.0, + "step": 1324 + }, + { + "epoch": 0.24605385329619314, + "grad_norm": 1.5658704042434692, + "learning_rate": 8.193069306930692e-07, + "loss": 0.4407, + "mean_token_accuracy": 0.8569360971450806, + "num_tokens": 46323860.0, + "step": 1325 + }, + { + "epoch": 0.24623955431754874, + "grad_norm": 1.4798405170440674, + "learning_rate": 8.199257425742574e-07, + "loss": 0.4362, + "mean_token_accuracy": 0.8540470004081726, + "num_tokens": 46359286.0, + "step": 1326 + }, + { + "epoch": 0.24642525533890436, + "grad_norm": 1.5508332252502441, + "learning_rate": 8.205445544554455e-07, + "loss": 0.4977, + "mean_token_accuracy": 0.8398610353469849, + "num_tokens": 46394086.0, + "step": 1327 + }, + { + "epoch": 0.24661095636025998, + "grad_norm": 1.426276683807373, + "learning_rate": 8.211633663366336e-07, + "loss": 0.4721, + "mean_token_accuracy": 0.8410028219223022, + "num_tokens": 46437205.0, + "step": 1328 + }, + { + "epoch": 0.2467966573816156, + "grad_norm": 1.3468936681747437, + "learning_rate": 8.217821782178217e-07, + "loss": 0.4161, + "mean_token_accuracy": 0.8623021245002747, + "num_tokens": 46476608.0, + "step": 1329 + }, + { + "epoch": 0.24698235840297122, + "grad_norm": 1.7193658351898193, + "learning_rate": 8.224009900990098e-07, + "loss": 0.4968, + "mean_token_accuracy": 0.845201849937439, + "num_tokens": 46507592.0, + "step": 1330 + }, + { + "epoch": 0.24716805942432685, + "grad_norm": 1.4408080577850342, + "learning_rate": 8.230198019801979e-07, + "loss": 0.4184, + "mean_token_accuracy": 0.8630061745643616, + "num_tokens": 46545064.0, + "step": 1331 + }, + { + "epoch": 0.24735376044568244, + "grad_norm": 1.5695360898971558, + "learning_rate": 8.236386138613861e-07, + "loss": 0.3916, + "mean_token_accuracy": 0.8694572448730469, + "num_tokens": 46580537.0, + "step": 1332 + }, + { + "epoch": 0.24753946146703806, + "grad_norm": 1.5859814882278442, + "learning_rate": 8.242574257425742e-07, + "loss": 0.447, + "mean_token_accuracy": 0.854317843914032, + "num_tokens": 46615307.0, + "step": 1333 + }, + { + "epoch": 0.24772516248839369, + "grad_norm": 1.606664776802063, + "learning_rate": 8.248762376237624e-07, + "loss": 0.4481, + "mean_token_accuracy": 0.8546162843704224, + "num_tokens": 46651517.0, + "step": 1334 + }, + { + "epoch": 0.2479108635097493, + "grad_norm": 1.5809502601623535, + "learning_rate": 8.254950495049504e-07, + "loss": 0.4619, + "mean_token_accuracy": 0.8465465307235718, + "num_tokens": 46684588.0, + "step": 1335 + }, + { + "epoch": 0.24809656453110493, + "grad_norm": 1.5259710550308228, + "learning_rate": 8.261138613861386e-07, + "loss": 0.4265, + "mean_token_accuracy": 0.8577753305435181, + "num_tokens": 46718938.0, + "step": 1336 + }, + { + "epoch": 0.24828226555246055, + "grad_norm": 1.5222784280776978, + "learning_rate": 8.267326732673267e-07, + "loss": 0.4069, + "mean_token_accuracy": 0.8651472926139832, + "num_tokens": 46751411.0, + "step": 1337 + }, + { + "epoch": 0.24846796657381615, + "grad_norm": 1.4592324495315552, + "learning_rate": 8.273514851485148e-07, + "loss": 0.4151, + "mean_token_accuracy": 0.8633458614349365, + "num_tokens": 46789004.0, + "step": 1338 + }, + { + "epoch": 0.24865366759517177, + "grad_norm": 1.5530774593353271, + "learning_rate": 8.279702970297029e-07, + "loss": 0.4604, + "mean_token_accuracy": 0.8472229242324829, + "num_tokens": 46823149.0, + "step": 1339 + }, + { + "epoch": 0.2488393686165274, + "grad_norm": 1.4502328634262085, + "learning_rate": 8.285891089108911e-07, + "loss": 0.416, + "mean_token_accuracy": 0.8654627203941345, + "num_tokens": 46857063.0, + "step": 1340 + }, + { + "epoch": 0.249025069637883, + "grad_norm": 1.5788930654525757, + "learning_rate": 8.292079207920791e-07, + "loss": 0.4648, + "mean_token_accuracy": 0.8476195335388184, + "num_tokens": 46890055.0, + "step": 1341 + }, + { + "epoch": 0.24921077065923863, + "grad_norm": 1.5927342176437378, + "learning_rate": 8.298267326732673e-07, + "loss": 0.446, + "mean_token_accuracy": 0.8522526621818542, + "num_tokens": 46920732.0, + "step": 1342 + }, + { + "epoch": 0.24939647168059426, + "grad_norm": 1.4258978366851807, + "learning_rate": 8.304455445544553e-07, + "loss": 0.4463, + "mean_token_accuracy": 0.8514683842658997, + "num_tokens": 46960903.0, + "step": 1343 + }, + { + "epoch": 0.24958217270194985, + "grad_norm": 1.4291619062423706, + "learning_rate": 8.310643564356435e-07, + "loss": 0.4852, + "mean_token_accuracy": 0.8433784246444702, + "num_tokens": 47001022.0, + "step": 1344 + }, + { + "epoch": 0.24976787372330547, + "grad_norm": 1.5138509273529053, + "learning_rate": 8.316831683168316e-07, + "loss": 0.4297, + "mean_token_accuracy": 0.8584686517715454, + "num_tokens": 47037007.0, + "step": 1345 + }, + { + "epoch": 0.2499535747446611, + "grad_norm": 1.540850043296814, + "learning_rate": 8.323019801980198e-07, + "loss": 0.3944, + "mean_token_accuracy": 0.8688094615936279, + "num_tokens": 47066851.0, + "step": 1346 + }, + { + "epoch": 0.2501392757660167, + "grad_norm": 1.5894583463668823, + "learning_rate": 8.329207920792079e-07, + "loss": 0.48, + "mean_token_accuracy": 0.8438023328781128, + "num_tokens": 47100078.0, + "step": 1347 + }, + { + "epoch": 0.2503249767873723, + "grad_norm": 1.6223466396331787, + "learning_rate": 8.33539603960396e-07, + "loss": 0.4454, + "mean_token_accuracy": 0.8514714241027832, + "num_tokens": 47130134.0, + "step": 1348 + }, + { + "epoch": 0.25051067780872793, + "grad_norm": 1.408476710319519, + "learning_rate": 8.341584158415841e-07, + "loss": 0.4465, + "mean_token_accuracy": 0.8526700139045715, + "num_tokens": 47168810.0, + "step": 1349 + }, + { + "epoch": 0.25069637883008355, + "grad_norm": 1.5059738159179688, + "learning_rate": 8.347772277227723e-07, + "loss": 0.458, + "mean_token_accuracy": 0.8535265326499939, + "num_tokens": 47206013.0, + "step": 1350 + }, + { + "epoch": 0.2508820798514392, + "grad_norm": 1.5220997333526611, + "learning_rate": 8.353960396039603e-07, + "loss": 0.4424, + "mean_token_accuracy": 0.8553261756896973, + "num_tokens": 47239064.0, + "step": 1351 + }, + { + "epoch": 0.2510677808727948, + "grad_norm": 1.6192196607589722, + "learning_rate": 8.360148514851485e-07, + "loss": 0.5051, + "mean_token_accuracy": 0.8402895927429199, + "num_tokens": 47276830.0, + "step": 1352 + }, + { + "epoch": 0.2512534818941504, + "grad_norm": 1.5581525564193726, + "learning_rate": 8.366336633663366e-07, + "loss": 0.4095, + "mean_token_accuracy": 0.8649485111236572, + "num_tokens": 47311889.0, + "step": 1353 + }, + { + "epoch": 0.25143918291550604, + "grad_norm": 1.9697415828704834, + "learning_rate": 8.372524752475247e-07, + "loss": 0.4044, + "mean_token_accuracy": 0.8652927875518799, + "num_tokens": 47334674.0, + "step": 1354 + }, + { + "epoch": 0.25162488393686167, + "grad_norm": 1.5674619674682617, + "learning_rate": 8.378712871287128e-07, + "loss": 0.4582, + "mean_token_accuracy": 0.8487696647644043, + "num_tokens": 47369928.0, + "step": 1355 + }, + { + "epoch": 0.2518105849582173, + "grad_norm": 1.8989965915679932, + "learning_rate": 8.384900990099009e-07, + "loss": 0.4877, + "mean_token_accuracy": 0.8430166244506836, + "num_tokens": 47397199.0, + "step": 1356 + }, + { + "epoch": 0.2519962859795729, + "grad_norm": 1.6410224437713623, + "learning_rate": 8.39108910891089e-07, + "loss": 0.5034, + "mean_token_accuracy": 0.8386853933334351, + "num_tokens": 47430837.0, + "step": 1357 + }, + { + "epoch": 0.25218198700092853, + "grad_norm": 1.536880373954773, + "learning_rate": 8.397277227722772e-07, + "loss": 0.4908, + "mean_token_accuracy": 0.8420157432556152, + "num_tokens": 47467654.0, + "step": 1358 + }, + { + "epoch": 0.2523676880222841, + "grad_norm": 1.6967430114746094, + "learning_rate": 8.403465346534653e-07, + "loss": 0.406, + "mean_token_accuracy": 0.8669016361236572, + "num_tokens": 47497253.0, + "step": 1359 + }, + { + "epoch": 0.2525533890436397, + "grad_norm": 1.7527426481246948, + "learning_rate": 8.409653465346535e-07, + "loss": 0.4884, + "mean_token_accuracy": 0.8412431478500366, + "num_tokens": 47534158.0, + "step": 1360 + }, + { + "epoch": 0.25273909006499534, + "grad_norm": 1.5567851066589355, + "learning_rate": 8.415841584158416e-07, + "loss": 0.4765, + "mean_token_accuracy": 0.8455890417098999, + "num_tokens": 47569831.0, + "step": 1361 + }, + { + "epoch": 0.25292479108635096, + "grad_norm": 1.716416358947754, + "learning_rate": 8.422029702970297e-07, + "loss": 0.4871, + "mean_token_accuracy": 0.8392882943153381, + "num_tokens": 47598146.0, + "step": 1362 + }, + { + "epoch": 0.2531104921077066, + "grad_norm": 1.5469748973846436, + "learning_rate": 8.428217821782178e-07, + "loss": 0.422, + "mean_token_accuracy": 0.8598909378051758, + "num_tokens": 47630924.0, + "step": 1363 + }, + { + "epoch": 0.2532961931290622, + "grad_norm": 1.5838449001312256, + "learning_rate": 8.434405940594059e-07, + "loss": 0.4243, + "mean_token_accuracy": 0.8607363700866699, + "num_tokens": 47668359.0, + "step": 1364 + }, + { + "epoch": 0.25348189415041783, + "grad_norm": 1.4944490194320679, + "learning_rate": 8.44059405940594e-07, + "loss": 0.4087, + "mean_token_accuracy": 0.8652008771896362, + "num_tokens": 47701565.0, + "step": 1365 + }, + { + "epoch": 0.25366759517177345, + "grad_norm": 1.4873143434524536, + "learning_rate": 8.446782178217822e-07, + "loss": 0.4312, + "mean_token_accuracy": 0.8578606843948364, + "num_tokens": 47739621.0, + "step": 1366 + }, + { + "epoch": 0.2538532961931291, + "grad_norm": 1.6791937351226807, + "learning_rate": 8.452970297029702e-07, + "loss": 0.4494, + "mean_token_accuracy": 0.8514049053192139, + "num_tokens": 47768807.0, + "step": 1367 + }, + { + "epoch": 0.2540389972144847, + "grad_norm": 1.4951730966567993, + "learning_rate": 8.459158415841584e-07, + "loss": 0.5255, + "mean_token_accuracy": 0.8320944905281067, + "num_tokens": 47809449.0, + "step": 1368 + }, + { + "epoch": 0.2542246982358403, + "grad_norm": 1.3945178985595703, + "learning_rate": 8.465346534653464e-07, + "loss": 0.3889, + "mean_token_accuracy": 0.8704866170883179, + "num_tokens": 47848563.0, + "step": 1369 + }, + { + "epoch": 0.25441039925719594, + "grad_norm": 1.5162242650985718, + "learning_rate": 8.471534653465346e-07, + "loss": 0.4269, + "mean_token_accuracy": 0.8600672483444214, + "num_tokens": 47881074.0, + "step": 1370 + }, + { + "epoch": 0.2545961002785515, + "grad_norm": 1.4527409076690674, + "learning_rate": 8.477722772277227e-07, + "loss": 0.4288, + "mean_token_accuracy": 0.8593575954437256, + "num_tokens": 47915955.0, + "step": 1371 + }, + { + "epoch": 0.25478180129990713, + "grad_norm": 1.4403079748153687, + "learning_rate": 8.483910891089109e-07, + "loss": 0.4331, + "mean_token_accuracy": 0.8558533787727356, + "num_tokens": 47958117.0, + "step": 1372 + }, + { + "epoch": 0.25496750232126275, + "grad_norm": 1.587363600730896, + "learning_rate": 8.49009900990099e-07, + "loss": 0.3938, + "mean_token_accuracy": 0.8714418411254883, + "num_tokens": 47990664.0, + "step": 1373 + }, + { + "epoch": 0.2551532033426184, + "grad_norm": 1.416898250579834, + "learning_rate": 8.496287128712872e-07, + "loss": 0.4344, + "mean_token_accuracy": 0.8587368130683899, + "num_tokens": 48027864.0, + "step": 1374 + }, + { + "epoch": 0.255338904363974, + "grad_norm": 1.4325182437896729, + "learning_rate": 8.502475247524752e-07, + "loss": 0.4109, + "mean_token_accuracy": 0.8629179000854492, + "num_tokens": 48066827.0, + "step": 1375 + }, + { + "epoch": 0.2555246053853296, + "grad_norm": 1.5281870365142822, + "learning_rate": 8.508663366336634e-07, + "loss": 0.4951, + "mean_token_accuracy": 0.8441951274871826, + "num_tokens": 48103566.0, + "step": 1376 + }, + { + "epoch": 0.25571030640668524, + "grad_norm": 1.5021936893463135, + "learning_rate": 8.514851485148514e-07, + "loss": 0.4505, + "mean_token_accuracy": 0.8519743084907532, + "num_tokens": 48145588.0, + "step": 1377 + }, + { + "epoch": 0.25589600742804086, + "grad_norm": 1.491221308708191, + "learning_rate": 8.521039603960396e-07, + "loss": 0.4616, + "mean_token_accuracy": 0.8482010960578918, + "num_tokens": 48182351.0, + "step": 1378 + }, + { + "epoch": 0.2560817084493965, + "grad_norm": 1.498822808265686, + "learning_rate": 8.527227722772277e-07, + "loss": 0.3514, + "mean_token_accuracy": 0.8796849250793457, + "num_tokens": 48214030.0, + "step": 1379 + }, + { + "epoch": 0.2562674094707521, + "grad_norm": 1.61158287525177, + "learning_rate": 8.533415841584158e-07, + "loss": 0.4363, + "mean_token_accuracy": 0.8566436767578125, + "num_tokens": 48247397.0, + "step": 1380 + }, + { + "epoch": 0.2564531104921077, + "grad_norm": 1.6768876314163208, + "learning_rate": 8.539603960396039e-07, + "loss": 0.4464, + "mean_token_accuracy": 0.8546525239944458, + "num_tokens": 48276806.0, + "step": 1381 + }, + { + "epoch": 0.25663881151346335, + "grad_norm": 1.5370402336120605, + "learning_rate": 8.54579207920792e-07, + "loss": 0.4807, + "mean_token_accuracy": 0.8413091897964478, + "num_tokens": 48316759.0, + "step": 1382 + }, + { + "epoch": 0.2568245125348189, + "grad_norm": 1.4544517993927002, + "learning_rate": 8.551980198019801e-07, + "loss": 0.4733, + "mean_token_accuracy": 0.840563178062439, + "num_tokens": 48359870.0, + "step": 1383 + }, + { + "epoch": 0.25701021355617454, + "grad_norm": 1.4629462957382202, + "learning_rate": 8.558168316831683e-07, + "loss": 0.4509, + "mean_token_accuracy": 0.8580068945884705, + "num_tokens": 48398926.0, + "step": 1384 + }, + { + "epoch": 0.25719591457753016, + "grad_norm": 1.4856401681900024, + "learning_rate": 8.564356435643563e-07, + "loss": 0.4811, + "mean_token_accuracy": 0.8414579629898071, + "num_tokens": 48439231.0, + "step": 1385 + }, + { + "epoch": 0.2573816155988858, + "grad_norm": 1.7489441633224487, + "learning_rate": 8.570544554455446e-07, + "loss": 0.4602, + "mean_token_accuracy": 0.847785472869873, + "num_tokens": 48471078.0, + "step": 1386 + }, + { + "epoch": 0.2575673166202414, + "grad_norm": 1.653593897819519, + "learning_rate": 8.576732673267327e-07, + "loss": 0.4188, + "mean_token_accuracy": 0.8594369888305664, + "num_tokens": 48504031.0, + "step": 1387 + }, + { + "epoch": 0.257753017641597, + "grad_norm": 1.4741642475128174, + "learning_rate": 8.582920792079208e-07, + "loss": 0.442, + "mean_token_accuracy": 0.8578109741210938, + "num_tokens": 48544727.0, + "step": 1388 + }, + { + "epoch": 0.25793871866295265, + "grad_norm": 1.6915634870529175, + "learning_rate": 8.589108910891089e-07, + "loss": 0.5097, + "mean_token_accuracy": 0.8417852520942688, + "num_tokens": 48576915.0, + "step": 1389 + }, + { + "epoch": 0.25812441968430827, + "grad_norm": 1.4245193004608154, + "learning_rate": 8.59529702970297e-07, + "loss": 0.4589, + "mean_token_accuracy": 0.8502602577209473, + "num_tokens": 48615447.0, + "step": 1390 + }, + { + "epoch": 0.2583101207056639, + "grad_norm": 1.823439121246338, + "learning_rate": 8.601485148514851e-07, + "loss": 0.4787, + "mean_token_accuracy": 0.8393340110778809, + "num_tokens": 48643407.0, + "step": 1391 + }, + { + "epoch": 0.2584958217270195, + "grad_norm": 1.430262565612793, + "learning_rate": 8.607673267326733e-07, + "loss": 0.4122, + "mean_token_accuracy": 0.865975022315979, + "num_tokens": 48684796.0, + "step": 1392 + }, + { + "epoch": 0.25868152274837514, + "grad_norm": 1.5810447931289673, + "learning_rate": 8.613861386138613e-07, + "loss": 0.4885, + "mean_token_accuracy": 0.8436577320098877, + "num_tokens": 48718173.0, + "step": 1393 + }, + { + "epoch": 0.25886722376973076, + "grad_norm": 1.5103425979614258, + "learning_rate": 8.620049504950495e-07, + "loss": 0.4573, + "mean_token_accuracy": 0.8525353670120239, + "num_tokens": 48753057.0, + "step": 1394 + }, + { + "epoch": 0.2590529247910863, + "grad_norm": 1.547748327255249, + "learning_rate": 8.626237623762375e-07, + "loss": 0.4693, + "mean_token_accuracy": 0.8442350625991821, + "num_tokens": 48784838.0, + "step": 1395 + }, + { + "epoch": 0.25923862581244195, + "grad_norm": 1.5470328330993652, + "learning_rate": 8.632425742574257e-07, + "loss": 0.4472, + "mean_token_accuracy": 0.854603111743927, + "num_tokens": 48819182.0, + "step": 1396 + }, + { + "epoch": 0.25942432683379757, + "grad_norm": 1.5690010786056519, + "learning_rate": 8.638613861386138e-07, + "loss": 0.4488, + "mean_token_accuracy": 0.853436291217804, + "num_tokens": 48853431.0, + "step": 1397 + }, + { + "epoch": 0.2596100278551532, + "grad_norm": 1.528228759765625, + "learning_rate": 8.644801980198019e-07, + "loss": 0.4021, + "mean_token_accuracy": 0.8700283765792847, + "num_tokens": 48884102.0, + "step": 1398 + }, + { + "epoch": 0.2597957288765088, + "grad_norm": 1.5522593259811401, + "learning_rate": 8.650990099009901e-07, + "loss": 0.4413, + "mean_token_accuracy": 0.8561354875564575, + "num_tokens": 48920799.0, + "step": 1399 + }, + { + "epoch": 0.25998142989786444, + "grad_norm": 1.5151212215423584, + "learning_rate": 8.657178217821783e-07, + "loss": 0.4306, + "mean_token_accuracy": 0.8613249063491821, + "num_tokens": 48956378.0, + "step": 1400 + }, + { + "epoch": 0.26016713091922006, + "grad_norm": 1.5874499082565308, + "learning_rate": 8.663366336633663e-07, + "loss": 0.4507, + "mean_token_accuracy": 0.8538218140602112, + "num_tokens": 48991487.0, + "step": 1401 + }, + { + "epoch": 0.2603528319405757, + "grad_norm": 1.4725996255874634, + "learning_rate": 8.669554455445545e-07, + "loss": 0.4622, + "mean_token_accuracy": 0.8496329188346863, + "num_tokens": 49030909.0, + "step": 1402 + }, + { + "epoch": 0.2605385329619313, + "grad_norm": 1.4474455118179321, + "learning_rate": 8.675742574257425e-07, + "loss": 0.4504, + "mean_token_accuracy": 0.8525441884994507, + "num_tokens": 49070395.0, + "step": 1403 + }, + { + "epoch": 0.2607242339832869, + "grad_norm": 1.5532361268997192, + "learning_rate": 8.681930693069307e-07, + "loss": 0.4523, + "mean_token_accuracy": 0.852595865726471, + "num_tokens": 49104665.0, + "step": 1404 + }, + { + "epoch": 0.26090993500464255, + "grad_norm": 1.6019502878189087, + "learning_rate": 8.688118811881188e-07, + "loss": 0.4811, + "mean_token_accuracy": 0.8475838303565979, + "num_tokens": 49138001.0, + "step": 1405 + }, + { + "epoch": 0.26109563602599817, + "grad_norm": 1.5660579204559326, + "learning_rate": 8.694306930693069e-07, + "loss": 0.4622, + "mean_token_accuracy": 0.8472909331321716, + "num_tokens": 49173765.0, + "step": 1406 + }, + { + "epoch": 0.26128133704735373, + "grad_norm": 1.6195484399795532, + "learning_rate": 8.70049504950495e-07, + "loss": 0.4809, + "mean_token_accuracy": 0.8443492650985718, + "num_tokens": 49209256.0, + "step": 1407 + }, + { + "epoch": 0.26146703806870936, + "grad_norm": 1.4491949081420898, + "learning_rate": 8.706683168316832e-07, + "loss": 0.3863, + "mean_token_accuracy": 0.8681790232658386, + "num_tokens": 49246023.0, + "step": 1408 + }, + { + "epoch": 0.261652739090065, + "grad_norm": 1.325027346611023, + "learning_rate": 8.712871287128712e-07, + "loss": 0.4594, + "mean_token_accuracy": 0.8495668172836304, + "num_tokens": 49293517.0, + "step": 1409 + }, + { + "epoch": 0.2618384401114206, + "grad_norm": 1.5661309957504272, + "learning_rate": 8.719059405940594e-07, + "loss": 0.4629, + "mean_token_accuracy": 0.8507068157196045, + "num_tokens": 49328674.0, + "step": 1410 + }, + { + "epoch": 0.2620241411327762, + "grad_norm": 1.4750173091888428, + "learning_rate": 8.725247524752474e-07, + "loss": 0.4883, + "mean_token_accuracy": 0.8404842615127563, + "num_tokens": 49365198.0, + "step": 1411 + }, + { + "epoch": 0.26220984215413184, + "grad_norm": 1.4841523170471191, + "learning_rate": 8.731435643564357e-07, + "loss": 0.4755, + "mean_token_accuracy": 0.8439075946807861, + "num_tokens": 49402178.0, + "step": 1412 + }, + { + "epoch": 0.26239554317548747, + "grad_norm": 1.6341428756713867, + "learning_rate": 8.737623762376238e-07, + "loss": 0.4389, + "mean_token_accuracy": 0.8551921844482422, + "num_tokens": 49431288.0, + "step": 1413 + }, + { + "epoch": 0.2625812441968431, + "grad_norm": 1.4472037553787231, + "learning_rate": 8.743811881188119e-07, + "loss": 0.3855, + "mean_token_accuracy": 0.8685099482536316, + "num_tokens": 49468916.0, + "step": 1414 + }, + { + "epoch": 0.2627669452181987, + "grad_norm": 1.4744393825531006, + "learning_rate": 8.75e-07, + "loss": 0.4685, + "mean_token_accuracy": 0.8464106917381287, + "num_tokens": 49508756.0, + "step": 1415 + }, + { + "epoch": 0.26295264623955433, + "grad_norm": 1.4399170875549316, + "learning_rate": 8.75618811881188e-07, + "loss": 0.4332, + "mean_token_accuracy": 0.8557384610176086, + "num_tokens": 49549949.0, + "step": 1416 + }, + { + "epoch": 0.26313834726090995, + "grad_norm": 1.4928193092346191, + "learning_rate": 8.762376237623762e-07, + "loss": 0.4461, + "mean_token_accuracy": 0.8567948937416077, + "num_tokens": 49587788.0, + "step": 1417 + }, + { + "epoch": 0.2633240482822656, + "grad_norm": 1.472657561302185, + "learning_rate": 8.768564356435643e-07, + "loss": 0.4405, + "mean_token_accuracy": 0.8533748388290405, + "num_tokens": 49621735.0, + "step": 1418 + }, + { + "epoch": 0.26350974930362114, + "grad_norm": 1.4798789024353027, + "learning_rate": 8.774752475247524e-07, + "loss": 0.4364, + "mean_token_accuracy": 0.8547202944755554, + "num_tokens": 49659869.0, + "step": 1419 + }, + { + "epoch": 0.26369545032497677, + "grad_norm": 1.417694330215454, + "learning_rate": 8.780940594059405e-07, + "loss": 0.4345, + "mean_token_accuracy": 0.8601188659667969, + "num_tokens": 49697516.0, + "step": 1420 + }, + { + "epoch": 0.2638811513463324, + "grad_norm": 1.4553139209747314, + "learning_rate": 8.787128712871287e-07, + "loss": 0.4672, + "mean_token_accuracy": 0.845236599445343, + "num_tokens": 49736768.0, + "step": 1421 + }, + { + "epoch": 0.264066852367688, + "grad_norm": 1.3964389562606812, + "learning_rate": 8.793316831683167e-07, + "loss": 0.4264, + "mean_token_accuracy": 0.8577466011047363, + "num_tokens": 49778940.0, + "step": 1422 + }, + { + "epoch": 0.26425255338904363, + "grad_norm": 1.5767652988433838, + "learning_rate": 8.799504950495049e-07, + "loss": 0.4169, + "mean_token_accuracy": 0.858770489692688, + "num_tokens": 49811835.0, + "step": 1423 + }, + { + "epoch": 0.26443825441039925, + "grad_norm": 1.498598337173462, + "learning_rate": 8.805693069306929e-07, + "loss": 0.5003, + "mean_token_accuracy": 0.8367262482643127, + "num_tokens": 49852298.0, + "step": 1424 + }, + { + "epoch": 0.2646239554317549, + "grad_norm": 1.6039597988128662, + "learning_rate": 8.811881188118812e-07, + "loss": 0.5033, + "mean_token_accuracy": 0.8371951580047607, + "num_tokens": 49885866.0, + "step": 1425 + }, + { + "epoch": 0.2648096564531105, + "grad_norm": 1.4469302892684937, + "learning_rate": 8.818069306930693e-07, + "loss": 0.4566, + "mean_token_accuracy": 0.8507590293884277, + "num_tokens": 49923531.0, + "step": 1426 + }, + { + "epoch": 0.2649953574744661, + "grad_norm": 1.4342410564422607, + "learning_rate": 8.824257425742574e-07, + "loss": 0.4586, + "mean_token_accuracy": 0.8524220585823059, + "num_tokens": 49964212.0, + "step": 1427 + }, + { + "epoch": 0.26518105849582174, + "grad_norm": 1.6602392196655273, + "learning_rate": 8.830445544554455e-07, + "loss": 0.4977, + "mean_token_accuracy": 0.8411914110183716, + "num_tokens": 49998133.0, + "step": 1428 + }, + { + "epoch": 0.26536675951717736, + "grad_norm": 1.4628303050994873, + "learning_rate": 8.836633663366337e-07, + "loss": 0.4567, + "mean_token_accuracy": 0.8505716919898987, + "num_tokens": 50039457.0, + "step": 1429 + }, + { + "epoch": 0.265552460538533, + "grad_norm": 1.4471473693847656, + "learning_rate": 8.842821782178217e-07, + "loss": 0.3939, + "mean_token_accuracy": 0.8654536008834839, + "num_tokens": 50074473.0, + "step": 1430 + }, + { + "epoch": 0.26573816155988855, + "grad_norm": 1.5760308504104614, + "learning_rate": 8.849009900990099e-07, + "loss": 0.4455, + "mean_token_accuracy": 0.8532618284225464, + "num_tokens": 50110596.0, + "step": 1431 + }, + { + "epoch": 0.2659238625812442, + "grad_norm": 1.4209058284759521, + "learning_rate": 8.855198019801979e-07, + "loss": 0.4185, + "mean_token_accuracy": 0.8626872301101685, + "num_tokens": 50144908.0, + "step": 1432 + }, + { + "epoch": 0.2661095636025998, + "grad_norm": 1.6985317468643188, + "learning_rate": 8.861386138613861e-07, + "loss": 0.4662, + "mean_token_accuracy": 0.8472907543182373, + "num_tokens": 50173252.0, + "step": 1433 + }, + { + "epoch": 0.2662952646239554, + "grad_norm": 1.530991554260254, + "learning_rate": 8.867574257425742e-07, + "loss": 0.4328, + "mean_token_accuracy": 0.8580247759819031, + "num_tokens": 50209461.0, + "step": 1434 + }, + { + "epoch": 0.26648096564531104, + "grad_norm": 1.489044427871704, + "learning_rate": 8.873762376237623e-07, + "loss": 0.4974, + "mean_token_accuracy": 0.8410601615905762, + "num_tokens": 50247794.0, + "step": 1435 + }, + { + "epoch": 0.26666666666666666, + "grad_norm": 1.3873835802078247, + "learning_rate": 8.879950495049504e-07, + "loss": 0.4047, + "mean_token_accuracy": 0.8661522269248962, + "num_tokens": 50286821.0, + "step": 1436 + }, + { + "epoch": 0.2668523676880223, + "grad_norm": 1.4041320085525513, + "learning_rate": 8.886138613861385e-07, + "loss": 0.4129, + "mean_token_accuracy": 0.8615561723709106, + "num_tokens": 50327492.0, + "step": 1437 + }, + { + "epoch": 0.2670380687093779, + "grad_norm": 1.4732288122177124, + "learning_rate": 8.892326732673266e-07, + "loss": 0.4002, + "mean_token_accuracy": 0.8666847944259644, + "num_tokens": 50362256.0, + "step": 1438 + }, + { + "epoch": 0.26722376973073353, + "grad_norm": 1.4535679817199707, + "learning_rate": 8.898514851485149e-07, + "loss": 0.4088, + "mean_token_accuracy": 0.8631860613822937, + "num_tokens": 50399807.0, + "step": 1439 + }, + { + "epoch": 0.26740947075208915, + "grad_norm": 1.7003095149993896, + "learning_rate": 8.904702970297029e-07, + "loss": 0.4238, + "mean_token_accuracy": 0.8646475672721863, + "num_tokens": 50429278.0, + "step": 1440 + }, + { + "epoch": 0.2675951717734448, + "grad_norm": 1.467283844947815, + "learning_rate": 8.910891089108911e-07, + "loss": 0.3739, + "mean_token_accuracy": 0.8732835054397583, + "num_tokens": 50461660.0, + "step": 1441 + }, + { + "epoch": 0.2677808727948004, + "grad_norm": 1.5931588411331177, + "learning_rate": 8.917079207920792e-07, + "loss": 0.4498, + "mean_token_accuracy": 0.8534858822822571, + "num_tokens": 50492498.0, + "step": 1442 + }, + { + "epoch": 0.26796657381615596, + "grad_norm": 1.356378436088562, + "learning_rate": 8.923267326732673e-07, + "loss": 0.4156, + "mean_token_accuracy": 0.8622472882270813, + "num_tokens": 50534223.0, + "step": 1443 + }, + { + "epoch": 0.2681522748375116, + "grad_norm": 1.6265054941177368, + "learning_rate": 8.929455445544554e-07, + "loss": 0.4079, + "mean_token_accuracy": 0.8664754629135132, + "num_tokens": 50568020.0, + "step": 1444 + }, + { + "epoch": 0.2683379758588672, + "grad_norm": 1.4900039434432983, + "learning_rate": 8.935643564356435e-07, + "loss": 0.4534, + "mean_token_accuracy": 0.8525292277336121, + "num_tokens": 50610708.0, + "step": 1445 + }, + { + "epoch": 0.26852367688022283, + "grad_norm": 1.7386462688446045, + "learning_rate": 8.941831683168316e-07, + "loss": 0.4054, + "mean_token_accuracy": 0.8649752736091614, + "num_tokens": 50638591.0, + "step": 1446 + }, + { + "epoch": 0.26870937790157845, + "grad_norm": 1.6514527797698975, + "learning_rate": 8.948019801980198e-07, + "loss": 0.4188, + "mean_token_accuracy": 0.8641461133956909, + "num_tokens": 50666849.0, + "step": 1447 + }, + { + "epoch": 0.26889507892293407, + "grad_norm": 1.6468006372451782, + "learning_rate": 8.954207920792078e-07, + "loss": 0.4922, + "mean_token_accuracy": 0.8375121355056763, + "num_tokens": 50700692.0, + "step": 1448 + }, + { + "epoch": 0.2690807799442897, + "grad_norm": 1.5594698190689087, + "learning_rate": 8.96039603960396e-07, + "loss": 0.4509, + "mean_token_accuracy": 0.8492756485939026, + "num_tokens": 50735796.0, + "step": 1449 + }, + { + "epoch": 0.2692664809656453, + "grad_norm": 1.5298559665679932, + "learning_rate": 8.96658415841584e-07, + "loss": 0.493, + "mean_token_accuracy": 0.8390528559684753, + "num_tokens": 50773356.0, + "step": 1450 + }, + { + "epoch": 0.26945218198700094, + "grad_norm": 1.530941128730774, + "learning_rate": 8.972772277227722e-07, + "loss": 0.4336, + "mean_token_accuracy": 0.856508195400238, + "num_tokens": 50806918.0, + "step": 1451 + }, + { + "epoch": 0.26963788300835656, + "grad_norm": 1.5975509881973267, + "learning_rate": 8.978960396039604e-07, + "loss": 0.4167, + "mean_token_accuracy": 0.8661869764328003, + "num_tokens": 50838247.0, + "step": 1452 + }, + { + "epoch": 0.2698235840297122, + "grad_norm": 1.5475746393203735, + "learning_rate": 8.985148514851485e-07, + "loss": 0.4569, + "mean_token_accuracy": 0.8528746366500854, + "num_tokens": 50875224.0, + "step": 1453 + }, + { + "epoch": 0.2700092850510678, + "grad_norm": 1.7636408805847168, + "learning_rate": 8.991336633663366e-07, + "loss": 0.4383, + "mean_token_accuracy": 0.862331748008728, + "num_tokens": 50904574.0, + "step": 1454 + }, + { + "epoch": 0.27019498607242337, + "grad_norm": 1.5989419221878052, + "learning_rate": 8.997524752475248e-07, + "loss": 0.4816, + "mean_token_accuracy": 0.8395925760269165, + "num_tokens": 50938970.0, + "step": 1455 + }, + { + "epoch": 0.270380687093779, + "grad_norm": 1.5030286312103271, + "learning_rate": 9.003712871287128e-07, + "loss": 0.4654, + "mean_token_accuracy": 0.8485912084579468, + "num_tokens": 50975191.0, + "step": 1456 + }, + { + "epoch": 0.2705663881151346, + "grad_norm": 1.4036712646484375, + "learning_rate": 9.00990099009901e-07, + "loss": 0.4389, + "mean_token_accuracy": 0.8518718481063843, + "num_tokens": 51014920.0, + "step": 1457 + }, + { + "epoch": 0.27075208913649024, + "grad_norm": 1.6511049270629883, + "learning_rate": 9.01608910891089e-07, + "loss": 0.4616, + "mean_token_accuracy": 0.8466647863388062, + "num_tokens": 51046724.0, + "step": 1458 + }, + { + "epoch": 0.27093779015784586, + "grad_norm": 1.459246039390564, + "learning_rate": 9.022277227722772e-07, + "loss": 0.4649, + "mean_token_accuracy": 0.8485897779464722, + "num_tokens": 51083738.0, + "step": 1459 + }, + { + "epoch": 0.2711234911792015, + "grad_norm": 1.3983758687973022, + "learning_rate": 9.028465346534653e-07, + "loss": 0.4268, + "mean_token_accuracy": 0.8591555953025818, + "num_tokens": 51123954.0, + "step": 1460 + }, + { + "epoch": 0.2713091922005571, + "grad_norm": 1.51852548122406, + "learning_rate": 9.034653465346534e-07, + "loss": 0.4568, + "mean_token_accuracy": 0.8509993553161621, + "num_tokens": 51163449.0, + "step": 1461 + }, + { + "epoch": 0.2714948932219127, + "grad_norm": 1.4862728118896484, + "learning_rate": 9.040841584158415e-07, + "loss": 0.4476, + "mean_token_accuracy": 0.8513936400413513, + "num_tokens": 51199162.0, + "step": 1462 + }, + { + "epoch": 0.27168059424326835, + "grad_norm": 1.5026495456695557, + "learning_rate": 9.047029702970296e-07, + "loss": 0.4504, + "mean_token_accuracy": 0.8531306385993958, + "num_tokens": 51234079.0, + "step": 1463 + }, + { + "epoch": 0.27186629526462397, + "grad_norm": 1.5032055377960205, + "learning_rate": 9.053217821782177e-07, + "loss": 0.441, + "mean_token_accuracy": 0.8566845655441284, + "num_tokens": 51271005.0, + "step": 1464 + }, + { + "epoch": 0.2720519962859796, + "grad_norm": 1.5385397672653198, + "learning_rate": 9.05940594059406e-07, + "loss": 0.3962, + "mean_token_accuracy": 0.8666113615036011, + "num_tokens": 51303920.0, + "step": 1465 + }, + { + "epoch": 0.2722376973073352, + "grad_norm": 1.5692951679229736, + "learning_rate": 9.06559405940594e-07, + "loss": 0.4458, + "mean_token_accuracy": 0.8531657457351685, + "num_tokens": 51338166.0, + "step": 1466 + }, + { + "epoch": 0.2724233983286908, + "grad_norm": 1.525141716003418, + "learning_rate": 9.071782178217822e-07, + "loss": 0.3959, + "mean_token_accuracy": 0.8677773475646973, + "num_tokens": 51370873.0, + "step": 1467 + }, + { + "epoch": 0.2726090993500464, + "grad_norm": 1.8117164373397827, + "learning_rate": 9.077970297029703e-07, + "loss": 0.4866, + "mean_token_accuracy": 0.8445343375205994, + "num_tokens": 51396153.0, + "step": 1468 + }, + { + "epoch": 0.272794800371402, + "grad_norm": 1.5056413412094116, + "learning_rate": 9.084158415841584e-07, + "loss": 0.407, + "mean_token_accuracy": 0.8637757301330566, + "num_tokens": 51429559.0, + "step": 1469 + }, + { + "epoch": 0.27298050139275765, + "grad_norm": 1.4282580614089966, + "learning_rate": 9.090346534653465e-07, + "loss": 0.3765, + "mean_token_accuracy": 0.8765481114387512, + "num_tokens": 51467907.0, + "step": 1470 + }, + { + "epoch": 0.27316620241411327, + "grad_norm": 1.4928301572799683, + "learning_rate": 9.096534653465346e-07, + "loss": 0.4355, + "mean_token_accuracy": 0.8599151968955994, + "num_tokens": 51505501.0, + "step": 1471 + }, + { + "epoch": 0.2733519034354689, + "grad_norm": 1.4856743812561035, + "learning_rate": 9.102722772277227e-07, + "loss": 0.4631, + "mean_token_accuracy": 0.8515770435333252, + "num_tokens": 51544624.0, + "step": 1472 + }, + { + "epoch": 0.2735376044568245, + "grad_norm": 1.529543399810791, + "learning_rate": 9.108910891089109e-07, + "loss": 0.4126, + "mean_token_accuracy": 0.8619292974472046, + "num_tokens": 51576819.0, + "step": 1473 + }, + { + "epoch": 0.27372330547818013, + "grad_norm": 1.6272032260894775, + "learning_rate": 9.115099009900989e-07, + "loss": 0.4739, + "mean_token_accuracy": 0.8445851802825928, + "num_tokens": 51612618.0, + "step": 1474 + }, + { + "epoch": 0.27390900649953576, + "grad_norm": 1.4890722036361694, + "learning_rate": 9.121287128712871e-07, + "loss": 0.4341, + "mean_token_accuracy": 0.8578072786331177, + "num_tokens": 51649510.0, + "step": 1475 + }, + { + "epoch": 0.2740947075208914, + "grad_norm": 1.7259491682052612, + "learning_rate": 9.127475247524751e-07, + "loss": 0.5467, + "mean_token_accuracy": 0.8289775848388672, + "num_tokens": 51680740.0, + "step": 1476 + }, + { + "epoch": 0.274280408542247, + "grad_norm": 1.4789459705352783, + "learning_rate": 9.133663366336633e-07, + "loss": 0.4054, + "mean_token_accuracy": 0.869196891784668, + "num_tokens": 51718407.0, + "step": 1477 + }, + { + "epoch": 0.2744661095636026, + "grad_norm": 1.5723801851272583, + "learning_rate": 9.139851485148514e-07, + "loss": 0.4685, + "mean_token_accuracy": 0.8472589254379272, + "num_tokens": 51755886.0, + "step": 1478 + }, + { + "epoch": 0.27465181058495824, + "grad_norm": 1.7206065654754639, + "learning_rate": 9.146039603960396e-07, + "loss": 0.4206, + "mean_token_accuracy": 0.861403226852417, + "num_tokens": 51788565.0, + "step": 1479 + }, + { + "epoch": 0.2748375116063138, + "grad_norm": 1.4118999242782593, + "learning_rate": 9.152227722772277e-07, + "loss": 0.4397, + "mean_token_accuracy": 0.8542691469192505, + "num_tokens": 51827383.0, + "step": 1480 + }, + { + "epoch": 0.27502321262766943, + "grad_norm": 1.5945686101913452, + "learning_rate": 9.158415841584159e-07, + "loss": 0.4702, + "mean_token_accuracy": 0.8460834622383118, + "num_tokens": 51859856.0, + "step": 1481 + }, + { + "epoch": 0.27520891364902506, + "grad_norm": 1.5686618089675903, + "learning_rate": 9.164603960396039e-07, + "loss": 0.4186, + "mean_token_accuracy": 0.8644663095474243, + "num_tokens": 51896677.0, + "step": 1482 + }, + { + "epoch": 0.2753946146703807, + "grad_norm": 1.5548361539840698, + "learning_rate": 9.170792079207921e-07, + "loss": 0.4474, + "mean_token_accuracy": 0.8505846261978149, + "num_tokens": 51932625.0, + "step": 1483 + }, + { + "epoch": 0.2755803156917363, + "grad_norm": 1.712693452835083, + "learning_rate": 9.176980198019801e-07, + "loss": 0.4892, + "mean_token_accuracy": 0.8411707878112793, + "num_tokens": 51964378.0, + "step": 1484 + }, + { + "epoch": 0.2757660167130919, + "grad_norm": 1.5779415369033813, + "learning_rate": 9.183168316831683e-07, + "loss": 0.4628, + "mean_token_accuracy": 0.8496559858322144, + "num_tokens": 51999273.0, + "step": 1485 + }, + { + "epoch": 0.27595171773444754, + "grad_norm": 1.5158637762069702, + "learning_rate": 9.189356435643564e-07, + "loss": 0.448, + "mean_token_accuracy": 0.8524520397186279, + "num_tokens": 52035763.0, + "step": 1486 + }, + { + "epoch": 0.27613741875580317, + "grad_norm": 1.7536405324935913, + "learning_rate": 9.195544554455445e-07, + "loss": 0.4844, + "mean_token_accuracy": 0.8372095227241516, + "num_tokens": 52064670.0, + "step": 1487 + }, + { + "epoch": 0.2763231197771588, + "grad_norm": 1.793054461479187, + "learning_rate": 9.201732673267326e-07, + "loss": 0.4078, + "mean_token_accuracy": 0.8689551949501038, + "num_tokens": 52087562.0, + "step": 1488 + }, + { + "epoch": 0.2765088207985144, + "grad_norm": 1.5813357830047607, + "learning_rate": 9.207920792079208e-07, + "loss": 0.4294, + "mean_token_accuracy": 0.8564854264259338, + "num_tokens": 52121830.0, + "step": 1489 + }, + { + "epoch": 0.27669452181987003, + "grad_norm": 1.5449336767196655, + "learning_rate": 9.214108910891088e-07, + "loss": 0.4465, + "mean_token_accuracy": 0.8553967475891113, + "num_tokens": 52156347.0, + "step": 1490 + }, + { + "epoch": 0.27688022284122565, + "grad_norm": 1.4513311386108398, + "learning_rate": 9.220297029702971e-07, + "loss": 0.4331, + "mean_token_accuracy": 0.8573923110961914, + "num_tokens": 52194864.0, + "step": 1491 + }, + { + "epoch": 0.2770659238625812, + "grad_norm": 1.5185099840164185, + "learning_rate": 9.226485148514851e-07, + "loss": 0.4467, + "mean_token_accuracy": 0.8532029390335083, + "num_tokens": 52226904.0, + "step": 1492 + }, + { + "epoch": 0.27725162488393684, + "grad_norm": 1.5594768524169922, + "learning_rate": 9.232673267326733e-07, + "loss": 0.4193, + "mean_token_accuracy": 0.8598694205284119, + "num_tokens": 52264595.0, + "step": 1493 + }, + { + "epoch": 0.27743732590529246, + "grad_norm": 1.492479920387268, + "learning_rate": 9.238861386138614e-07, + "loss": 0.4124, + "mean_token_accuracy": 0.8641833066940308, + "num_tokens": 52299075.0, + "step": 1494 + }, + { + "epoch": 0.2776230269266481, + "grad_norm": 1.6539863348007202, + "learning_rate": 9.245049504950495e-07, + "loss": 0.4222, + "mean_token_accuracy": 0.8619927167892456, + "num_tokens": 52328162.0, + "step": 1495 + }, + { + "epoch": 0.2778087279480037, + "grad_norm": 1.4840972423553467, + "learning_rate": 9.251237623762376e-07, + "loss": 0.428, + "mean_token_accuracy": 0.8626567125320435, + "num_tokens": 52369105.0, + "step": 1496 + }, + { + "epoch": 0.27799442896935933, + "grad_norm": 1.7274118661880493, + "learning_rate": 9.257425742574257e-07, + "loss": 0.4155, + "mean_token_accuracy": 0.8653984665870667, + "num_tokens": 52398015.0, + "step": 1497 + }, + { + "epoch": 0.27818012999071495, + "grad_norm": 1.4730151891708374, + "learning_rate": 9.263613861386138e-07, + "loss": 0.4695, + "mean_token_accuracy": 0.8470500707626343, + "num_tokens": 52435501.0, + "step": 1498 + }, + { + "epoch": 0.2783658310120706, + "grad_norm": 1.578109860420227, + "learning_rate": 9.26980198019802e-07, + "loss": 0.4179, + "mean_token_accuracy": 0.8623670339584351, + "num_tokens": 52465856.0, + "step": 1499 + }, + { + "epoch": 0.2785515320334262, + "grad_norm": 1.613891839981079, + "learning_rate": 9.2759900990099e-07, + "loss": 0.4401, + "mean_token_accuracy": 0.8552052974700928, + "num_tokens": 52498311.0, + "step": 1500 + }, + { + "epoch": 0.2787372330547818, + "grad_norm": 1.426173448562622, + "learning_rate": 9.282178217821782e-07, + "loss": 0.3821, + "mean_token_accuracy": 0.8686399459838867, + "num_tokens": 52533062.0, + "step": 1501 + }, + { + "epoch": 0.27892293407613744, + "grad_norm": 1.516185998916626, + "learning_rate": 9.288366336633663e-07, + "loss": 0.432, + "mean_token_accuracy": 0.8539249897003174, + "num_tokens": 52570703.0, + "step": 1502 + }, + { + "epoch": 0.27910863509749306, + "grad_norm": 1.5095221996307373, + "learning_rate": 9.294554455445544e-07, + "loss": 0.47, + "mean_token_accuracy": 0.8413677215576172, + "num_tokens": 52604040.0, + "step": 1503 + }, + { + "epoch": 0.27929433611884863, + "grad_norm": 1.3718719482421875, + "learning_rate": 9.300742574257425e-07, + "loss": 0.391, + "mean_token_accuracy": 0.8711386919021606, + "num_tokens": 52642872.0, + "step": 1504 + }, + { + "epoch": 0.27948003714020425, + "grad_norm": 1.4269976615905762, + "learning_rate": 9.306930693069307e-07, + "loss": 0.4205, + "mean_token_accuracy": 0.8610111474990845, + "num_tokens": 52684488.0, + "step": 1505 + }, + { + "epoch": 0.2796657381615599, + "grad_norm": 1.5485813617706299, + "learning_rate": 9.313118811881188e-07, + "loss": 0.4602, + "mean_token_accuracy": 0.851600170135498, + "num_tokens": 52718916.0, + "step": 1506 + }, + { + "epoch": 0.2798514391829155, + "grad_norm": 1.5032802820205688, + "learning_rate": 9.31930693069307e-07, + "loss": 0.4388, + "mean_token_accuracy": 0.858457088470459, + "num_tokens": 52755279.0, + "step": 1507 + }, + { + "epoch": 0.2800371402042711, + "grad_norm": 1.611086368560791, + "learning_rate": 9.32549504950495e-07, + "loss": 0.4502, + "mean_token_accuracy": 0.852430522441864, + "num_tokens": 52789257.0, + "step": 1508 + }, + { + "epoch": 0.28022284122562674, + "grad_norm": 1.5035001039505005, + "learning_rate": 9.331683168316832e-07, + "loss": 0.4257, + "mean_token_accuracy": 0.8563481569290161, + "num_tokens": 52824666.0, + "step": 1509 + }, + { + "epoch": 0.28040854224698236, + "grad_norm": 1.5756480693817139, + "learning_rate": 9.337871287128712e-07, + "loss": 0.4431, + "mean_token_accuracy": 0.8549472093582153, + "num_tokens": 52856671.0, + "step": 1510 + }, + { + "epoch": 0.280594243268338, + "grad_norm": 1.4780583381652832, + "learning_rate": 9.344059405940594e-07, + "loss": 0.4402, + "mean_token_accuracy": 0.855491042137146, + "num_tokens": 52892658.0, + "step": 1511 + }, + { + "epoch": 0.2807799442896936, + "grad_norm": 1.4846328496932983, + "learning_rate": 9.350247524752475e-07, + "loss": 0.43, + "mean_token_accuracy": 0.8618993759155273, + "num_tokens": 52928764.0, + "step": 1512 + }, + { + "epoch": 0.28096564531104923, + "grad_norm": 1.571308970451355, + "learning_rate": 9.356435643564356e-07, + "loss": 0.4456, + "mean_token_accuracy": 0.8515738248825073, + "num_tokens": 52964529.0, + "step": 1513 + }, + { + "epoch": 0.28115134633240485, + "grad_norm": 1.5817793607711792, + "learning_rate": 9.362623762376237e-07, + "loss": 0.4733, + "mean_token_accuracy": 0.8451015949249268, + "num_tokens": 52999056.0, + "step": 1514 + }, + { + "epoch": 0.28133704735376047, + "grad_norm": 1.5001447200775146, + "learning_rate": 9.368811881188119e-07, + "loss": 0.4562, + "mean_token_accuracy": 0.851202666759491, + "num_tokens": 53039774.0, + "step": 1515 + }, + { + "epoch": 0.28152274837511604, + "grad_norm": 1.4666624069213867, + "learning_rate": 9.374999999999999e-07, + "loss": 0.4061, + "mean_token_accuracy": 0.864986777305603, + "num_tokens": 53075099.0, + "step": 1516 + }, + { + "epoch": 0.28170844939647166, + "grad_norm": 1.675106406211853, + "learning_rate": 9.38118811881188e-07, + "loss": 0.4339, + "mean_token_accuracy": 0.8576343655586243, + "num_tokens": 53101224.0, + "step": 1517 + }, + { + "epoch": 0.2818941504178273, + "grad_norm": 1.6061646938323975, + "learning_rate": 9.387376237623762e-07, + "loss": 0.4067, + "mean_token_accuracy": 0.8635230660438538, + "num_tokens": 53128776.0, + "step": 1518 + }, + { + "epoch": 0.2820798514391829, + "grad_norm": 1.7335054874420166, + "learning_rate": 9.393564356435643e-07, + "loss": 0.4405, + "mean_token_accuracy": 0.8524109125137329, + "num_tokens": 53158095.0, + "step": 1519 + }, + { + "epoch": 0.2822655524605385, + "grad_norm": 1.5614218711853027, + "learning_rate": 9.399752475247525e-07, + "loss": 0.4415, + "mean_token_accuracy": 0.8526186347007751, + "num_tokens": 53191459.0, + "step": 1520 + }, + { + "epoch": 0.28245125348189415, + "grad_norm": 1.3699127435684204, + "learning_rate": 9.405940594059405e-07, + "loss": 0.418, + "mean_token_accuracy": 0.8594187498092651, + "num_tokens": 53230299.0, + "step": 1521 + }, + { + "epoch": 0.28263695450324977, + "grad_norm": 1.439695954322815, + "learning_rate": 9.412128712871287e-07, + "loss": 0.4252, + "mean_token_accuracy": 0.8595598936080933, + "num_tokens": 53268954.0, + "step": 1522 + }, + { + "epoch": 0.2828226555246054, + "grad_norm": 1.6610459089279175, + "learning_rate": 9.418316831683168e-07, + "loss": 0.4917, + "mean_token_accuracy": 0.8404057025909424, + "num_tokens": 53300756.0, + "step": 1523 + }, + { + "epoch": 0.283008356545961, + "grad_norm": 1.4894428253173828, + "learning_rate": 9.424504950495049e-07, + "loss": 0.4372, + "mean_token_accuracy": 0.8594642877578735, + "num_tokens": 53336897.0, + "step": 1524 + }, + { + "epoch": 0.28319405756731664, + "grad_norm": 1.48512864112854, + "learning_rate": 9.43069306930693e-07, + "loss": 0.4033, + "mean_token_accuracy": 0.8685078620910645, + "num_tokens": 53370821.0, + "step": 1525 + }, + { + "epoch": 0.28337975858867226, + "grad_norm": 1.5870473384857178, + "learning_rate": 9.436881188118811e-07, + "loss": 0.4575, + "mean_token_accuracy": 0.8511216640472412, + "num_tokens": 53401871.0, + "step": 1526 + }, + { + "epoch": 0.2835654596100279, + "grad_norm": 1.449839472770691, + "learning_rate": 9.443069306930692e-07, + "loss": 0.3742, + "mean_token_accuracy": 0.8745566010475159, + "num_tokens": 53435438.0, + "step": 1527 + }, + { + "epoch": 0.28375116063138345, + "grad_norm": 1.661141037940979, + "learning_rate": 9.449257425742574e-07, + "loss": 0.4519, + "mean_token_accuracy": 0.8545752763748169, + "num_tokens": 53468306.0, + "step": 1528 + }, + { + "epoch": 0.28393686165273907, + "grad_norm": 1.5872410535812378, + "learning_rate": 9.455445544554454e-07, + "loss": 0.488, + "mean_token_accuracy": 0.8426923751831055, + "num_tokens": 53506986.0, + "step": 1529 + }, + { + "epoch": 0.2841225626740947, + "grad_norm": 1.6342860460281372, + "learning_rate": 9.461633663366336e-07, + "loss": 0.4355, + "mean_token_accuracy": 0.8557997941970825, + "num_tokens": 53538367.0, + "step": 1530 + }, + { + "epoch": 0.2843082636954503, + "grad_norm": 1.4321366548538208, + "learning_rate": 9.467821782178216e-07, + "loss": 0.4229, + "mean_token_accuracy": 0.8589252233505249, + "num_tokens": 53578638.0, + "step": 1531 + }, + { + "epoch": 0.28449396471680594, + "grad_norm": 1.3684135675430298, + "learning_rate": 9.474009900990099e-07, + "loss": 0.3541, + "mean_token_accuracy": 0.8812299966812134, + "num_tokens": 53615462.0, + "step": 1532 + }, + { + "epoch": 0.28467966573816156, + "grad_norm": 1.672816514968872, + "learning_rate": 9.48019801980198e-07, + "loss": 0.4461, + "mean_token_accuracy": 0.8504103422164917, + "num_tokens": 53646311.0, + "step": 1533 + }, + { + "epoch": 0.2848653667595172, + "grad_norm": 1.4530136585235596, + "learning_rate": 9.486386138613861e-07, + "loss": 0.4163, + "mean_token_accuracy": 0.8606462478637695, + "num_tokens": 53684455.0, + "step": 1534 + }, + { + "epoch": 0.2850510677808728, + "grad_norm": 1.7575000524520874, + "learning_rate": 9.492574257425742e-07, + "loss": 0.44, + "mean_token_accuracy": 0.8550870418548584, + "num_tokens": 53715654.0, + "step": 1535 + }, + { + "epoch": 0.2852367688022284, + "grad_norm": 1.6600617170333862, + "learning_rate": 9.498762376237624e-07, + "loss": 0.4608, + "mean_token_accuracy": 0.8499401807785034, + "num_tokens": 53745266.0, + "step": 1536 + }, + { + "epoch": 0.28542246982358405, + "grad_norm": 1.5962915420532227, + "learning_rate": 9.504950495049504e-07, + "loss": 0.4331, + "mean_token_accuracy": 0.8627437949180603, + "num_tokens": 53775456.0, + "step": 1537 + }, + { + "epoch": 0.28560817084493967, + "grad_norm": 1.5717862844467163, + "learning_rate": 9.511138613861386e-07, + "loss": 0.4163, + "mean_token_accuracy": 0.8633236289024353, + "num_tokens": 53809283.0, + "step": 1538 + }, + { + "epoch": 0.2857938718662953, + "grad_norm": 1.7224396467208862, + "learning_rate": 9.517326732673266e-07, + "loss": 0.4612, + "mean_token_accuracy": 0.8553170561790466, + "num_tokens": 53837788.0, + "step": 1539 + }, + { + "epoch": 0.28597957288765086, + "grad_norm": 1.4541943073272705, + "learning_rate": 9.523514851485148e-07, + "loss": 0.4205, + "mean_token_accuracy": 0.861474335193634, + "num_tokens": 53874487.0, + "step": 1540 + }, + { + "epoch": 0.2861652739090065, + "grad_norm": 1.6414973735809326, + "learning_rate": 9.529702970297029e-07, + "loss": 0.4536, + "mean_token_accuracy": 0.8505200743675232, + "num_tokens": 53909145.0, + "step": 1541 + }, + { + "epoch": 0.2863509749303621, + "grad_norm": 1.3890860080718994, + "learning_rate": 9.53589108910891e-07, + "loss": 0.4119, + "mean_token_accuracy": 0.8641673922538757, + "num_tokens": 53949788.0, + "step": 1542 + }, + { + "epoch": 0.2865366759517177, + "grad_norm": 1.4007387161254883, + "learning_rate": 9.542079207920792e-07, + "loss": 0.3808, + "mean_token_accuracy": 0.8742229342460632, + "num_tokens": 53985770.0, + "step": 1543 + }, + { + "epoch": 0.28672237697307335, + "grad_norm": 1.6550242900848389, + "learning_rate": 9.548267326732672e-07, + "loss": 0.4439, + "mean_token_accuracy": 0.856737494468689, + "num_tokens": 54015546.0, + "step": 1544 + }, + { + "epoch": 0.28690807799442897, + "grad_norm": 1.5667964220046997, + "learning_rate": 9.554455445544553e-07, + "loss": 0.4274, + "mean_token_accuracy": 0.8575329780578613, + "num_tokens": 54047635.0, + "step": 1545 + }, + { + "epoch": 0.2870937790157846, + "grad_norm": 1.4510881900787354, + "learning_rate": 9.560643564356436e-07, + "loss": 0.4513, + "mean_token_accuracy": 0.8517715334892273, + "num_tokens": 54085314.0, + "step": 1546 + }, + { + "epoch": 0.2872794800371402, + "grad_norm": 1.8121535778045654, + "learning_rate": 9.566831683168316e-07, + "loss": 0.4746, + "mean_token_accuracy": 0.8481579422950745, + "num_tokens": 54112414.0, + "step": 1547 + }, + { + "epoch": 0.28746518105849583, + "grad_norm": 1.5707803964614868, + "learning_rate": 9.573019801980197e-07, + "loss": 0.3954, + "mean_token_accuracy": 0.8703607320785522, + "num_tokens": 54146573.0, + "step": 1548 + }, + { + "epoch": 0.28765088207985146, + "grad_norm": 1.474380373954773, + "learning_rate": 9.579207920792078e-07, + "loss": 0.4052, + "mean_token_accuracy": 0.8625823259353638, + "num_tokens": 54181626.0, + "step": 1549 + }, + { + "epoch": 0.2878365831012071, + "grad_norm": 1.4933247566223145, + "learning_rate": 9.58539603960396e-07, + "loss": 0.4661, + "mean_token_accuracy": 0.8467296361923218, + "num_tokens": 54221751.0, + "step": 1550 + }, + { + "epoch": 0.2880222841225627, + "grad_norm": 1.5694446563720703, + "learning_rate": 9.59158415841584e-07, + "loss": 0.4006, + "mean_token_accuracy": 0.8680856823921204, + "num_tokens": 54255962.0, + "step": 1551 + }, + { + "epoch": 0.28820798514391827, + "grad_norm": 1.4132055044174194, + "learning_rate": 9.597772277227723e-07, + "loss": 0.4147, + "mean_token_accuracy": 0.8618658185005188, + "num_tokens": 54296318.0, + "step": 1552 + }, + { + "epoch": 0.2883936861652739, + "grad_norm": 1.6007359027862549, + "learning_rate": 9.603960396039604e-07, + "loss": 0.4307, + "mean_token_accuracy": 0.8561346530914307, + "num_tokens": 54327808.0, + "step": 1553 + }, + { + "epoch": 0.2885793871866295, + "grad_norm": 1.728103756904602, + "learning_rate": 9.610148514851485e-07, + "loss": 0.5275, + "mean_token_accuracy": 0.8339389562606812, + "num_tokens": 54360019.0, + "step": 1554 + }, + { + "epoch": 0.28876508820798513, + "grad_norm": 1.5151594877243042, + "learning_rate": 9.616336633663365e-07, + "loss": 0.4141, + "mean_token_accuracy": 0.8612159490585327, + "num_tokens": 54393103.0, + "step": 1555 + }, + { + "epoch": 0.28895078922934075, + "grad_norm": 1.3529525995254517, + "learning_rate": 9.622524752475248e-07, + "loss": 0.3827, + "mean_token_accuracy": 0.8709834814071655, + "num_tokens": 54436927.0, + "step": 1556 + }, + { + "epoch": 0.2891364902506964, + "grad_norm": 1.5018423795700073, + "learning_rate": 9.628712871287129e-07, + "loss": 0.4563, + "mean_token_accuracy": 0.8565351963043213, + "num_tokens": 54476839.0, + "step": 1557 + }, + { + "epoch": 0.289322191272052, + "grad_norm": 1.7331037521362305, + "learning_rate": 9.63490099009901e-07, + "loss": 0.461, + "mean_token_accuracy": 0.849906861782074, + "num_tokens": 54507443.0, + "step": 1558 + }, + { + "epoch": 0.2895078922934076, + "grad_norm": 1.6436303853988647, + "learning_rate": 9.64108910891089e-07, + "loss": 0.48, + "mean_token_accuracy": 0.843866765499115, + "num_tokens": 54539462.0, + "step": 1559 + }, + { + "epoch": 0.28969359331476324, + "grad_norm": 1.5277581214904785, + "learning_rate": 9.647277227722772e-07, + "loss": 0.4337, + "mean_token_accuracy": 0.8604641556739807, + "num_tokens": 54574015.0, + "step": 1560 + }, + { + "epoch": 0.28987929433611886, + "grad_norm": 1.4287694692611694, + "learning_rate": 9.653465346534653e-07, + "loss": 0.4137, + "mean_token_accuracy": 0.8645443320274353, + "num_tokens": 54610671.0, + "step": 1561 + }, + { + "epoch": 0.2900649953574745, + "grad_norm": 1.4826065301895142, + "learning_rate": 9.659653465346534e-07, + "loss": 0.4515, + "mean_token_accuracy": 0.851372480392456, + "num_tokens": 54649359.0, + "step": 1562 + }, + { + "epoch": 0.2902506963788301, + "grad_norm": 1.468782901763916, + "learning_rate": 9.665841584158414e-07, + "loss": 0.401, + "mean_token_accuracy": 0.8666048049926758, + "num_tokens": 54683960.0, + "step": 1563 + }, + { + "epoch": 0.2904363974001857, + "grad_norm": 1.5480655431747437, + "learning_rate": 9.672029702970297e-07, + "loss": 0.4802, + "mean_token_accuracy": 0.8420024514198303, + "num_tokens": 54720119.0, + "step": 1564 + }, + { + "epoch": 0.2906220984215413, + "grad_norm": 1.6276694536209106, + "learning_rate": 9.678217821782177e-07, + "loss": 0.4542, + "mean_token_accuracy": 0.8535575270652771, + "num_tokens": 54754605.0, + "step": 1565 + }, + { + "epoch": 0.2908077994428969, + "grad_norm": 1.4845515489578247, + "learning_rate": 9.68440594059406e-07, + "loss": 0.4278, + "mean_token_accuracy": 0.8598823547363281, + "num_tokens": 54786755.0, + "step": 1566 + }, + { + "epoch": 0.29099350046425254, + "grad_norm": 1.6596391201019287, + "learning_rate": 9.69059405940594e-07, + "loss": 0.3948, + "mean_token_accuracy": 0.8689005374908447, + "num_tokens": 54811945.0, + "step": 1567 + }, + { + "epoch": 0.29117920148560816, + "grad_norm": 1.4590229988098145, + "learning_rate": 9.696782178217821e-07, + "loss": 0.4273, + "mean_token_accuracy": 0.8586587905883789, + "num_tokens": 54850438.0, + "step": 1568 + }, + { + "epoch": 0.2913649025069638, + "grad_norm": 1.4800279140472412, + "learning_rate": 9.702970297029702e-07, + "loss": 0.3853, + "mean_token_accuracy": 0.8715823888778687, + "num_tokens": 54884143.0, + "step": 1569 + }, + { + "epoch": 0.2915506035283194, + "grad_norm": 1.4115290641784668, + "learning_rate": 9.709158415841585e-07, + "loss": 0.4544, + "mean_token_accuracy": 0.8492481112480164, + "num_tokens": 54923415.0, + "step": 1570 + }, + { + "epoch": 0.29173630454967503, + "grad_norm": 1.4010848999023438, + "learning_rate": 9.715346534653465e-07, + "loss": 0.3887, + "mean_token_accuracy": 0.8648460507392883, + "num_tokens": 54959976.0, + "step": 1571 + }, + { + "epoch": 0.29192200557103065, + "grad_norm": 1.7679675817489624, + "learning_rate": 9.721534653465346e-07, + "loss": 0.4694, + "mean_token_accuracy": 0.8448246717453003, + "num_tokens": 54990210.0, + "step": 1572 + }, + { + "epoch": 0.2921077065923863, + "grad_norm": 1.556943655014038, + "learning_rate": 9.727722772277226e-07, + "loss": 0.4189, + "mean_token_accuracy": 0.8585307598114014, + "num_tokens": 55026582.0, + "step": 1573 + }, + { + "epoch": 0.2922934076137419, + "grad_norm": 1.4236646890640259, + "learning_rate": 9.73391089108911e-07, + "loss": 0.3708, + "mean_token_accuracy": 0.8785300850868225, + "num_tokens": 55062965.0, + "step": 1574 + }, + { + "epoch": 0.2924791086350975, + "grad_norm": 1.5029484033584595, + "learning_rate": 9.74009900990099e-07, + "loss": 0.4701, + "mean_token_accuracy": 0.8497666120529175, + "num_tokens": 55099909.0, + "step": 1575 + }, + { + "epoch": 0.2926648096564531, + "grad_norm": 1.6213923692703247, + "learning_rate": 9.74628712871287e-07, + "loss": 0.4392, + "mean_token_accuracy": 0.856601357460022, + "num_tokens": 55133113.0, + "step": 1576 + }, + { + "epoch": 0.2928505106778087, + "grad_norm": 1.609934687614441, + "learning_rate": 9.75247524752475e-07, + "loss": 0.4657, + "mean_token_accuracy": 0.8494648933410645, + "num_tokens": 55165560.0, + "step": 1577 + }, + { + "epoch": 0.29303621169916433, + "grad_norm": 1.6292712688446045, + "learning_rate": 9.758663366336633e-07, + "loss": 0.4444, + "mean_token_accuracy": 0.856266975402832, + "num_tokens": 55195832.0, + "step": 1578 + }, + { + "epoch": 0.29322191272051995, + "grad_norm": 1.5239847898483276, + "learning_rate": 9.764851485148514e-07, + "loss": 0.4298, + "mean_token_accuracy": 0.8603898286819458, + "num_tokens": 55231884.0, + "step": 1579 + }, + { + "epoch": 0.2934076137418756, + "grad_norm": 1.6834583282470703, + "learning_rate": 9.771039603960397e-07, + "loss": 0.4299, + "mean_token_accuracy": 0.8560813665390015, + "num_tokens": 55265865.0, + "step": 1580 + }, + { + "epoch": 0.2935933147632312, + "grad_norm": 1.5403635501861572, + "learning_rate": 9.777227722772277e-07, + "loss": 0.4433, + "mean_token_accuracy": 0.8642576932907104, + "num_tokens": 55298544.0, + "step": 1581 + }, + { + "epoch": 0.2937790157845868, + "grad_norm": 1.534626841545105, + "learning_rate": 9.783415841584158e-07, + "loss": 0.42, + "mean_token_accuracy": 0.8654195070266724, + "num_tokens": 55335666.0, + "step": 1582 + }, + { + "epoch": 0.29396471680594244, + "grad_norm": 1.5618922710418701, + "learning_rate": 9.789603960396039e-07, + "loss": 0.3741, + "mean_token_accuracy": 0.8740040063858032, + "num_tokens": 55366991.0, + "step": 1583 + }, + { + "epoch": 0.29415041782729806, + "grad_norm": 1.4958016872406006, + "learning_rate": 9.795792079207921e-07, + "loss": 0.3781, + "mean_token_accuracy": 0.8713468313217163, + "num_tokens": 55405628.0, + "step": 1584 + }, + { + "epoch": 0.2943361188486537, + "grad_norm": 1.670203447341919, + "learning_rate": 9.801980198019802e-07, + "loss": 0.4732, + "mean_token_accuracy": 0.8431588411331177, + "num_tokens": 55433457.0, + "step": 1585 + }, + { + "epoch": 0.2945218198700093, + "grad_norm": 1.6822932958602905, + "learning_rate": 9.808168316831682e-07, + "loss": 0.4659, + "mean_token_accuracy": 0.8443244695663452, + "num_tokens": 55463215.0, + "step": 1586 + }, + { + "epoch": 0.2947075208913649, + "grad_norm": 1.547047734260559, + "learning_rate": 9.814356435643563e-07, + "loss": 0.4447, + "mean_token_accuracy": 0.8517414331436157, + "num_tokens": 55500407.0, + "step": 1587 + }, + { + "epoch": 0.2948932219127205, + "grad_norm": 1.5476795434951782, + "learning_rate": 9.820544554455446e-07, + "loss": 0.4296, + "mean_token_accuracy": 0.8584789037704468, + "num_tokens": 55537037.0, + "step": 1588 + }, + { + "epoch": 0.2950789229340761, + "grad_norm": 1.6714086532592773, + "learning_rate": 9.826732673267326e-07, + "loss": 0.4205, + "mean_token_accuracy": 0.8606149554252625, + "num_tokens": 55566716.0, + "step": 1589 + }, + { + "epoch": 0.29526462395543174, + "grad_norm": 1.533531904220581, + "learning_rate": 9.832920792079207e-07, + "loss": 0.3997, + "mean_token_accuracy": 0.8687119483947754, + "num_tokens": 55597985.0, + "step": 1590 + }, + { + "epoch": 0.29545032497678736, + "grad_norm": 1.6861385107040405, + "learning_rate": 9.83910891089109e-07, + "loss": 0.4395, + "mean_token_accuracy": 0.8631574511528015, + "num_tokens": 55625069.0, + "step": 1591 + }, + { + "epoch": 0.295636025998143, + "grad_norm": 1.3538998365402222, + "learning_rate": 9.84529702970297e-07, + "loss": 0.3901, + "mean_token_accuracy": 0.867942214012146, + "num_tokens": 55666168.0, + "step": 1592 + }, + { + "epoch": 0.2958217270194986, + "grad_norm": 1.7186963558197021, + "learning_rate": 9.85148514851485e-07, + "loss": 0.3931, + "mean_token_accuracy": 0.8657875061035156, + "num_tokens": 55700681.0, + "step": 1593 + }, + { + "epoch": 0.2960074280408542, + "grad_norm": 1.639439344406128, + "learning_rate": 9.857673267326733e-07, + "loss": 0.4614, + "mean_token_accuracy": 0.8460537791252136, + "num_tokens": 55736330.0, + "step": 1594 + }, + { + "epoch": 0.29619312906220985, + "grad_norm": 1.5330473184585571, + "learning_rate": 9.863861386138614e-07, + "loss": 0.4282, + "mean_token_accuracy": 0.859062671661377, + "num_tokens": 55771067.0, + "step": 1595 + }, + { + "epoch": 0.29637883008356547, + "grad_norm": 1.4155361652374268, + "learning_rate": 9.870049504950495e-07, + "loss": 0.3873, + "mean_token_accuracy": 0.869199812412262, + "num_tokens": 55807012.0, + "step": 1596 + }, + { + "epoch": 0.2965645311049211, + "grad_norm": 1.6204242706298828, + "learning_rate": 9.876237623762375e-07, + "loss": 0.4323, + "mean_token_accuracy": 0.8581369519233704, + "num_tokens": 55843305.0, + "step": 1597 + }, + { + "epoch": 0.2967502321262767, + "grad_norm": 1.6206178665161133, + "learning_rate": 9.882425742574258e-07, + "loss": 0.4627, + "mean_token_accuracy": 0.848499596118927, + "num_tokens": 55877410.0, + "step": 1598 + }, + { + "epoch": 0.29693593314763234, + "grad_norm": 1.6946940422058105, + "learning_rate": 9.888613861386138e-07, + "loss": 0.3977, + "mean_token_accuracy": 0.8707640171051025, + "num_tokens": 55906627.0, + "step": 1599 + }, + { + "epoch": 0.2971216341689879, + "grad_norm": 1.500959873199463, + "learning_rate": 9.89480198019802e-07, + "loss": 0.4002, + "mean_token_accuracy": 0.8686003088951111, + "num_tokens": 55941809.0, + "step": 1600 + }, + { + "epoch": 0.2973073351903435, + "grad_norm": 1.5606625080108643, + "learning_rate": 9.9009900990099e-07, + "loss": 0.5103, + "mean_token_accuracy": 0.8393486142158508, + "num_tokens": 55982239.0, + "step": 1601 + }, + { + "epoch": 0.29749303621169915, + "grad_norm": 1.5814257860183716, + "learning_rate": 9.907178217821782e-07, + "loss": 0.4334, + "mean_token_accuracy": 0.8555238246917725, + "num_tokens": 56014468.0, + "step": 1602 + }, + { + "epoch": 0.29767873723305477, + "grad_norm": 1.4883886575698853, + "learning_rate": 9.913366336633663e-07, + "loss": 0.4364, + "mean_token_accuracy": 0.8550728559494019, + "num_tokens": 56054214.0, + "step": 1603 + }, + { + "epoch": 0.2978644382544104, + "grad_norm": 1.5653598308563232, + "learning_rate": 9.919554455445546e-07, + "loss": 0.4963, + "mean_token_accuracy": 0.8402402400970459, + "num_tokens": 56088550.0, + "step": 1604 + }, + { + "epoch": 0.298050139275766, + "grad_norm": 1.611480712890625, + "learning_rate": 9.925742574257426e-07, + "loss": 0.4852, + "mean_token_accuracy": 0.8444615602493286, + "num_tokens": 56125464.0, + "step": 1605 + }, + { + "epoch": 0.29823584029712163, + "grad_norm": 1.330021619796753, + "learning_rate": 9.931930693069307e-07, + "loss": 0.4175, + "mean_token_accuracy": 0.8625831604003906, + "num_tokens": 56170478.0, + "step": 1606 + }, + { + "epoch": 0.29842154131847726, + "grad_norm": 1.5186073780059814, + "learning_rate": 9.938118811881187e-07, + "loss": 0.4089, + "mean_token_accuracy": 0.8640386462211609, + "num_tokens": 56207602.0, + "step": 1607 + }, + { + "epoch": 0.2986072423398329, + "grad_norm": 1.41278076171875, + "learning_rate": 9.94430693069307e-07, + "loss": 0.4034, + "mean_token_accuracy": 0.8651757836341858, + "num_tokens": 56248159.0, + "step": 1608 + }, + { + "epoch": 0.2987929433611885, + "grad_norm": 1.6612638235092163, + "learning_rate": 9.95049504950495e-07, + "loss": 0.4091, + "mean_token_accuracy": 0.865652322769165, + "num_tokens": 56274458.0, + "step": 1609 + }, + { + "epoch": 0.2989786443825441, + "grad_norm": 1.4682730436325073, + "learning_rate": 9.956683168316831e-07, + "loss": 0.3981, + "mean_token_accuracy": 0.8666194081306458, + "num_tokens": 56310129.0, + "step": 1610 + }, + { + "epoch": 0.29916434540389975, + "grad_norm": 1.6186355352401733, + "learning_rate": 9.962871287128712e-07, + "loss": 0.4433, + "mean_token_accuracy": 0.8551703095436096, + "num_tokens": 56342607.0, + "step": 1611 + }, + { + "epoch": 0.2993500464252553, + "grad_norm": 1.5895990133285522, + "learning_rate": 9.969059405940595e-07, + "loss": 0.4061, + "mean_token_accuracy": 0.8654018640518188, + "num_tokens": 56378135.0, + "step": 1612 + }, + { + "epoch": 0.29953574744661093, + "grad_norm": 1.5378464460372925, + "learning_rate": 9.975247524752475e-07, + "loss": 0.4461, + "mean_token_accuracy": 0.8526206612586975, + "num_tokens": 56413721.0, + "step": 1613 + }, + { + "epoch": 0.29972144846796656, + "grad_norm": 1.5448552370071411, + "learning_rate": 9.981435643564356e-07, + "loss": 0.3932, + "mean_token_accuracy": 0.8706504106521606, + "num_tokens": 56448071.0, + "step": 1614 + }, + { + "epoch": 0.2999071494893222, + "grad_norm": 1.6239324808120728, + "learning_rate": 9.987623762376236e-07, + "loss": 0.4542, + "mean_token_accuracy": 0.8559266924858093, + "num_tokens": 56477294.0, + "step": 1615 + }, + { + "epoch": 0.3000928505106778, + "grad_norm": 1.6168546676635742, + "learning_rate": 9.99381188118812e-07, + "loss": 0.4485, + "mean_token_accuracy": 0.857121467590332, + "num_tokens": 56509727.0, + "step": 1616 + }, + { + "epoch": 0.3002785515320334, + "grad_norm": 1.5543378591537476, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8634119033813477, + "num_tokens": 56539615.0, + "step": 1617 + }, + { + "epoch": 0.30046425255338904, + "grad_norm": 1.4405062198638916, + "learning_rate": 1e-06, + "loss": 0.4711, + "mean_token_accuracy": 0.8427993059158325, + "num_tokens": 56579968.0, + "step": 1618 + }, + { + "epoch": 0.30064995357474467, + "grad_norm": 1.4893944263458252, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8683080673217773, + "num_tokens": 56612968.0, + "step": 1619 + }, + { + "epoch": 0.3008356545961003, + "grad_norm": 1.4802569150924683, + "learning_rate": 1e-06, + "loss": 0.4829, + "mean_token_accuracy": 0.8459994792938232, + "num_tokens": 56651356.0, + "step": 1620 + }, + { + "epoch": 0.3010213556174559, + "grad_norm": 1.6509395837783813, + "learning_rate": 1e-06, + "loss": 0.4724, + "mean_token_accuracy": 0.844693660736084, + "num_tokens": 56688049.0, + "step": 1621 + }, + { + "epoch": 0.30120705663881153, + "grad_norm": 1.4815282821655273, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.8584504723548889, + "num_tokens": 56725594.0, + "step": 1622 + }, + { + "epoch": 0.30139275766016715, + "grad_norm": 1.459821105003357, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8675680160522461, + "num_tokens": 56760804.0, + "step": 1623 + }, + { + "epoch": 0.3015784586815227, + "grad_norm": 1.560685396194458, + "learning_rate": 1e-06, + "loss": 0.4354, + "mean_token_accuracy": 0.861061155796051, + "num_tokens": 56795830.0, + "step": 1624 + }, + { + "epoch": 0.30176415970287834, + "grad_norm": 1.5116486549377441, + "learning_rate": 1e-06, + "loss": 0.4408, + "mean_token_accuracy": 0.855659544467926, + "num_tokens": 56829413.0, + "step": 1625 + }, + { + "epoch": 0.30194986072423396, + "grad_norm": 1.3964406251907349, + "learning_rate": 1e-06, + "loss": 0.4047, + "mean_token_accuracy": 0.864668607711792, + "num_tokens": 56869013.0, + "step": 1626 + }, + { + "epoch": 0.3021355617455896, + "grad_norm": 1.589084267616272, + "learning_rate": 1e-06, + "loss": 0.4498, + "mean_token_accuracy": 0.8528802990913391, + "num_tokens": 56904118.0, + "step": 1627 + }, + { + "epoch": 0.3023212627669452, + "grad_norm": 1.512141227722168, + "learning_rate": 1e-06, + "loss": 0.4472, + "mean_token_accuracy": 0.851587176322937, + "num_tokens": 56943163.0, + "step": 1628 + }, + { + "epoch": 0.30250696378830083, + "grad_norm": 1.4653162956237793, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8706449866294861, + "num_tokens": 56976539.0, + "step": 1629 + }, + { + "epoch": 0.30269266480965645, + "grad_norm": 1.5607165098190308, + "learning_rate": 1e-06, + "loss": 0.4554, + "mean_token_accuracy": 0.8509483933448792, + "num_tokens": 57016037.0, + "step": 1630 + }, + { + "epoch": 0.3028783658310121, + "grad_norm": 1.5425045490264893, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8633661270141602, + "num_tokens": 57046525.0, + "step": 1631 + }, + { + "epoch": 0.3030640668523677, + "grad_norm": 1.4595264196395874, + "learning_rate": 1e-06, + "loss": 0.4232, + "mean_token_accuracy": 0.8580418825149536, + "num_tokens": 57081279.0, + "step": 1632 + }, + { + "epoch": 0.3032497678737233, + "grad_norm": 1.4674370288848877, + "learning_rate": 1e-06, + "loss": 0.4726, + "mean_token_accuracy": 0.8471568822860718, + "num_tokens": 57122962.0, + "step": 1633 + }, + { + "epoch": 0.30343546889507894, + "grad_norm": 1.6555662155151367, + "learning_rate": 1e-06, + "loss": 0.4152, + "mean_token_accuracy": 0.8645358085632324, + "num_tokens": 57152286.0, + "step": 1634 + }, + { + "epoch": 0.30362116991643456, + "grad_norm": 1.8875700235366821, + "learning_rate": 1e-06, + "loss": 0.4894, + "mean_token_accuracy": 0.8401870727539062, + "num_tokens": 57178001.0, + "step": 1635 + }, + { + "epoch": 0.3038068709377902, + "grad_norm": 1.5084458589553833, + "learning_rate": 1e-06, + "loss": 0.437, + "mean_token_accuracy": 0.8569900989532471, + "num_tokens": 57214188.0, + "step": 1636 + }, + { + "epoch": 0.30399257195914575, + "grad_norm": 1.4139595031738281, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8656353950500488, + "num_tokens": 57251450.0, + "step": 1637 + }, + { + "epoch": 0.3041782729805014, + "grad_norm": 1.478265404701233, + "learning_rate": 1e-06, + "loss": 0.4594, + "mean_token_accuracy": 0.8486478328704834, + "num_tokens": 57285563.0, + "step": 1638 + }, + { + "epoch": 0.304363974001857, + "grad_norm": 1.4411859512329102, + "learning_rate": 1e-06, + "loss": 0.4682, + "mean_token_accuracy": 0.8486006259918213, + "num_tokens": 57329830.0, + "step": 1639 + }, + { + "epoch": 0.3045496750232126, + "grad_norm": 1.4842255115509033, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8617316484451294, + "num_tokens": 57363284.0, + "step": 1640 + }, + { + "epoch": 0.30473537604456824, + "grad_norm": 1.3972530364990234, + "learning_rate": 1e-06, + "loss": 0.424, + "mean_token_accuracy": 0.8587045669555664, + "num_tokens": 57401808.0, + "step": 1641 + }, + { + "epoch": 0.30492107706592386, + "grad_norm": 1.5352070331573486, + "learning_rate": 1e-06, + "loss": 0.4514, + "mean_token_accuracy": 0.8492481708526611, + "num_tokens": 57436689.0, + "step": 1642 + }, + { + "epoch": 0.3051067780872795, + "grad_norm": 1.5568561553955078, + "learning_rate": 1e-06, + "loss": 0.4417, + "mean_token_accuracy": 0.8577834367752075, + "num_tokens": 57471158.0, + "step": 1643 + }, + { + "epoch": 0.3052924791086351, + "grad_norm": 1.5961956977844238, + "learning_rate": 1e-06, + "loss": 0.4539, + "mean_token_accuracy": 0.8511844873428345, + "num_tokens": 57504848.0, + "step": 1644 + }, + { + "epoch": 0.30547818012999073, + "grad_norm": 1.5183498859405518, + "learning_rate": 1e-06, + "loss": 0.4163, + "mean_token_accuracy": 0.8584042191505432, + "num_tokens": 57539553.0, + "step": 1645 + }, + { + "epoch": 0.30566388115134635, + "grad_norm": 1.859368085861206, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8639082908630371, + "num_tokens": 57563731.0, + "step": 1646 + }, + { + "epoch": 0.305849582172702, + "grad_norm": 1.572654128074646, + "learning_rate": 1e-06, + "loss": 0.4841, + "mean_token_accuracy": 0.8418017625808716, + "num_tokens": 57598621.0, + "step": 1647 + }, + { + "epoch": 0.3060352831940576, + "grad_norm": 1.674100399017334, + "learning_rate": 1e-06, + "loss": 0.4688, + "mean_token_accuracy": 0.845137357711792, + "num_tokens": 57630570.0, + "step": 1648 + }, + { + "epoch": 0.30622098421541316, + "grad_norm": 1.644781231880188, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8630449175834656, + "num_tokens": 57662933.0, + "step": 1649 + }, + { + "epoch": 0.3064066852367688, + "grad_norm": 1.4413316249847412, + "learning_rate": 1e-06, + "loss": 0.428, + "mean_token_accuracy": 0.8610890507698059, + "num_tokens": 57701795.0, + "step": 1650 + }, + { + "epoch": 0.3065923862581244, + "grad_norm": 1.5769250392913818, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8628755807876587, + "num_tokens": 57734555.0, + "step": 1651 + }, + { + "epoch": 0.30677808727948, + "grad_norm": 1.4799555540084839, + "learning_rate": 1e-06, + "loss": 0.4452, + "mean_token_accuracy": 0.8547792434692383, + "num_tokens": 57773954.0, + "step": 1652 + }, + { + "epoch": 0.30696378830083565, + "grad_norm": 1.5432796478271484, + "learning_rate": 1e-06, + "loss": 0.4283, + "mean_token_accuracy": 0.8563814759254456, + "num_tokens": 57810440.0, + "step": 1653 + }, + { + "epoch": 0.30714948932219127, + "grad_norm": 1.4128707647323608, + "learning_rate": 1e-06, + "loss": 0.4621, + "mean_token_accuracy": 0.8509210348129272, + "num_tokens": 57851947.0, + "step": 1654 + }, + { + "epoch": 0.3073351903435469, + "grad_norm": 1.7714433670043945, + "learning_rate": 1e-06, + "loss": 0.4583, + "mean_token_accuracy": 0.8565492033958435, + "num_tokens": 57881349.0, + "step": 1655 + }, + { + "epoch": 0.3075208913649025, + "grad_norm": 1.4040998220443726, + "learning_rate": 1e-06, + "loss": 0.4299, + "mean_token_accuracy": 0.8575085401535034, + "num_tokens": 57921952.0, + "step": 1656 + }, + { + "epoch": 0.30770659238625814, + "grad_norm": 1.5213514566421509, + "learning_rate": 1e-06, + "loss": 0.4642, + "mean_token_accuracy": 0.8525325655937195, + "num_tokens": 57962259.0, + "step": 1657 + }, + { + "epoch": 0.30789229340761376, + "grad_norm": 1.5162429809570312, + "learning_rate": 1e-06, + "loss": 0.4503, + "mean_token_accuracy": 0.8506382703781128, + "num_tokens": 57999046.0, + "step": 1658 + }, + { + "epoch": 0.3080779944289694, + "grad_norm": 1.5391820669174194, + "learning_rate": 1e-06, + "loss": 0.409, + "mean_token_accuracy": 0.8720799684524536, + "num_tokens": 58031699.0, + "step": 1659 + }, + { + "epoch": 0.308263695450325, + "grad_norm": 1.5344858169555664, + "learning_rate": 1e-06, + "loss": 0.4766, + "mean_token_accuracy": 0.8431444764137268, + "num_tokens": 58069920.0, + "step": 1660 + }, + { + "epoch": 0.30844939647168057, + "grad_norm": 1.6749367713928223, + "learning_rate": 1e-06, + "loss": 0.5207, + "mean_token_accuracy": 0.8314937353134155, + "num_tokens": 58100815.0, + "step": 1661 + }, + { + "epoch": 0.3086350974930362, + "grad_norm": 1.583752989768982, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8567386269569397, + "num_tokens": 58134611.0, + "step": 1662 + }, + { + "epoch": 0.3088207985143918, + "grad_norm": 1.646231770515442, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8669236898422241, + "num_tokens": 58166425.0, + "step": 1663 + }, + { + "epoch": 0.30900649953574744, + "grad_norm": 1.461545467376709, + "learning_rate": 1e-06, + "loss": 0.4185, + "mean_token_accuracy": 0.8616006374359131, + "num_tokens": 58204833.0, + "step": 1664 + }, + { + "epoch": 0.30919220055710306, + "grad_norm": 1.5072970390319824, + "learning_rate": 1e-06, + "loss": 0.4215, + "mean_token_accuracy": 0.862246572971344, + "num_tokens": 58241544.0, + "step": 1665 + }, + { + "epoch": 0.3093779015784587, + "grad_norm": 1.5116217136383057, + "learning_rate": 1e-06, + "loss": 0.4426, + "mean_token_accuracy": 0.8524839878082275, + "num_tokens": 58278208.0, + "step": 1666 + }, + { + "epoch": 0.3095636025998143, + "grad_norm": 1.536603569984436, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8687254190444946, + "num_tokens": 58308903.0, + "step": 1667 + }, + { + "epoch": 0.3097493036211699, + "grad_norm": 1.5170613527297974, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8624758720397949, + "num_tokens": 58350717.0, + "step": 1668 + }, + { + "epoch": 0.30993500464252555, + "grad_norm": 1.4065885543823242, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8609445095062256, + "num_tokens": 58392058.0, + "step": 1669 + }, + { + "epoch": 0.31012070566388117, + "grad_norm": 1.520171046257019, + "learning_rate": 1e-06, + "loss": 0.4667, + "mean_token_accuracy": 0.8459460735321045, + "num_tokens": 58427364.0, + "step": 1670 + }, + { + "epoch": 0.3103064066852368, + "grad_norm": 1.8639144897460938, + "learning_rate": 1e-06, + "loss": 0.5069, + "mean_token_accuracy": 0.8341977596282959, + "num_tokens": 58458160.0, + "step": 1671 + }, + { + "epoch": 0.3104921077065924, + "grad_norm": 1.5471233129501343, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8660556674003601, + "num_tokens": 58498671.0, + "step": 1672 + }, + { + "epoch": 0.310677808727948, + "grad_norm": 1.5133026838302612, + "learning_rate": 1e-06, + "loss": 0.457, + "mean_token_accuracy": 0.8489772081375122, + "num_tokens": 58536420.0, + "step": 1673 + }, + { + "epoch": 0.3108635097493036, + "grad_norm": 1.4571624994277954, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.861428439617157, + "num_tokens": 58572463.0, + "step": 1674 + }, + { + "epoch": 0.3110492107706592, + "grad_norm": 1.2878341674804688, + "learning_rate": 1e-06, + "loss": 0.3986, + "mean_token_accuracy": 0.8686972260475159, + "num_tokens": 58616985.0, + "step": 1675 + }, + { + "epoch": 0.31123491179201485, + "grad_norm": 1.4416040182113647, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8570865988731384, + "num_tokens": 58656877.0, + "step": 1676 + }, + { + "epoch": 0.31142061281337047, + "grad_norm": 1.4380083084106445, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8697413206100464, + "num_tokens": 58698320.0, + "step": 1677 + }, + { + "epoch": 0.3116063138347261, + "grad_norm": 1.5481507778167725, + "learning_rate": 1e-06, + "loss": 0.4552, + "mean_token_accuracy": 0.8490429520606995, + "num_tokens": 58734096.0, + "step": 1678 + }, + { + "epoch": 0.3117920148560817, + "grad_norm": 1.4624625444412231, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8652640581130981, + "num_tokens": 58767407.0, + "step": 1679 + }, + { + "epoch": 0.31197771587743733, + "grad_norm": 1.7049224376678467, + "learning_rate": 1e-06, + "loss": 0.4755, + "mean_token_accuracy": 0.8474990129470825, + "num_tokens": 58798650.0, + "step": 1680 + }, + { + "epoch": 0.31216341689879296, + "grad_norm": 1.4926105737686157, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8570334911346436, + "num_tokens": 58834500.0, + "step": 1681 + }, + { + "epoch": 0.3123491179201486, + "grad_norm": 1.4484786987304688, + "learning_rate": 1e-06, + "loss": 0.4305, + "mean_token_accuracy": 0.8603463768959045, + "num_tokens": 58872883.0, + "step": 1682 + }, + { + "epoch": 0.3125348189415042, + "grad_norm": 1.6523302793502808, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.8646427392959595, + "num_tokens": 58901884.0, + "step": 1683 + }, + { + "epoch": 0.3127205199628598, + "grad_norm": 1.5030864477157593, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.869928240776062, + "num_tokens": 58938671.0, + "step": 1684 + }, + { + "epoch": 0.3129062209842154, + "grad_norm": 1.5394700765609741, + "learning_rate": 1e-06, + "loss": 0.516, + "mean_token_accuracy": 0.8292665481567383, + "num_tokens": 58975816.0, + "step": 1685 + }, + { + "epoch": 0.313091922005571, + "grad_norm": 1.4731863737106323, + "learning_rate": 1e-06, + "loss": 0.4326, + "mean_token_accuracy": 0.8603096008300781, + "num_tokens": 59012505.0, + "step": 1686 + }, + { + "epoch": 0.31327762302692663, + "grad_norm": 1.3482294082641602, + "learning_rate": 1e-06, + "loss": 0.4458, + "mean_token_accuracy": 0.8532778024673462, + "num_tokens": 59056243.0, + "step": 1687 + }, + { + "epoch": 0.31346332404828225, + "grad_norm": 1.4174628257751465, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.868371307849884, + "num_tokens": 59096681.0, + "step": 1688 + }, + { + "epoch": 0.3136490250696379, + "grad_norm": 1.5204188823699951, + "learning_rate": 1e-06, + "loss": 0.4333, + "mean_token_accuracy": 0.8539268970489502, + "num_tokens": 59132231.0, + "step": 1689 + }, + { + "epoch": 0.3138347260909935, + "grad_norm": 1.5358664989471436, + "learning_rate": 1e-06, + "loss": 0.4895, + "mean_token_accuracy": 0.8440294861793518, + "num_tokens": 59171599.0, + "step": 1690 + }, + { + "epoch": 0.3140204271123491, + "grad_norm": 1.5015370845794678, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.864499032497406, + "num_tokens": 59206545.0, + "step": 1691 + }, + { + "epoch": 0.31420612813370474, + "grad_norm": 1.5311903953552246, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.86368727684021, + "num_tokens": 59240874.0, + "step": 1692 + }, + { + "epoch": 0.31439182915506036, + "grad_norm": 1.53179931640625, + "learning_rate": 1e-06, + "loss": 0.4181, + "mean_token_accuracy": 0.8631262183189392, + "num_tokens": 59277222.0, + "step": 1693 + }, + { + "epoch": 0.314577530176416, + "grad_norm": 1.7327464818954468, + "learning_rate": 1e-06, + "loss": 0.4733, + "mean_token_accuracy": 0.8445505499839783, + "num_tokens": 59306161.0, + "step": 1694 + }, + { + "epoch": 0.3147632311977716, + "grad_norm": 1.5943623781204224, + "learning_rate": 1e-06, + "loss": 0.4507, + "mean_token_accuracy": 0.8489984273910522, + "num_tokens": 59342379.0, + "step": 1695 + }, + { + "epoch": 0.31494893221912723, + "grad_norm": 1.4550734758377075, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8729597330093384, + "num_tokens": 59381620.0, + "step": 1696 + }, + { + "epoch": 0.3151346332404828, + "grad_norm": 1.4952325820922852, + "learning_rate": 1e-06, + "loss": 0.4809, + "mean_token_accuracy": 0.8411613702774048, + "num_tokens": 59420489.0, + "step": 1697 + }, + { + "epoch": 0.3153203342618384, + "grad_norm": 1.665337085723877, + "learning_rate": 1e-06, + "loss": 0.4632, + "mean_token_accuracy": 0.8496883511543274, + "num_tokens": 59452525.0, + "step": 1698 + }, + { + "epoch": 0.31550603528319404, + "grad_norm": 1.5086630582809448, + "learning_rate": 1e-06, + "loss": 0.4254, + "mean_token_accuracy": 0.8631183505058289, + "num_tokens": 59486651.0, + "step": 1699 + }, + { + "epoch": 0.31569173630454966, + "grad_norm": 1.7828588485717773, + "learning_rate": 1e-06, + "loss": 0.4397, + "mean_token_accuracy": 0.854856014251709, + "num_tokens": 59517333.0, + "step": 1700 + }, + { + "epoch": 0.3158774373259053, + "grad_norm": 1.522973895072937, + "learning_rate": 1e-06, + "loss": 0.4507, + "mean_token_accuracy": 0.849651575088501, + "num_tokens": 59553294.0, + "step": 1701 + }, + { + "epoch": 0.3160631383472609, + "grad_norm": 1.5955721139907837, + "learning_rate": 1e-06, + "loss": 0.4609, + "mean_token_accuracy": 0.8534390330314636, + "num_tokens": 59585623.0, + "step": 1702 + }, + { + "epoch": 0.31624883936861653, + "grad_norm": 1.6272917985916138, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8535008430480957, + "num_tokens": 59621054.0, + "step": 1703 + }, + { + "epoch": 0.31643454038997215, + "grad_norm": 1.4856189489364624, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8649671077728271, + "num_tokens": 59656783.0, + "step": 1704 + }, + { + "epoch": 0.3166202414113278, + "grad_norm": 1.5260576009750366, + "learning_rate": 1e-06, + "loss": 0.463, + "mean_token_accuracy": 0.8479565382003784, + "num_tokens": 59693670.0, + "step": 1705 + }, + { + "epoch": 0.3168059424326834, + "grad_norm": 1.6364401578903198, + "learning_rate": 1e-06, + "loss": 0.4472, + "mean_token_accuracy": 0.8536770343780518, + "num_tokens": 59723429.0, + "step": 1706 + }, + { + "epoch": 0.316991643454039, + "grad_norm": 1.4804905652999878, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.8613379597663879, + "num_tokens": 59758552.0, + "step": 1707 + }, + { + "epoch": 0.31717734447539464, + "grad_norm": 1.4870637655258179, + "learning_rate": 1e-06, + "loss": 0.4519, + "mean_token_accuracy": 0.8526321053504944, + "num_tokens": 59792467.0, + "step": 1708 + }, + { + "epoch": 0.3173630454967502, + "grad_norm": 1.478058099746704, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8638929724693298, + "num_tokens": 59827099.0, + "step": 1709 + }, + { + "epoch": 0.31754874651810583, + "grad_norm": 1.4770978689193726, + "learning_rate": 1e-06, + "loss": 0.4984, + "mean_token_accuracy": 0.8344418406486511, + "num_tokens": 59865096.0, + "step": 1710 + }, + { + "epoch": 0.31773444753946145, + "grad_norm": 1.6567498445510864, + "learning_rate": 1e-06, + "loss": 0.4809, + "mean_token_accuracy": 0.8429871797561646, + "num_tokens": 59897664.0, + "step": 1711 + }, + { + "epoch": 0.3179201485608171, + "grad_norm": 1.480493426322937, + "learning_rate": 1e-06, + "loss": 0.4623, + "mean_token_accuracy": 0.8479610085487366, + "num_tokens": 59935493.0, + "step": 1712 + }, + { + "epoch": 0.3181058495821727, + "grad_norm": 1.5792442560195923, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8681235313415527, + "num_tokens": 59971565.0, + "step": 1713 + }, + { + "epoch": 0.3182915506035283, + "grad_norm": 1.5522863864898682, + "learning_rate": 1e-06, + "loss": 0.4633, + "mean_token_accuracy": 0.8507362604141235, + "num_tokens": 60007424.0, + "step": 1714 + }, + { + "epoch": 0.31847725162488394, + "grad_norm": 1.6082853078842163, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.8542745113372803, + "num_tokens": 60042622.0, + "step": 1715 + }, + { + "epoch": 0.31866295264623956, + "grad_norm": 1.5999380350112915, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8678378462791443, + "num_tokens": 60074199.0, + "step": 1716 + }, + { + "epoch": 0.3188486536675952, + "grad_norm": 1.5651909112930298, + "learning_rate": 1e-06, + "loss": 0.4313, + "mean_token_accuracy": 0.8591835498809814, + "num_tokens": 60105001.0, + "step": 1717 + }, + { + "epoch": 0.3190343546889508, + "grad_norm": 1.5018720626831055, + "learning_rate": 1e-06, + "loss": 0.4575, + "mean_token_accuracy": 0.8537580966949463, + "num_tokens": 60141748.0, + "step": 1718 + }, + { + "epoch": 0.3192200557103064, + "grad_norm": 1.6565182209014893, + "learning_rate": 1e-06, + "loss": 0.4778, + "mean_token_accuracy": 0.8446900248527527, + "num_tokens": 60176801.0, + "step": 1719 + }, + { + "epoch": 0.31940575673166205, + "grad_norm": 1.5557109117507935, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8527481555938721, + "num_tokens": 60212735.0, + "step": 1720 + }, + { + "epoch": 0.3195914577530176, + "grad_norm": 1.552089810371399, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8644066452980042, + "num_tokens": 60244306.0, + "step": 1721 + }, + { + "epoch": 0.31977715877437324, + "grad_norm": 1.5188472270965576, + "learning_rate": 1e-06, + "loss": 0.4891, + "mean_token_accuracy": 0.83940589427948, + "num_tokens": 60283912.0, + "step": 1722 + }, + { + "epoch": 0.31996285979572886, + "grad_norm": 1.5217000246047974, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8591500520706177, + "num_tokens": 60316648.0, + "step": 1723 + }, + { + "epoch": 0.3201485608170845, + "grad_norm": 1.5071386098861694, + "learning_rate": 1e-06, + "loss": 0.4627, + "mean_token_accuracy": 0.848303496837616, + "num_tokens": 60353456.0, + "step": 1724 + }, + { + "epoch": 0.3203342618384401, + "grad_norm": 1.5441787242889404, + "learning_rate": 1e-06, + "loss": 0.4285, + "mean_token_accuracy": 0.858265221118927, + "num_tokens": 60387634.0, + "step": 1725 + }, + { + "epoch": 0.3205199628597957, + "grad_norm": 1.5798171758651733, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8565313816070557, + "num_tokens": 60421184.0, + "step": 1726 + }, + { + "epoch": 0.32070566388115135, + "grad_norm": 1.6504154205322266, + "learning_rate": 1e-06, + "loss": 0.4653, + "mean_token_accuracy": 0.8500717878341675, + "num_tokens": 60456293.0, + "step": 1727 + }, + { + "epoch": 0.32089136490250697, + "grad_norm": 1.509863257408142, + "learning_rate": 1e-06, + "loss": 0.4428, + "mean_token_accuracy": 0.8537490963935852, + "num_tokens": 60494946.0, + "step": 1728 + }, + { + "epoch": 0.3210770659238626, + "grad_norm": 1.4568895101547241, + "learning_rate": 1e-06, + "loss": 0.4116, + "mean_token_accuracy": 0.8625075221061707, + "num_tokens": 60530338.0, + "step": 1729 + }, + { + "epoch": 0.3212627669452182, + "grad_norm": 1.4918732643127441, + "learning_rate": 1e-06, + "loss": 0.4397, + "mean_token_accuracy": 0.8556802868843079, + "num_tokens": 60568223.0, + "step": 1730 + }, + { + "epoch": 0.32144846796657384, + "grad_norm": 1.5694712400436401, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8712852597236633, + "num_tokens": 60600650.0, + "step": 1731 + }, + { + "epoch": 0.32163416898792946, + "grad_norm": 1.4711233377456665, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8664193749427795, + "num_tokens": 60640110.0, + "step": 1732 + }, + { + "epoch": 0.321819870009285, + "grad_norm": 1.5743112564086914, + "learning_rate": 1e-06, + "loss": 0.4532, + "mean_token_accuracy": 0.8497553467750549, + "num_tokens": 60673837.0, + "step": 1733 + }, + { + "epoch": 0.32200557103064065, + "grad_norm": 1.5909329652786255, + "learning_rate": 1e-06, + "loss": 0.4528, + "mean_token_accuracy": 0.8513588309288025, + "num_tokens": 60705637.0, + "step": 1734 + }, + { + "epoch": 0.32219127205199627, + "grad_norm": 1.7041449546813965, + "learning_rate": 1e-06, + "loss": 0.4523, + "mean_token_accuracy": 0.8531888127326965, + "num_tokens": 60733249.0, + "step": 1735 + }, + { + "epoch": 0.3223769730733519, + "grad_norm": 1.4706296920776367, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8585919737815857, + "num_tokens": 60770234.0, + "step": 1736 + }, + { + "epoch": 0.3225626740947075, + "grad_norm": 1.5656665563583374, + "learning_rate": 1e-06, + "loss": 0.4405, + "mean_token_accuracy": 0.8543603420257568, + "num_tokens": 60803029.0, + "step": 1737 + }, + { + "epoch": 0.32274837511606314, + "grad_norm": 1.7084187269210815, + "learning_rate": 1e-06, + "loss": 0.4523, + "mean_token_accuracy": 0.849282443523407, + "num_tokens": 60831359.0, + "step": 1738 + }, + { + "epoch": 0.32293407613741876, + "grad_norm": 1.4505765438079834, + "learning_rate": 1e-06, + "loss": 0.458, + "mean_token_accuracy": 0.8524437546730042, + "num_tokens": 60870855.0, + "step": 1739 + }, + { + "epoch": 0.3231197771587744, + "grad_norm": 1.6197799444198608, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8716694116592407, + "num_tokens": 60901799.0, + "step": 1740 + }, + { + "epoch": 0.32330547818013, + "grad_norm": 1.410717487335205, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.8633990287780762, + "num_tokens": 60939657.0, + "step": 1741 + }, + { + "epoch": 0.3234911792014856, + "grad_norm": 1.629542350769043, + "learning_rate": 1e-06, + "loss": 0.4692, + "mean_token_accuracy": 0.8432320356369019, + "num_tokens": 60972047.0, + "step": 1742 + }, + { + "epoch": 0.32367688022284125, + "grad_norm": 1.4664524793624878, + "learning_rate": 1e-06, + "loss": 0.4349, + "mean_token_accuracy": 0.8569701313972473, + "num_tokens": 61011306.0, + "step": 1743 + }, + { + "epoch": 0.32386258124419687, + "grad_norm": 1.5241972208023071, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8547742366790771, + "num_tokens": 61045700.0, + "step": 1744 + }, + { + "epoch": 0.32404828226555243, + "grad_norm": 1.516318917274475, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8646037578582764, + "num_tokens": 61079105.0, + "step": 1745 + }, + { + "epoch": 0.32423398328690806, + "grad_norm": 1.5860997438430786, + "learning_rate": 1e-06, + "loss": 0.4171, + "mean_token_accuracy": 0.8625878095626831, + "num_tokens": 61113068.0, + "step": 1746 + }, + { + "epoch": 0.3244196843082637, + "grad_norm": 1.538398265838623, + "learning_rate": 1e-06, + "loss": 0.439, + "mean_token_accuracy": 0.8541632890701294, + "num_tokens": 61152836.0, + "step": 1747 + }, + { + "epoch": 0.3246053853296193, + "grad_norm": 1.5592436790466309, + "learning_rate": 1e-06, + "loss": 0.4623, + "mean_token_accuracy": 0.8480463027954102, + "num_tokens": 61187128.0, + "step": 1748 + }, + { + "epoch": 0.3247910863509749, + "grad_norm": 1.5097843408584595, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8659324645996094, + "num_tokens": 61220320.0, + "step": 1749 + }, + { + "epoch": 0.32497678737233054, + "grad_norm": 1.3899948596954346, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.8719143867492676, + "num_tokens": 61257489.0, + "step": 1750 + }, + { + "epoch": 0.32516248839368617, + "grad_norm": 1.3910530805587769, + "learning_rate": 1e-06, + "loss": 0.4308, + "mean_token_accuracy": 0.857122540473938, + "num_tokens": 61295087.0, + "step": 1751 + }, + { + "epoch": 0.3253481894150418, + "grad_norm": 1.4540519714355469, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.8649152517318726, + "num_tokens": 61332568.0, + "step": 1752 + }, + { + "epoch": 0.3255338904363974, + "grad_norm": 1.532502293586731, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8661547899246216, + "num_tokens": 61367192.0, + "step": 1753 + }, + { + "epoch": 0.32571959145775303, + "grad_norm": 1.3582303524017334, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8727871179580688, + "num_tokens": 61406454.0, + "step": 1754 + }, + { + "epoch": 0.32590529247910865, + "grad_norm": 1.6805272102355957, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8603492975234985, + "num_tokens": 61436207.0, + "step": 1755 + }, + { + "epoch": 0.3260909935004643, + "grad_norm": 1.5258994102478027, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8698727488517761, + "num_tokens": 61467761.0, + "step": 1756 + }, + { + "epoch": 0.32627669452181984, + "grad_norm": 1.4434293508529663, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.860092282295227, + "num_tokens": 61505376.0, + "step": 1757 + }, + { + "epoch": 0.32646239554317547, + "grad_norm": 1.678689956665039, + "learning_rate": 1e-06, + "loss": 0.5031, + "mean_token_accuracy": 0.8356155157089233, + "num_tokens": 61537692.0, + "step": 1758 + }, + { + "epoch": 0.3266480965645311, + "grad_norm": 1.4817051887512207, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8704943060874939, + "num_tokens": 61575374.0, + "step": 1759 + }, + { + "epoch": 0.3268337975858867, + "grad_norm": 1.5250294208526611, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.863163411617279, + "num_tokens": 61610934.0, + "step": 1760 + }, + { + "epoch": 0.32701949860724233, + "grad_norm": 1.4759430885314941, + "learning_rate": 1e-06, + "loss": 0.4399, + "mean_token_accuracy": 0.8537850975990295, + "num_tokens": 61645827.0, + "step": 1761 + }, + { + "epoch": 0.32720519962859795, + "grad_norm": 1.4234809875488281, + "learning_rate": 1e-06, + "loss": 0.428, + "mean_token_accuracy": 0.85713791847229, + "num_tokens": 61687525.0, + "step": 1762 + }, + { + "epoch": 0.3273909006499536, + "grad_norm": 1.4828853607177734, + "learning_rate": 1e-06, + "loss": 0.4288, + "mean_token_accuracy": 0.8609960675239563, + "num_tokens": 61723337.0, + "step": 1763 + }, + { + "epoch": 0.3275766016713092, + "grad_norm": 1.5383785963058472, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.8618932962417603, + "num_tokens": 61759211.0, + "step": 1764 + }, + { + "epoch": 0.3277623026926648, + "grad_norm": 1.469753384590149, + "learning_rate": 1e-06, + "loss": 0.4559, + "mean_token_accuracy": 0.8539955019950867, + "num_tokens": 61802990.0, + "step": 1765 + }, + { + "epoch": 0.32794800371402044, + "grad_norm": 1.5169090032577515, + "learning_rate": 1e-06, + "loss": 0.4298, + "mean_token_accuracy": 0.8562780618667603, + "num_tokens": 61838317.0, + "step": 1766 + }, + { + "epoch": 0.32813370473537606, + "grad_norm": 1.5524104833602905, + "learning_rate": 1e-06, + "loss": 0.4243, + "mean_token_accuracy": 0.8616149425506592, + "num_tokens": 61873505.0, + "step": 1767 + }, + { + "epoch": 0.3283194057567317, + "grad_norm": 1.5539828538894653, + "learning_rate": 1e-06, + "loss": 0.4484, + "mean_token_accuracy": 0.8490002155303955, + "num_tokens": 61905314.0, + "step": 1768 + }, + { + "epoch": 0.32850510677808725, + "grad_norm": 1.5445994138717651, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8684746026992798, + "num_tokens": 61939377.0, + "step": 1769 + }, + { + "epoch": 0.3286908077994429, + "grad_norm": 1.5674831867218018, + "learning_rate": 1e-06, + "loss": 0.4758, + "mean_token_accuracy": 0.8467673063278198, + "num_tokens": 61973693.0, + "step": 1770 + }, + { + "epoch": 0.3288765088207985, + "grad_norm": 1.5261191129684448, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8709299564361572, + "num_tokens": 62006570.0, + "step": 1771 + }, + { + "epoch": 0.3290622098421541, + "grad_norm": 1.5404332876205444, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8678625822067261, + "num_tokens": 62037488.0, + "step": 1772 + }, + { + "epoch": 0.32924791086350974, + "grad_norm": 1.6109520196914673, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8743327260017395, + "num_tokens": 62066277.0, + "step": 1773 + }, + { + "epoch": 0.32943361188486536, + "grad_norm": 1.6299563646316528, + "learning_rate": 1e-06, + "loss": 0.449, + "mean_token_accuracy": 0.8530781269073486, + "num_tokens": 62098008.0, + "step": 1774 + }, + { + "epoch": 0.329619312906221, + "grad_norm": 1.5364227294921875, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8640176057815552, + "num_tokens": 62137904.0, + "step": 1775 + }, + { + "epoch": 0.3298050139275766, + "grad_norm": 1.5853265523910522, + "learning_rate": 1e-06, + "loss": 0.4442, + "mean_token_accuracy": 0.8593281507492065, + "num_tokens": 62174117.0, + "step": 1776 + }, + { + "epoch": 0.32999071494893223, + "grad_norm": 1.7072818279266357, + "learning_rate": 1e-06, + "loss": 0.4598, + "mean_token_accuracy": 0.8475721478462219, + "num_tokens": 62216459.0, + "step": 1777 + }, + { + "epoch": 0.33017641597028785, + "grad_norm": 1.4528857469558716, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.8693890571594238, + "num_tokens": 62254503.0, + "step": 1778 + }, + { + "epoch": 0.3303621169916435, + "grad_norm": 1.3752050399780273, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8667471408843994, + "num_tokens": 62295355.0, + "step": 1779 + }, + { + "epoch": 0.3305478180129991, + "grad_norm": 1.678675651550293, + "learning_rate": 1e-06, + "loss": 0.4496, + "mean_token_accuracy": 0.8496007919311523, + "num_tokens": 62322825.0, + "step": 1780 + }, + { + "epoch": 0.33073351903435466, + "grad_norm": 1.5946705341339111, + "learning_rate": 1e-06, + "loss": 0.4359, + "mean_token_accuracy": 0.8558199405670166, + "num_tokens": 62355855.0, + "step": 1781 + }, + { + "epoch": 0.3309192200557103, + "grad_norm": 1.4570132493972778, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8799049854278564, + "num_tokens": 62388729.0, + "step": 1782 + }, + { + "epoch": 0.3311049210770659, + "grad_norm": 1.5060076713562012, + "learning_rate": 1e-06, + "loss": 0.4453, + "mean_token_accuracy": 0.8514188528060913, + "num_tokens": 62425347.0, + "step": 1783 + }, + { + "epoch": 0.3312906220984215, + "grad_norm": 1.4194294214248657, + "learning_rate": 1e-06, + "loss": 0.4357, + "mean_token_accuracy": 0.8553436994552612, + "num_tokens": 62464094.0, + "step": 1784 + }, + { + "epoch": 0.33147632311977715, + "grad_norm": 1.5092835426330566, + "learning_rate": 1e-06, + "loss": 0.4654, + "mean_token_accuracy": 0.8517933487892151, + "num_tokens": 62497843.0, + "step": 1785 + }, + { + "epoch": 0.33166202414113277, + "grad_norm": 1.6498925685882568, + "learning_rate": 1e-06, + "loss": 0.435, + "mean_token_accuracy": 0.8550419807434082, + "num_tokens": 62533971.0, + "step": 1786 + }, + { + "epoch": 0.3318477251624884, + "grad_norm": 1.442495346069336, + "learning_rate": 1e-06, + "loss": 0.4445, + "mean_token_accuracy": 0.8521337509155273, + "num_tokens": 62573961.0, + "step": 1787 + }, + { + "epoch": 0.332033426183844, + "grad_norm": 1.499240517616272, + "learning_rate": 1e-06, + "loss": 0.4721, + "mean_token_accuracy": 0.8461043834686279, + "num_tokens": 62614335.0, + "step": 1788 + }, + { + "epoch": 0.33221912720519964, + "grad_norm": 1.5407966375350952, + "learning_rate": 1e-06, + "loss": 0.4163, + "mean_token_accuracy": 0.8594061732292175, + "num_tokens": 62646029.0, + "step": 1789 + }, + { + "epoch": 0.33240482822655526, + "grad_norm": 1.5924750566482544, + "learning_rate": 1e-06, + "loss": 0.4583, + "mean_token_accuracy": 0.8499199151992798, + "num_tokens": 62678528.0, + "step": 1790 + }, + { + "epoch": 0.3325905292479109, + "grad_norm": 1.628823161125183, + "learning_rate": 1e-06, + "loss": 0.4363, + "mean_token_accuracy": 0.8572312593460083, + "num_tokens": 62709270.0, + "step": 1791 + }, + { + "epoch": 0.3327762302692665, + "grad_norm": 1.544783592224121, + "learning_rate": 1e-06, + "loss": 0.4248, + "mean_token_accuracy": 0.8595670461654663, + "num_tokens": 62742144.0, + "step": 1792 + }, + { + "epoch": 0.33296193129062207, + "grad_norm": 1.4455556869506836, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.8638553619384766, + "num_tokens": 62778226.0, + "step": 1793 + }, + { + "epoch": 0.3331476323119777, + "grad_norm": 1.641465425491333, + "learning_rate": 1e-06, + "loss": 0.4325, + "mean_token_accuracy": 0.8578181862831116, + "num_tokens": 62808429.0, + "step": 1794 + }, + { + "epoch": 0.3333333333333333, + "grad_norm": 1.8818819522857666, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.860837459564209, + "num_tokens": 62850301.0, + "step": 1795 + }, + { + "epoch": 0.33351903435468894, + "grad_norm": 1.4037342071533203, + "learning_rate": 1e-06, + "loss": 0.4443, + "mean_token_accuracy": 0.8519860506057739, + "num_tokens": 62894156.0, + "step": 1796 + }, + { + "epoch": 0.33370473537604456, + "grad_norm": 1.686745047569275, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8728407621383667, + "num_tokens": 62923252.0, + "step": 1797 + }, + { + "epoch": 0.3338904363974002, + "grad_norm": 1.5659093856811523, + "learning_rate": 1e-06, + "loss": 0.4695, + "mean_token_accuracy": 0.8489916324615479, + "num_tokens": 62959865.0, + "step": 1798 + }, + { + "epoch": 0.3340761374187558, + "grad_norm": 1.5694046020507812, + "learning_rate": 1e-06, + "loss": 0.492, + "mean_token_accuracy": 0.8441715240478516, + "num_tokens": 62993872.0, + "step": 1799 + }, + { + "epoch": 0.3342618384401114, + "grad_norm": 1.468281626701355, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.866634726524353, + "num_tokens": 63028708.0, + "step": 1800 + }, + { + "epoch": 0.33444753946146705, + "grad_norm": 1.5017386674880981, + "learning_rate": 1e-06, + "loss": 0.4547, + "mean_token_accuracy": 0.8524482250213623, + "num_tokens": 63067526.0, + "step": 1801 + }, + { + "epoch": 0.33463324048282267, + "grad_norm": 1.4528508186340332, + "learning_rate": 1e-06, + "loss": 0.4622, + "mean_token_accuracy": 0.8523439168930054, + "num_tokens": 63108683.0, + "step": 1802 + }, + { + "epoch": 0.3348189415041783, + "grad_norm": 1.4740887880325317, + "learning_rate": 1e-06, + "loss": 0.4477, + "mean_token_accuracy": 0.8564759492874146, + "num_tokens": 63144666.0, + "step": 1803 + }, + { + "epoch": 0.3350046425255339, + "grad_norm": 1.442632794380188, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8712954521179199, + "num_tokens": 63176800.0, + "step": 1804 + }, + { + "epoch": 0.33519034354688954, + "grad_norm": 1.4102741479873657, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8690510392189026, + "num_tokens": 63214156.0, + "step": 1805 + }, + { + "epoch": 0.3353760445682451, + "grad_norm": 1.5048096179962158, + "learning_rate": 1e-06, + "loss": 0.4793, + "mean_token_accuracy": 0.8410329818725586, + "num_tokens": 63255753.0, + "step": 1806 + }, + { + "epoch": 0.3355617455896007, + "grad_norm": 1.5646846294403076, + "learning_rate": 1e-06, + "loss": 0.4746, + "mean_token_accuracy": 0.845414936542511, + "num_tokens": 63289415.0, + "step": 1807 + }, + { + "epoch": 0.33574744661095635, + "grad_norm": 1.3774133920669556, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8711740970611572, + "num_tokens": 63327472.0, + "step": 1808 + }, + { + "epoch": 0.33593314763231197, + "grad_norm": 1.4405019283294678, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.8698753118515015, + "num_tokens": 63360336.0, + "step": 1809 + }, + { + "epoch": 0.3361188486536676, + "grad_norm": 1.6605031490325928, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8691872954368591, + "num_tokens": 63389185.0, + "step": 1810 + }, + { + "epoch": 0.3363045496750232, + "grad_norm": 1.4646199941635132, + "learning_rate": 1e-06, + "loss": 0.429, + "mean_token_accuracy": 0.8610854148864746, + "num_tokens": 63428044.0, + "step": 1811 + }, + { + "epoch": 0.33649025069637883, + "grad_norm": 1.5750023126602173, + "learning_rate": 1e-06, + "loss": 0.4422, + "mean_token_accuracy": 0.8597804307937622, + "num_tokens": 63461550.0, + "step": 1812 + }, + { + "epoch": 0.33667595171773446, + "grad_norm": 1.5905485153198242, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8630889654159546, + "num_tokens": 63492914.0, + "step": 1813 + }, + { + "epoch": 0.3368616527390901, + "grad_norm": 1.5014303922653198, + "learning_rate": 1e-06, + "loss": 0.4304, + "mean_token_accuracy": 0.8600784540176392, + "num_tokens": 63527989.0, + "step": 1814 + }, + { + "epoch": 0.3370473537604457, + "grad_norm": 1.4971706867218018, + "learning_rate": 1e-06, + "loss": 0.4307, + "mean_token_accuracy": 0.8609307408332825, + "num_tokens": 63560907.0, + "step": 1815 + }, + { + "epoch": 0.3372330547818013, + "grad_norm": 1.551119089126587, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8524354696273804, + "num_tokens": 63594440.0, + "step": 1816 + }, + { + "epoch": 0.33741875580315694, + "grad_norm": 1.5107189416885376, + "learning_rate": 1e-06, + "loss": 0.4388, + "mean_token_accuracy": 0.8549512624740601, + "num_tokens": 63630431.0, + "step": 1817 + }, + { + "epoch": 0.3376044568245125, + "grad_norm": 1.4841560125350952, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8661081790924072, + "num_tokens": 63669257.0, + "step": 1818 + }, + { + "epoch": 0.33779015784586813, + "grad_norm": 1.582062840461731, + "learning_rate": 1e-06, + "loss": 0.4388, + "mean_token_accuracy": 0.8556985855102539, + "num_tokens": 63703902.0, + "step": 1819 + }, + { + "epoch": 0.33797585886722376, + "grad_norm": 1.7234331369400024, + "learning_rate": 1e-06, + "loss": 0.4347, + "mean_token_accuracy": 0.8553867340087891, + "num_tokens": 63730252.0, + "step": 1820 + }, + { + "epoch": 0.3381615598885794, + "grad_norm": 1.600346565246582, + "learning_rate": 1e-06, + "loss": 0.4333, + "mean_token_accuracy": 0.8561975359916687, + "num_tokens": 63762974.0, + "step": 1821 + }, + { + "epoch": 0.338347260909935, + "grad_norm": 1.5464622974395752, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8738075494766235, + "num_tokens": 63793780.0, + "step": 1822 + }, + { + "epoch": 0.3385329619312906, + "grad_norm": 1.6711946725845337, + "learning_rate": 1e-06, + "loss": 0.5047, + "mean_token_accuracy": 0.8364061117172241, + "num_tokens": 63825158.0, + "step": 1823 + }, + { + "epoch": 0.33871866295264624, + "grad_norm": 1.4610806703567505, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8618518114089966, + "num_tokens": 63859806.0, + "step": 1824 + }, + { + "epoch": 0.33890436397400187, + "grad_norm": 1.5029758214950562, + "learning_rate": 1e-06, + "loss": 0.4165, + "mean_token_accuracy": 0.8684275150299072, + "num_tokens": 63893415.0, + "step": 1825 + }, + { + "epoch": 0.3390900649953575, + "grad_norm": 1.549454689025879, + "learning_rate": 1e-06, + "loss": 0.4163, + "mean_token_accuracy": 0.8619430065155029, + "num_tokens": 63925954.0, + "step": 1826 + }, + { + "epoch": 0.3392757660167131, + "grad_norm": 1.4174131155014038, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8666231632232666, + "num_tokens": 63964561.0, + "step": 1827 + }, + { + "epoch": 0.33946146703806873, + "grad_norm": 1.5986289978027344, + "learning_rate": 1e-06, + "loss": 0.4398, + "mean_token_accuracy": 0.8547945022583008, + "num_tokens": 63997906.0, + "step": 1828 + }, + { + "epoch": 0.33964716805942435, + "grad_norm": 1.5560327768325806, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8714302182197571, + "num_tokens": 64028323.0, + "step": 1829 + }, + { + "epoch": 0.3398328690807799, + "grad_norm": 1.5130674839019775, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.8588449954986572, + "num_tokens": 64062146.0, + "step": 1830 + }, + { + "epoch": 0.34001857010213554, + "grad_norm": 1.6116094589233398, + "learning_rate": 1e-06, + "loss": 0.4282, + "mean_token_accuracy": 0.8605822324752808, + "num_tokens": 64092937.0, + "step": 1831 + }, + { + "epoch": 0.34020427112349116, + "grad_norm": 1.4761496782302856, + "learning_rate": 1e-06, + "loss": 0.4389, + "mean_token_accuracy": 0.8542084693908691, + "num_tokens": 64131298.0, + "step": 1832 + }, + { + "epoch": 0.3403899721448468, + "grad_norm": 1.6639108657836914, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8692331910133362, + "num_tokens": 64161520.0, + "step": 1833 + }, + { + "epoch": 0.3405756731662024, + "grad_norm": 1.502945899963379, + "learning_rate": 1e-06, + "loss": 0.4257, + "mean_token_accuracy": 0.8587135672569275, + "num_tokens": 64196948.0, + "step": 1834 + }, + { + "epoch": 0.34076137418755803, + "grad_norm": 1.4814515113830566, + "learning_rate": 1e-06, + "loss": 0.4269, + "mean_token_accuracy": 0.8621983528137207, + "num_tokens": 64232604.0, + "step": 1835 + }, + { + "epoch": 0.34094707520891365, + "grad_norm": 1.5517629384994507, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.8605861663818359, + "num_tokens": 64264899.0, + "step": 1836 + }, + { + "epoch": 0.3411327762302693, + "grad_norm": 1.6677480936050415, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8695602416992188, + "num_tokens": 64297604.0, + "step": 1837 + }, + { + "epoch": 0.3413184772516249, + "grad_norm": 1.495172142982483, + "learning_rate": 1e-06, + "loss": 0.4412, + "mean_token_accuracy": 0.852025032043457, + "num_tokens": 64335436.0, + "step": 1838 + }, + { + "epoch": 0.3415041782729805, + "grad_norm": 1.445390224456787, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8712378144264221, + "num_tokens": 64368963.0, + "step": 1839 + }, + { + "epoch": 0.34168987929433614, + "grad_norm": 1.4869310855865479, + "learning_rate": 1e-06, + "loss": 0.4576, + "mean_token_accuracy": 0.8510847091674805, + "num_tokens": 64406155.0, + "step": 1840 + }, + { + "epoch": 0.34187558031569176, + "grad_norm": 1.3374781608581543, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.8655794858932495, + "num_tokens": 64449401.0, + "step": 1841 + }, + { + "epoch": 0.34206128133704733, + "grad_norm": 1.7899194955825806, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8831571340560913, + "num_tokens": 64477569.0, + "step": 1842 + }, + { + "epoch": 0.34224698235840295, + "grad_norm": 1.6050437688827515, + "learning_rate": 1e-06, + "loss": 0.4633, + "mean_token_accuracy": 0.848410964012146, + "num_tokens": 64511537.0, + "step": 1843 + }, + { + "epoch": 0.3424326833797586, + "grad_norm": 1.6402045488357544, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8760271668434143, + "num_tokens": 64538214.0, + "step": 1844 + }, + { + "epoch": 0.3426183844011142, + "grad_norm": 1.6647700071334839, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8613686561584473, + "num_tokens": 64566152.0, + "step": 1845 + }, + { + "epoch": 0.3428040854224698, + "grad_norm": 1.5104539394378662, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8575778007507324, + "num_tokens": 64600770.0, + "step": 1846 + }, + { + "epoch": 0.34298978644382544, + "grad_norm": 1.588360071182251, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8722773790359497, + "num_tokens": 64631901.0, + "step": 1847 + }, + { + "epoch": 0.34317548746518106, + "grad_norm": 1.5209547281265259, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8696889877319336, + "num_tokens": 64664890.0, + "step": 1848 + }, + { + "epoch": 0.3433611884865367, + "grad_norm": 1.4395710229873657, + "learning_rate": 1e-06, + "loss": 0.3911, + "mean_token_accuracy": 0.869163453578949, + "num_tokens": 64700329.0, + "step": 1849 + }, + { + "epoch": 0.3435468895078923, + "grad_norm": 1.4456180334091187, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8733664751052856, + "num_tokens": 64733787.0, + "step": 1850 + }, + { + "epoch": 0.34373259052924793, + "grad_norm": 1.6657922267913818, + "learning_rate": 1e-06, + "loss": 0.4323, + "mean_token_accuracy": 0.8534562587738037, + "num_tokens": 64765025.0, + "step": 1851 + }, + { + "epoch": 0.34391829155060355, + "grad_norm": 1.780037522315979, + "learning_rate": 1e-06, + "loss": 0.5067, + "mean_token_accuracy": 0.8383269309997559, + "num_tokens": 64802431.0, + "step": 1852 + }, + { + "epoch": 0.34410399257195917, + "grad_norm": 1.4671366214752197, + "learning_rate": 1e-06, + "loss": 0.4078, + "mean_token_accuracy": 0.8634614944458008, + "num_tokens": 64836059.0, + "step": 1853 + }, + { + "epoch": 0.34428969359331474, + "grad_norm": 1.51897132396698, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.8631435632705688, + "num_tokens": 64876496.0, + "step": 1854 + }, + { + "epoch": 0.34447539461467036, + "grad_norm": 1.504019856452942, + "learning_rate": 1e-06, + "loss": 0.433, + "mean_token_accuracy": 0.8600783944129944, + "num_tokens": 64909595.0, + "step": 1855 + }, + { + "epoch": 0.344661095636026, + "grad_norm": 1.52414071559906, + "learning_rate": 1e-06, + "loss": 0.4427, + "mean_token_accuracy": 0.8536222577095032, + "num_tokens": 64945075.0, + "step": 1856 + }, + { + "epoch": 0.3448467966573816, + "grad_norm": 1.4693392515182495, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.8725544214248657, + "num_tokens": 64981329.0, + "step": 1857 + }, + { + "epoch": 0.3450324976787372, + "grad_norm": 1.4178072214126587, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8658558130264282, + "num_tokens": 65022097.0, + "step": 1858 + }, + { + "epoch": 0.34521819870009285, + "grad_norm": 1.4957941770553589, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8654299974441528, + "num_tokens": 65057363.0, + "step": 1859 + }, + { + "epoch": 0.34540389972144847, + "grad_norm": 1.5769904851913452, + "learning_rate": 1e-06, + "loss": 0.4684, + "mean_token_accuracy": 0.8455446362495422, + "num_tokens": 65091511.0, + "step": 1860 + }, + { + "epoch": 0.3455896007428041, + "grad_norm": 1.4741101264953613, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.8606466054916382, + "num_tokens": 65128154.0, + "step": 1861 + }, + { + "epoch": 0.3457753017641597, + "grad_norm": 1.5284017324447632, + "learning_rate": 1e-06, + "loss": 0.4318, + "mean_token_accuracy": 0.8562877774238586, + "num_tokens": 65165248.0, + "step": 1862 + }, + { + "epoch": 0.34596100278551534, + "grad_norm": 1.5195281505584717, + "learning_rate": 1e-06, + "loss": 0.4269, + "mean_token_accuracy": 0.8563129305839539, + "num_tokens": 65197973.0, + "step": 1863 + }, + { + "epoch": 0.34614670380687096, + "grad_norm": 1.4249272346496582, + "learning_rate": 1e-06, + "loss": 0.4042, + "mean_token_accuracy": 0.8619723320007324, + "num_tokens": 65233446.0, + "step": 1864 + }, + { + "epoch": 0.3463324048282266, + "grad_norm": 1.682133674621582, + "learning_rate": 1e-06, + "loss": 0.4588, + "mean_token_accuracy": 0.8501092791557312, + "num_tokens": 65261144.0, + "step": 1865 + }, + { + "epoch": 0.34651810584958215, + "grad_norm": 1.7427499294281006, + "learning_rate": 1e-06, + "loss": 0.5104, + "mean_token_accuracy": 0.8360714912414551, + "num_tokens": 65290335.0, + "step": 1866 + }, + { + "epoch": 0.34670380687093777, + "grad_norm": 1.51658296585083, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8730964660644531, + "num_tokens": 65329197.0, + "step": 1867 + }, + { + "epoch": 0.3468895078922934, + "grad_norm": 1.52835214138031, + "learning_rate": 1e-06, + "loss": 0.4165, + "mean_token_accuracy": 0.8640042543411255, + "num_tokens": 65362868.0, + "step": 1868 + }, + { + "epoch": 0.347075208913649, + "grad_norm": 1.4899357557296753, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8674741983413696, + "num_tokens": 65396221.0, + "step": 1869 + }, + { + "epoch": 0.34726090993500464, + "grad_norm": 1.4882550239562988, + "learning_rate": 1e-06, + "loss": 0.4126, + "mean_token_accuracy": 0.8591153025627136, + "num_tokens": 65431506.0, + "step": 1870 + }, + { + "epoch": 0.34744661095636026, + "grad_norm": 1.512646198272705, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8671766519546509, + "num_tokens": 65465078.0, + "step": 1871 + }, + { + "epoch": 0.3476323119777159, + "grad_norm": 1.4008517265319824, + "learning_rate": 1e-06, + "loss": 0.4424, + "mean_token_accuracy": 0.859102189540863, + "num_tokens": 65503957.0, + "step": 1872 + }, + { + "epoch": 0.3478180129990715, + "grad_norm": 1.5363503694534302, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8604005575180054, + "num_tokens": 65541783.0, + "step": 1873 + }, + { + "epoch": 0.3480037140204271, + "grad_norm": 1.435004472732544, + "learning_rate": 1e-06, + "loss": 0.4638, + "mean_token_accuracy": 0.8474623560905457, + "num_tokens": 65581860.0, + "step": 1874 + }, + { + "epoch": 0.34818941504178275, + "grad_norm": 1.5680781602859497, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.8642940521240234, + "num_tokens": 65612317.0, + "step": 1875 + }, + { + "epoch": 0.34837511606313837, + "grad_norm": 1.5809333324432373, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8589259386062622, + "num_tokens": 65645876.0, + "step": 1876 + }, + { + "epoch": 0.348560817084494, + "grad_norm": 1.5826244354248047, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.8593196272850037, + "num_tokens": 65677544.0, + "step": 1877 + }, + { + "epoch": 0.34874651810584956, + "grad_norm": 1.557020902633667, + "learning_rate": 1e-06, + "loss": 0.4486, + "mean_token_accuracy": 0.8472495675086975, + "num_tokens": 65713056.0, + "step": 1878 + }, + { + "epoch": 0.3489322191272052, + "grad_norm": 1.4341334104537964, + "learning_rate": 1e-06, + "loss": 0.4184, + "mean_token_accuracy": 0.8614373207092285, + "num_tokens": 65750315.0, + "step": 1879 + }, + { + "epoch": 0.3491179201485608, + "grad_norm": 1.4872734546661377, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8691883087158203, + "num_tokens": 65788169.0, + "step": 1880 + }, + { + "epoch": 0.3493036211699164, + "grad_norm": 1.7384345531463623, + "learning_rate": 1e-06, + "loss": 0.4317, + "mean_token_accuracy": 0.8609569072723389, + "num_tokens": 65816785.0, + "step": 1881 + }, + { + "epoch": 0.34948932219127204, + "grad_norm": 1.4652098417282104, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8698607683181763, + "num_tokens": 65850335.0, + "step": 1882 + }, + { + "epoch": 0.34967502321262767, + "grad_norm": 1.5058259963989258, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.8621323704719543, + "num_tokens": 65885848.0, + "step": 1883 + }, + { + "epoch": 0.3498607242339833, + "grad_norm": 1.5180528163909912, + "learning_rate": 1e-06, + "loss": 0.4212, + "mean_token_accuracy": 0.8611996173858643, + "num_tokens": 65919169.0, + "step": 1884 + }, + { + "epoch": 0.3500464252553389, + "grad_norm": 1.564733862876892, + "learning_rate": 1e-06, + "loss": 0.4506, + "mean_token_accuracy": 0.8488278985023499, + "num_tokens": 65956731.0, + "step": 1885 + }, + { + "epoch": 0.35023212627669453, + "grad_norm": 1.5023515224456787, + "learning_rate": 1e-06, + "loss": 0.4346, + "mean_token_accuracy": 0.8558511734008789, + "num_tokens": 65990849.0, + "step": 1886 + }, + { + "epoch": 0.35041782729805016, + "grad_norm": 1.423554539680481, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.8673397302627563, + "num_tokens": 66029862.0, + "step": 1887 + }, + { + "epoch": 0.3506035283194058, + "grad_norm": 1.5329591035842896, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8746967911720276, + "num_tokens": 66060581.0, + "step": 1888 + }, + { + "epoch": 0.3507892293407614, + "grad_norm": 1.609537959098816, + "learning_rate": 1e-06, + "loss": 0.4766, + "mean_token_accuracy": 0.8484514951705933, + "num_tokens": 66095147.0, + "step": 1889 + }, + { + "epoch": 0.35097493036211697, + "grad_norm": 1.460242748260498, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.8595843315124512, + "num_tokens": 66133071.0, + "step": 1890 + }, + { + "epoch": 0.3511606313834726, + "grad_norm": 1.4119150638580322, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8667291402816772, + "num_tokens": 66170449.0, + "step": 1891 + }, + { + "epoch": 0.3513463324048282, + "grad_norm": 1.4524800777435303, + "learning_rate": 1e-06, + "loss": 0.4355, + "mean_token_accuracy": 0.8547634482383728, + "num_tokens": 66207447.0, + "step": 1892 + }, + { + "epoch": 0.35153203342618383, + "grad_norm": 1.5687941312789917, + "learning_rate": 1e-06, + "loss": 0.411, + "mean_token_accuracy": 0.8640525341033936, + "num_tokens": 66240652.0, + "step": 1893 + }, + { + "epoch": 0.35171773444753945, + "grad_norm": 1.739362359046936, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.862747311592102, + "num_tokens": 66274138.0, + "step": 1894 + }, + { + "epoch": 0.3519034354688951, + "grad_norm": 1.4758318662643433, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.864656925201416, + "num_tokens": 66309496.0, + "step": 1895 + }, + { + "epoch": 0.3520891364902507, + "grad_norm": 1.5179075002670288, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8673615455627441, + "num_tokens": 66341662.0, + "step": 1896 + }, + { + "epoch": 0.3522748375116063, + "grad_norm": 1.6273822784423828, + "learning_rate": 1e-06, + "loss": 0.4682, + "mean_token_accuracy": 0.8451401591300964, + "num_tokens": 66372870.0, + "step": 1897 + }, + { + "epoch": 0.35246053853296194, + "grad_norm": 1.4410762786865234, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8606282472610474, + "num_tokens": 66409784.0, + "step": 1898 + }, + { + "epoch": 0.35264623955431756, + "grad_norm": 1.6566357612609863, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.860551118850708, + "num_tokens": 66440304.0, + "step": 1899 + }, + { + "epoch": 0.3528319405756732, + "grad_norm": 1.5254205465316772, + "learning_rate": 1e-06, + "loss": 0.4643, + "mean_token_accuracy": 0.8471803665161133, + "num_tokens": 66474736.0, + "step": 1900 + }, + { + "epoch": 0.3530176415970288, + "grad_norm": 1.5221585035324097, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8514996767044067, + "num_tokens": 66512216.0, + "step": 1901 + }, + { + "epoch": 0.3532033426183844, + "grad_norm": 1.4193897247314453, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8686783313751221, + "num_tokens": 66549884.0, + "step": 1902 + }, + { + "epoch": 0.35338904363974, + "grad_norm": 1.4688562154769897, + "learning_rate": 1e-06, + "loss": 0.4608, + "mean_token_accuracy": 0.8463554382324219, + "num_tokens": 66588316.0, + "step": 1903 + }, + { + "epoch": 0.3535747446610956, + "grad_norm": 1.5124198198318481, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.882394552230835, + "num_tokens": 66620732.0, + "step": 1904 + }, + { + "epoch": 0.35376044568245124, + "grad_norm": 1.650603175163269, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8603726625442505, + "num_tokens": 66651698.0, + "step": 1905 + }, + { + "epoch": 0.35394614670380686, + "grad_norm": 1.4125490188598633, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8687479496002197, + "num_tokens": 66691322.0, + "step": 1906 + }, + { + "epoch": 0.3541318477251625, + "grad_norm": 1.5036801099777222, + "learning_rate": 1e-06, + "loss": 0.4378, + "mean_token_accuracy": 0.8480973839759827, + "num_tokens": 66725132.0, + "step": 1907 + }, + { + "epoch": 0.3543175487465181, + "grad_norm": 1.375853419303894, + "learning_rate": 1e-06, + "loss": 0.4494, + "mean_token_accuracy": 0.8533514738082886, + "num_tokens": 66771078.0, + "step": 1908 + }, + { + "epoch": 0.35450324976787373, + "grad_norm": 1.4666898250579834, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8611356019973755, + "num_tokens": 66807717.0, + "step": 1909 + }, + { + "epoch": 0.35468895078922935, + "grad_norm": 1.4743008613586426, + "learning_rate": 1e-06, + "loss": 0.4577, + "mean_token_accuracy": 0.8482332229614258, + "num_tokens": 66848180.0, + "step": 1910 + }, + { + "epoch": 0.354874651810585, + "grad_norm": 1.4635993242263794, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8753758668899536, + "num_tokens": 66882151.0, + "step": 1911 + }, + { + "epoch": 0.3550603528319406, + "grad_norm": 1.475674033164978, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8664228320121765, + "num_tokens": 66916302.0, + "step": 1912 + }, + { + "epoch": 0.3552460538532962, + "grad_norm": 1.5382531881332397, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8738139867782593, + "num_tokens": 66950346.0, + "step": 1913 + }, + { + "epoch": 0.3554317548746518, + "grad_norm": 1.4989148378372192, + "learning_rate": 1e-06, + "loss": 0.4321, + "mean_token_accuracy": 0.8550775647163391, + "num_tokens": 66987113.0, + "step": 1914 + }, + { + "epoch": 0.3556174558960074, + "grad_norm": 1.4227027893066406, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8561776280403137, + "num_tokens": 67024819.0, + "step": 1915 + }, + { + "epoch": 0.35580315691736303, + "grad_norm": 1.5459542274475098, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8603999614715576, + "num_tokens": 67059814.0, + "step": 1916 + }, + { + "epoch": 0.35598885793871865, + "grad_norm": 1.6548784971237183, + "learning_rate": 1e-06, + "loss": 0.4485, + "mean_token_accuracy": 0.8515487909317017, + "num_tokens": 67088992.0, + "step": 1917 + }, + { + "epoch": 0.3561745589600743, + "grad_norm": 1.4730230569839478, + "learning_rate": 1e-06, + "loss": 0.4498, + "mean_token_accuracy": 0.8547544479370117, + "num_tokens": 67128284.0, + "step": 1918 + }, + { + "epoch": 0.3563602599814299, + "grad_norm": 1.6397958993911743, + "learning_rate": 1e-06, + "loss": 0.4868, + "mean_token_accuracy": 0.8483278751373291, + "num_tokens": 67164599.0, + "step": 1919 + }, + { + "epoch": 0.3565459610027855, + "grad_norm": 1.4399608373641968, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8658971786499023, + "num_tokens": 67200028.0, + "step": 1920 + }, + { + "epoch": 0.35673166202414114, + "grad_norm": 1.5201572179794312, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8609126806259155, + "num_tokens": 67231788.0, + "step": 1921 + }, + { + "epoch": 0.35691736304549676, + "grad_norm": 1.566017985343933, + "learning_rate": 1e-06, + "loss": 0.4551, + "mean_token_accuracy": 0.8513752222061157, + "num_tokens": 67265721.0, + "step": 1922 + }, + { + "epoch": 0.3571030640668524, + "grad_norm": 1.5281440019607544, + "learning_rate": 1e-06, + "loss": 0.425, + "mean_token_accuracy": 0.8594613075256348, + "num_tokens": 67306307.0, + "step": 1923 + }, + { + "epoch": 0.357288765088208, + "grad_norm": 1.5092915296554565, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8654131889343262, + "num_tokens": 67339857.0, + "step": 1924 + }, + { + "epoch": 0.3574744661095636, + "grad_norm": 1.6021714210510254, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8504624366760254, + "num_tokens": 67369578.0, + "step": 1925 + }, + { + "epoch": 0.3576601671309192, + "grad_norm": 1.4728647470474243, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8632175922393799, + "num_tokens": 67404108.0, + "step": 1926 + }, + { + "epoch": 0.3578458681522748, + "grad_norm": 1.5087928771972656, + "learning_rate": 1e-06, + "loss": 0.4408, + "mean_token_accuracy": 0.8568121194839478, + "num_tokens": 67442442.0, + "step": 1927 + }, + { + "epoch": 0.35803156917363044, + "grad_norm": 1.5189310312271118, + "learning_rate": 1e-06, + "loss": 0.448, + "mean_token_accuracy": 0.8533027172088623, + "num_tokens": 67475301.0, + "step": 1928 + }, + { + "epoch": 0.35821727019498606, + "grad_norm": 1.5124787092208862, + "learning_rate": 1e-06, + "loss": 0.4295, + "mean_token_accuracy": 0.8566209077835083, + "num_tokens": 67511461.0, + "step": 1929 + }, + { + "epoch": 0.3584029712163417, + "grad_norm": 1.4371237754821777, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8637794256210327, + "num_tokens": 67549498.0, + "step": 1930 + }, + { + "epoch": 0.3585886722376973, + "grad_norm": 1.4338678121566772, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8656445741653442, + "num_tokens": 67584705.0, + "step": 1931 + }, + { + "epoch": 0.3587743732590529, + "grad_norm": 1.5096583366394043, + "learning_rate": 1e-06, + "loss": 0.413, + "mean_token_accuracy": 0.8628032207489014, + "num_tokens": 67617971.0, + "step": 1932 + }, + { + "epoch": 0.35896007428040855, + "grad_norm": 1.514837384223938, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8610292673110962, + "num_tokens": 67655553.0, + "step": 1933 + }, + { + "epoch": 0.35914577530176417, + "grad_norm": 1.545779824256897, + "learning_rate": 1e-06, + "loss": 0.4202, + "mean_token_accuracy": 0.8620808124542236, + "num_tokens": 67689796.0, + "step": 1934 + }, + { + "epoch": 0.3593314763231198, + "grad_norm": 1.4959959983825684, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.8585851192474365, + "num_tokens": 67725940.0, + "step": 1935 + }, + { + "epoch": 0.3595171773444754, + "grad_norm": 1.546059012413025, + "learning_rate": 1e-06, + "loss": 0.4338, + "mean_token_accuracy": 0.8541196584701538, + "num_tokens": 67760354.0, + "step": 1936 + }, + { + "epoch": 0.35970287836583104, + "grad_norm": 1.5533552169799805, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8523461222648621, + "num_tokens": 67795123.0, + "step": 1937 + }, + { + "epoch": 0.3598885793871866, + "grad_norm": 1.4577994346618652, + "learning_rate": 1e-06, + "loss": 0.4175, + "mean_token_accuracy": 0.8597401976585388, + "num_tokens": 67831612.0, + "step": 1938 + }, + { + "epoch": 0.3600742804085422, + "grad_norm": 1.4232017993927002, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8654211163520813, + "num_tokens": 67868191.0, + "step": 1939 + }, + { + "epoch": 0.36025998142989785, + "grad_norm": 1.5913947820663452, + "learning_rate": 1e-06, + "loss": 0.4264, + "mean_token_accuracy": 0.8578208684921265, + "num_tokens": 67903062.0, + "step": 1940 + }, + { + "epoch": 0.36044568245125347, + "grad_norm": 1.429963231086731, + "learning_rate": 1e-06, + "loss": 0.4234, + "mean_token_accuracy": 0.8586819171905518, + "num_tokens": 67941346.0, + "step": 1941 + }, + { + "epoch": 0.3606313834726091, + "grad_norm": 1.4903870820999146, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.8595297932624817, + "num_tokens": 67976188.0, + "step": 1942 + }, + { + "epoch": 0.3608170844939647, + "grad_norm": 1.54731285572052, + "learning_rate": 1e-06, + "loss": 0.4362, + "mean_token_accuracy": 0.855522871017456, + "num_tokens": 68006762.0, + "step": 1943 + }, + { + "epoch": 0.36100278551532033, + "grad_norm": 1.5549482107162476, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8741250038146973, + "num_tokens": 68038449.0, + "step": 1944 + }, + { + "epoch": 0.36118848653667596, + "grad_norm": 1.4862769842147827, + "learning_rate": 1e-06, + "loss": 0.4624, + "mean_token_accuracy": 0.8488435745239258, + "num_tokens": 68076671.0, + "step": 1945 + }, + { + "epoch": 0.3613741875580316, + "grad_norm": 1.4744654893875122, + "learning_rate": 1e-06, + "loss": 0.4307, + "mean_token_accuracy": 0.8586490154266357, + "num_tokens": 68113289.0, + "step": 1946 + }, + { + "epoch": 0.3615598885793872, + "grad_norm": 1.4720749855041504, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8594793677330017, + "num_tokens": 68151021.0, + "step": 1947 + }, + { + "epoch": 0.3617455896007428, + "grad_norm": 1.4887512922286987, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8656191825866699, + "num_tokens": 68189647.0, + "step": 1948 + }, + { + "epoch": 0.36193129062209844, + "grad_norm": 1.6292575597763062, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.869530200958252, + "num_tokens": 68215602.0, + "step": 1949 + }, + { + "epoch": 0.362116991643454, + "grad_norm": 1.5178142786026, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.8483288884162903, + "num_tokens": 68252007.0, + "step": 1950 + }, + { + "epoch": 0.36230269266480963, + "grad_norm": 1.4848501682281494, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8660359382629395, + "num_tokens": 68284015.0, + "step": 1951 + }, + { + "epoch": 0.36248839368616526, + "grad_norm": 1.5006519556045532, + "learning_rate": 1e-06, + "loss": 0.4248, + "mean_token_accuracy": 0.8593457937240601, + "num_tokens": 68319712.0, + "step": 1952 + }, + { + "epoch": 0.3626740947075209, + "grad_norm": 1.4581413269042969, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8694740533828735, + "num_tokens": 68359226.0, + "step": 1953 + }, + { + "epoch": 0.3628597957288765, + "grad_norm": 1.4640636444091797, + "learning_rate": 1e-06, + "loss": 0.4051, + "mean_token_accuracy": 0.8673040270805359, + "num_tokens": 68393524.0, + "step": 1954 + }, + { + "epoch": 0.3630454967502321, + "grad_norm": 1.4662249088287354, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8637771606445312, + "num_tokens": 68427222.0, + "step": 1955 + }, + { + "epoch": 0.36323119777158774, + "grad_norm": 1.4335095882415771, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8616026639938354, + "num_tokens": 68466881.0, + "step": 1956 + }, + { + "epoch": 0.36341689879294337, + "grad_norm": 1.4662280082702637, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8681622743606567, + "num_tokens": 68502508.0, + "step": 1957 + }, + { + "epoch": 0.363602599814299, + "grad_norm": 1.5016862154006958, + "learning_rate": 1e-06, + "loss": 0.4264, + "mean_token_accuracy": 0.8604012727737427, + "num_tokens": 68540136.0, + "step": 1958 + }, + { + "epoch": 0.3637883008356546, + "grad_norm": 1.6385469436645508, + "learning_rate": 1e-06, + "loss": 0.425, + "mean_token_accuracy": 0.8589919209480286, + "num_tokens": 68573926.0, + "step": 1959 + }, + { + "epoch": 0.36397400185701023, + "grad_norm": 1.4419500827789307, + "learning_rate": 1e-06, + "loss": 0.4231, + "mean_token_accuracy": 0.8613222241401672, + "num_tokens": 68611394.0, + "step": 1960 + }, + { + "epoch": 0.36415970287836585, + "grad_norm": 1.6190252304077148, + "learning_rate": 1e-06, + "loss": 0.4672, + "mean_token_accuracy": 0.8472920060157776, + "num_tokens": 68644886.0, + "step": 1961 + }, + { + "epoch": 0.3643454038997215, + "grad_norm": 1.3572970628738403, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8686342835426331, + "num_tokens": 68685010.0, + "step": 1962 + }, + { + "epoch": 0.36453110492107704, + "grad_norm": 1.4942030906677246, + "learning_rate": 1e-06, + "loss": 0.4341, + "mean_token_accuracy": 0.8576827049255371, + "num_tokens": 68720184.0, + "step": 1963 + }, + { + "epoch": 0.36471680594243266, + "grad_norm": 1.6160048246383667, + "learning_rate": 1e-06, + "loss": 0.4095, + "mean_token_accuracy": 0.8630235195159912, + "num_tokens": 68751604.0, + "step": 1964 + }, + { + "epoch": 0.3649025069637883, + "grad_norm": 1.3544654846191406, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8638806939125061, + "num_tokens": 68791875.0, + "step": 1965 + }, + { + "epoch": 0.3650882079851439, + "grad_norm": 1.4990652799606323, + "learning_rate": 1e-06, + "loss": 0.4281, + "mean_token_accuracy": 0.8568214178085327, + "num_tokens": 68825683.0, + "step": 1966 + }, + { + "epoch": 0.36527390900649953, + "grad_norm": 1.564738154411316, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.861197292804718, + "num_tokens": 68858245.0, + "step": 1967 + }, + { + "epoch": 0.36545961002785515, + "grad_norm": 1.4866799116134644, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.8644055128097534, + "num_tokens": 68893067.0, + "step": 1968 + }, + { + "epoch": 0.3656453110492108, + "grad_norm": 1.5330114364624023, + "learning_rate": 1e-06, + "loss": 0.4498, + "mean_token_accuracy": 0.8516196608543396, + "num_tokens": 68929581.0, + "step": 1969 + }, + { + "epoch": 0.3658310120705664, + "grad_norm": 1.525044322013855, + "learning_rate": 1e-06, + "loss": 0.4523, + "mean_token_accuracy": 0.8516432642936707, + "num_tokens": 68966505.0, + "step": 1970 + }, + { + "epoch": 0.366016713091922, + "grad_norm": 1.5566868782043457, + "learning_rate": 1e-06, + "loss": 0.4613, + "mean_token_accuracy": 0.8485319018363953, + "num_tokens": 69004804.0, + "step": 1971 + }, + { + "epoch": 0.36620241411327764, + "grad_norm": 1.6098878383636475, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.8672997951507568, + "num_tokens": 69035267.0, + "step": 1972 + }, + { + "epoch": 0.36638811513463326, + "grad_norm": 1.477525234222412, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8469004034996033, + "num_tokens": 69070275.0, + "step": 1973 + }, + { + "epoch": 0.3665738161559889, + "grad_norm": 1.4726932048797607, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8648360967636108, + "num_tokens": 69103987.0, + "step": 1974 + }, + { + "epoch": 0.36675951717734445, + "grad_norm": 1.4402647018432617, + "learning_rate": 1e-06, + "loss": 0.4293, + "mean_token_accuracy": 0.8621937036514282, + "num_tokens": 69140806.0, + "step": 1975 + }, + { + "epoch": 0.3669452181987001, + "grad_norm": 1.497756004333496, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8609951734542847, + "num_tokens": 69174434.0, + "step": 1976 + }, + { + "epoch": 0.3671309192200557, + "grad_norm": 1.4715901613235474, + "learning_rate": 1e-06, + "loss": 0.4557, + "mean_token_accuracy": 0.8527199029922485, + "num_tokens": 69214655.0, + "step": 1977 + }, + { + "epoch": 0.3673166202414113, + "grad_norm": 1.38639497756958, + "learning_rate": 1e-06, + "loss": 0.4533, + "mean_token_accuracy": 0.8513882160186768, + "num_tokens": 69257002.0, + "step": 1978 + }, + { + "epoch": 0.36750232126276694, + "grad_norm": 1.5225636959075928, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.8645420074462891, + "num_tokens": 69290243.0, + "step": 1979 + }, + { + "epoch": 0.36768802228412256, + "grad_norm": 1.4847826957702637, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8704600930213928, + "num_tokens": 69326017.0, + "step": 1980 + }, + { + "epoch": 0.3678737233054782, + "grad_norm": 1.5596950054168701, + "learning_rate": 1e-06, + "loss": 0.4348, + "mean_token_accuracy": 0.8623834848403931, + "num_tokens": 69357236.0, + "step": 1981 + }, + { + "epoch": 0.3680594243268338, + "grad_norm": 1.6192271709442139, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8578095436096191, + "num_tokens": 69389722.0, + "step": 1982 + }, + { + "epoch": 0.36824512534818943, + "grad_norm": 1.455767273902893, + "learning_rate": 1e-06, + "loss": 0.437, + "mean_token_accuracy": 0.8575344681739807, + "num_tokens": 69425333.0, + "step": 1983 + }, + { + "epoch": 0.36843082636954505, + "grad_norm": 1.6382830142974854, + "learning_rate": 1e-06, + "loss": 0.4549, + "mean_token_accuracy": 0.8487056493759155, + "num_tokens": 69456989.0, + "step": 1984 + }, + { + "epoch": 0.3686165273909007, + "grad_norm": 1.5510362386703491, + "learning_rate": 1e-06, + "loss": 0.442, + "mean_token_accuracy": 0.8564236760139465, + "num_tokens": 69492487.0, + "step": 1985 + }, + { + "epoch": 0.3688022284122563, + "grad_norm": 1.5682501792907715, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8755344152450562, + "num_tokens": 69522868.0, + "step": 1986 + }, + { + "epoch": 0.36898792943361186, + "grad_norm": 1.6648257970809937, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.861609935760498, + "num_tokens": 69554581.0, + "step": 1987 + }, + { + "epoch": 0.3691736304549675, + "grad_norm": 1.4384299516677856, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8629989624023438, + "num_tokens": 69593744.0, + "step": 1988 + }, + { + "epoch": 0.3693593314763231, + "grad_norm": 1.532914161682129, + "learning_rate": 1e-06, + "loss": 0.4787, + "mean_token_accuracy": 0.8519243001937866, + "num_tokens": 69630585.0, + "step": 1989 + }, + { + "epoch": 0.3695450324976787, + "grad_norm": 1.3609392642974854, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.866057276725769, + "num_tokens": 69673104.0, + "step": 1990 + }, + { + "epoch": 0.36973073351903435, + "grad_norm": 1.6157506704330444, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8677269220352173, + "num_tokens": 69706663.0, + "step": 1991 + }, + { + "epoch": 0.36991643454038997, + "grad_norm": 1.536501407623291, + "learning_rate": 1e-06, + "loss": 0.401, + "mean_token_accuracy": 0.8635283708572388, + "num_tokens": 69742078.0, + "step": 1992 + }, + { + "epoch": 0.3701021355617456, + "grad_norm": 1.4462776184082031, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.872424304485321, + "num_tokens": 69776501.0, + "step": 1993 + }, + { + "epoch": 0.3702878365831012, + "grad_norm": 1.437761664390564, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8614247441291809, + "num_tokens": 69813879.0, + "step": 1994 + }, + { + "epoch": 0.37047353760445684, + "grad_norm": 1.526183009147644, + "learning_rate": 1e-06, + "loss": 0.4228, + "mean_token_accuracy": 0.8603379726409912, + "num_tokens": 69850914.0, + "step": 1995 + }, + { + "epoch": 0.37065923862581246, + "grad_norm": 1.5609785318374634, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8636568784713745, + "num_tokens": 69884475.0, + "step": 1996 + }, + { + "epoch": 0.3708449396471681, + "grad_norm": 1.4851924180984497, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8712906837463379, + "num_tokens": 69920446.0, + "step": 1997 + }, + { + "epoch": 0.3710306406685237, + "grad_norm": 1.4615347385406494, + "learning_rate": 1e-06, + "loss": 0.4134, + "mean_token_accuracy": 0.8599172830581665, + "num_tokens": 69957327.0, + "step": 1998 + }, + { + "epoch": 0.37121634168987927, + "grad_norm": 1.5382705926895142, + "learning_rate": 1e-06, + "loss": 0.4437, + "mean_token_accuracy": 0.8515366315841675, + "num_tokens": 69990275.0, + "step": 1999 + }, + { + "epoch": 0.3714020427112349, + "grad_norm": 1.4024258852005005, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.8684459924697876, + "num_tokens": 70027826.0, + "step": 2000 + }, + { + "epoch": 0.3715877437325905, + "grad_norm": 1.77097487449646, + "learning_rate": 1e-06, + "loss": 0.4538, + "mean_token_accuracy": 0.8573542833328247, + "num_tokens": 70054263.0, + "step": 2001 + }, + { + "epoch": 0.37177344475394614, + "grad_norm": 1.4021966457366943, + "learning_rate": 1e-06, + "loss": 0.4486, + "mean_token_accuracy": 0.8526350855827332, + "num_tokens": 70097071.0, + "step": 2002 + }, + { + "epoch": 0.37195914577530176, + "grad_norm": 1.4879956245422363, + "learning_rate": 1e-06, + "loss": 0.4207, + "mean_token_accuracy": 0.8589869737625122, + "num_tokens": 70136611.0, + "step": 2003 + }, + { + "epoch": 0.3721448467966574, + "grad_norm": 1.457133173942566, + "learning_rate": 1e-06, + "loss": 0.4638, + "mean_token_accuracy": 0.8489217758178711, + "num_tokens": 70173723.0, + "step": 2004 + }, + { + "epoch": 0.372330547818013, + "grad_norm": 1.5071114301681519, + "learning_rate": 1e-06, + "loss": 0.4154, + "mean_token_accuracy": 0.8634328842163086, + "num_tokens": 70203629.0, + "step": 2005 + }, + { + "epoch": 0.3725162488393686, + "grad_norm": 1.8292945623397827, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8644724488258362, + "num_tokens": 70227403.0, + "step": 2006 + }, + { + "epoch": 0.37270194986072425, + "grad_norm": 1.5687564611434937, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8518202304840088, + "num_tokens": 70261910.0, + "step": 2007 + }, + { + "epoch": 0.37288765088207987, + "grad_norm": 1.5745301246643066, + "learning_rate": 1e-06, + "loss": 0.4475, + "mean_token_accuracy": 0.853303074836731, + "num_tokens": 70299764.0, + "step": 2008 + }, + { + "epoch": 0.3730733519034355, + "grad_norm": 1.4526134729385376, + "learning_rate": 1e-06, + "loss": 0.4433, + "mean_token_accuracy": 0.8526024222373962, + "num_tokens": 70339569.0, + "step": 2009 + }, + { + "epoch": 0.3732590529247911, + "grad_norm": 1.8296455144882202, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8490771055221558, + "num_tokens": 70367767.0, + "step": 2010 + }, + { + "epoch": 0.3734447539461467, + "grad_norm": 1.701008677482605, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8704737424850464, + "num_tokens": 70399211.0, + "step": 2011 + }, + { + "epoch": 0.3736304549675023, + "grad_norm": 1.3406370878219604, + "learning_rate": 1e-06, + "loss": 0.4434, + "mean_token_accuracy": 0.8515900373458862, + "num_tokens": 70444089.0, + "step": 2012 + }, + { + "epoch": 0.3738161559888579, + "grad_norm": 1.5431911945343018, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.8632215261459351, + "num_tokens": 70479613.0, + "step": 2013 + }, + { + "epoch": 0.37400185701021355, + "grad_norm": 1.597588300704956, + "learning_rate": 1e-06, + "loss": 0.424, + "mean_token_accuracy": 0.8621445894241333, + "num_tokens": 70512912.0, + "step": 2014 + }, + { + "epoch": 0.37418755803156917, + "grad_norm": 1.5474992990493774, + "learning_rate": 1e-06, + "loss": 0.4152, + "mean_token_accuracy": 0.8620667457580566, + "num_tokens": 70545046.0, + "step": 2015 + }, + { + "epoch": 0.3743732590529248, + "grad_norm": 1.4569411277770996, + "learning_rate": 1e-06, + "loss": 0.4534, + "mean_token_accuracy": 0.8562732338905334, + "num_tokens": 70583964.0, + "step": 2016 + }, + { + "epoch": 0.3745589600742804, + "grad_norm": 1.5472736358642578, + "learning_rate": 1e-06, + "loss": 0.5149, + "mean_token_accuracy": 0.8333771228790283, + "num_tokens": 70621952.0, + "step": 2017 + }, + { + "epoch": 0.37474466109563603, + "grad_norm": 1.485944151878357, + "learning_rate": 1e-06, + "loss": 0.4545, + "mean_token_accuracy": 0.8480466604232788, + "num_tokens": 70657561.0, + "step": 2018 + }, + { + "epoch": 0.37493036211699166, + "grad_norm": 1.6013867855072021, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.8641496896743774, + "num_tokens": 70688137.0, + "step": 2019 + }, + { + "epoch": 0.3751160631383473, + "grad_norm": 1.641396164894104, + "learning_rate": 1e-06, + "loss": 0.4665, + "mean_token_accuracy": 0.8449675440788269, + "num_tokens": 70720943.0, + "step": 2020 + }, + { + "epoch": 0.3753017641597029, + "grad_norm": 1.4754064083099365, + "learning_rate": 1e-06, + "loss": 0.4623, + "mean_token_accuracy": 0.8464043140411377, + "num_tokens": 70756904.0, + "step": 2021 + }, + { + "epoch": 0.3754874651810585, + "grad_norm": 1.5450352430343628, + "learning_rate": 1e-06, + "loss": 0.433, + "mean_token_accuracy": 0.8598864078521729, + "num_tokens": 70794693.0, + "step": 2022 + }, + { + "epoch": 0.3756731662024141, + "grad_norm": 1.50320565700531, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.8684187531471252, + "num_tokens": 70828979.0, + "step": 2023 + }, + { + "epoch": 0.3758588672237697, + "grad_norm": 1.5348087549209595, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.8639534115791321, + "num_tokens": 70861479.0, + "step": 2024 + }, + { + "epoch": 0.37604456824512533, + "grad_norm": 1.7791788578033447, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8580945730209351, + "num_tokens": 70887678.0, + "step": 2025 + }, + { + "epoch": 0.37623026926648095, + "grad_norm": 1.4633861780166626, + "learning_rate": 1e-06, + "loss": 0.4435, + "mean_token_accuracy": 0.8551385998725891, + "num_tokens": 70924087.0, + "step": 2026 + }, + { + "epoch": 0.3764159702878366, + "grad_norm": 1.4974408149719238, + "learning_rate": 1e-06, + "loss": 0.4228, + "mean_token_accuracy": 0.8613078594207764, + "num_tokens": 70957887.0, + "step": 2027 + }, + { + "epoch": 0.3766016713091922, + "grad_norm": 1.4268214702606201, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8599686026573181, + "num_tokens": 70994188.0, + "step": 2028 + }, + { + "epoch": 0.3767873723305478, + "grad_norm": 1.5641268491744995, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.8633458614349365, + "num_tokens": 71031066.0, + "step": 2029 + }, + { + "epoch": 0.37697307335190344, + "grad_norm": 1.4331095218658447, + "learning_rate": 1e-06, + "loss": 0.439, + "mean_token_accuracy": 0.85038161277771, + "num_tokens": 71068792.0, + "step": 2030 + }, + { + "epoch": 0.37715877437325906, + "grad_norm": 1.4966340065002441, + "learning_rate": 1e-06, + "loss": 0.4218, + "mean_token_accuracy": 0.8586074113845825, + "num_tokens": 71106525.0, + "step": 2031 + }, + { + "epoch": 0.3773444753946147, + "grad_norm": 1.4429064989089966, + "learning_rate": 1e-06, + "loss": 0.4638, + "mean_token_accuracy": 0.8471068739891052, + "num_tokens": 71147475.0, + "step": 2032 + }, + { + "epoch": 0.3775301764159703, + "grad_norm": 1.4661568403244019, + "learning_rate": 1e-06, + "loss": 0.4448, + "mean_token_accuracy": 0.850278377532959, + "num_tokens": 71182949.0, + "step": 2033 + }, + { + "epoch": 0.37771587743732593, + "grad_norm": 1.594050645828247, + "learning_rate": 1e-06, + "loss": 0.4265, + "mean_token_accuracy": 0.8591982126235962, + "num_tokens": 71217382.0, + "step": 2034 + }, + { + "epoch": 0.3779015784586815, + "grad_norm": 1.3682624101638794, + "learning_rate": 1e-06, + "loss": 0.4582, + "mean_token_accuracy": 0.8491363525390625, + "num_tokens": 71260625.0, + "step": 2035 + }, + { + "epoch": 0.3780872794800371, + "grad_norm": 1.4664009809494019, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8628356456756592, + "num_tokens": 71295520.0, + "step": 2036 + }, + { + "epoch": 0.37827298050139274, + "grad_norm": 1.4188778400421143, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8554346561431885, + "num_tokens": 71334185.0, + "step": 2037 + }, + { + "epoch": 0.37845868152274836, + "grad_norm": 1.643048644065857, + "learning_rate": 1e-06, + "loss": 0.4376, + "mean_token_accuracy": 0.8533178567886353, + "num_tokens": 71366059.0, + "step": 2038 + }, + { + "epoch": 0.378644382544104, + "grad_norm": 1.5103436708450317, + "learning_rate": 1e-06, + "loss": 0.4126, + "mean_token_accuracy": 0.8617043495178223, + "num_tokens": 71400383.0, + "step": 2039 + }, + { + "epoch": 0.3788300835654596, + "grad_norm": 1.4945847988128662, + "learning_rate": 1e-06, + "loss": 0.4438, + "mean_token_accuracy": 0.8518601655960083, + "num_tokens": 71435463.0, + "step": 2040 + }, + { + "epoch": 0.37901578458681523, + "grad_norm": 1.5910401344299316, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8700852394104004, + "num_tokens": 71469200.0, + "step": 2041 + }, + { + "epoch": 0.37920148560817085, + "grad_norm": 1.5350276231765747, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8734077215194702, + "num_tokens": 71502413.0, + "step": 2042 + }, + { + "epoch": 0.3793871866295265, + "grad_norm": 1.405204176902771, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8690690398216248, + "num_tokens": 71540034.0, + "step": 2043 + }, + { + "epoch": 0.3795728876508821, + "grad_norm": 1.514243245124817, + "learning_rate": 1e-06, + "loss": 0.4471, + "mean_token_accuracy": 0.8555134534835815, + "num_tokens": 71577943.0, + "step": 2044 + }, + { + "epoch": 0.3797585886722377, + "grad_norm": 1.59137761592865, + "learning_rate": 1e-06, + "loss": 0.4589, + "mean_token_accuracy": 0.8505561947822571, + "num_tokens": 71612564.0, + "step": 2045 + }, + { + "epoch": 0.37994428969359334, + "grad_norm": 1.5095181465148926, + "learning_rate": 1e-06, + "loss": 0.4264, + "mean_token_accuracy": 0.8587474822998047, + "num_tokens": 71649410.0, + "step": 2046 + }, + { + "epoch": 0.3801299907149489, + "grad_norm": 1.577156662940979, + "learning_rate": 1e-06, + "loss": 0.4247, + "mean_token_accuracy": 0.8574645519256592, + "num_tokens": 71681845.0, + "step": 2047 + }, + { + "epoch": 0.38031569173630453, + "grad_norm": 1.4923161268234253, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.8641303777694702, + "num_tokens": 71717025.0, + "step": 2048 + }, + { + "epoch": 0.38050139275766015, + "grad_norm": 1.508707046508789, + "learning_rate": 1e-06, + "loss": 0.4748, + "mean_token_accuracy": 0.8455837965011597, + "num_tokens": 71757054.0, + "step": 2049 + }, + { + "epoch": 0.3806870937790158, + "grad_norm": 1.3886675834655762, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8610944747924805, + "num_tokens": 71798205.0, + "step": 2050 + }, + { + "epoch": 0.3808727948003714, + "grad_norm": 1.710563063621521, + "learning_rate": 1e-06, + "loss": 0.472, + "mean_token_accuracy": 0.8465185761451721, + "num_tokens": 71831567.0, + "step": 2051 + }, + { + "epoch": 0.381058495821727, + "grad_norm": 1.503162145614624, + "learning_rate": 1e-06, + "loss": 0.4629, + "mean_token_accuracy": 0.8533999919891357, + "num_tokens": 71869289.0, + "step": 2052 + }, + { + "epoch": 0.38124419684308264, + "grad_norm": 1.5211987495422363, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.8688756227493286, + "num_tokens": 71903293.0, + "step": 2053 + }, + { + "epoch": 0.38142989786443826, + "grad_norm": 1.4757524728775024, + "learning_rate": 1e-06, + "loss": 0.4086, + "mean_token_accuracy": 0.8615344166755676, + "num_tokens": 71942078.0, + "step": 2054 + }, + { + "epoch": 0.3816155988857939, + "grad_norm": 1.5026471614837646, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8757519125938416, + "num_tokens": 71971623.0, + "step": 2055 + }, + { + "epoch": 0.3818012999071495, + "grad_norm": 1.4951789379119873, + "learning_rate": 1e-06, + "loss": 0.461, + "mean_token_accuracy": 0.8458845615386963, + "num_tokens": 72005008.0, + "step": 2056 + }, + { + "epoch": 0.3819870009285051, + "grad_norm": 1.4374549388885498, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.8614695072174072, + "num_tokens": 72039745.0, + "step": 2057 + }, + { + "epoch": 0.38217270194986075, + "grad_norm": 1.4982874393463135, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8609805107116699, + "num_tokens": 72073639.0, + "step": 2058 + }, + { + "epoch": 0.3823584029712163, + "grad_norm": 1.4738795757293701, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8554753065109253, + "num_tokens": 72110225.0, + "step": 2059 + }, + { + "epoch": 0.38254410399257194, + "grad_norm": 1.6125316619873047, + "learning_rate": 1e-06, + "loss": 0.4288, + "mean_token_accuracy": 0.8554853200912476, + "num_tokens": 72139715.0, + "step": 2060 + }, + { + "epoch": 0.38272980501392756, + "grad_norm": 1.5784270763397217, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8784635066986084, + "num_tokens": 72168566.0, + "step": 2061 + }, + { + "epoch": 0.3829155060352832, + "grad_norm": 1.40202796459198, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8661320209503174, + "num_tokens": 72208252.0, + "step": 2062 + }, + { + "epoch": 0.3831012070566388, + "grad_norm": 1.4005008935928345, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8727002739906311, + "num_tokens": 72248460.0, + "step": 2063 + }, + { + "epoch": 0.3832869080779944, + "grad_norm": 1.4698028564453125, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.867644190788269, + "num_tokens": 72283296.0, + "step": 2064 + }, + { + "epoch": 0.38347260909935005, + "grad_norm": 1.5080724954605103, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8702948093414307, + "num_tokens": 72319812.0, + "step": 2065 + }, + { + "epoch": 0.38365831012070567, + "grad_norm": 1.505305290222168, + "learning_rate": 1e-06, + "loss": 0.4411, + "mean_token_accuracy": 0.8561013340950012, + "num_tokens": 72359176.0, + "step": 2066 + }, + { + "epoch": 0.3838440111420613, + "grad_norm": 1.4759796857833862, + "learning_rate": 1e-06, + "loss": 0.4319, + "mean_token_accuracy": 0.8566848635673523, + "num_tokens": 72395859.0, + "step": 2067 + }, + { + "epoch": 0.3840297121634169, + "grad_norm": 1.5711156129837036, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8705040216445923, + "num_tokens": 72431043.0, + "step": 2068 + }, + { + "epoch": 0.38421541318477254, + "grad_norm": 1.445177674293518, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8665282130241394, + "num_tokens": 72467498.0, + "step": 2069 + }, + { + "epoch": 0.38440111420612816, + "grad_norm": 1.4286565780639648, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8624555468559265, + "num_tokens": 72503857.0, + "step": 2070 + }, + { + "epoch": 0.3845868152274837, + "grad_norm": 1.5148416757583618, + "learning_rate": 1e-06, + "loss": 0.4588, + "mean_token_accuracy": 0.8509774208068848, + "num_tokens": 72539748.0, + "step": 2071 + }, + { + "epoch": 0.38477251624883935, + "grad_norm": 1.5726277828216553, + "learning_rate": 1e-06, + "loss": 0.4415, + "mean_token_accuracy": 0.8584720492362976, + "num_tokens": 72572514.0, + "step": 2072 + }, + { + "epoch": 0.38495821727019497, + "grad_norm": 1.5606677532196045, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.8565226197242737, + "num_tokens": 72605114.0, + "step": 2073 + }, + { + "epoch": 0.3851439182915506, + "grad_norm": 1.3937087059020996, + "learning_rate": 1e-06, + "loss": 0.4255, + "mean_token_accuracy": 0.8589361906051636, + "num_tokens": 72644952.0, + "step": 2074 + }, + { + "epoch": 0.3853296193129062, + "grad_norm": 1.4402035474777222, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8694500923156738, + "num_tokens": 72680923.0, + "step": 2075 + }, + { + "epoch": 0.38551532033426184, + "grad_norm": 1.5823780298233032, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8665342926979065, + "num_tokens": 72709423.0, + "step": 2076 + }, + { + "epoch": 0.38570102135561746, + "grad_norm": 1.4581035375595093, + "learning_rate": 1e-06, + "loss": 0.4252, + "mean_token_accuracy": 0.861960768699646, + "num_tokens": 72744677.0, + "step": 2077 + }, + { + "epoch": 0.3858867223769731, + "grad_norm": 1.352131962776184, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8664793968200684, + "num_tokens": 72788643.0, + "step": 2078 + }, + { + "epoch": 0.3860724233983287, + "grad_norm": 1.493558645248413, + "learning_rate": 1e-06, + "loss": 0.4746, + "mean_token_accuracy": 0.8437997698783875, + "num_tokens": 72826050.0, + "step": 2079 + }, + { + "epoch": 0.3862581244196843, + "grad_norm": 1.4778950214385986, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.858687162399292, + "num_tokens": 72860012.0, + "step": 2080 + }, + { + "epoch": 0.38644382544103995, + "grad_norm": 1.5232971906661987, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8775749206542969, + "num_tokens": 72893303.0, + "step": 2081 + }, + { + "epoch": 0.38662952646239557, + "grad_norm": 1.4593466520309448, + "learning_rate": 1e-06, + "loss": 0.41, + "mean_token_accuracy": 0.8629660606384277, + "num_tokens": 72929888.0, + "step": 2082 + }, + { + "epoch": 0.38681522748375113, + "grad_norm": 1.4649041891098022, + "learning_rate": 1e-06, + "loss": 0.4079, + "mean_token_accuracy": 0.8628748059272766, + "num_tokens": 72966382.0, + "step": 2083 + }, + { + "epoch": 0.38700092850510676, + "grad_norm": 1.5683996677398682, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.866640567779541, + "num_tokens": 72998788.0, + "step": 2084 + }, + { + "epoch": 0.3871866295264624, + "grad_norm": 1.4538766145706177, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8638035655021667, + "num_tokens": 73034034.0, + "step": 2085 + }, + { + "epoch": 0.387372330547818, + "grad_norm": 1.4113373756408691, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8757737874984741, + "num_tokens": 73066403.0, + "step": 2086 + }, + { + "epoch": 0.3875580315691736, + "grad_norm": 1.4703459739685059, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8631999492645264, + "num_tokens": 73102771.0, + "step": 2087 + }, + { + "epoch": 0.38774373259052924, + "grad_norm": 1.5008260011672974, + "learning_rate": 1e-06, + "loss": 0.4415, + "mean_token_accuracy": 0.857938289642334, + "num_tokens": 73142061.0, + "step": 2088 + }, + { + "epoch": 0.38792943361188487, + "grad_norm": 1.5039530992507935, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8690224885940552, + "num_tokens": 73173001.0, + "step": 2089 + }, + { + "epoch": 0.3881151346332405, + "grad_norm": 1.565228819847107, + "learning_rate": 1e-06, + "loss": 0.4599, + "mean_token_accuracy": 0.8425330519676208, + "num_tokens": 73208903.0, + "step": 2090 + }, + { + "epoch": 0.3883008356545961, + "grad_norm": 1.5131417512893677, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8705989122390747, + "num_tokens": 73243743.0, + "step": 2091 + }, + { + "epoch": 0.38848653667595173, + "grad_norm": 1.4566746950149536, + "learning_rate": 1e-06, + "loss": 0.4238, + "mean_token_accuracy": 0.8601860404014587, + "num_tokens": 73279225.0, + "step": 2092 + }, + { + "epoch": 0.38867223769730735, + "grad_norm": 1.4429998397827148, + "learning_rate": 1e-06, + "loss": 0.4399, + "mean_token_accuracy": 0.8552642464637756, + "num_tokens": 73319042.0, + "step": 2093 + }, + { + "epoch": 0.388857938718663, + "grad_norm": 1.6835880279541016, + "learning_rate": 1e-06, + "loss": 0.431, + "mean_token_accuracy": 0.857147216796875, + "num_tokens": 73347842.0, + "step": 2094 + }, + { + "epoch": 0.38904363974001854, + "grad_norm": 1.500995397567749, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8687282204627991, + "num_tokens": 73379955.0, + "step": 2095 + }, + { + "epoch": 0.38922934076137417, + "grad_norm": 1.590558409690857, + "learning_rate": 1e-06, + "loss": 0.4501, + "mean_token_accuracy": 0.8496624231338501, + "num_tokens": 73411808.0, + "step": 2096 + }, + { + "epoch": 0.3894150417827298, + "grad_norm": 1.4003963470458984, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.867430567741394, + "num_tokens": 73452934.0, + "step": 2097 + }, + { + "epoch": 0.3896007428040854, + "grad_norm": 1.5770554542541504, + "learning_rate": 1e-06, + "loss": 0.4538, + "mean_token_accuracy": 0.8513075113296509, + "num_tokens": 73484843.0, + "step": 2098 + }, + { + "epoch": 0.38978644382544103, + "grad_norm": 1.530663013458252, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8646838665008545, + "num_tokens": 73518554.0, + "step": 2099 + }, + { + "epoch": 0.38997214484679665, + "grad_norm": 1.6776164770126343, + "learning_rate": 1e-06, + "loss": 0.4381, + "mean_token_accuracy": 0.8592982292175293, + "num_tokens": 73548397.0, + "step": 2100 + }, + { + "epoch": 0.3901578458681523, + "grad_norm": 1.5429108142852783, + "learning_rate": 1e-06, + "loss": 0.4458, + "mean_token_accuracy": 0.8502755165100098, + "num_tokens": 73585402.0, + "step": 2101 + }, + { + "epoch": 0.3903435468895079, + "grad_norm": 1.4186594486236572, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8510699272155762, + "num_tokens": 73624316.0, + "step": 2102 + }, + { + "epoch": 0.3905292479108635, + "grad_norm": 1.4765156507492065, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.8712839484214783, + "num_tokens": 73658960.0, + "step": 2103 + }, + { + "epoch": 0.39071494893221914, + "grad_norm": 1.3371925354003906, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8751012086868286, + "num_tokens": 73700016.0, + "step": 2104 + }, + { + "epoch": 0.39090064995357476, + "grad_norm": 1.4209413528442383, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8769661784172058, + "num_tokens": 73737250.0, + "step": 2105 + }, + { + "epoch": 0.3910863509749304, + "grad_norm": 1.5050941705703735, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8735489845275879, + "num_tokens": 73770990.0, + "step": 2106 + }, + { + "epoch": 0.39127205199628595, + "grad_norm": 1.5397793054580688, + "learning_rate": 1e-06, + "loss": 0.4154, + "mean_token_accuracy": 0.8597525954246521, + "num_tokens": 73800375.0, + "step": 2107 + }, + { + "epoch": 0.3914577530176416, + "grad_norm": 1.3281289339065552, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8623849749565125, + "num_tokens": 73844850.0, + "step": 2108 + }, + { + "epoch": 0.3916434540389972, + "grad_norm": 1.5012186765670776, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8645447492599487, + "num_tokens": 73883536.0, + "step": 2109 + }, + { + "epoch": 0.3918291550603528, + "grad_norm": 1.5999186038970947, + "learning_rate": 1e-06, + "loss": 0.4212, + "mean_token_accuracy": 0.8613094091415405, + "num_tokens": 73914136.0, + "step": 2110 + }, + { + "epoch": 0.39201485608170844, + "grad_norm": 1.4869345426559448, + "learning_rate": 1e-06, + "loss": 0.4181, + "mean_token_accuracy": 0.8642357587814331, + "num_tokens": 73950876.0, + "step": 2111 + }, + { + "epoch": 0.39220055710306406, + "grad_norm": 1.4815990924835205, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8569751977920532, + "num_tokens": 73987991.0, + "step": 2112 + }, + { + "epoch": 0.3923862581244197, + "grad_norm": 1.7283730506896973, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.8601893186569214, + "num_tokens": 74024026.0, + "step": 2113 + }, + { + "epoch": 0.3925719591457753, + "grad_norm": 1.6212704181671143, + "learning_rate": 1e-06, + "loss": 0.4378, + "mean_token_accuracy": 0.8510680198669434, + "num_tokens": 74053793.0, + "step": 2114 + }, + { + "epoch": 0.39275766016713093, + "grad_norm": 1.5469903945922852, + "learning_rate": 1e-06, + "loss": 0.4199, + "mean_token_accuracy": 0.8561362624168396, + "num_tokens": 74091025.0, + "step": 2115 + }, + { + "epoch": 0.39294336118848655, + "grad_norm": 1.4125502109527588, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.870840847492218, + "num_tokens": 74126269.0, + "step": 2116 + }, + { + "epoch": 0.3931290622098422, + "grad_norm": 1.589316725730896, + "learning_rate": 1e-06, + "loss": 0.4398, + "mean_token_accuracy": 0.8544966578483582, + "num_tokens": 74158160.0, + "step": 2117 + }, + { + "epoch": 0.3933147632311978, + "grad_norm": 1.4618576765060425, + "learning_rate": 1e-06, + "loss": 0.4226, + "mean_token_accuracy": 0.8629814386367798, + "num_tokens": 74197465.0, + "step": 2118 + }, + { + "epoch": 0.39350046425255336, + "grad_norm": 1.457069993019104, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.8592070937156677, + "num_tokens": 74231867.0, + "step": 2119 + }, + { + "epoch": 0.393686165273909, + "grad_norm": 1.5999577045440674, + "learning_rate": 1e-06, + "loss": 0.4512, + "mean_token_accuracy": 0.8518919348716736, + "num_tokens": 74264216.0, + "step": 2120 + }, + { + "epoch": 0.3938718662952646, + "grad_norm": 1.372070550918579, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8753238320350647, + "num_tokens": 74304831.0, + "step": 2121 + }, + { + "epoch": 0.3940575673166202, + "grad_norm": 1.5649163722991943, + "learning_rate": 1e-06, + "loss": 0.4298, + "mean_token_accuracy": 0.856542706489563, + "num_tokens": 74337711.0, + "step": 2122 + }, + { + "epoch": 0.39424326833797585, + "grad_norm": 1.658698320388794, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8739713430404663, + "num_tokens": 74363149.0, + "step": 2123 + }, + { + "epoch": 0.39442896935933147, + "grad_norm": 1.6808619499206543, + "learning_rate": 1e-06, + "loss": 0.4437, + "mean_token_accuracy": 0.8499336242675781, + "num_tokens": 74389374.0, + "step": 2124 + }, + { + "epoch": 0.3946146703806871, + "grad_norm": 1.345047950744629, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8727480173110962, + "num_tokens": 74426190.0, + "step": 2125 + }, + { + "epoch": 0.3948003714020427, + "grad_norm": 1.456748127937317, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8722532391548157, + "num_tokens": 74458864.0, + "step": 2126 + }, + { + "epoch": 0.39498607242339834, + "grad_norm": 1.5286129713058472, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8692506551742554, + "num_tokens": 74491418.0, + "step": 2127 + }, + { + "epoch": 0.39517177344475396, + "grad_norm": 1.6054872274398804, + "learning_rate": 1e-06, + "loss": 0.4137, + "mean_token_accuracy": 0.8612788915634155, + "num_tokens": 74521938.0, + "step": 2128 + }, + { + "epoch": 0.3953574744661096, + "grad_norm": 1.5023940801620483, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8605448603630066, + "num_tokens": 74556770.0, + "step": 2129 + }, + { + "epoch": 0.3955431754874652, + "grad_norm": 1.6755062341690063, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8706235885620117, + "num_tokens": 74582343.0, + "step": 2130 + }, + { + "epoch": 0.3957288765088208, + "grad_norm": 1.5137131214141846, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8611986637115479, + "num_tokens": 74617535.0, + "step": 2131 + }, + { + "epoch": 0.3959145775301764, + "grad_norm": 1.3839902877807617, + "learning_rate": 1e-06, + "loss": 0.444, + "mean_token_accuracy": 0.8531948328018188, + "num_tokens": 74659474.0, + "step": 2132 + }, + { + "epoch": 0.396100278551532, + "grad_norm": 1.5840309858322144, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8615765571594238, + "num_tokens": 74694427.0, + "step": 2133 + }, + { + "epoch": 0.39628597957288764, + "grad_norm": 1.5973970890045166, + "learning_rate": 1e-06, + "loss": 0.4201, + "mean_token_accuracy": 0.8601748943328857, + "num_tokens": 74729295.0, + "step": 2134 + }, + { + "epoch": 0.39647168059424326, + "grad_norm": 1.4746325016021729, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8655370473861694, + "num_tokens": 74764730.0, + "step": 2135 + }, + { + "epoch": 0.3966573816155989, + "grad_norm": 1.4085839986801147, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.861125111579895, + "num_tokens": 74803271.0, + "step": 2136 + }, + { + "epoch": 0.3968430826369545, + "grad_norm": 1.5624080896377563, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8648526668548584, + "num_tokens": 74844884.0, + "step": 2137 + }, + { + "epoch": 0.3970287836583101, + "grad_norm": 1.549611210823059, + "learning_rate": 1e-06, + "loss": 0.4568, + "mean_token_accuracy": 0.8474988341331482, + "num_tokens": 74877881.0, + "step": 2138 + }, + { + "epoch": 0.39721448467966575, + "grad_norm": 1.5076360702514648, + "learning_rate": 1e-06, + "loss": 0.4403, + "mean_token_accuracy": 0.853126585483551, + "num_tokens": 74913497.0, + "step": 2139 + }, + { + "epoch": 0.39740018570102137, + "grad_norm": 1.4136319160461426, + "learning_rate": 1e-06, + "loss": 0.4351, + "mean_token_accuracy": 0.8553962111473083, + "num_tokens": 74953134.0, + "step": 2140 + }, + { + "epoch": 0.397585886722377, + "grad_norm": 1.5777826309204102, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8766607046127319, + "num_tokens": 74986608.0, + "step": 2141 + }, + { + "epoch": 0.3977715877437326, + "grad_norm": 1.4027040004730225, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8682495355606079, + "num_tokens": 75025580.0, + "step": 2142 + }, + { + "epoch": 0.39795728876508824, + "grad_norm": 1.3365294933319092, + "learning_rate": 1e-06, + "loss": 0.4452, + "mean_token_accuracy": 0.8505905866622925, + "num_tokens": 75072695.0, + "step": 2143 + }, + { + "epoch": 0.3981429897864438, + "grad_norm": 1.3674890995025635, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8739674091339111, + "num_tokens": 75109264.0, + "step": 2144 + }, + { + "epoch": 0.3983286908077994, + "grad_norm": 1.551684856414795, + "learning_rate": 1e-06, + "loss": 0.4383, + "mean_token_accuracy": 0.8573981523513794, + "num_tokens": 75142819.0, + "step": 2145 + }, + { + "epoch": 0.39851439182915505, + "grad_norm": 1.4830467700958252, + "learning_rate": 1e-06, + "loss": 0.4541, + "mean_token_accuracy": 0.8510196208953857, + "num_tokens": 75180230.0, + "step": 2146 + }, + { + "epoch": 0.39870009285051067, + "grad_norm": 1.665677547454834, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8594304323196411, + "num_tokens": 75212217.0, + "step": 2147 + }, + { + "epoch": 0.3988857938718663, + "grad_norm": 1.5506478548049927, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8758153915405273, + "num_tokens": 75245181.0, + "step": 2148 + }, + { + "epoch": 0.3990714948932219, + "grad_norm": 1.6420315504074097, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8777062892913818, + "num_tokens": 75274243.0, + "step": 2149 + }, + { + "epoch": 0.39925719591457753, + "grad_norm": 1.6124924421310425, + "learning_rate": 1e-06, + "loss": 0.4202, + "mean_token_accuracy": 0.8611014485359192, + "num_tokens": 75305460.0, + "step": 2150 + }, + { + "epoch": 0.39944289693593316, + "grad_norm": 1.4375920295715332, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8705695867538452, + "num_tokens": 75340789.0, + "step": 2151 + }, + { + "epoch": 0.3996285979572888, + "grad_norm": 1.385588526725769, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8604432344436646, + "num_tokens": 75380479.0, + "step": 2152 + }, + { + "epoch": 0.3998142989786444, + "grad_norm": 1.549307942390442, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.8657839298248291, + "num_tokens": 75411555.0, + "step": 2153 + }, + { + "epoch": 0.4, + "grad_norm": 1.6263316869735718, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.8605843782424927, + "num_tokens": 75440048.0, + "step": 2154 + }, + { + "epoch": 0.40018570102135564, + "grad_norm": 1.3388173580169678, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8702026009559631, + "num_tokens": 75481652.0, + "step": 2155 + }, + { + "epoch": 0.4003714020427112, + "grad_norm": 1.5519787073135376, + "learning_rate": 1e-06, + "loss": 0.431, + "mean_token_accuracy": 0.8561418056488037, + "num_tokens": 75517384.0, + "step": 2156 + }, + { + "epoch": 0.40055710306406683, + "grad_norm": 1.5882935523986816, + "learning_rate": 1e-06, + "loss": 0.4518, + "mean_token_accuracy": 0.8511797189712524, + "num_tokens": 75553033.0, + "step": 2157 + }, + { + "epoch": 0.40074280408542245, + "grad_norm": 1.695165991783142, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8613157272338867, + "num_tokens": 75581608.0, + "step": 2158 + }, + { + "epoch": 0.4009285051067781, + "grad_norm": 1.5523276329040527, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8697055578231812, + "num_tokens": 75614157.0, + "step": 2159 + }, + { + "epoch": 0.4011142061281337, + "grad_norm": 1.4932801723480225, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.869652509689331, + "num_tokens": 75649939.0, + "step": 2160 + }, + { + "epoch": 0.4012999071494893, + "grad_norm": 1.424483299255371, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.864153265953064, + "num_tokens": 75690531.0, + "step": 2161 + }, + { + "epoch": 0.40148560817084494, + "grad_norm": 1.6199901103973389, + "learning_rate": 1e-06, + "loss": 0.4631, + "mean_token_accuracy": 0.8440892696380615, + "num_tokens": 75726997.0, + "step": 2162 + }, + { + "epoch": 0.40167130919220057, + "grad_norm": 1.5854792594909668, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8735454678535461, + "num_tokens": 75757986.0, + "step": 2163 + }, + { + "epoch": 0.4018570102135562, + "grad_norm": 1.5552103519439697, + "learning_rate": 1e-06, + "loss": 0.4085, + "mean_token_accuracy": 0.8609669804573059, + "num_tokens": 75793575.0, + "step": 2164 + }, + { + "epoch": 0.4020427112349118, + "grad_norm": 1.5007766485214233, + "learning_rate": 1e-06, + "loss": 0.4258, + "mean_token_accuracy": 0.856601357460022, + "num_tokens": 75826073.0, + "step": 2165 + }, + { + "epoch": 0.40222841225626743, + "grad_norm": 1.4752215147018433, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8697991371154785, + "num_tokens": 75863084.0, + "step": 2166 + }, + { + "epoch": 0.40241411327762305, + "grad_norm": 1.4471538066864014, + "learning_rate": 1e-06, + "loss": 0.4745, + "mean_token_accuracy": 0.8441563248634338, + "num_tokens": 75905090.0, + "step": 2167 + }, + { + "epoch": 0.4025998142989786, + "grad_norm": 1.5474811792373657, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8675174713134766, + "num_tokens": 75938773.0, + "step": 2168 + }, + { + "epoch": 0.40278551532033424, + "grad_norm": 1.5529063940048218, + "learning_rate": 1e-06, + "loss": 0.4226, + "mean_token_accuracy": 0.8618487119674683, + "num_tokens": 75973662.0, + "step": 2169 + }, + { + "epoch": 0.40297121634168986, + "grad_norm": 1.393893837928772, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.866751492023468, + "num_tokens": 76011581.0, + "step": 2170 + }, + { + "epoch": 0.4031569173630455, + "grad_norm": 1.5109730958938599, + "learning_rate": 1e-06, + "loss": 0.4508, + "mean_token_accuracy": 0.8531588912010193, + "num_tokens": 76046514.0, + "step": 2171 + }, + { + "epoch": 0.4033426183844011, + "grad_norm": 1.5243006944656372, + "learning_rate": 1e-06, + "loss": 0.4791, + "mean_token_accuracy": 0.8420442342758179, + "num_tokens": 76084669.0, + "step": 2172 + }, + { + "epoch": 0.40352831940575673, + "grad_norm": 1.5271575450897217, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.8635615110397339, + "num_tokens": 76118217.0, + "step": 2173 + }, + { + "epoch": 0.40371402042711235, + "grad_norm": 1.71104097366333, + "learning_rate": 1e-06, + "loss": 0.4474, + "mean_token_accuracy": 0.8538703918457031, + "num_tokens": 76145054.0, + "step": 2174 + }, + { + "epoch": 0.403899721448468, + "grad_norm": 1.5265072584152222, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8620550632476807, + "num_tokens": 76177816.0, + "step": 2175 + }, + { + "epoch": 0.4040854224698236, + "grad_norm": 1.7049952745437622, + "learning_rate": 1e-06, + "loss": 0.4437, + "mean_token_accuracy": 0.852063000202179, + "num_tokens": 76209023.0, + "step": 2176 + }, + { + "epoch": 0.4042711234911792, + "grad_norm": 1.547023057937622, + "learning_rate": 1e-06, + "loss": 0.4456, + "mean_token_accuracy": 0.8520840406417847, + "num_tokens": 76246571.0, + "step": 2177 + }, + { + "epoch": 0.40445682451253484, + "grad_norm": 1.456323504447937, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8659483194351196, + "num_tokens": 76284247.0, + "step": 2178 + }, + { + "epoch": 0.40464252553389046, + "grad_norm": 1.3669273853302002, + "learning_rate": 1e-06, + "loss": 0.4547, + "mean_token_accuracy": 0.8484398126602173, + "num_tokens": 76328690.0, + "step": 2179 + }, + { + "epoch": 0.40482822655524603, + "grad_norm": 1.530212640762329, + "learning_rate": 1e-06, + "loss": 0.411, + "mean_token_accuracy": 0.861995279788971, + "num_tokens": 76365091.0, + "step": 2180 + }, + { + "epoch": 0.40501392757660165, + "grad_norm": 1.4143688678741455, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8634757995605469, + "num_tokens": 76402120.0, + "step": 2181 + }, + { + "epoch": 0.4051996285979573, + "grad_norm": 1.4494788646697998, + "learning_rate": 1e-06, + "loss": 0.4429, + "mean_token_accuracy": 0.8555829524993896, + "num_tokens": 76443689.0, + "step": 2182 + }, + { + "epoch": 0.4053853296193129, + "grad_norm": 1.6176306009292603, + "learning_rate": 1e-06, + "loss": 0.4736, + "mean_token_accuracy": 0.8451254963874817, + "num_tokens": 76477057.0, + "step": 2183 + }, + { + "epoch": 0.4055710306406685, + "grad_norm": 1.5171897411346436, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8713308572769165, + "num_tokens": 76509062.0, + "step": 2184 + }, + { + "epoch": 0.40575673166202414, + "grad_norm": 1.4929367303848267, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8592268228530884, + "num_tokens": 76545749.0, + "step": 2185 + }, + { + "epoch": 0.40594243268337976, + "grad_norm": 1.4668911695480347, + "learning_rate": 1e-06, + "loss": 0.4244, + "mean_token_accuracy": 0.8597654700279236, + "num_tokens": 76583213.0, + "step": 2186 + }, + { + "epoch": 0.4061281337047354, + "grad_norm": 1.4970711469650269, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8627749681472778, + "num_tokens": 76615688.0, + "step": 2187 + }, + { + "epoch": 0.406313834726091, + "grad_norm": 1.561384677886963, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.8529455661773682, + "num_tokens": 76650093.0, + "step": 2188 + }, + { + "epoch": 0.4064995357474466, + "grad_norm": 1.413590431213379, + "learning_rate": 1e-06, + "loss": 0.4224, + "mean_token_accuracy": 0.8637830018997192, + "num_tokens": 76688868.0, + "step": 2189 + }, + { + "epoch": 0.40668523676880225, + "grad_norm": 1.4027608633041382, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8695093393325806, + "num_tokens": 76727207.0, + "step": 2190 + }, + { + "epoch": 0.40687093779015787, + "grad_norm": 1.521270751953125, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8693163990974426, + "num_tokens": 76763209.0, + "step": 2191 + }, + { + "epoch": 0.40705663881151344, + "grad_norm": 1.414566993713379, + "learning_rate": 1e-06, + "loss": 0.4552, + "mean_token_accuracy": 0.8505616188049316, + "num_tokens": 76802611.0, + "step": 2192 + }, + { + "epoch": 0.40724233983286906, + "grad_norm": 1.5788862705230713, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8681086301803589, + "num_tokens": 76838901.0, + "step": 2193 + }, + { + "epoch": 0.4074280408542247, + "grad_norm": 1.3885860443115234, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8608994483947754, + "num_tokens": 76878587.0, + "step": 2194 + }, + { + "epoch": 0.4076137418755803, + "grad_norm": 1.4688622951507568, + "learning_rate": 1e-06, + "loss": 0.4402, + "mean_token_accuracy": 0.8582550883293152, + "num_tokens": 76913292.0, + "step": 2195 + }, + { + "epoch": 0.4077994428969359, + "grad_norm": 1.615222454071045, + "learning_rate": 1e-06, + "loss": 0.4495, + "mean_token_accuracy": 0.8515563011169434, + "num_tokens": 76944321.0, + "step": 2196 + }, + { + "epoch": 0.40798514391829155, + "grad_norm": 1.3502998352050781, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8736506700515747, + "num_tokens": 76984942.0, + "step": 2197 + }, + { + "epoch": 0.40817084493964717, + "grad_norm": 1.5206140279769897, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8727788329124451, + "num_tokens": 77019460.0, + "step": 2198 + }, + { + "epoch": 0.4083565459610028, + "grad_norm": 1.4663560390472412, + "learning_rate": 1e-06, + "loss": 0.4516, + "mean_token_accuracy": 0.8487054109573364, + "num_tokens": 77058909.0, + "step": 2199 + }, + { + "epoch": 0.4085422469823584, + "grad_norm": 1.5516215562820435, + "learning_rate": 1e-06, + "loss": 0.4476, + "mean_token_accuracy": 0.8514924049377441, + "num_tokens": 77091716.0, + "step": 2200 + }, + { + "epoch": 0.40872794800371404, + "grad_norm": 1.616768479347229, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8571727275848389, + "num_tokens": 77120372.0, + "step": 2201 + }, + { + "epoch": 0.40891364902506966, + "grad_norm": 1.57411527633667, + "learning_rate": 1e-06, + "loss": 0.416, + "mean_token_accuracy": 0.8629180788993835, + "num_tokens": 77151340.0, + "step": 2202 + }, + { + "epoch": 0.4090993500464253, + "grad_norm": 1.5695358514785767, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8727561235427856, + "num_tokens": 77183286.0, + "step": 2203 + }, + { + "epoch": 0.40928505106778085, + "grad_norm": 1.6552258729934692, + "learning_rate": 1e-06, + "loss": 0.4213, + "mean_token_accuracy": 0.859082818031311, + "num_tokens": 77215069.0, + "step": 2204 + }, + { + "epoch": 0.40947075208913647, + "grad_norm": 1.470754861831665, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8738046884536743, + "num_tokens": 77250587.0, + "step": 2205 + }, + { + "epoch": 0.4096564531104921, + "grad_norm": 1.503579020500183, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8661359548568726, + "num_tokens": 77286354.0, + "step": 2206 + }, + { + "epoch": 0.4098421541318477, + "grad_norm": 1.626939296722412, + "learning_rate": 1e-06, + "loss": 0.4311, + "mean_token_accuracy": 0.8550019860267639, + "num_tokens": 77319045.0, + "step": 2207 + }, + { + "epoch": 0.41002785515320334, + "grad_norm": 1.6494171619415283, + "learning_rate": 1e-06, + "loss": 0.4503, + "mean_token_accuracy": 0.8483279943466187, + "num_tokens": 77349648.0, + "step": 2208 + }, + { + "epoch": 0.41021355617455896, + "grad_norm": 1.5230634212493896, + "learning_rate": 1e-06, + "loss": 0.4038, + "mean_token_accuracy": 0.8623947501182556, + "num_tokens": 77383783.0, + "step": 2209 + }, + { + "epoch": 0.4103992571959146, + "grad_norm": 1.6072133779525757, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8662575483322144, + "num_tokens": 77417311.0, + "step": 2210 + }, + { + "epoch": 0.4105849582172702, + "grad_norm": 1.4735944271087646, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8635959625244141, + "num_tokens": 77452147.0, + "step": 2211 + }, + { + "epoch": 0.4107706592386258, + "grad_norm": 1.5271339416503906, + "learning_rate": 1e-06, + "loss": 0.4519, + "mean_token_accuracy": 0.8512125611305237, + "num_tokens": 77489856.0, + "step": 2212 + }, + { + "epoch": 0.41095636025998145, + "grad_norm": 1.5107630491256714, + "learning_rate": 1e-06, + "loss": 0.4287, + "mean_token_accuracy": 0.8578200340270996, + "num_tokens": 77524518.0, + "step": 2213 + }, + { + "epoch": 0.41114206128133707, + "grad_norm": 1.5175602436065674, + "learning_rate": 1e-06, + "loss": 0.4983, + "mean_token_accuracy": 0.8345287442207336, + "num_tokens": 77560612.0, + "step": 2214 + }, + { + "epoch": 0.4113277623026927, + "grad_norm": 1.4436047077178955, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8634923696517944, + "num_tokens": 77594436.0, + "step": 2215 + }, + { + "epoch": 0.41151346332404826, + "grad_norm": 1.5005123615264893, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8643556833267212, + "num_tokens": 77626980.0, + "step": 2216 + }, + { + "epoch": 0.4116991643454039, + "grad_norm": 1.432523488998413, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8703734874725342, + "num_tokens": 77662460.0, + "step": 2217 + }, + { + "epoch": 0.4118848653667595, + "grad_norm": 1.414419412612915, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8612759709358215, + "num_tokens": 77701565.0, + "step": 2218 + }, + { + "epoch": 0.4120705663881151, + "grad_norm": 1.6723741292953491, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8617509603500366, + "num_tokens": 77734642.0, + "step": 2219 + }, + { + "epoch": 0.41225626740947074, + "grad_norm": 1.54433274269104, + "learning_rate": 1e-06, + "loss": 0.4506, + "mean_token_accuracy": 0.8539661169052124, + "num_tokens": 77767518.0, + "step": 2220 + }, + { + "epoch": 0.41244196843082637, + "grad_norm": 1.6018940210342407, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.864702582359314, + "num_tokens": 77800071.0, + "step": 2221 + }, + { + "epoch": 0.412627669452182, + "grad_norm": 1.6338614225387573, + "learning_rate": 1e-06, + "loss": 0.4685, + "mean_token_accuracy": 0.84624183177948, + "num_tokens": 77831534.0, + "step": 2222 + }, + { + "epoch": 0.4128133704735376, + "grad_norm": 1.463741660118103, + "learning_rate": 1e-06, + "loss": 0.4261, + "mean_token_accuracy": 0.8562565445899963, + "num_tokens": 77871105.0, + "step": 2223 + }, + { + "epoch": 0.41299907149489323, + "grad_norm": 1.5082896947860718, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8577113151550293, + "num_tokens": 77904571.0, + "step": 2224 + }, + { + "epoch": 0.41318477251624885, + "grad_norm": 1.5995526313781738, + "learning_rate": 1e-06, + "loss": 0.4113, + "mean_token_accuracy": 0.8583372831344604, + "num_tokens": 77935206.0, + "step": 2225 + }, + { + "epoch": 0.4133704735376045, + "grad_norm": 1.4676008224487305, + "learning_rate": 1e-06, + "loss": 0.4429, + "mean_token_accuracy": 0.8537606000900269, + "num_tokens": 77972151.0, + "step": 2226 + }, + { + "epoch": 0.4135561745589601, + "grad_norm": 1.6164473295211792, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.867307186126709, + "num_tokens": 78004231.0, + "step": 2227 + }, + { + "epoch": 0.41374187558031567, + "grad_norm": 1.512703776359558, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8710156083106995, + "num_tokens": 78035791.0, + "step": 2228 + }, + { + "epoch": 0.4139275766016713, + "grad_norm": 1.5514575242996216, + "learning_rate": 1e-06, + "loss": 0.4627, + "mean_token_accuracy": 0.8470067381858826, + "num_tokens": 78070639.0, + "step": 2229 + }, + { + "epoch": 0.4141132776230269, + "grad_norm": 1.4677621126174927, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8682578802108765, + "num_tokens": 78104168.0, + "step": 2230 + }, + { + "epoch": 0.41429897864438253, + "grad_norm": 1.4279496669769287, + "learning_rate": 1e-06, + "loss": 0.446, + "mean_token_accuracy": 0.850069522857666, + "num_tokens": 78147004.0, + "step": 2231 + }, + { + "epoch": 0.41448467966573815, + "grad_norm": 1.6286948919296265, + "learning_rate": 1e-06, + "loss": 0.4624, + "mean_token_accuracy": 0.8474217653274536, + "num_tokens": 78177566.0, + "step": 2232 + }, + { + "epoch": 0.4146703806870938, + "grad_norm": 1.4794875383377075, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.856334924697876, + "num_tokens": 78212794.0, + "step": 2233 + }, + { + "epoch": 0.4148560817084494, + "grad_norm": 1.4344462156295776, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8608248233795166, + "num_tokens": 78248630.0, + "step": 2234 + }, + { + "epoch": 0.415041782729805, + "grad_norm": 1.519555926322937, + "learning_rate": 1e-06, + "loss": 0.4457, + "mean_token_accuracy": 0.8518345952033997, + "num_tokens": 78281318.0, + "step": 2235 + }, + { + "epoch": 0.41522748375116064, + "grad_norm": 1.509445309638977, + "learning_rate": 1e-06, + "loss": 0.4218, + "mean_token_accuracy": 0.8589358329772949, + "num_tokens": 78313857.0, + "step": 2236 + }, + { + "epoch": 0.41541318477251626, + "grad_norm": 1.4782711267471313, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.853969395160675, + "num_tokens": 78353633.0, + "step": 2237 + }, + { + "epoch": 0.4155988857938719, + "grad_norm": 1.4216127395629883, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8716800212860107, + "num_tokens": 78392826.0, + "step": 2238 + }, + { + "epoch": 0.4157845868152275, + "grad_norm": 1.6873036623001099, + "learning_rate": 1e-06, + "loss": 0.4174, + "mean_token_accuracy": 0.8624223470687866, + "num_tokens": 78421390.0, + "step": 2239 + }, + { + "epoch": 0.4159702878365831, + "grad_norm": 1.4916871786117554, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8781588077545166, + "num_tokens": 78458649.0, + "step": 2240 + }, + { + "epoch": 0.4161559888579387, + "grad_norm": 1.371782660484314, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8701444864273071, + "num_tokens": 78495996.0, + "step": 2241 + }, + { + "epoch": 0.4163416898792943, + "grad_norm": 1.4135866165161133, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.8659583330154419, + "num_tokens": 78534091.0, + "step": 2242 + }, + { + "epoch": 0.41652739090064994, + "grad_norm": 1.3929269313812256, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8736367225646973, + "num_tokens": 78570419.0, + "step": 2243 + }, + { + "epoch": 0.41671309192200556, + "grad_norm": 1.621450424194336, + "learning_rate": 1e-06, + "loss": 0.4186, + "mean_token_accuracy": 0.8598943948745728, + "num_tokens": 78602687.0, + "step": 2244 + }, + { + "epoch": 0.4168987929433612, + "grad_norm": 1.5226422548294067, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8572850227355957, + "num_tokens": 78639549.0, + "step": 2245 + }, + { + "epoch": 0.4170844939647168, + "grad_norm": 1.5367035865783691, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8616808652877808, + "num_tokens": 78674769.0, + "step": 2246 + }, + { + "epoch": 0.41727019498607243, + "grad_norm": 1.5549731254577637, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8692256808280945, + "num_tokens": 78709784.0, + "step": 2247 + }, + { + "epoch": 0.41745589600742805, + "grad_norm": 1.7643957138061523, + "learning_rate": 1e-06, + "loss": 0.4703, + "mean_token_accuracy": 0.8482863903045654, + "num_tokens": 78739716.0, + "step": 2248 + }, + { + "epoch": 0.4176415970287837, + "grad_norm": 1.4660886526107788, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.857519268989563, + "num_tokens": 78774952.0, + "step": 2249 + }, + { + "epoch": 0.4178272980501393, + "grad_norm": 1.5844820737838745, + "learning_rate": 1e-06, + "loss": 0.4526, + "mean_token_accuracy": 0.8510065674781799, + "num_tokens": 78811407.0, + "step": 2250 + }, + { + "epoch": 0.4180129990714949, + "grad_norm": 1.5885454416275024, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.8656876087188721, + "num_tokens": 78842417.0, + "step": 2251 + }, + { + "epoch": 0.4181987000928505, + "grad_norm": 1.5229231119155884, + "learning_rate": 1e-06, + "loss": 0.5022, + "mean_token_accuracy": 0.8355837464332581, + "num_tokens": 78882341.0, + "step": 2252 + }, + { + "epoch": 0.4183844011142061, + "grad_norm": 1.405764102935791, + "learning_rate": 1e-06, + "loss": 0.4387, + "mean_token_accuracy": 0.8543325066566467, + "num_tokens": 78923774.0, + "step": 2253 + }, + { + "epoch": 0.41857010213556173, + "grad_norm": 1.593640923500061, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8618325591087341, + "num_tokens": 78957291.0, + "step": 2254 + }, + { + "epoch": 0.41875580315691735, + "grad_norm": 1.808652639389038, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8729749917984009, + "num_tokens": 78982594.0, + "step": 2255 + }, + { + "epoch": 0.41894150417827297, + "grad_norm": 1.5491797924041748, + "learning_rate": 1e-06, + "loss": 0.4167, + "mean_token_accuracy": 0.861284613609314, + "num_tokens": 79015630.0, + "step": 2256 + }, + { + "epoch": 0.4191272051996286, + "grad_norm": 1.4733682870864868, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8655555248260498, + "num_tokens": 79054517.0, + "step": 2257 + }, + { + "epoch": 0.4193129062209842, + "grad_norm": 1.4606409072875977, + "learning_rate": 1e-06, + "loss": 0.4278, + "mean_token_accuracy": 0.8615905046463013, + "num_tokens": 79091519.0, + "step": 2258 + }, + { + "epoch": 0.41949860724233984, + "grad_norm": 1.578480839729309, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.880325198173523, + "num_tokens": 79119187.0, + "step": 2259 + }, + { + "epoch": 0.41968430826369546, + "grad_norm": 1.5611623525619507, + "learning_rate": 1e-06, + "loss": 0.4346, + "mean_token_accuracy": 0.8584137558937073, + "num_tokens": 79153439.0, + "step": 2260 + }, + { + "epoch": 0.4198700092850511, + "grad_norm": 1.4353408813476562, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8728803396224976, + "num_tokens": 79192217.0, + "step": 2261 + }, + { + "epoch": 0.4200557103064067, + "grad_norm": 1.478575587272644, + "learning_rate": 1e-06, + "loss": 0.4509, + "mean_token_accuracy": 0.8495621085166931, + "num_tokens": 79231362.0, + "step": 2262 + }, + { + "epoch": 0.4202414113277623, + "grad_norm": 1.4209222793579102, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8602232933044434, + "num_tokens": 79272450.0, + "step": 2263 + }, + { + "epoch": 0.4204271123491179, + "grad_norm": 1.4943749904632568, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8673874735832214, + "num_tokens": 79305616.0, + "step": 2264 + }, + { + "epoch": 0.4206128133704735, + "grad_norm": 1.6972622871398926, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.8625386953353882, + "num_tokens": 79336668.0, + "step": 2265 + }, + { + "epoch": 0.42079851439182914, + "grad_norm": 1.5795979499816895, + "learning_rate": 1e-06, + "loss": 0.4256, + "mean_token_accuracy": 0.8675872683525085, + "num_tokens": 79369210.0, + "step": 2266 + }, + { + "epoch": 0.42098421541318476, + "grad_norm": 1.462014079093933, + "learning_rate": 1e-06, + "loss": 0.4475, + "mean_token_accuracy": 0.852395236492157, + "num_tokens": 79408324.0, + "step": 2267 + }, + { + "epoch": 0.4211699164345404, + "grad_norm": 1.4603685140609741, + "learning_rate": 1e-06, + "loss": 0.457, + "mean_token_accuracy": 0.8477811217308044, + "num_tokens": 79446839.0, + "step": 2268 + }, + { + "epoch": 0.421355617455896, + "grad_norm": 1.439588189125061, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8611558675765991, + "num_tokens": 79483583.0, + "step": 2269 + }, + { + "epoch": 0.4215413184772516, + "grad_norm": 1.425365686416626, + "learning_rate": 1e-06, + "loss": 0.401, + "mean_token_accuracy": 0.8661057949066162, + "num_tokens": 79523541.0, + "step": 2270 + }, + { + "epoch": 0.42172701949860725, + "grad_norm": 1.417320728302002, + "learning_rate": 1e-06, + "loss": 0.4045, + "mean_token_accuracy": 0.8655117750167847, + "num_tokens": 79560606.0, + "step": 2271 + }, + { + "epoch": 0.42191272051996287, + "grad_norm": 1.4948481321334839, + "learning_rate": 1e-06, + "loss": 0.4291, + "mean_token_accuracy": 0.858274519443512, + "num_tokens": 79597651.0, + "step": 2272 + }, + { + "epoch": 0.4220984215413185, + "grad_norm": 1.4855492115020752, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8604104518890381, + "num_tokens": 79631410.0, + "step": 2273 + }, + { + "epoch": 0.4222841225626741, + "grad_norm": 1.4478654861450195, + "learning_rate": 1e-06, + "loss": 0.4245, + "mean_token_accuracy": 0.8580931425094604, + "num_tokens": 79672513.0, + "step": 2274 + }, + { + "epoch": 0.42246982358402974, + "grad_norm": 1.5210360288619995, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8740667104721069, + "num_tokens": 79704468.0, + "step": 2275 + }, + { + "epoch": 0.4226555246053853, + "grad_norm": 1.4845690727233887, + "learning_rate": 1e-06, + "loss": 0.4353, + "mean_token_accuracy": 0.8575680255889893, + "num_tokens": 79742363.0, + "step": 2276 + }, + { + "epoch": 0.4228412256267409, + "grad_norm": 1.43793523311615, + "learning_rate": 1e-06, + "loss": 0.4283, + "mean_token_accuracy": 0.8569451570510864, + "num_tokens": 79778212.0, + "step": 2277 + }, + { + "epoch": 0.42302692664809655, + "grad_norm": 1.506606936454773, + "learning_rate": 1e-06, + "loss": 0.458, + "mean_token_accuracy": 0.8494991064071655, + "num_tokens": 79819488.0, + "step": 2278 + }, + { + "epoch": 0.42321262766945217, + "grad_norm": 1.7017146348953247, + "learning_rate": 1e-06, + "loss": 0.4877, + "mean_token_accuracy": 0.8313509225845337, + "num_tokens": 79853324.0, + "step": 2279 + }, + { + "epoch": 0.4233983286908078, + "grad_norm": 1.5558401346206665, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8605089783668518, + "num_tokens": 79889266.0, + "step": 2280 + }, + { + "epoch": 0.4235840297121634, + "grad_norm": 1.4461195468902588, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8746984601020813, + "num_tokens": 79922926.0, + "step": 2281 + }, + { + "epoch": 0.42376973073351903, + "grad_norm": 1.3432035446166992, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8742228150367737, + "num_tokens": 79963110.0, + "step": 2282 + }, + { + "epoch": 0.42395543175487466, + "grad_norm": 1.3501485586166382, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8740954399108887, + "num_tokens": 80002146.0, + "step": 2283 + }, + { + "epoch": 0.4241411327762303, + "grad_norm": 1.5177336931228638, + "learning_rate": 1e-06, + "loss": 0.4441, + "mean_token_accuracy": 0.8488144278526306, + "num_tokens": 80036018.0, + "step": 2284 + }, + { + "epoch": 0.4243268337975859, + "grad_norm": 1.5070863962173462, + "learning_rate": 1e-06, + "loss": 0.4538, + "mean_token_accuracy": 0.8505208492279053, + "num_tokens": 80070912.0, + "step": 2285 + }, + { + "epoch": 0.4245125348189415, + "grad_norm": 1.545440912246704, + "learning_rate": 1e-06, + "loss": 0.3962, + "mean_token_accuracy": 0.8653286695480347, + "num_tokens": 80106449.0, + "step": 2286 + }, + { + "epoch": 0.42469823584029714, + "grad_norm": 1.4564417600631714, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8725564479827881, + "num_tokens": 80140767.0, + "step": 2287 + }, + { + "epoch": 0.42488393686165277, + "grad_norm": 1.5130854845046997, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.8571704626083374, + "num_tokens": 80176317.0, + "step": 2288 + }, + { + "epoch": 0.42506963788300833, + "grad_norm": 1.3689225912094116, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.875042200088501, + "num_tokens": 80213265.0, + "step": 2289 + }, + { + "epoch": 0.42525533890436396, + "grad_norm": 1.4805978536605835, + "learning_rate": 1e-06, + "loss": 0.4435, + "mean_token_accuracy": 0.8513628244400024, + "num_tokens": 80255083.0, + "step": 2290 + }, + { + "epoch": 0.4254410399257196, + "grad_norm": 1.4991254806518555, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8524801731109619, + "num_tokens": 80293037.0, + "step": 2291 + }, + { + "epoch": 0.4256267409470752, + "grad_norm": 1.4785199165344238, + "learning_rate": 1e-06, + "loss": 0.4248, + "mean_token_accuracy": 0.8583821654319763, + "num_tokens": 80328327.0, + "step": 2292 + }, + { + "epoch": 0.4258124419684308, + "grad_norm": 1.554269790649414, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.8519208431243896, + "num_tokens": 80362894.0, + "step": 2293 + }, + { + "epoch": 0.42599814298978644, + "grad_norm": 1.5681222677230835, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8697916865348816, + "num_tokens": 80395631.0, + "step": 2294 + }, + { + "epoch": 0.42618384401114207, + "grad_norm": 1.5185201168060303, + "learning_rate": 1e-06, + "loss": 0.4573, + "mean_token_accuracy": 0.8478525280952454, + "num_tokens": 80433482.0, + "step": 2295 + }, + { + "epoch": 0.4263695450324977, + "grad_norm": 1.541056752204895, + "learning_rate": 1e-06, + "loss": 0.4424, + "mean_token_accuracy": 0.8543820381164551, + "num_tokens": 80469831.0, + "step": 2296 + }, + { + "epoch": 0.4265552460538533, + "grad_norm": 1.3457056283950806, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8669097423553467, + "num_tokens": 80508133.0, + "step": 2297 + }, + { + "epoch": 0.42674094707520893, + "grad_norm": 1.4131067991256714, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8631656765937805, + "num_tokens": 80543914.0, + "step": 2298 + }, + { + "epoch": 0.42692664809656455, + "grad_norm": 1.4931081533432007, + "learning_rate": 1e-06, + "loss": 0.4691, + "mean_token_accuracy": 0.8457969427108765, + "num_tokens": 80584570.0, + "step": 2299 + }, + { + "epoch": 0.4271123491179202, + "grad_norm": 1.6303699016571045, + "learning_rate": 1e-06, + "loss": 0.4917, + "mean_token_accuracy": 0.8428412675857544, + "num_tokens": 80619777.0, + "step": 2300 + }, + { + "epoch": 0.42729805013927574, + "grad_norm": 1.4439654350280762, + "learning_rate": 1e-06, + "loss": 0.4627, + "mean_token_accuracy": 0.846703290939331, + "num_tokens": 80662162.0, + "step": 2301 + }, + { + "epoch": 0.42748375116063136, + "grad_norm": 1.5948982238769531, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.855822741985321, + "num_tokens": 80693443.0, + "step": 2302 + }, + { + "epoch": 0.427669452181987, + "grad_norm": 1.610352635383606, + "learning_rate": 1e-06, + "loss": 0.472, + "mean_token_accuracy": 0.8464388847351074, + "num_tokens": 80723307.0, + "step": 2303 + }, + { + "epoch": 0.4278551532033426, + "grad_norm": 1.526180386543274, + "learning_rate": 1e-06, + "loss": 0.4323, + "mean_token_accuracy": 0.8582639694213867, + "num_tokens": 80757029.0, + "step": 2304 + }, + { + "epoch": 0.42804085422469823, + "grad_norm": 1.4182132482528687, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8654006123542786, + "num_tokens": 80793595.0, + "step": 2305 + }, + { + "epoch": 0.42822655524605385, + "grad_norm": 1.5384448766708374, + "learning_rate": 1e-06, + "loss": 0.4338, + "mean_token_accuracy": 0.8560763597488403, + "num_tokens": 80827227.0, + "step": 2306 + }, + { + "epoch": 0.4284122562674095, + "grad_norm": 1.4583367109298706, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.858709990978241, + "num_tokens": 80863602.0, + "step": 2307 + }, + { + "epoch": 0.4285979572887651, + "grad_norm": 1.3654264211654663, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8602679967880249, + "num_tokens": 80906038.0, + "step": 2308 + }, + { + "epoch": 0.4287836583101207, + "grad_norm": 1.4272576570510864, + "learning_rate": 1e-06, + "loss": 0.4072, + "mean_token_accuracy": 0.8653708696365356, + "num_tokens": 80946801.0, + "step": 2309 + }, + { + "epoch": 0.42896935933147634, + "grad_norm": 1.5694975852966309, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8592864871025085, + "num_tokens": 80980452.0, + "step": 2310 + }, + { + "epoch": 0.42915506035283196, + "grad_norm": 1.6141777038574219, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8652303814888, + "num_tokens": 81011922.0, + "step": 2311 + }, + { + "epoch": 0.4293407613741876, + "grad_norm": 1.4545879364013672, + "learning_rate": 1e-06, + "loss": 0.4254, + "mean_token_accuracy": 0.8562798500061035, + "num_tokens": 81050409.0, + "step": 2312 + }, + { + "epoch": 0.42952646239554315, + "grad_norm": 1.5179975032806396, + "learning_rate": 1e-06, + "loss": 0.436, + "mean_token_accuracy": 0.8554127216339111, + "num_tokens": 81090663.0, + "step": 2313 + }, + { + "epoch": 0.4297121634168988, + "grad_norm": 1.3956043720245361, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8714308142662048, + "num_tokens": 81127196.0, + "step": 2314 + }, + { + "epoch": 0.4298978644382544, + "grad_norm": 1.546492338180542, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8603385090827942, + "num_tokens": 81160155.0, + "step": 2315 + }, + { + "epoch": 0.43008356545961, + "grad_norm": 1.4371720552444458, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8663552403450012, + "num_tokens": 81196872.0, + "step": 2316 + }, + { + "epoch": 0.43026926648096564, + "grad_norm": 1.4308956861495972, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8648539185523987, + "num_tokens": 81231567.0, + "step": 2317 + }, + { + "epoch": 0.43045496750232126, + "grad_norm": 1.431848406791687, + "learning_rate": 1e-06, + "loss": 0.4487, + "mean_token_accuracy": 0.853451132774353, + "num_tokens": 81272175.0, + "step": 2318 + }, + { + "epoch": 0.4306406685236769, + "grad_norm": 1.4604295492172241, + "learning_rate": 1e-06, + "loss": 0.4288, + "mean_token_accuracy": 0.8579349517822266, + "num_tokens": 81306664.0, + "step": 2319 + }, + { + "epoch": 0.4308263695450325, + "grad_norm": 1.5439116954803467, + "learning_rate": 1e-06, + "loss": 0.4505, + "mean_token_accuracy": 0.8535072803497314, + "num_tokens": 81340048.0, + "step": 2320 + }, + { + "epoch": 0.43101207056638813, + "grad_norm": 1.505818247795105, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8705029487609863, + "num_tokens": 81373134.0, + "step": 2321 + }, + { + "epoch": 0.43119777158774375, + "grad_norm": 1.5887576341629028, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.873319685459137, + "num_tokens": 81400336.0, + "step": 2322 + }, + { + "epoch": 0.43138347260909937, + "grad_norm": 1.5511945486068726, + "learning_rate": 1e-06, + "loss": 0.4339, + "mean_token_accuracy": 0.8550850749015808, + "num_tokens": 81432761.0, + "step": 2323 + }, + { + "epoch": 0.431569173630455, + "grad_norm": 1.486513376235962, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.855614960193634, + "num_tokens": 81468507.0, + "step": 2324 + }, + { + "epoch": 0.43175487465181056, + "grad_norm": 1.467840313911438, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8654413223266602, + "num_tokens": 81502848.0, + "step": 2325 + }, + { + "epoch": 0.4319405756731662, + "grad_norm": 1.5436608791351318, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8702378273010254, + "num_tokens": 81536398.0, + "step": 2326 + }, + { + "epoch": 0.4321262766945218, + "grad_norm": 1.6115731000900269, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.864815354347229, + "num_tokens": 81569126.0, + "step": 2327 + }, + { + "epoch": 0.4323119777158774, + "grad_norm": 1.5124821662902832, + "learning_rate": 1e-06, + "loss": 0.5151, + "mean_token_accuracy": 0.8336427211761475, + "num_tokens": 81608846.0, + "step": 2328 + }, + { + "epoch": 0.43249767873723305, + "grad_norm": 1.6477880477905273, + "learning_rate": 1e-06, + "loss": 0.4813, + "mean_token_accuracy": 0.8424865007400513, + "num_tokens": 81642028.0, + "step": 2329 + }, + { + "epoch": 0.43268337975858867, + "grad_norm": 1.586564302444458, + "learning_rate": 1e-06, + "loss": 0.407, + "mean_token_accuracy": 0.8614081144332886, + "num_tokens": 81673890.0, + "step": 2330 + }, + { + "epoch": 0.4328690807799443, + "grad_norm": 1.5589650869369507, + "learning_rate": 1e-06, + "loss": 0.4389, + "mean_token_accuracy": 0.8541638851165771, + "num_tokens": 81710032.0, + "step": 2331 + }, + { + "epoch": 0.4330547818012999, + "grad_norm": 1.4028093814849854, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8781932592391968, + "num_tokens": 81748492.0, + "step": 2332 + }, + { + "epoch": 0.43324048282265554, + "grad_norm": 1.470849871635437, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8669264316558838, + "num_tokens": 81784605.0, + "step": 2333 + }, + { + "epoch": 0.43342618384401116, + "grad_norm": 1.4879037141799927, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8596112728118896, + "num_tokens": 81819490.0, + "step": 2334 + }, + { + "epoch": 0.4336118848653668, + "grad_norm": 1.4947255849838257, + "learning_rate": 1e-06, + "loss": 0.4229, + "mean_token_accuracy": 0.8584553003311157, + "num_tokens": 81852539.0, + "step": 2335 + }, + { + "epoch": 0.4337975858867224, + "grad_norm": 1.511177897453308, + "learning_rate": 1e-06, + "loss": 0.4075, + "mean_token_accuracy": 0.8625169992446899, + "num_tokens": 81885733.0, + "step": 2336 + }, + { + "epoch": 0.43398328690807797, + "grad_norm": 1.5588115453720093, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.86496901512146, + "num_tokens": 81918009.0, + "step": 2337 + }, + { + "epoch": 0.4341689879294336, + "grad_norm": 1.566316843032837, + "learning_rate": 1e-06, + "loss": 0.4516, + "mean_token_accuracy": 0.8540223836898804, + "num_tokens": 81950263.0, + "step": 2338 + }, + { + "epoch": 0.4343546889507892, + "grad_norm": 1.408238410949707, + "learning_rate": 1e-06, + "loss": 0.4639, + "mean_token_accuracy": 0.8483420610427856, + "num_tokens": 81991577.0, + "step": 2339 + }, + { + "epoch": 0.43454038997214484, + "grad_norm": 1.5673390626907349, + "learning_rate": 1e-06, + "loss": 0.4698, + "mean_token_accuracy": 0.8463852405548096, + "num_tokens": 82025464.0, + "step": 2340 + }, + { + "epoch": 0.43472609099350046, + "grad_norm": 1.5822865962982178, + "learning_rate": 1e-06, + "loss": 0.4313, + "mean_token_accuracy": 0.8548243641853333, + "num_tokens": 82055792.0, + "step": 2341 + }, + { + "epoch": 0.4349117920148561, + "grad_norm": 1.4586784839630127, + "learning_rate": 1e-06, + "loss": 0.4431, + "mean_token_accuracy": 0.8491590023040771, + "num_tokens": 82095848.0, + "step": 2342 + }, + { + "epoch": 0.4350974930362117, + "grad_norm": 1.4361605644226074, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8678044080734253, + "num_tokens": 82131045.0, + "step": 2343 + }, + { + "epoch": 0.4352831940575673, + "grad_norm": 1.4581104516983032, + "learning_rate": 1e-06, + "loss": 0.4427, + "mean_token_accuracy": 0.851597785949707, + "num_tokens": 82172335.0, + "step": 2344 + }, + { + "epoch": 0.43546889507892295, + "grad_norm": 1.6466050148010254, + "learning_rate": 1e-06, + "loss": 0.4661, + "mean_token_accuracy": 0.8438565731048584, + "num_tokens": 82202629.0, + "step": 2345 + }, + { + "epoch": 0.43565459610027857, + "grad_norm": 1.519052267074585, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8632367849349976, + "num_tokens": 82235770.0, + "step": 2346 + }, + { + "epoch": 0.4358402971216342, + "grad_norm": 1.464905858039856, + "learning_rate": 1e-06, + "loss": 0.486, + "mean_token_accuracy": 0.8416933417320251, + "num_tokens": 82273422.0, + "step": 2347 + }, + { + "epoch": 0.4360259981429898, + "grad_norm": 1.518622875213623, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8661142587661743, + "num_tokens": 82306319.0, + "step": 2348 + }, + { + "epoch": 0.4362116991643454, + "grad_norm": 1.5914671421051025, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8637841939926147, + "num_tokens": 82339672.0, + "step": 2349 + }, + { + "epoch": 0.436397400185701, + "grad_norm": 1.4785267114639282, + "learning_rate": 1e-06, + "loss": 0.4229, + "mean_token_accuracy": 0.8578956127166748, + "num_tokens": 82375296.0, + "step": 2350 + }, + { + "epoch": 0.4365831012070566, + "grad_norm": 1.6009279489517212, + "learning_rate": 1e-06, + "loss": 0.4628, + "mean_token_accuracy": 0.8453953862190247, + "num_tokens": 82406014.0, + "step": 2351 + }, + { + "epoch": 0.43676880222841225, + "grad_norm": 1.5309157371520996, + "learning_rate": 1e-06, + "loss": 0.4758, + "mean_token_accuracy": 0.8457018136978149, + "num_tokens": 82440715.0, + "step": 2352 + }, + { + "epoch": 0.43695450324976787, + "grad_norm": 1.4424450397491455, + "learning_rate": 1e-06, + "loss": 0.4208, + "mean_token_accuracy": 0.8582680821418762, + "num_tokens": 82477126.0, + "step": 2353 + }, + { + "epoch": 0.4371402042711235, + "grad_norm": 1.386620283126831, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8747718930244446, + "num_tokens": 82516933.0, + "step": 2354 + }, + { + "epoch": 0.4373259052924791, + "grad_norm": 1.5741766691207886, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8711337447166443, + "num_tokens": 82545130.0, + "step": 2355 + }, + { + "epoch": 0.43751160631383473, + "grad_norm": 1.5622141361236572, + "learning_rate": 1e-06, + "loss": 0.4239, + "mean_token_accuracy": 0.8639384508132935, + "num_tokens": 82579287.0, + "step": 2356 + }, + { + "epoch": 0.43769730733519036, + "grad_norm": 1.5403872728347778, + "learning_rate": 1e-06, + "loss": 0.4215, + "mean_token_accuracy": 0.8568980693817139, + "num_tokens": 82614782.0, + "step": 2357 + }, + { + "epoch": 0.437883008356546, + "grad_norm": 1.3933223485946655, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8659394383430481, + "num_tokens": 82653093.0, + "step": 2358 + }, + { + "epoch": 0.4380687093779016, + "grad_norm": 1.460450530052185, + "learning_rate": 1e-06, + "loss": 0.4165, + "mean_token_accuracy": 0.858734130859375, + "num_tokens": 82690527.0, + "step": 2359 + }, + { + "epoch": 0.4382544103992572, + "grad_norm": 1.415854573249817, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8683621883392334, + "num_tokens": 82729125.0, + "step": 2360 + }, + { + "epoch": 0.4384401114206128, + "grad_norm": 1.478596568107605, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8625432252883911, + "num_tokens": 82765162.0, + "step": 2361 + }, + { + "epoch": 0.4386258124419684, + "grad_norm": 1.5485371351242065, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8654879331588745, + "num_tokens": 82798795.0, + "step": 2362 + }, + { + "epoch": 0.43881151346332403, + "grad_norm": 1.403460144996643, + "learning_rate": 1e-06, + "loss": 0.4325, + "mean_token_accuracy": 0.8586317300796509, + "num_tokens": 82840690.0, + "step": 2363 + }, + { + "epoch": 0.43899721448467965, + "grad_norm": 1.5702996253967285, + "learning_rate": 1e-06, + "loss": 0.4259, + "mean_token_accuracy": 0.8581881523132324, + "num_tokens": 82879009.0, + "step": 2364 + }, + { + "epoch": 0.4391829155060353, + "grad_norm": 1.4749394655227661, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8620129823684692, + "num_tokens": 82914295.0, + "step": 2365 + }, + { + "epoch": 0.4393686165273909, + "grad_norm": 1.4158644676208496, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8649228811264038, + "num_tokens": 82950251.0, + "step": 2366 + }, + { + "epoch": 0.4395543175487465, + "grad_norm": 1.4598106145858765, + "learning_rate": 1e-06, + "loss": 0.4487, + "mean_token_accuracy": 0.852914035320282, + "num_tokens": 82985977.0, + "step": 2367 + }, + { + "epoch": 0.43974001857010214, + "grad_norm": 1.574563980102539, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8670052289962769, + "num_tokens": 83020090.0, + "step": 2368 + }, + { + "epoch": 0.43992571959145776, + "grad_norm": 1.3961182832717896, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8699949979782104, + "num_tokens": 83060046.0, + "step": 2369 + }, + { + "epoch": 0.4401114206128134, + "grad_norm": 1.3908379077911377, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8594145774841309, + "num_tokens": 83102702.0, + "step": 2370 + }, + { + "epoch": 0.440297121634169, + "grad_norm": 1.500160574913025, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.860200047492981, + "num_tokens": 83136209.0, + "step": 2371 + }, + { + "epoch": 0.44048282265552463, + "grad_norm": 1.4006847143173218, + "learning_rate": 1e-06, + "loss": 0.4242, + "mean_token_accuracy": 0.8596017360687256, + "num_tokens": 83177746.0, + "step": 2372 + }, + { + "epoch": 0.4406685236768802, + "grad_norm": 1.4984707832336426, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8588310480117798, + "num_tokens": 83215021.0, + "step": 2373 + }, + { + "epoch": 0.4408542246982358, + "grad_norm": 1.5266573429107666, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.8672966361045837, + "num_tokens": 83250439.0, + "step": 2374 + }, + { + "epoch": 0.44103992571959144, + "grad_norm": 1.5719242095947266, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8661326169967651, + "num_tokens": 83284246.0, + "step": 2375 + }, + { + "epoch": 0.44122562674094706, + "grad_norm": 1.5343258380889893, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8669912815093994, + "num_tokens": 83314828.0, + "step": 2376 + }, + { + "epoch": 0.4414113277623027, + "grad_norm": 1.4861875772476196, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8662639260292053, + "num_tokens": 83350294.0, + "step": 2377 + }, + { + "epoch": 0.4415970287836583, + "grad_norm": 1.5106120109558105, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8644335865974426, + "num_tokens": 83385940.0, + "step": 2378 + }, + { + "epoch": 0.44178272980501393, + "grad_norm": 1.625943899154663, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.8632057905197144, + "num_tokens": 83414483.0, + "step": 2379 + }, + { + "epoch": 0.44196843082636955, + "grad_norm": 1.4939887523651123, + "learning_rate": 1e-06, + "loss": 0.4603, + "mean_token_accuracy": 0.8453172445297241, + "num_tokens": 83455737.0, + "step": 2380 + }, + { + "epoch": 0.4421541318477252, + "grad_norm": 1.620082139968872, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8711071014404297, + "num_tokens": 83484903.0, + "step": 2381 + }, + { + "epoch": 0.4423398328690808, + "grad_norm": 1.4627320766448975, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8618077635765076, + "num_tokens": 83518794.0, + "step": 2382 + }, + { + "epoch": 0.4425255338904364, + "grad_norm": 1.421656608581543, + "learning_rate": 1e-06, + "loss": 0.431, + "mean_token_accuracy": 0.857316255569458, + "num_tokens": 83559290.0, + "step": 2383 + }, + { + "epoch": 0.44271123491179204, + "grad_norm": 1.4091402292251587, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8695386648178101, + "num_tokens": 83600544.0, + "step": 2384 + }, + { + "epoch": 0.4428969359331476, + "grad_norm": 1.4952951669692993, + "learning_rate": 1e-06, + "loss": 0.4407, + "mean_token_accuracy": 0.8507809042930603, + "num_tokens": 83641940.0, + "step": 2385 + }, + { + "epoch": 0.44308263695450323, + "grad_norm": 1.5573254823684692, + "learning_rate": 1e-06, + "loss": 0.4599, + "mean_token_accuracy": 0.8500012159347534, + "num_tokens": 83676662.0, + "step": 2386 + }, + { + "epoch": 0.44326833797585885, + "grad_norm": 1.4312947988510132, + "learning_rate": 1e-06, + "loss": 0.4241, + "mean_token_accuracy": 0.8587735891342163, + "num_tokens": 83714216.0, + "step": 2387 + }, + { + "epoch": 0.4434540389972145, + "grad_norm": 1.4245418310165405, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8542236089706421, + "num_tokens": 83754774.0, + "step": 2388 + }, + { + "epoch": 0.4436397400185701, + "grad_norm": 1.3652441501617432, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.8684703707695007, + "num_tokens": 83791388.0, + "step": 2389 + }, + { + "epoch": 0.4438254410399257, + "grad_norm": 1.534096598625183, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8666307926177979, + "num_tokens": 83825228.0, + "step": 2390 + }, + { + "epoch": 0.44401114206128134, + "grad_norm": 1.6127989292144775, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.8663692474365234, + "num_tokens": 83853893.0, + "step": 2391 + }, + { + "epoch": 0.44419684308263696, + "grad_norm": 1.5308977365493774, + "learning_rate": 1e-06, + "loss": 0.433, + "mean_token_accuracy": 0.8552558422088623, + "num_tokens": 83888590.0, + "step": 2392 + }, + { + "epoch": 0.4443825441039926, + "grad_norm": 1.547349452972412, + "learning_rate": 1e-06, + "loss": 0.4376, + "mean_token_accuracy": 0.8551598787307739, + "num_tokens": 83922979.0, + "step": 2393 + }, + { + "epoch": 0.4445682451253482, + "grad_norm": 1.5156264305114746, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.8571580648422241, + "num_tokens": 83955322.0, + "step": 2394 + }, + { + "epoch": 0.4447539461467038, + "grad_norm": 1.5293571949005127, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8747327327728271, + "num_tokens": 83986044.0, + "step": 2395 + }, + { + "epoch": 0.44493964716805945, + "grad_norm": 1.4644513130187988, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8719320297241211, + "num_tokens": 84021557.0, + "step": 2396 + }, + { + "epoch": 0.445125348189415, + "grad_norm": 1.4935531616210938, + "learning_rate": 1e-06, + "loss": 0.4507, + "mean_token_accuracy": 0.8533415794372559, + "num_tokens": 84059064.0, + "step": 2397 + }, + { + "epoch": 0.44531104921077064, + "grad_norm": 1.5178346633911133, + "learning_rate": 1e-06, + "loss": 0.4565, + "mean_token_accuracy": 0.8496292233467102, + "num_tokens": 84095780.0, + "step": 2398 + }, + { + "epoch": 0.44549675023212626, + "grad_norm": 1.448272943496704, + "learning_rate": 1e-06, + "loss": 0.4846, + "mean_token_accuracy": 0.8476003408432007, + "num_tokens": 84140585.0, + "step": 2399 + }, + { + "epoch": 0.4456824512534819, + "grad_norm": 1.5579348802566528, + "learning_rate": 1e-06, + "loss": 0.4367, + "mean_token_accuracy": 0.8578819632530212, + "num_tokens": 84175189.0, + "step": 2400 + }, + { + "epoch": 0.4458681522748375, + "grad_norm": 1.4839118719100952, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8613462448120117, + "num_tokens": 84211077.0, + "step": 2401 + }, + { + "epoch": 0.4460538532961931, + "grad_norm": 1.4871851205825806, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8738015294075012, + "num_tokens": 84247294.0, + "step": 2402 + }, + { + "epoch": 0.44623955431754875, + "grad_norm": 1.5346440076828003, + "learning_rate": 1e-06, + "loss": 0.4484, + "mean_token_accuracy": 0.8485682606697083, + "num_tokens": 84282364.0, + "step": 2403 + }, + { + "epoch": 0.44642525533890437, + "grad_norm": 1.414935827255249, + "learning_rate": 1e-06, + "loss": 0.4492, + "mean_token_accuracy": 0.8508847951889038, + "num_tokens": 84319968.0, + "step": 2404 + }, + { + "epoch": 0.44661095636026, + "grad_norm": 1.462037444114685, + "learning_rate": 1e-06, + "loss": 0.431, + "mean_token_accuracy": 0.8553879261016846, + "num_tokens": 84357640.0, + "step": 2405 + }, + { + "epoch": 0.4467966573816156, + "grad_norm": 1.4036312103271484, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8659964799880981, + "num_tokens": 84398075.0, + "step": 2406 + }, + { + "epoch": 0.44698235840297124, + "grad_norm": 1.5684242248535156, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.8678674697875977, + "num_tokens": 84432666.0, + "step": 2407 + }, + { + "epoch": 0.44716805942432686, + "grad_norm": 1.7552410364151, + "learning_rate": 1e-06, + "loss": 0.4473, + "mean_token_accuracy": 0.8473052978515625, + "num_tokens": 84460274.0, + "step": 2408 + }, + { + "epoch": 0.4473537604456824, + "grad_norm": 1.4788256883621216, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8748723268508911, + "num_tokens": 84495779.0, + "step": 2409 + }, + { + "epoch": 0.44753946146703805, + "grad_norm": 1.4725010395050049, + "learning_rate": 1e-06, + "loss": 0.4208, + "mean_token_accuracy": 0.8599086999893188, + "num_tokens": 84532505.0, + "step": 2410 + }, + { + "epoch": 0.44772516248839367, + "grad_norm": 1.402777910232544, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.860640287399292, + "num_tokens": 84569965.0, + "step": 2411 + }, + { + "epoch": 0.4479108635097493, + "grad_norm": 1.4406204223632812, + "learning_rate": 1e-06, + "loss": 0.425, + "mean_token_accuracy": 0.8544445633888245, + "num_tokens": 84606005.0, + "step": 2412 + }, + { + "epoch": 0.4480965645311049, + "grad_norm": 1.5826053619384766, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8610270023345947, + "num_tokens": 84637712.0, + "step": 2413 + }, + { + "epoch": 0.44828226555246053, + "grad_norm": 1.469897747039795, + "learning_rate": 1e-06, + "loss": 0.4353, + "mean_token_accuracy": 0.8547906279563904, + "num_tokens": 84674566.0, + "step": 2414 + }, + { + "epoch": 0.44846796657381616, + "grad_norm": 1.437757968902588, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8729946613311768, + "num_tokens": 84711478.0, + "step": 2415 + }, + { + "epoch": 0.4486536675951718, + "grad_norm": 1.5199230909347534, + "learning_rate": 1e-06, + "loss": 0.4448, + "mean_token_accuracy": 0.8530582189559937, + "num_tokens": 84748803.0, + "step": 2416 + }, + { + "epoch": 0.4488393686165274, + "grad_norm": 1.5731433629989624, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8531706929206848, + "num_tokens": 84781536.0, + "step": 2417 + }, + { + "epoch": 0.449025069637883, + "grad_norm": 1.459581732749939, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8691073656082153, + "num_tokens": 84815116.0, + "step": 2418 + }, + { + "epoch": 0.44921077065923865, + "grad_norm": 1.4146993160247803, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8706499338150024, + "num_tokens": 84851465.0, + "step": 2419 + }, + { + "epoch": 0.44939647168059427, + "grad_norm": 1.5684208869934082, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8573378324508667, + "num_tokens": 84882566.0, + "step": 2420 + }, + { + "epoch": 0.44958217270194983, + "grad_norm": 1.8440369367599487, + "learning_rate": 1e-06, + "loss": 0.4384, + "mean_token_accuracy": 0.8516747951507568, + "num_tokens": 84922264.0, + "step": 2421 + }, + { + "epoch": 0.44976787372330546, + "grad_norm": 1.4660608768463135, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8724751472473145, + "num_tokens": 84954684.0, + "step": 2422 + }, + { + "epoch": 0.4499535747446611, + "grad_norm": 1.4839510917663574, + "learning_rate": 1e-06, + "loss": 0.4195, + "mean_token_accuracy": 0.8560579419136047, + "num_tokens": 84991935.0, + "step": 2423 + }, + { + "epoch": 0.4501392757660167, + "grad_norm": 1.4339783191680908, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8773965835571289, + "num_tokens": 85025898.0, + "step": 2424 + }, + { + "epoch": 0.4503249767873723, + "grad_norm": 1.3732370138168335, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8771541118621826, + "num_tokens": 85058309.0, + "step": 2425 + }, + { + "epoch": 0.45051067780872794, + "grad_norm": 1.5452195405960083, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8660644292831421, + "num_tokens": 85089621.0, + "step": 2426 + }, + { + "epoch": 0.45069637883008357, + "grad_norm": 1.5176016092300415, + "learning_rate": 1e-06, + "loss": 0.4704, + "mean_token_accuracy": 0.8511195182800293, + "num_tokens": 85124840.0, + "step": 2427 + }, + { + "epoch": 0.4508820798514392, + "grad_norm": 1.4464876651763916, + "learning_rate": 1e-06, + "loss": 0.4081, + "mean_token_accuracy": 0.8589603900909424, + "num_tokens": 85160497.0, + "step": 2428 + }, + { + "epoch": 0.4510677808727948, + "grad_norm": 1.457090139389038, + "learning_rate": 1e-06, + "loss": 0.4097, + "mean_token_accuracy": 0.8637880086898804, + "num_tokens": 85196108.0, + "step": 2429 + }, + { + "epoch": 0.45125348189415043, + "grad_norm": 1.3451400995254517, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.8653224110603333, + "num_tokens": 85241019.0, + "step": 2430 + }, + { + "epoch": 0.45143918291550605, + "grad_norm": 1.5301971435546875, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8641135096549988, + "num_tokens": 85274553.0, + "step": 2431 + }, + { + "epoch": 0.4516248839368617, + "grad_norm": 1.464095950126648, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8640246391296387, + "num_tokens": 85312921.0, + "step": 2432 + }, + { + "epoch": 0.45181058495821724, + "grad_norm": 1.6216598749160767, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8703896999359131, + "num_tokens": 85340959.0, + "step": 2433 + }, + { + "epoch": 0.45199628597957286, + "grad_norm": 1.6088610887527466, + "learning_rate": 1e-06, + "loss": 0.4867, + "mean_token_accuracy": 0.8412518501281738, + "num_tokens": 85375545.0, + "step": 2434 + }, + { + "epoch": 0.4521819870009285, + "grad_norm": 1.3685050010681152, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8648180961608887, + "num_tokens": 85417054.0, + "step": 2435 + }, + { + "epoch": 0.4523676880222841, + "grad_norm": 1.6345781087875366, + "learning_rate": 1e-06, + "loss": 0.4745, + "mean_token_accuracy": 0.8464803099632263, + "num_tokens": 85451254.0, + "step": 2436 + }, + { + "epoch": 0.45255338904363973, + "grad_norm": 1.4866764545440674, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8580523133277893, + "num_tokens": 85488367.0, + "step": 2437 + }, + { + "epoch": 0.45273909006499535, + "grad_norm": 1.366472840309143, + "learning_rate": 1e-06, + "loss": 0.4291, + "mean_token_accuracy": 0.858551561832428, + "num_tokens": 85528661.0, + "step": 2438 + }, + { + "epoch": 0.452924791086351, + "grad_norm": 1.43303382396698, + "learning_rate": 1e-06, + "loss": 0.4363, + "mean_token_accuracy": 0.8558070659637451, + "num_tokens": 85568683.0, + "step": 2439 + }, + { + "epoch": 0.4531104921077066, + "grad_norm": 1.35835599899292, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.862135112285614, + "num_tokens": 85608526.0, + "step": 2440 + }, + { + "epoch": 0.4532961931290622, + "grad_norm": 1.5916708707809448, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8699842691421509, + "num_tokens": 85638267.0, + "step": 2441 + }, + { + "epoch": 0.45348189415041784, + "grad_norm": 1.548505187034607, + "learning_rate": 1e-06, + "loss": 0.4042, + "mean_token_accuracy": 0.8656298518180847, + "num_tokens": 85667550.0, + "step": 2442 + }, + { + "epoch": 0.45366759517177346, + "grad_norm": 1.4688242673873901, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8622063994407654, + "num_tokens": 85705218.0, + "step": 2443 + }, + { + "epoch": 0.4538532961931291, + "grad_norm": 1.4861711263656616, + "learning_rate": 1e-06, + "loss": 0.4235, + "mean_token_accuracy": 0.8582806587219238, + "num_tokens": 85741994.0, + "step": 2444 + }, + { + "epoch": 0.45403899721448465, + "grad_norm": 1.465917944908142, + "learning_rate": 1e-06, + "loss": 0.471, + "mean_token_accuracy": 0.8448015451431274, + "num_tokens": 85779255.0, + "step": 2445 + }, + { + "epoch": 0.4542246982358403, + "grad_norm": 1.3868238925933838, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8684431314468384, + "num_tokens": 85817858.0, + "step": 2446 + }, + { + "epoch": 0.4544103992571959, + "grad_norm": 1.4095103740692139, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.858763575553894, + "num_tokens": 85859989.0, + "step": 2447 + }, + { + "epoch": 0.4545961002785515, + "grad_norm": 1.5200406312942505, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.8638933897018433, + "num_tokens": 85892561.0, + "step": 2448 + }, + { + "epoch": 0.45478180129990714, + "grad_norm": 1.4528388977050781, + "learning_rate": 1e-06, + "loss": 0.4644, + "mean_token_accuracy": 0.8475608229637146, + "num_tokens": 85931818.0, + "step": 2449 + }, + { + "epoch": 0.45496750232126276, + "grad_norm": 1.4964499473571777, + "learning_rate": 1e-06, + "loss": 0.4148, + "mean_token_accuracy": 0.861017107963562, + "num_tokens": 85970561.0, + "step": 2450 + }, + { + "epoch": 0.4551532033426184, + "grad_norm": 1.4655119180679321, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8630103468894958, + "num_tokens": 86008265.0, + "step": 2451 + }, + { + "epoch": 0.455338904363974, + "grad_norm": 1.4757204055786133, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8648664951324463, + "num_tokens": 86046274.0, + "step": 2452 + }, + { + "epoch": 0.45552460538532963, + "grad_norm": 1.8834272623062134, + "learning_rate": 1e-06, + "loss": 0.4515, + "mean_token_accuracy": 0.8559134602546692, + "num_tokens": 86070354.0, + "step": 2453 + }, + { + "epoch": 0.45571030640668525, + "grad_norm": 1.5329300165176392, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8609356880187988, + "num_tokens": 86104405.0, + "step": 2454 + }, + { + "epoch": 0.4558960074280409, + "grad_norm": 1.4656931161880493, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.863172173500061, + "num_tokens": 86139980.0, + "step": 2455 + }, + { + "epoch": 0.4560817084493965, + "grad_norm": 1.4853748083114624, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8671835064888, + "num_tokens": 86174079.0, + "step": 2456 + }, + { + "epoch": 0.4562674094707521, + "grad_norm": 1.450992465019226, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8868921995162964, + "num_tokens": 86206672.0, + "step": 2457 + }, + { + "epoch": 0.4564531104921077, + "grad_norm": 1.5911920070648193, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8615017533302307, + "num_tokens": 86239610.0, + "step": 2458 + }, + { + "epoch": 0.4566388115134633, + "grad_norm": 1.5610053539276123, + "learning_rate": 1e-06, + "loss": 0.4045, + "mean_token_accuracy": 0.8651065230369568, + "num_tokens": 86272787.0, + "step": 2459 + }, + { + "epoch": 0.4568245125348189, + "grad_norm": 1.468224287033081, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8675971031188965, + "num_tokens": 86312524.0, + "step": 2460 + }, + { + "epoch": 0.45701021355617455, + "grad_norm": 1.5340744256973267, + "learning_rate": 1e-06, + "loss": 0.403, + "mean_token_accuracy": 0.8631789684295654, + "num_tokens": 86345430.0, + "step": 2461 + }, + { + "epoch": 0.45719591457753017, + "grad_norm": 1.5933054685592651, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.861029326915741, + "num_tokens": 86376382.0, + "step": 2462 + }, + { + "epoch": 0.4573816155988858, + "grad_norm": 1.4776517152786255, + "learning_rate": 1e-06, + "loss": 0.4388, + "mean_token_accuracy": 0.85845947265625, + "num_tokens": 86411668.0, + "step": 2463 + }, + { + "epoch": 0.4575673166202414, + "grad_norm": 1.4907588958740234, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8591699600219727, + "num_tokens": 86446445.0, + "step": 2464 + }, + { + "epoch": 0.45775301764159704, + "grad_norm": 1.4242488145828247, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8604831695556641, + "num_tokens": 86484152.0, + "step": 2465 + }, + { + "epoch": 0.45793871866295266, + "grad_norm": 1.541614055633545, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.8525397181510925, + "num_tokens": 86518068.0, + "step": 2466 + }, + { + "epoch": 0.4581244196843083, + "grad_norm": 1.5524877309799194, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8771829605102539, + "num_tokens": 86550171.0, + "step": 2467 + }, + { + "epoch": 0.4583101207056639, + "grad_norm": 1.6289595365524292, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8762025237083435, + "num_tokens": 86577555.0, + "step": 2468 + }, + { + "epoch": 0.4584958217270195, + "grad_norm": 1.4377392530441284, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.86209636926651, + "num_tokens": 86616589.0, + "step": 2469 + }, + { + "epoch": 0.4586815227483751, + "grad_norm": 1.4372398853302002, + "learning_rate": 1e-06, + "loss": 0.4151, + "mean_token_accuracy": 0.8609991669654846, + "num_tokens": 86651797.0, + "step": 2470 + }, + { + "epoch": 0.4588672237697307, + "grad_norm": 1.3643382787704468, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.867077112197876, + "num_tokens": 86692573.0, + "step": 2471 + }, + { + "epoch": 0.45905292479108634, + "grad_norm": 1.390125036239624, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.865172266960144, + "num_tokens": 86731219.0, + "step": 2472 + }, + { + "epoch": 0.45923862581244196, + "grad_norm": 1.5582287311553955, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8745397925376892, + "num_tokens": 86765647.0, + "step": 2473 + }, + { + "epoch": 0.4594243268337976, + "grad_norm": 1.66092050075531, + "learning_rate": 1e-06, + "loss": 0.4811, + "mean_token_accuracy": 0.8412500619888306, + "num_tokens": 86799524.0, + "step": 2474 + }, + { + "epoch": 0.4596100278551532, + "grad_norm": 1.5855779647827148, + "learning_rate": 1e-06, + "loss": 0.4401, + "mean_token_accuracy": 0.8498329520225525, + "num_tokens": 86830322.0, + "step": 2475 + }, + { + "epoch": 0.4597957288765088, + "grad_norm": 1.4401882886886597, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.862006664276123, + "num_tokens": 86867697.0, + "step": 2476 + }, + { + "epoch": 0.45998142989786445, + "grad_norm": 1.5627820491790771, + "learning_rate": 1e-06, + "loss": 0.4797, + "mean_token_accuracy": 0.8436713218688965, + "num_tokens": 86901352.0, + "step": 2477 + }, + { + "epoch": 0.46016713091922007, + "grad_norm": 1.4606778621673584, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8701509833335876, + "num_tokens": 86933307.0, + "step": 2478 + }, + { + "epoch": 0.4603528319405757, + "grad_norm": 1.5053160190582275, + "learning_rate": 1e-06, + "loss": 0.4355, + "mean_token_accuracy": 0.853316605091095, + "num_tokens": 86970139.0, + "step": 2479 + }, + { + "epoch": 0.4605385329619313, + "grad_norm": 1.4795019626617432, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.8661980032920837, + "num_tokens": 87002715.0, + "step": 2480 + }, + { + "epoch": 0.46072423398328693, + "grad_norm": 1.3801062107086182, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8777704834938049, + "num_tokens": 87039488.0, + "step": 2481 + }, + { + "epoch": 0.4609099350046425, + "grad_norm": 1.383520483970642, + "learning_rate": 1e-06, + "loss": 0.3967, + "mean_token_accuracy": 0.8669884204864502, + "num_tokens": 87077529.0, + "step": 2482 + }, + { + "epoch": 0.4610956360259981, + "grad_norm": 1.5298317670822144, + "learning_rate": 1e-06, + "loss": 0.4421, + "mean_token_accuracy": 0.850768506526947, + "num_tokens": 87113123.0, + "step": 2483 + }, + { + "epoch": 0.46128133704735375, + "grad_norm": 1.4836244583129883, + "learning_rate": 1e-06, + "loss": 0.4617, + "mean_token_accuracy": 0.8476598858833313, + "num_tokens": 87152831.0, + "step": 2484 + }, + { + "epoch": 0.46146703806870937, + "grad_norm": 1.5131338834762573, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.858535647392273, + "num_tokens": 87189203.0, + "step": 2485 + }, + { + "epoch": 0.461652739090065, + "grad_norm": 1.356130838394165, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.8616273999214172, + "num_tokens": 87230056.0, + "step": 2486 + }, + { + "epoch": 0.4618384401114206, + "grad_norm": 1.5865458250045776, + "learning_rate": 1e-06, + "loss": 0.4497, + "mean_token_accuracy": 0.8513744473457336, + "num_tokens": 87262553.0, + "step": 2487 + }, + { + "epoch": 0.46202414113277623, + "grad_norm": 1.6070624589920044, + "learning_rate": 1e-06, + "loss": 0.4379, + "mean_token_accuracy": 0.8567514419555664, + "num_tokens": 87295500.0, + "step": 2488 + }, + { + "epoch": 0.46220984215413186, + "grad_norm": 1.5897507667541504, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8790621757507324, + "num_tokens": 87324504.0, + "step": 2489 + }, + { + "epoch": 0.4623955431754875, + "grad_norm": 1.4487372636795044, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.8654292821884155, + "num_tokens": 87360722.0, + "step": 2490 + }, + { + "epoch": 0.4625812441968431, + "grad_norm": 1.5613819360733032, + "learning_rate": 1e-06, + "loss": 0.4473, + "mean_token_accuracy": 0.8540773391723633, + "num_tokens": 87393896.0, + "step": 2491 + }, + { + "epoch": 0.4627669452181987, + "grad_norm": 1.3599189519882202, + "learning_rate": 1e-06, + "loss": 0.4698, + "mean_token_accuracy": 0.8440497517585754, + "num_tokens": 87437587.0, + "step": 2492 + }, + { + "epoch": 0.46295264623955434, + "grad_norm": 1.573360800743103, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8686662912368774, + "num_tokens": 87469324.0, + "step": 2493 + }, + { + "epoch": 0.4631383472609099, + "grad_norm": 1.481627106666565, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8583273887634277, + "num_tokens": 87503782.0, + "step": 2494 + }, + { + "epoch": 0.46332404828226553, + "grad_norm": 1.566823959350586, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8590216040611267, + "num_tokens": 87535446.0, + "step": 2495 + }, + { + "epoch": 0.46350974930362115, + "grad_norm": 1.5659066438674927, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8617625832557678, + "num_tokens": 87566345.0, + "step": 2496 + }, + { + "epoch": 0.4636954503249768, + "grad_norm": 1.5369820594787598, + "learning_rate": 1e-06, + "loss": 0.396, + "mean_token_accuracy": 0.8692784309387207, + "num_tokens": 87600590.0, + "step": 2497 + }, + { + "epoch": 0.4638811513463324, + "grad_norm": 1.5154471397399902, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8597373366355896, + "num_tokens": 87632883.0, + "step": 2498 + }, + { + "epoch": 0.464066852367688, + "grad_norm": 1.5177812576293945, + "learning_rate": 1e-06, + "loss": 0.4583, + "mean_token_accuracy": 0.8448863625526428, + "num_tokens": 87667312.0, + "step": 2499 + }, + { + "epoch": 0.46425255338904364, + "grad_norm": 1.5454360246658325, + "learning_rate": 1e-06, + "loss": 0.4154, + "mean_token_accuracy": 0.8646376132965088, + "num_tokens": 87702301.0, + "step": 2500 + }, + { + "epoch": 0.46443825441039926, + "grad_norm": 1.494596242904663, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8711572885513306, + "num_tokens": 87740715.0, + "step": 2501 + }, + { + "epoch": 0.4646239554317549, + "grad_norm": 1.4031568765640259, + "learning_rate": 1e-06, + "loss": 0.4254, + "mean_token_accuracy": 0.8590559363365173, + "num_tokens": 87780473.0, + "step": 2502 + }, + { + "epoch": 0.4648096564531105, + "grad_norm": 1.5804309844970703, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8664569854736328, + "num_tokens": 87814160.0, + "step": 2503 + }, + { + "epoch": 0.46499535747446613, + "grad_norm": 1.6165639162063599, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.872130811214447, + "num_tokens": 87844265.0, + "step": 2504 + }, + { + "epoch": 0.46518105849582175, + "grad_norm": 1.4245567321777344, + "learning_rate": 1e-06, + "loss": 0.4624, + "mean_token_accuracy": 0.8486779928207397, + "num_tokens": 87883258.0, + "step": 2505 + }, + { + "epoch": 0.4653667595171773, + "grad_norm": 1.4393160343170166, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8719208240509033, + "num_tokens": 87919815.0, + "step": 2506 + }, + { + "epoch": 0.46555246053853294, + "grad_norm": 1.425118327140808, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8729130029678345, + "num_tokens": 87961170.0, + "step": 2507 + }, + { + "epoch": 0.46573816155988856, + "grad_norm": 1.6091951131820679, + "learning_rate": 1e-06, + "loss": 0.4597, + "mean_token_accuracy": 0.8387770652770996, + "num_tokens": 87992920.0, + "step": 2508 + }, + { + "epoch": 0.4659238625812442, + "grad_norm": 1.582646369934082, + "learning_rate": 1e-06, + "loss": 0.4638, + "mean_token_accuracy": 0.8474709391593933, + "num_tokens": 88028621.0, + "step": 2509 + }, + { + "epoch": 0.4661095636025998, + "grad_norm": 1.5337815284729004, + "learning_rate": 1e-06, + "loss": 0.4664, + "mean_token_accuracy": 0.8516600131988525, + "num_tokens": 88065756.0, + "step": 2510 + }, + { + "epoch": 0.46629526462395543, + "grad_norm": 1.554282546043396, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8742197155952454, + "num_tokens": 88098980.0, + "step": 2511 + }, + { + "epoch": 0.46648096564531105, + "grad_norm": 1.4044800996780396, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8706529140472412, + "num_tokens": 88138401.0, + "step": 2512 + }, + { + "epoch": 0.4666666666666667, + "grad_norm": 1.5884159803390503, + "learning_rate": 1e-06, + "loss": 0.4358, + "mean_token_accuracy": 0.8559880256652832, + "num_tokens": 88174361.0, + "step": 2513 + }, + { + "epoch": 0.4668523676880223, + "grad_norm": 1.3940374851226807, + "learning_rate": 1e-06, + "loss": 0.4086, + "mean_token_accuracy": 0.8620785474777222, + "num_tokens": 88213050.0, + "step": 2514 + }, + { + "epoch": 0.4670380687093779, + "grad_norm": 1.4849919080734253, + "learning_rate": 1e-06, + "loss": 0.4577, + "mean_token_accuracy": 0.8505963087081909, + "num_tokens": 88251681.0, + "step": 2515 + }, + { + "epoch": 0.46722376973073354, + "grad_norm": 1.5279732942581177, + "learning_rate": 1e-06, + "loss": 0.4585, + "mean_token_accuracy": 0.8537750840187073, + "num_tokens": 88287598.0, + "step": 2516 + }, + { + "epoch": 0.46740947075208916, + "grad_norm": 1.346777319908142, + "learning_rate": 1e-06, + "loss": 0.4, + "mean_token_accuracy": 0.8661630153656006, + "num_tokens": 88327701.0, + "step": 2517 + }, + { + "epoch": 0.46759517177344473, + "grad_norm": 1.4722541570663452, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8712408542633057, + "num_tokens": 88364983.0, + "step": 2518 + }, + { + "epoch": 0.46778087279480035, + "grad_norm": 1.3367047309875488, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8623330593109131, + "num_tokens": 88404276.0, + "step": 2519 + }, + { + "epoch": 0.467966573816156, + "grad_norm": 1.4007004499435425, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.868821918964386, + "num_tokens": 88441822.0, + "step": 2520 + }, + { + "epoch": 0.4681522748375116, + "grad_norm": 1.5977963209152222, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8749204874038696, + "num_tokens": 88469294.0, + "step": 2521 + }, + { + "epoch": 0.4683379758588672, + "grad_norm": 1.4257557392120361, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8577686548233032, + "num_tokens": 88508481.0, + "step": 2522 + }, + { + "epoch": 0.46852367688022284, + "grad_norm": 1.501001000404358, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.8686826229095459, + "num_tokens": 88541930.0, + "step": 2523 + }, + { + "epoch": 0.46870937790157846, + "grad_norm": 1.3211603164672852, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8742642998695374, + "num_tokens": 88580294.0, + "step": 2524 + }, + { + "epoch": 0.4688950789229341, + "grad_norm": 1.3363105058670044, + "learning_rate": 1e-06, + "loss": 0.4595, + "mean_token_accuracy": 0.8477613925933838, + "num_tokens": 88624146.0, + "step": 2525 + }, + { + "epoch": 0.4690807799442897, + "grad_norm": 1.5119783878326416, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8661558628082275, + "num_tokens": 88656482.0, + "step": 2526 + }, + { + "epoch": 0.4692664809656453, + "grad_norm": 1.4647839069366455, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8610567450523376, + "num_tokens": 88693777.0, + "step": 2527 + }, + { + "epoch": 0.46945218198700095, + "grad_norm": 1.6524624824523926, + "learning_rate": 1e-06, + "loss": 0.4516, + "mean_token_accuracy": 0.8476250171661377, + "num_tokens": 88725225.0, + "step": 2528 + }, + { + "epoch": 0.46963788300835657, + "grad_norm": 1.448157548904419, + "learning_rate": 1e-06, + "loss": 0.4612, + "mean_token_accuracy": 0.8507533073425293, + "num_tokens": 88765967.0, + "step": 2529 + }, + { + "epoch": 0.46982358402971214, + "grad_norm": 1.5165479183197021, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8547589778900146, + "num_tokens": 88801961.0, + "step": 2530 + }, + { + "epoch": 0.47000928505106776, + "grad_norm": 1.2470537424087524, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8810343146324158, + "num_tokens": 88844085.0, + "step": 2531 + }, + { + "epoch": 0.4701949860724234, + "grad_norm": 1.4075225591659546, + "learning_rate": 1e-06, + "loss": 0.429, + "mean_token_accuracy": 0.8547277450561523, + "num_tokens": 88885741.0, + "step": 2532 + }, + { + "epoch": 0.470380687093779, + "grad_norm": 1.5414222478866577, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.8569295406341553, + "num_tokens": 88918566.0, + "step": 2533 + }, + { + "epoch": 0.4705663881151346, + "grad_norm": 1.4784563779830933, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8768579363822937, + "num_tokens": 88952485.0, + "step": 2534 + }, + { + "epoch": 0.47075208913649025, + "grad_norm": 1.4678250551223755, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8705698847770691, + "num_tokens": 88988901.0, + "step": 2535 + }, + { + "epoch": 0.47093779015784587, + "grad_norm": 1.5004098415374756, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.868446946144104, + "num_tokens": 89023322.0, + "step": 2536 + }, + { + "epoch": 0.4711234911792015, + "grad_norm": 1.4914199113845825, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8707663416862488, + "num_tokens": 89058592.0, + "step": 2537 + }, + { + "epoch": 0.4713091922005571, + "grad_norm": 1.701850414276123, + "learning_rate": 1e-06, + "loss": 0.445, + "mean_token_accuracy": 0.8524088859558105, + "num_tokens": 89084615.0, + "step": 2538 + }, + { + "epoch": 0.47149489322191274, + "grad_norm": 1.4684958457946777, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8711682558059692, + "num_tokens": 89123029.0, + "step": 2539 + }, + { + "epoch": 0.47168059424326836, + "grad_norm": 1.4813146591186523, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.8696566820144653, + "num_tokens": 89157853.0, + "step": 2540 + }, + { + "epoch": 0.471866295264624, + "grad_norm": 1.6928389072418213, + "learning_rate": 1e-06, + "loss": 0.4759, + "mean_token_accuracy": 0.8508162498474121, + "num_tokens": 89188584.0, + "step": 2541 + }, + { + "epoch": 0.47205199628597955, + "grad_norm": 1.5039715766906738, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.864319920539856, + "num_tokens": 89222000.0, + "step": 2542 + }, + { + "epoch": 0.47223769730733517, + "grad_norm": 1.5124908685684204, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.866178572177887, + "num_tokens": 89259480.0, + "step": 2543 + }, + { + "epoch": 0.4724233983286908, + "grad_norm": 1.4612921476364136, + "learning_rate": 1e-06, + "loss": 0.4678, + "mean_token_accuracy": 0.8527716398239136, + "num_tokens": 89303171.0, + "step": 2544 + }, + { + "epoch": 0.4726090993500464, + "grad_norm": 1.4997532367706299, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8669683933258057, + "num_tokens": 89337761.0, + "step": 2545 + }, + { + "epoch": 0.47279480037140204, + "grad_norm": 1.5933492183685303, + "learning_rate": 1e-06, + "loss": 0.446, + "mean_token_accuracy": 0.8528951406478882, + "num_tokens": 89369936.0, + "step": 2546 + }, + { + "epoch": 0.47298050139275766, + "grad_norm": 1.634321928024292, + "learning_rate": 1e-06, + "loss": 0.423, + "mean_token_accuracy": 0.8590560555458069, + "num_tokens": 89404552.0, + "step": 2547 + }, + { + "epoch": 0.4731662024141133, + "grad_norm": 1.4682289361953735, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8612713813781738, + "num_tokens": 89444373.0, + "step": 2548 + }, + { + "epoch": 0.4733519034354689, + "grad_norm": 1.4707536697387695, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8680206537246704, + "num_tokens": 89477084.0, + "step": 2549 + }, + { + "epoch": 0.4735376044568245, + "grad_norm": 1.56661057472229, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8638094067573547, + "num_tokens": 89510667.0, + "step": 2550 + }, + { + "epoch": 0.47372330547818015, + "grad_norm": 1.400651454925537, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8576402068138123, + "num_tokens": 89551059.0, + "step": 2551 + }, + { + "epoch": 0.47390900649953577, + "grad_norm": 1.363368034362793, + "learning_rate": 1e-06, + "loss": 0.4267, + "mean_token_accuracy": 0.8545540571212769, + "num_tokens": 89588848.0, + "step": 2552 + }, + { + "epoch": 0.4740947075208914, + "grad_norm": 1.449669599533081, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8740086555480957, + "num_tokens": 89624074.0, + "step": 2553 + }, + { + "epoch": 0.47428040854224696, + "grad_norm": 1.5449193716049194, + "learning_rate": 1e-06, + "loss": 0.4564, + "mean_token_accuracy": 0.8480198383331299, + "num_tokens": 89656793.0, + "step": 2554 + }, + { + "epoch": 0.4744661095636026, + "grad_norm": 1.5113657712936401, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8689112663269043, + "num_tokens": 89689228.0, + "step": 2555 + }, + { + "epoch": 0.4746518105849582, + "grad_norm": 1.5010606050491333, + "learning_rate": 1e-06, + "loss": 0.4459, + "mean_token_accuracy": 0.853752613067627, + "num_tokens": 89725217.0, + "step": 2556 + }, + { + "epoch": 0.4748375116063138, + "grad_norm": 1.641926646232605, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8545328378677368, + "num_tokens": 89757293.0, + "step": 2557 + }, + { + "epoch": 0.47502321262766944, + "grad_norm": 1.5911822319030762, + "learning_rate": 1e-06, + "loss": 0.4391, + "mean_token_accuracy": 0.8576808571815491, + "num_tokens": 89788289.0, + "step": 2558 + }, + { + "epoch": 0.47520891364902507, + "grad_norm": 1.3912014961242676, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.859630823135376, + "num_tokens": 89827484.0, + "step": 2559 + }, + { + "epoch": 0.4753946146703807, + "grad_norm": 1.4709088802337646, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8711508512496948, + "num_tokens": 89858870.0, + "step": 2560 + }, + { + "epoch": 0.4755803156917363, + "grad_norm": 1.559514045715332, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8606584668159485, + "num_tokens": 89891477.0, + "step": 2561 + }, + { + "epoch": 0.47576601671309193, + "grad_norm": 1.3934398889541626, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.868933916091919, + "num_tokens": 89930420.0, + "step": 2562 + }, + { + "epoch": 0.47595171773444755, + "grad_norm": 1.6340734958648682, + "learning_rate": 1e-06, + "loss": 0.4926, + "mean_token_accuracy": 0.8411630392074585, + "num_tokens": 89964968.0, + "step": 2563 + }, + { + "epoch": 0.4761374187558032, + "grad_norm": 1.4596500396728516, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8649501800537109, + "num_tokens": 89998630.0, + "step": 2564 + }, + { + "epoch": 0.4763231197771588, + "grad_norm": 1.5405915975570679, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8689857721328735, + "num_tokens": 90026850.0, + "step": 2565 + }, + { + "epoch": 0.47650882079851437, + "grad_norm": 2.0400633811950684, + "learning_rate": 1e-06, + "loss": 0.4255, + "mean_token_accuracy": 0.8650224208831787, + "num_tokens": 90062174.0, + "step": 2566 + }, + { + "epoch": 0.47669452181987, + "grad_norm": 1.3834760189056396, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8882327079772949, + "num_tokens": 90099013.0, + "step": 2567 + }, + { + "epoch": 0.4768802228412256, + "grad_norm": 1.6574649810791016, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8672301769256592, + "num_tokens": 90126769.0, + "step": 2568 + }, + { + "epoch": 0.47706592386258123, + "grad_norm": 1.6811808347702026, + "learning_rate": 1e-06, + "loss": 0.4393, + "mean_token_accuracy": 0.8570358157157898, + "num_tokens": 90155839.0, + "step": 2569 + }, + { + "epoch": 0.47725162488393685, + "grad_norm": 1.4058279991149902, + "learning_rate": 1e-06, + "loss": 0.4289, + "mean_token_accuracy": 0.8599660992622375, + "num_tokens": 90194635.0, + "step": 2570 + }, + { + "epoch": 0.4774373259052925, + "grad_norm": 1.4678170680999756, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8626461029052734, + "num_tokens": 90230436.0, + "step": 2571 + }, + { + "epoch": 0.4776230269266481, + "grad_norm": 1.4298079013824463, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8744279146194458, + "num_tokens": 90266342.0, + "step": 2572 + }, + { + "epoch": 0.4778087279480037, + "grad_norm": 1.6594334840774536, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.873316764831543, + "num_tokens": 90294995.0, + "step": 2573 + }, + { + "epoch": 0.47799442896935934, + "grad_norm": 1.5744959115982056, + "learning_rate": 1e-06, + "loss": 0.4571, + "mean_token_accuracy": 0.8502171039581299, + "num_tokens": 90328771.0, + "step": 2574 + }, + { + "epoch": 0.47818012999071496, + "grad_norm": 1.388726830482483, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.8727038502693176, + "num_tokens": 90365287.0, + "step": 2575 + }, + { + "epoch": 0.4783658310120706, + "grad_norm": 1.365822434425354, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8570282459259033, + "num_tokens": 90407253.0, + "step": 2576 + }, + { + "epoch": 0.4785515320334262, + "grad_norm": 1.4160575866699219, + "learning_rate": 1e-06, + "loss": 0.4059, + "mean_token_accuracy": 0.8591973185539246, + "num_tokens": 90446214.0, + "step": 2577 + }, + { + "epoch": 0.4787372330547818, + "grad_norm": 1.4974498748779297, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8597740530967712, + "num_tokens": 90481899.0, + "step": 2578 + }, + { + "epoch": 0.4789229340761374, + "grad_norm": 1.429589033126831, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8692036867141724, + "num_tokens": 90517877.0, + "step": 2579 + }, + { + "epoch": 0.479108635097493, + "grad_norm": 1.4435206651687622, + "learning_rate": 1e-06, + "loss": 0.4068, + "mean_token_accuracy": 0.8655058145523071, + "num_tokens": 90554423.0, + "step": 2580 + }, + { + "epoch": 0.47929433611884864, + "grad_norm": 1.5820152759552002, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.856938362121582, + "num_tokens": 90584142.0, + "step": 2581 + }, + { + "epoch": 0.47948003714020426, + "grad_norm": 1.4213058948516846, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8642102479934692, + "num_tokens": 90627728.0, + "step": 2582 + }, + { + "epoch": 0.4796657381615599, + "grad_norm": 1.4971308708190918, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8627070188522339, + "num_tokens": 90663902.0, + "step": 2583 + }, + { + "epoch": 0.4798514391829155, + "grad_norm": 1.541616678237915, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8753902912139893, + "num_tokens": 90695463.0, + "step": 2584 + }, + { + "epoch": 0.48003714020427113, + "grad_norm": 1.3784794807434082, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8617372512817383, + "num_tokens": 90735310.0, + "step": 2585 + }, + { + "epoch": 0.48022284122562675, + "grad_norm": 1.5275440216064453, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8550150394439697, + "num_tokens": 90770766.0, + "step": 2586 + }, + { + "epoch": 0.4804085422469824, + "grad_norm": 1.4483306407928467, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8646427392959595, + "num_tokens": 90805954.0, + "step": 2587 + }, + { + "epoch": 0.480594243268338, + "grad_norm": 1.613889217376709, + "learning_rate": 1e-06, + "loss": 0.482, + "mean_token_accuracy": 0.8385450839996338, + "num_tokens": 90840515.0, + "step": 2588 + }, + { + "epoch": 0.4807799442896936, + "grad_norm": 1.441367268562317, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8708995580673218, + "num_tokens": 90873955.0, + "step": 2589 + }, + { + "epoch": 0.4809656453110492, + "grad_norm": 1.4434128999710083, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8741246461868286, + "num_tokens": 90911979.0, + "step": 2590 + }, + { + "epoch": 0.4811513463324048, + "grad_norm": 1.6191387176513672, + "learning_rate": 1e-06, + "loss": 0.4365, + "mean_token_accuracy": 0.8540451526641846, + "num_tokens": 90946614.0, + "step": 2591 + }, + { + "epoch": 0.4813370473537604, + "grad_norm": 1.5657039880752563, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8618946075439453, + "num_tokens": 90981730.0, + "step": 2592 + }, + { + "epoch": 0.48152274837511605, + "grad_norm": 1.619057536125183, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8674812912940979, + "num_tokens": 91008920.0, + "step": 2593 + }, + { + "epoch": 0.48170844939647167, + "grad_norm": 1.4293707609176636, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8764635324478149, + "num_tokens": 91045383.0, + "step": 2594 + }, + { + "epoch": 0.4818941504178273, + "grad_norm": 1.4820388555526733, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8668786287307739, + "num_tokens": 91080017.0, + "step": 2595 + }, + { + "epoch": 0.4820798514391829, + "grad_norm": 1.50338876247406, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8747151494026184, + "num_tokens": 91114156.0, + "step": 2596 + }, + { + "epoch": 0.48226555246053854, + "grad_norm": 1.5121597051620483, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8705427646636963, + "num_tokens": 91149840.0, + "step": 2597 + }, + { + "epoch": 0.48245125348189416, + "grad_norm": 1.3979123830795288, + "learning_rate": 1e-06, + "loss": 0.4049, + "mean_token_accuracy": 0.8650203943252563, + "num_tokens": 91187827.0, + "step": 2598 + }, + { + "epoch": 0.4826369545032498, + "grad_norm": 1.6654176712036133, + "learning_rate": 1e-06, + "loss": 0.4359, + "mean_token_accuracy": 0.8545058965682983, + "num_tokens": 91217123.0, + "step": 2599 + }, + { + "epoch": 0.4828226555246054, + "grad_norm": 1.504232406616211, + "learning_rate": 1e-06, + "loss": 0.4786, + "mean_token_accuracy": 0.8413379192352295, + "num_tokens": 91256598.0, + "step": 2600 + }, + { + "epoch": 0.483008356545961, + "grad_norm": 1.4664846658706665, + "learning_rate": 1e-06, + "loss": 0.4403, + "mean_token_accuracy": 0.8578125238418579, + "num_tokens": 91294013.0, + "step": 2601 + }, + { + "epoch": 0.4831940575673166, + "grad_norm": 1.618309497833252, + "learning_rate": 1e-06, + "loss": 0.4542, + "mean_token_accuracy": 0.8493440747261047, + "num_tokens": 91325126.0, + "step": 2602 + }, + { + "epoch": 0.4833797585886722, + "grad_norm": 1.6653375625610352, + "learning_rate": 1e-06, + "loss": 0.396, + "mean_token_accuracy": 0.8667588233947754, + "num_tokens": 91354811.0, + "step": 2603 + }, + { + "epoch": 0.48356545961002784, + "grad_norm": 1.516506314277649, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8637232184410095, + "num_tokens": 91390679.0, + "step": 2604 + }, + { + "epoch": 0.48375116063138346, + "grad_norm": 1.5724977254867554, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8599812388420105, + "num_tokens": 91420825.0, + "step": 2605 + }, + { + "epoch": 0.4839368616527391, + "grad_norm": 1.5889486074447632, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.861338198184967, + "num_tokens": 91453353.0, + "step": 2606 + }, + { + "epoch": 0.4841225626740947, + "grad_norm": 1.4551883935928345, + "learning_rate": 1e-06, + "loss": 0.4297, + "mean_token_accuracy": 0.8582713603973389, + "num_tokens": 91490965.0, + "step": 2607 + }, + { + "epoch": 0.4843082636954503, + "grad_norm": 1.5152446031570435, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8635826110839844, + "num_tokens": 91523218.0, + "step": 2608 + }, + { + "epoch": 0.48449396471680595, + "grad_norm": 1.565429449081421, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.871026337146759, + "num_tokens": 91557345.0, + "step": 2609 + }, + { + "epoch": 0.48467966573816157, + "grad_norm": 1.6530617475509644, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8450087308883667, + "num_tokens": 91587523.0, + "step": 2610 + }, + { + "epoch": 0.4848653667595172, + "grad_norm": 1.4197094440460205, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8739557266235352, + "num_tokens": 91621338.0, + "step": 2611 + }, + { + "epoch": 0.4850510677808728, + "grad_norm": 1.4980764389038086, + "learning_rate": 1e-06, + "loss": 0.4215, + "mean_token_accuracy": 0.8565571308135986, + "num_tokens": 91655547.0, + "step": 2612 + }, + { + "epoch": 0.48523676880222844, + "grad_norm": 1.4939073324203491, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8676969408988953, + "num_tokens": 91685744.0, + "step": 2613 + }, + { + "epoch": 0.48542246982358406, + "grad_norm": 1.7670314311981201, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8786425590515137, + "num_tokens": 91711539.0, + "step": 2614 + }, + { + "epoch": 0.4856081708449396, + "grad_norm": 1.6390371322631836, + "learning_rate": 1e-06, + "loss": 0.4432, + "mean_token_accuracy": 0.8534732460975647, + "num_tokens": 91743283.0, + "step": 2615 + }, + { + "epoch": 0.48579387186629525, + "grad_norm": 1.478054165840149, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8738813400268555, + "num_tokens": 91774845.0, + "step": 2616 + }, + { + "epoch": 0.48597957288765087, + "grad_norm": 1.397748589515686, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8707140684127808, + "num_tokens": 91813170.0, + "step": 2617 + }, + { + "epoch": 0.4861652739090065, + "grad_norm": 1.5856060981750488, + "learning_rate": 1e-06, + "loss": 0.4842, + "mean_token_accuracy": 0.842841386795044, + "num_tokens": 91851604.0, + "step": 2618 + }, + { + "epoch": 0.4863509749303621, + "grad_norm": 1.5889742374420166, + "learning_rate": 1e-06, + "loss": 0.4213, + "mean_token_accuracy": 0.8584812879562378, + "num_tokens": 91881815.0, + "step": 2619 + }, + { + "epoch": 0.48653667595171773, + "grad_norm": 1.4204061031341553, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8725164532661438, + "num_tokens": 91917795.0, + "step": 2620 + }, + { + "epoch": 0.48672237697307336, + "grad_norm": 1.4666719436645508, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8609468340873718, + "num_tokens": 91952763.0, + "step": 2621 + }, + { + "epoch": 0.486908077994429, + "grad_norm": 1.3834543228149414, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8617545962333679, + "num_tokens": 91990043.0, + "step": 2622 + }, + { + "epoch": 0.4870937790157846, + "grad_norm": 1.5457967519760132, + "learning_rate": 1e-06, + "loss": 0.429, + "mean_token_accuracy": 0.8603996634483337, + "num_tokens": 92026959.0, + "step": 2623 + }, + { + "epoch": 0.4872794800371402, + "grad_norm": 1.4135406017303467, + "learning_rate": 1e-06, + "loss": 0.4441, + "mean_token_accuracy": 0.8527542352676392, + "num_tokens": 92069491.0, + "step": 2624 + }, + { + "epoch": 0.48746518105849584, + "grad_norm": 1.6589027643203735, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8684489130973816, + "num_tokens": 92094212.0, + "step": 2625 + }, + { + "epoch": 0.48765088207985147, + "grad_norm": 1.5443412065505981, + "learning_rate": 1e-06, + "loss": 0.4049, + "mean_token_accuracy": 0.8655952215194702, + "num_tokens": 92125600.0, + "step": 2626 + }, + { + "epoch": 0.48783658310120703, + "grad_norm": 1.4438087940216064, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8634176850318909, + "num_tokens": 92165233.0, + "step": 2627 + }, + { + "epoch": 0.48802228412256266, + "grad_norm": 1.4429517984390259, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8637276291847229, + "num_tokens": 92204928.0, + "step": 2628 + }, + { + "epoch": 0.4882079851439183, + "grad_norm": 1.4260573387145996, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.8636137247085571, + "num_tokens": 92240415.0, + "step": 2629 + }, + { + "epoch": 0.4883936861652739, + "grad_norm": 1.6315685510635376, + "learning_rate": 1e-06, + "loss": 0.4475, + "mean_token_accuracy": 0.8541661500930786, + "num_tokens": 92275515.0, + "step": 2630 + }, + { + "epoch": 0.4885793871866295, + "grad_norm": 1.4892228841781616, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.865375816822052, + "num_tokens": 92308221.0, + "step": 2631 + }, + { + "epoch": 0.48876508820798514, + "grad_norm": 1.5244691371917725, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.8687262535095215, + "num_tokens": 92340601.0, + "step": 2632 + }, + { + "epoch": 0.48895078922934077, + "grad_norm": 1.5299665927886963, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8705880641937256, + "num_tokens": 92376614.0, + "step": 2633 + }, + { + "epoch": 0.4891364902506964, + "grad_norm": 1.3344030380249023, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.879468560218811, + "num_tokens": 92413736.0, + "step": 2634 + }, + { + "epoch": 0.489322191272052, + "grad_norm": 1.4809081554412842, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.8610503077507019, + "num_tokens": 92447370.0, + "step": 2635 + }, + { + "epoch": 0.48950789229340763, + "grad_norm": 1.5268681049346924, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8630837202072144, + "num_tokens": 92484075.0, + "step": 2636 + }, + { + "epoch": 0.48969359331476325, + "grad_norm": 1.5533102750778198, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8603289723396301, + "num_tokens": 92517989.0, + "step": 2637 + }, + { + "epoch": 0.4898792943361189, + "grad_norm": 1.462488055229187, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8696579337120056, + "num_tokens": 92558699.0, + "step": 2638 + }, + { + "epoch": 0.49006499535747444, + "grad_norm": 1.3771576881408691, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8683327436447144, + "num_tokens": 92597925.0, + "step": 2639 + }, + { + "epoch": 0.49025069637883006, + "grad_norm": 1.5688563585281372, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8629269599914551, + "num_tokens": 92633563.0, + "step": 2640 + }, + { + "epoch": 0.4904363974001857, + "grad_norm": 1.646621823310852, + "learning_rate": 1e-06, + "loss": 0.4396, + "mean_token_accuracy": 0.8551692962646484, + "num_tokens": 92667437.0, + "step": 2641 + }, + { + "epoch": 0.4906220984215413, + "grad_norm": 1.553342342376709, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.861452579498291, + "num_tokens": 92701304.0, + "step": 2642 + }, + { + "epoch": 0.49080779944289693, + "grad_norm": 1.4636950492858887, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8605291843414307, + "num_tokens": 92736277.0, + "step": 2643 + }, + { + "epoch": 0.49099350046425255, + "grad_norm": 1.4770394563674927, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8623285293579102, + "num_tokens": 92775125.0, + "step": 2644 + }, + { + "epoch": 0.4911792014856082, + "grad_norm": 1.4447542428970337, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8674204349517822, + "num_tokens": 92813856.0, + "step": 2645 + }, + { + "epoch": 0.4913649025069638, + "grad_norm": 1.5248390436172485, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8756914138793945, + "num_tokens": 92845042.0, + "step": 2646 + }, + { + "epoch": 0.4915506035283194, + "grad_norm": 1.422006368637085, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8635538816452026, + "num_tokens": 92879797.0, + "step": 2647 + }, + { + "epoch": 0.49173630454967504, + "grad_norm": 1.5875643491744995, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8705478310585022, + "num_tokens": 92910890.0, + "step": 2648 + }, + { + "epoch": 0.49192200557103066, + "grad_norm": 1.5002249479293823, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8697046041488647, + "num_tokens": 92949522.0, + "step": 2649 + }, + { + "epoch": 0.4921077065923863, + "grad_norm": 1.3493015766143799, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8761622905731201, + "num_tokens": 92986125.0, + "step": 2650 + }, + { + "epoch": 0.49229340761374185, + "grad_norm": 1.4426521062850952, + "learning_rate": 1e-06, + "loss": 0.4206, + "mean_token_accuracy": 0.859929084777832, + "num_tokens": 93024544.0, + "step": 2651 + }, + { + "epoch": 0.4924791086350975, + "grad_norm": 1.5345590114593506, + "learning_rate": 1e-06, + "loss": 0.4336, + "mean_token_accuracy": 0.856209397315979, + "num_tokens": 93058426.0, + "step": 2652 + }, + { + "epoch": 0.4926648096564531, + "grad_norm": 1.498377799987793, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8611187934875488, + "num_tokens": 93095385.0, + "step": 2653 + }, + { + "epoch": 0.4928505106778087, + "grad_norm": 1.5159891843795776, + "learning_rate": 1e-06, + "loss": 0.4326, + "mean_token_accuracy": 0.8545771241188049, + "num_tokens": 93130263.0, + "step": 2654 + }, + { + "epoch": 0.49303621169916434, + "grad_norm": 1.356076717376709, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.8624763488769531, + "num_tokens": 93168870.0, + "step": 2655 + }, + { + "epoch": 0.49322191272051996, + "grad_norm": 1.4474780559539795, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8692432641983032, + "num_tokens": 93204873.0, + "step": 2656 + }, + { + "epoch": 0.4934076137418756, + "grad_norm": 1.6731818914413452, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8624089956283569, + "num_tokens": 93231618.0, + "step": 2657 + }, + { + "epoch": 0.4935933147632312, + "grad_norm": 1.4050297737121582, + "learning_rate": 1e-06, + "loss": 0.4454, + "mean_token_accuracy": 0.8548339605331421, + "num_tokens": 93272876.0, + "step": 2658 + }, + { + "epoch": 0.4937790157845868, + "grad_norm": 1.4807590246200562, + "learning_rate": 1e-06, + "loss": 0.4257, + "mean_token_accuracy": 0.8605111241340637, + "num_tokens": 93311015.0, + "step": 2659 + }, + { + "epoch": 0.49396471680594245, + "grad_norm": 1.360215663909912, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8798970580101013, + "num_tokens": 93348595.0, + "step": 2660 + }, + { + "epoch": 0.49415041782729807, + "grad_norm": 1.4209285974502563, + "learning_rate": 1e-06, + "loss": 0.4127, + "mean_token_accuracy": 0.8623921871185303, + "num_tokens": 93385221.0, + "step": 2661 + }, + { + "epoch": 0.4943361188486537, + "grad_norm": 1.624630331993103, + "learning_rate": 1e-06, + "loss": 0.4442, + "mean_token_accuracy": 0.849568784236908, + "num_tokens": 93415858.0, + "step": 2662 + }, + { + "epoch": 0.49452181987000926, + "grad_norm": 1.5735071897506714, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8607091307640076, + "num_tokens": 93448346.0, + "step": 2663 + }, + { + "epoch": 0.4947075208913649, + "grad_norm": 1.4812334775924683, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8643404245376587, + "num_tokens": 93484150.0, + "step": 2664 + }, + { + "epoch": 0.4948932219127205, + "grad_norm": 1.4048843383789062, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8738645315170288, + "num_tokens": 93525314.0, + "step": 2665 + }, + { + "epoch": 0.4950789229340761, + "grad_norm": 1.6040507555007935, + "learning_rate": 1e-06, + "loss": 0.414, + "mean_token_accuracy": 0.8615419864654541, + "num_tokens": 93553142.0, + "step": 2666 + }, + { + "epoch": 0.49526462395543175, + "grad_norm": 1.4907336235046387, + "learning_rate": 1e-06, + "loss": 0.4521, + "mean_token_accuracy": 0.8519784808158875, + "num_tokens": 93593152.0, + "step": 2667 + }, + { + "epoch": 0.49545032497678737, + "grad_norm": 1.4018791913986206, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.8620016574859619, + "num_tokens": 93632488.0, + "step": 2668 + }, + { + "epoch": 0.495636025998143, + "grad_norm": 1.4835443496704102, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8754651546478271, + "num_tokens": 93665090.0, + "step": 2669 + }, + { + "epoch": 0.4958217270194986, + "grad_norm": 1.4998944997787476, + "learning_rate": 1e-06, + "loss": 0.4323, + "mean_token_accuracy": 0.8605825304985046, + "num_tokens": 93706674.0, + "step": 2670 + }, + { + "epoch": 0.49600742804085424, + "grad_norm": 1.4937657117843628, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.8582532405853271, + "num_tokens": 93740988.0, + "step": 2671 + }, + { + "epoch": 0.49619312906220986, + "grad_norm": 1.4863625764846802, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8809077739715576, + "num_tokens": 93775645.0, + "step": 2672 + }, + { + "epoch": 0.4963788300835655, + "grad_norm": 1.4719414710998535, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8767092823982239, + "num_tokens": 93809263.0, + "step": 2673 + }, + { + "epoch": 0.4965645311049211, + "grad_norm": 1.5224683284759521, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.8645666241645813, + "num_tokens": 93844415.0, + "step": 2674 + }, + { + "epoch": 0.49675023212627667, + "grad_norm": 1.6604007482528687, + "learning_rate": 1e-06, + "loss": 0.4578, + "mean_token_accuracy": 0.8525040149688721, + "num_tokens": 93874138.0, + "step": 2675 + }, + { + "epoch": 0.4969359331476323, + "grad_norm": 1.595662236213684, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8645385503768921, + "num_tokens": 93902239.0, + "step": 2676 + }, + { + "epoch": 0.4971216341689879, + "grad_norm": 1.6517102718353271, + "learning_rate": 1e-06, + "loss": 0.4375, + "mean_token_accuracy": 0.8570516109466553, + "num_tokens": 93933354.0, + "step": 2677 + }, + { + "epoch": 0.49730733519034354, + "grad_norm": 1.395294427871704, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8647891283035278, + "num_tokens": 93972678.0, + "step": 2678 + }, + { + "epoch": 0.49749303621169916, + "grad_norm": 1.5951828956604004, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8667887449264526, + "num_tokens": 94006977.0, + "step": 2679 + }, + { + "epoch": 0.4976787372330548, + "grad_norm": 1.6740915775299072, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8616102933883667, + "num_tokens": 94035320.0, + "step": 2680 + }, + { + "epoch": 0.4978644382544104, + "grad_norm": 1.4598037004470825, + "learning_rate": 1e-06, + "loss": 0.4494, + "mean_token_accuracy": 0.8521785736083984, + "num_tokens": 94072055.0, + "step": 2681 + }, + { + "epoch": 0.498050139275766, + "grad_norm": 1.6685864925384521, + "learning_rate": 1e-06, + "loss": 0.4287, + "mean_token_accuracy": 0.8575061559677124, + "num_tokens": 94102796.0, + "step": 2682 + }, + { + "epoch": 0.49823584029712165, + "grad_norm": 1.5592100620269775, + "learning_rate": 1e-06, + "loss": 0.4265, + "mean_token_accuracy": 0.8582031726837158, + "num_tokens": 94135558.0, + "step": 2683 + }, + { + "epoch": 0.49842154131847727, + "grad_norm": 1.675024390220642, + "learning_rate": 1e-06, + "loss": 0.4506, + "mean_token_accuracy": 0.8580220937728882, + "num_tokens": 94162959.0, + "step": 2684 + }, + { + "epoch": 0.4986072423398329, + "grad_norm": 1.4974746704101562, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8765175342559814, + "num_tokens": 94198396.0, + "step": 2685 + }, + { + "epoch": 0.4987929433611885, + "grad_norm": 1.3946555852890015, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.854236364364624, + "num_tokens": 94239993.0, + "step": 2686 + }, + { + "epoch": 0.4989786443825441, + "grad_norm": 1.3927749395370483, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.870291531085968, + "num_tokens": 94279782.0, + "step": 2687 + }, + { + "epoch": 0.4991643454038997, + "grad_norm": 1.4203121662139893, + "learning_rate": 1e-06, + "loss": 0.3962, + "mean_token_accuracy": 0.8674761652946472, + "num_tokens": 94315519.0, + "step": 2688 + }, + { + "epoch": 0.4993500464252553, + "grad_norm": 1.5408014059066772, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8839525580406189, + "num_tokens": 94347434.0, + "step": 2689 + }, + { + "epoch": 0.49953574744661094, + "grad_norm": 1.6530927419662476, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.8599555492401123, + "num_tokens": 94375678.0, + "step": 2690 + }, + { + "epoch": 0.49972144846796657, + "grad_norm": 1.4387009143829346, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8660791516304016, + "num_tokens": 94418986.0, + "step": 2691 + }, + { + "epoch": 0.4999071494893222, + "grad_norm": 1.3524316549301147, + "learning_rate": 1e-06, + "loss": 0.498, + "mean_token_accuracy": 0.8374987840652466, + "num_tokens": 94466787.0, + "step": 2692 + }, + { + "epoch": 0.5000928505106778, + "grad_norm": 1.3955061435699463, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8829703330993652, + "num_tokens": 94498780.0, + "step": 2693 + }, + { + "epoch": 0.5002785515320334, + "grad_norm": 1.4054442644119263, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8721341490745544, + "num_tokens": 94536251.0, + "step": 2694 + }, + { + "epoch": 0.500464252553389, + "grad_norm": 1.532150149345398, + "learning_rate": 1e-06, + "loss": 0.403, + "mean_token_accuracy": 0.8649934530258179, + "num_tokens": 94571315.0, + "step": 2695 + }, + { + "epoch": 0.5006499535747446, + "grad_norm": 1.489815592765808, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8695462346076965, + "num_tokens": 94608226.0, + "step": 2696 + }, + { + "epoch": 0.5008356545961002, + "grad_norm": 1.46337890625, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8645042181015015, + "num_tokens": 94645710.0, + "step": 2697 + }, + { + "epoch": 0.5010213556174559, + "grad_norm": 1.3972535133361816, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8704222440719604, + "num_tokens": 94682249.0, + "step": 2698 + }, + { + "epoch": 0.5012070566388115, + "grad_norm": 1.4048371315002441, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8601183891296387, + "num_tokens": 94718380.0, + "step": 2699 + }, + { + "epoch": 0.5013927576601671, + "grad_norm": 1.511292576789856, + "learning_rate": 1e-06, + "loss": 0.4055, + "mean_token_accuracy": 0.8629968166351318, + "num_tokens": 94752925.0, + "step": 2700 + }, + { + "epoch": 0.5015784586815227, + "grad_norm": 1.5343236923217773, + "learning_rate": 1e-06, + "loss": 0.4055, + "mean_token_accuracy": 0.8642288446426392, + "num_tokens": 94791581.0, + "step": 2701 + }, + { + "epoch": 0.5017641597028784, + "grad_norm": 1.3786842823028564, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8719695806503296, + "num_tokens": 94832688.0, + "step": 2702 + }, + { + "epoch": 0.501949860724234, + "grad_norm": 1.6386289596557617, + "learning_rate": 1e-06, + "loss": 0.4754, + "mean_token_accuracy": 0.8440830111503601, + "num_tokens": 94869992.0, + "step": 2703 + }, + { + "epoch": 0.5021355617455896, + "grad_norm": 1.4698646068572998, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8707809448242188, + "num_tokens": 94906921.0, + "step": 2704 + }, + { + "epoch": 0.5023212627669452, + "grad_norm": 1.4080573320388794, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8771954774856567, + "num_tokens": 94943095.0, + "step": 2705 + }, + { + "epoch": 0.5025069637883008, + "grad_norm": 1.4284769296646118, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8661705851554871, + "num_tokens": 94976041.0, + "step": 2706 + }, + { + "epoch": 0.5026926648096565, + "grad_norm": 1.4685393571853638, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8791689872741699, + "num_tokens": 95010408.0, + "step": 2707 + }, + { + "epoch": 0.5028783658310121, + "grad_norm": 1.5282262563705444, + "learning_rate": 1e-06, + "loss": 0.4399, + "mean_token_accuracy": 0.8524528741836548, + "num_tokens": 95045301.0, + "step": 2708 + }, + { + "epoch": 0.5030640668523677, + "grad_norm": 1.4849567413330078, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.877903938293457, + "num_tokens": 95080708.0, + "step": 2709 + }, + { + "epoch": 0.5032497678737233, + "grad_norm": 1.5328129529953003, + "learning_rate": 1e-06, + "loss": 0.4357, + "mean_token_accuracy": 0.8580701351165771, + "num_tokens": 95117574.0, + "step": 2710 + }, + { + "epoch": 0.503435468895079, + "grad_norm": 1.622604250907898, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8728406429290771, + "num_tokens": 95146656.0, + "step": 2711 + }, + { + "epoch": 0.5036211699164346, + "grad_norm": 1.4957051277160645, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8730969429016113, + "num_tokens": 95178110.0, + "step": 2712 + }, + { + "epoch": 0.5038068709377902, + "grad_norm": 1.6044880151748657, + "learning_rate": 1e-06, + "loss": 0.4852, + "mean_token_accuracy": 0.8378579616546631, + "num_tokens": 95212976.0, + "step": 2713 + }, + { + "epoch": 0.5039925719591458, + "grad_norm": 1.3787964582443237, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.8599925637245178, + "num_tokens": 95255228.0, + "step": 2714 + }, + { + "epoch": 0.5041782729805014, + "grad_norm": 1.4904403686523438, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.863143801689148, + "num_tokens": 95288515.0, + "step": 2715 + }, + { + "epoch": 0.5043639740018571, + "grad_norm": 1.4774651527404785, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8674163222312927, + "num_tokens": 95324910.0, + "step": 2716 + }, + { + "epoch": 0.5045496750232126, + "grad_norm": 1.5284286737442017, + "learning_rate": 1e-06, + "loss": 0.4346, + "mean_token_accuracy": 0.8547815084457397, + "num_tokens": 95360730.0, + "step": 2717 + }, + { + "epoch": 0.5047353760445682, + "grad_norm": 1.5318951606750488, + "learning_rate": 1e-06, + "loss": 0.4573, + "mean_token_accuracy": 0.8530949950218201, + "num_tokens": 95395875.0, + "step": 2718 + }, + { + "epoch": 0.5049210770659238, + "grad_norm": 1.4922213554382324, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8726115822792053, + "num_tokens": 95428086.0, + "step": 2719 + }, + { + "epoch": 0.5051067780872794, + "grad_norm": 1.6913596391677856, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8692286014556885, + "num_tokens": 95454493.0, + "step": 2720 + }, + { + "epoch": 0.5052924791086351, + "grad_norm": 1.4938992261886597, + "learning_rate": 1e-06, + "loss": 0.4222, + "mean_token_accuracy": 0.8577357530593872, + "num_tokens": 95489761.0, + "step": 2721 + }, + { + "epoch": 0.5054781801299907, + "grad_norm": 1.453064203262329, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8634828329086304, + "num_tokens": 95524926.0, + "step": 2722 + }, + { + "epoch": 0.5056638811513463, + "grad_norm": 1.3274073600769043, + "learning_rate": 1e-06, + "loss": 0.4351, + "mean_token_accuracy": 0.8548353314399719, + "num_tokens": 95565680.0, + "step": 2723 + }, + { + "epoch": 0.5058495821727019, + "grad_norm": 1.4796313047409058, + "learning_rate": 1e-06, + "loss": 0.4488, + "mean_token_accuracy": 0.8501246571540833, + "num_tokens": 95602803.0, + "step": 2724 + }, + { + "epoch": 0.5060352831940576, + "grad_norm": 1.4113188982009888, + "learning_rate": 1e-06, + "loss": 0.4408, + "mean_token_accuracy": 0.850377082824707, + "num_tokens": 95642822.0, + "step": 2725 + }, + { + "epoch": 0.5062209842154132, + "grad_norm": 1.6251024007797241, + "learning_rate": 1e-06, + "loss": 0.4321, + "mean_token_accuracy": 0.855429470539093, + "num_tokens": 95673965.0, + "step": 2726 + }, + { + "epoch": 0.5064066852367688, + "grad_norm": 1.7309576272964478, + "learning_rate": 1e-06, + "loss": 0.4874, + "mean_token_accuracy": 0.8375149965286255, + "num_tokens": 95705625.0, + "step": 2727 + }, + { + "epoch": 0.5065923862581244, + "grad_norm": 1.639548897743225, + "learning_rate": 1e-06, + "loss": 0.4686, + "mean_token_accuracy": 0.8516359329223633, + "num_tokens": 95736021.0, + "step": 2728 + }, + { + "epoch": 0.50677808727948, + "grad_norm": 1.4131954908370972, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8708487749099731, + "num_tokens": 95775950.0, + "step": 2729 + }, + { + "epoch": 0.5069637883008357, + "grad_norm": 1.3572804927825928, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8743259310722351, + "num_tokens": 95814061.0, + "step": 2730 + }, + { + "epoch": 0.5071494893221913, + "grad_norm": 2.301055669784546, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8587286472320557, + "num_tokens": 95844058.0, + "step": 2731 + }, + { + "epoch": 0.5073351903435469, + "grad_norm": 1.3949183225631714, + "learning_rate": 1e-06, + "loss": 0.4276, + "mean_token_accuracy": 0.860533595085144, + "num_tokens": 95881623.0, + "step": 2732 + }, + { + "epoch": 0.5075208913649025, + "grad_norm": 1.6209468841552734, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.86482834815979, + "num_tokens": 95915381.0, + "step": 2733 + }, + { + "epoch": 0.5077065923862581, + "grad_norm": 1.4691020250320435, + "learning_rate": 1e-06, + "loss": 0.4372, + "mean_token_accuracy": 0.852059543132782, + "num_tokens": 95955873.0, + "step": 2734 + }, + { + "epoch": 0.5078922934076138, + "grad_norm": 1.4992706775665283, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.8689953684806824, + "num_tokens": 95991174.0, + "step": 2735 + }, + { + "epoch": 0.5080779944289694, + "grad_norm": 1.501584529876709, + "learning_rate": 1e-06, + "loss": 0.4158, + "mean_token_accuracy": 0.8611438870429993, + "num_tokens": 96024551.0, + "step": 2736 + }, + { + "epoch": 0.508263695450325, + "grad_norm": 1.4245351552963257, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8711936473846436, + "num_tokens": 96063519.0, + "step": 2737 + }, + { + "epoch": 0.5084493964716806, + "grad_norm": 1.4339728355407715, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8716801404953003, + "num_tokens": 96100383.0, + "step": 2738 + }, + { + "epoch": 0.5086350974930363, + "grad_norm": 1.5324584245681763, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8669718503952026, + "num_tokens": 96133297.0, + "step": 2739 + }, + { + "epoch": 0.5088207985143919, + "grad_norm": 1.4205949306488037, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8662561178207397, + "num_tokens": 96169939.0, + "step": 2740 + }, + { + "epoch": 0.5090064995357474, + "grad_norm": 1.4530054330825806, + "learning_rate": 1e-06, + "loss": 0.4089, + "mean_token_accuracy": 0.8645051717758179, + "num_tokens": 96207069.0, + "step": 2741 + }, + { + "epoch": 0.509192200557103, + "grad_norm": 1.4272520542144775, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8636921644210815, + "num_tokens": 96249864.0, + "step": 2742 + }, + { + "epoch": 0.5093779015784586, + "grad_norm": 1.5679502487182617, + "learning_rate": 1e-06, + "loss": 0.4501, + "mean_token_accuracy": 0.8537758588790894, + "num_tokens": 96288710.0, + "step": 2743 + }, + { + "epoch": 0.5095636025998143, + "grad_norm": 1.4103351831436157, + "learning_rate": 1e-06, + "loss": 0.4127, + "mean_token_accuracy": 0.8633043169975281, + "num_tokens": 96330096.0, + "step": 2744 + }, + { + "epoch": 0.5097493036211699, + "grad_norm": 1.4745720624923706, + "learning_rate": 1e-06, + "loss": 0.4174, + "mean_token_accuracy": 0.8605821132659912, + "num_tokens": 96368429.0, + "step": 2745 + }, + { + "epoch": 0.5099350046425255, + "grad_norm": 1.487459421157837, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8653109669685364, + "num_tokens": 96403855.0, + "step": 2746 + }, + { + "epoch": 0.5101207056638811, + "grad_norm": 1.6859005689620972, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.8571619987487793, + "num_tokens": 96433549.0, + "step": 2747 + }, + { + "epoch": 0.5103064066852367, + "grad_norm": 1.420685887336731, + "learning_rate": 1e-06, + "loss": 0.4248, + "mean_token_accuracy": 0.8613853454589844, + "num_tokens": 96472239.0, + "step": 2748 + }, + { + "epoch": 0.5104921077065924, + "grad_norm": 1.4123468399047852, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8776309490203857, + "num_tokens": 96508031.0, + "step": 2749 + }, + { + "epoch": 0.510677808727948, + "grad_norm": 1.5344020128250122, + "learning_rate": 1e-06, + "loss": 0.4081, + "mean_token_accuracy": 0.8697299957275391, + "num_tokens": 96539975.0, + "step": 2750 + }, + { + "epoch": 0.5108635097493036, + "grad_norm": 1.4326459169387817, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.864234209060669, + "num_tokens": 96581053.0, + "step": 2751 + }, + { + "epoch": 0.5110492107706592, + "grad_norm": 1.5950133800506592, + "learning_rate": 1e-06, + "loss": 0.458, + "mean_token_accuracy": 0.85136878490448, + "num_tokens": 96612421.0, + "step": 2752 + }, + { + "epoch": 0.5112349117920149, + "grad_norm": 1.4947936534881592, + "learning_rate": 1e-06, + "loss": 0.3966, + "mean_token_accuracy": 0.8700083494186401, + "num_tokens": 96644502.0, + "step": 2753 + }, + { + "epoch": 0.5114206128133705, + "grad_norm": 1.3284991979599, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8877002000808716, + "num_tokens": 96683141.0, + "step": 2754 + }, + { + "epoch": 0.5116063138347261, + "grad_norm": 1.646831750869751, + "learning_rate": 1e-06, + "loss": 0.4739, + "mean_token_accuracy": 0.844474196434021, + "num_tokens": 96719752.0, + "step": 2755 + }, + { + "epoch": 0.5117920148560817, + "grad_norm": 1.5222828388214111, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8638266324996948, + "num_tokens": 96754581.0, + "step": 2756 + }, + { + "epoch": 0.5119777158774373, + "grad_norm": 1.504738211631775, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8858845829963684, + "num_tokens": 96785440.0, + "step": 2757 + }, + { + "epoch": 0.512163416898793, + "grad_norm": 1.5581573247909546, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8662732839584351, + "num_tokens": 96819333.0, + "step": 2758 + }, + { + "epoch": 0.5123491179201486, + "grad_norm": 1.6468526124954224, + "learning_rate": 1e-06, + "loss": 0.4764, + "mean_token_accuracy": 0.8481137752532959, + "num_tokens": 96854038.0, + "step": 2759 + }, + { + "epoch": 0.5125348189415042, + "grad_norm": 1.3085070848464966, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8748772144317627, + "num_tokens": 96895239.0, + "step": 2760 + }, + { + "epoch": 0.5127205199628598, + "grad_norm": 1.342944860458374, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8564016819000244, + "num_tokens": 96937727.0, + "step": 2761 + }, + { + "epoch": 0.5129062209842155, + "grad_norm": 1.4705023765563965, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8704526424407959, + "num_tokens": 96971048.0, + "step": 2762 + }, + { + "epoch": 0.5130919220055711, + "grad_norm": 1.5485576391220093, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8634169101715088, + "num_tokens": 97004149.0, + "step": 2763 + }, + { + "epoch": 0.5132776230269267, + "grad_norm": 1.5625605583190918, + "learning_rate": 1e-06, + "loss": 0.4516, + "mean_token_accuracy": 0.8509611487388611, + "num_tokens": 97039774.0, + "step": 2764 + }, + { + "epoch": 0.5134633240482822, + "grad_norm": 1.4887278079986572, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8829286694526672, + "num_tokens": 97070732.0, + "step": 2765 + }, + { + "epoch": 0.5136490250696378, + "grad_norm": 1.4470566511154175, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.8667762279510498, + "num_tokens": 97106092.0, + "step": 2766 + }, + { + "epoch": 0.5138347260909935, + "grad_norm": 1.492836833000183, + "learning_rate": 1e-06, + "loss": 0.4075, + "mean_token_accuracy": 0.8648823499679565, + "num_tokens": 97139334.0, + "step": 2767 + }, + { + "epoch": 0.5140204271123491, + "grad_norm": 1.4816792011260986, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8721985816955566, + "num_tokens": 97178904.0, + "step": 2768 + }, + { + "epoch": 0.5142061281337047, + "grad_norm": 1.4440577030181885, + "learning_rate": 1e-06, + "loss": 0.4096, + "mean_token_accuracy": 0.8641182780265808, + "num_tokens": 97217735.0, + "step": 2769 + }, + { + "epoch": 0.5143918291550603, + "grad_norm": 1.523701548576355, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8706085681915283, + "num_tokens": 97258398.0, + "step": 2770 + }, + { + "epoch": 0.5145775301764159, + "grad_norm": 1.5636467933654785, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8753305077552795, + "num_tokens": 97292483.0, + "step": 2771 + }, + { + "epoch": 0.5147632311977716, + "grad_norm": 1.536442756652832, + "learning_rate": 1e-06, + "loss": 0.4625, + "mean_token_accuracy": 0.8473843336105347, + "num_tokens": 97327696.0, + "step": 2772 + }, + { + "epoch": 0.5149489322191272, + "grad_norm": 1.4428423643112183, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8791294097900391, + "num_tokens": 97362643.0, + "step": 2773 + }, + { + "epoch": 0.5151346332404828, + "grad_norm": 1.4295687675476074, + "learning_rate": 1e-06, + "loss": 0.4038, + "mean_token_accuracy": 0.8640819787979126, + "num_tokens": 97399102.0, + "step": 2774 + }, + { + "epoch": 0.5153203342618384, + "grad_norm": 1.427910566329956, + "learning_rate": 1e-06, + "loss": 0.4107, + "mean_token_accuracy": 0.8622925877571106, + "num_tokens": 97437199.0, + "step": 2775 + }, + { + "epoch": 0.515506035283194, + "grad_norm": 1.6482133865356445, + "learning_rate": 1e-06, + "loss": 0.4363, + "mean_token_accuracy": 0.8500854969024658, + "num_tokens": 97465389.0, + "step": 2776 + }, + { + "epoch": 0.5156917363045497, + "grad_norm": 1.4060946702957153, + "learning_rate": 1e-06, + "loss": 0.4314, + "mean_token_accuracy": 0.8567745685577393, + "num_tokens": 97505477.0, + "step": 2777 + }, + { + "epoch": 0.5158774373259053, + "grad_norm": 1.4394522905349731, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8666188716888428, + "num_tokens": 97541260.0, + "step": 2778 + }, + { + "epoch": 0.5160631383472609, + "grad_norm": 1.55527663230896, + "learning_rate": 1e-06, + "loss": 0.4211, + "mean_token_accuracy": 0.8617368340492249, + "num_tokens": 97579499.0, + "step": 2779 + }, + { + "epoch": 0.5162488393686165, + "grad_norm": 1.5200484991073608, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.8740461468696594, + "num_tokens": 97613609.0, + "step": 2780 + }, + { + "epoch": 0.5164345403899722, + "grad_norm": 1.4035545587539673, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.867576003074646, + "num_tokens": 97649129.0, + "step": 2781 + }, + { + "epoch": 0.5166202414113278, + "grad_norm": 1.514119267463684, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8619137406349182, + "num_tokens": 97683277.0, + "step": 2782 + }, + { + "epoch": 0.5168059424326834, + "grad_norm": 1.496389389038086, + "learning_rate": 1e-06, + "loss": 0.4598, + "mean_token_accuracy": 0.8510138392448425, + "num_tokens": 97719549.0, + "step": 2783 + }, + { + "epoch": 0.516991643454039, + "grad_norm": 1.331726312637329, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8689367175102234, + "num_tokens": 97759389.0, + "step": 2784 + }, + { + "epoch": 0.5171773444753947, + "grad_norm": 1.5976088047027588, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8606573343276978, + "num_tokens": 97786309.0, + "step": 2785 + }, + { + "epoch": 0.5173630454967503, + "grad_norm": 1.5815244913101196, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8727672100067139, + "num_tokens": 97816812.0, + "step": 2786 + }, + { + "epoch": 0.5175487465181059, + "grad_norm": 1.438882827758789, + "learning_rate": 1e-06, + "loss": 0.4234, + "mean_token_accuracy": 0.8624279499053955, + "num_tokens": 97855183.0, + "step": 2787 + }, + { + "epoch": 0.5177344475394615, + "grad_norm": 1.4312134981155396, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8685779571533203, + "num_tokens": 97891780.0, + "step": 2788 + }, + { + "epoch": 0.5179201485608171, + "grad_norm": 1.5153249502182007, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8718773126602173, + "num_tokens": 97926041.0, + "step": 2789 + }, + { + "epoch": 0.5181058495821727, + "grad_norm": 1.4589272737503052, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8712048530578613, + "num_tokens": 97962160.0, + "step": 2790 + }, + { + "epoch": 0.5182915506035283, + "grad_norm": 1.5109189748764038, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8590848445892334, + "num_tokens": 97993928.0, + "step": 2791 + }, + { + "epoch": 0.5184772516248839, + "grad_norm": 1.6693792343139648, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8605005741119385, + "num_tokens": 98023072.0, + "step": 2792 + }, + { + "epoch": 0.5186629526462395, + "grad_norm": 1.7925031185150146, + "learning_rate": 1e-06, + "loss": 0.4278, + "mean_token_accuracy": 0.8569933176040649, + "num_tokens": 98051011.0, + "step": 2793 + }, + { + "epoch": 0.5188486536675951, + "grad_norm": 1.53766930103302, + "learning_rate": 1e-06, + "loss": 0.4205, + "mean_token_accuracy": 0.863263726234436, + "num_tokens": 98086802.0, + "step": 2794 + }, + { + "epoch": 0.5190343546889508, + "grad_norm": 1.4847631454467773, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8706277012825012, + "num_tokens": 98126940.0, + "step": 2795 + }, + { + "epoch": 0.5192200557103064, + "grad_norm": 1.5137970447540283, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.866354763507843, + "num_tokens": 98159597.0, + "step": 2796 + }, + { + "epoch": 0.519405756731662, + "grad_norm": 1.5518172979354858, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8733009099960327, + "num_tokens": 98189876.0, + "step": 2797 + }, + { + "epoch": 0.5195914577530176, + "grad_norm": 1.6765879392623901, + "learning_rate": 1e-06, + "loss": 0.433, + "mean_token_accuracy": 0.8571234941482544, + "num_tokens": 98225848.0, + "step": 2798 + }, + { + "epoch": 0.5197771587743732, + "grad_norm": 1.6063041687011719, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8814468383789062, + "num_tokens": 98257970.0, + "step": 2799 + }, + { + "epoch": 0.5199628597957289, + "grad_norm": 1.5403403043746948, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8767699003219604, + "num_tokens": 98286753.0, + "step": 2800 + }, + { + "epoch": 0.5201485608170845, + "grad_norm": 1.4845603704452515, + "learning_rate": 1e-06, + "loss": 0.4254, + "mean_token_accuracy": 0.8601207733154297, + "num_tokens": 98325737.0, + "step": 2801 + }, + { + "epoch": 0.5203342618384401, + "grad_norm": 1.6194267272949219, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8749319314956665, + "num_tokens": 98357412.0, + "step": 2802 + }, + { + "epoch": 0.5205199628597957, + "grad_norm": 1.7699378728866577, + "learning_rate": 1e-06, + "loss": 0.4419, + "mean_token_accuracy": 0.8568093776702881, + "num_tokens": 98384084.0, + "step": 2803 + }, + { + "epoch": 0.5207056638811514, + "grad_norm": 1.4828588962554932, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8672924041748047, + "num_tokens": 98420765.0, + "step": 2804 + }, + { + "epoch": 0.520891364902507, + "grad_norm": 1.485520601272583, + "learning_rate": 1e-06, + "loss": 0.4206, + "mean_token_accuracy": 0.8556164503097534, + "num_tokens": 98455895.0, + "step": 2805 + }, + { + "epoch": 0.5210770659238626, + "grad_norm": 1.519424557685852, + "learning_rate": 1e-06, + "loss": 0.4275, + "mean_token_accuracy": 0.8599147796630859, + "num_tokens": 98492061.0, + "step": 2806 + }, + { + "epoch": 0.5212627669452182, + "grad_norm": 1.513927936553955, + "learning_rate": 1e-06, + "loss": 0.4047, + "mean_token_accuracy": 0.8665043711662292, + "num_tokens": 98529390.0, + "step": 2807 + }, + { + "epoch": 0.5214484679665738, + "grad_norm": 1.545373558998108, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8639451265335083, + "num_tokens": 98565567.0, + "step": 2808 + }, + { + "epoch": 0.5216341689879295, + "grad_norm": 1.532437801361084, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8696414828300476, + "num_tokens": 98601159.0, + "step": 2809 + }, + { + "epoch": 0.5218198700092851, + "grad_norm": 1.5197538137435913, + "learning_rate": 1e-06, + "loss": 0.4222, + "mean_token_accuracy": 0.8593789339065552, + "num_tokens": 98636984.0, + "step": 2810 + }, + { + "epoch": 0.5220055710306407, + "grad_norm": 1.5652062892913818, + "learning_rate": 1e-06, + "loss": 0.4205, + "mean_token_accuracy": 0.856431245803833, + "num_tokens": 98669723.0, + "step": 2811 + }, + { + "epoch": 0.5221912720519963, + "grad_norm": 1.5988577604293823, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8767547607421875, + "num_tokens": 98698165.0, + "step": 2812 + }, + { + "epoch": 0.522376973073352, + "grad_norm": 1.5075581073760986, + "learning_rate": 1e-06, + "loss": 0.4478, + "mean_token_accuracy": 0.852096438407898, + "num_tokens": 98733802.0, + "step": 2813 + }, + { + "epoch": 0.5225626740947075, + "grad_norm": 1.5590540170669556, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.866679310798645, + "num_tokens": 98766777.0, + "step": 2814 + }, + { + "epoch": 0.5227483751160631, + "grad_norm": 1.6663734912872314, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8692021369934082, + "num_tokens": 98795020.0, + "step": 2815 + }, + { + "epoch": 0.5229340761374187, + "grad_norm": 1.6390012502670288, + "learning_rate": 1e-06, + "loss": 0.4222, + "mean_token_accuracy": 0.8609719276428223, + "num_tokens": 98830053.0, + "step": 2816 + }, + { + "epoch": 0.5231197771587743, + "grad_norm": 1.4794840812683105, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8661746382713318, + "num_tokens": 98865309.0, + "step": 2817 + }, + { + "epoch": 0.52330547818013, + "grad_norm": 1.4594751596450806, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8604671955108643, + "num_tokens": 98900461.0, + "step": 2818 + }, + { + "epoch": 0.5234911792014856, + "grad_norm": 1.5772100687026978, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.8555963635444641, + "num_tokens": 98934900.0, + "step": 2819 + }, + { + "epoch": 0.5236768802228412, + "grad_norm": 1.584656000137329, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8715929388999939, + "num_tokens": 98961231.0, + "step": 2820 + }, + { + "epoch": 0.5238625812441968, + "grad_norm": 1.4823226928710938, + "learning_rate": 1e-06, + "loss": 0.4431, + "mean_token_accuracy": 0.8533055782318115, + "num_tokens": 99002453.0, + "step": 2821 + }, + { + "epoch": 0.5240482822655524, + "grad_norm": 1.467124342918396, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8602015972137451, + "num_tokens": 99040610.0, + "step": 2822 + }, + { + "epoch": 0.5242339832869081, + "grad_norm": 1.3733586072921753, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.871048629283905, + "num_tokens": 99081834.0, + "step": 2823 + }, + { + "epoch": 0.5244196843082637, + "grad_norm": 1.523071527481079, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.8561851978302002, + "num_tokens": 99117793.0, + "step": 2824 + }, + { + "epoch": 0.5246053853296193, + "grad_norm": 1.3626434803009033, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8663718700408936, + "num_tokens": 99156955.0, + "step": 2825 + }, + { + "epoch": 0.5247910863509749, + "grad_norm": 1.4786266088485718, + "learning_rate": 1e-06, + "loss": 0.4304, + "mean_token_accuracy": 0.8594401478767395, + "num_tokens": 99195989.0, + "step": 2826 + }, + { + "epoch": 0.5249767873723306, + "grad_norm": 1.524824619293213, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8620700240135193, + "num_tokens": 99234665.0, + "step": 2827 + }, + { + "epoch": 0.5251624883936862, + "grad_norm": 1.2807172536849976, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.883529543876648, + "num_tokens": 99276088.0, + "step": 2828 + }, + { + "epoch": 0.5253481894150418, + "grad_norm": 1.3612576723098755, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8644214868545532, + "num_tokens": 99316760.0, + "step": 2829 + }, + { + "epoch": 0.5255338904363974, + "grad_norm": 1.5651264190673828, + "learning_rate": 1e-06, + "loss": 0.4377, + "mean_token_accuracy": 0.8561026453971863, + "num_tokens": 99350416.0, + "step": 2830 + }, + { + "epoch": 0.525719591457753, + "grad_norm": 1.4885773658752441, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8792563676834106, + "num_tokens": 99383159.0, + "step": 2831 + }, + { + "epoch": 0.5259052924791087, + "grad_norm": 1.464453935623169, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8767808675765991, + "num_tokens": 99420115.0, + "step": 2832 + }, + { + "epoch": 0.5260909935004643, + "grad_norm": 1.6062545776367188, + "learning_rate": 1e-06, + "loss": 0.4337, + "mean_token_accuracy": 0.8558799028396606, + "num_tokens": 99453520.0, + "step": 2833 + }, + { + "epoch": 0.5262766945218199, + "grad_norm": 1.5906263589859009, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.867525577545166, + "num_tokens": 99483996.0, + "step": 2834 + }, + { + "epoch": 0.5264623955431755, + "grad_norm": 1.6648082733154297, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8538985848426819, + "num_tokens": 99515084.0, + "step": 2835 + }, + { + "epoch": 0.5266480965645312, + "grad_norm": 1.391054391860962, + "learning_rate": 1e-06, + "loss": 0.4148, + "mean_token_accuracy": 0.8616426587104797, + "num_tokens": 99553996.0, + "step": 2836 + }, + { + "epoch": 0.5268337975858868, + "grad_norm": 1.3256843090057373, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8727680444717407, + "num_tokens": 99595203.0, + "step": 2837 + }, + { + "epoch": 0.5270194986072423, + "grad_norm": 1.3681919574737549, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8748379945755005, + "num_tokens": 99635244.0, + "step": 2838 + }, + { + "epoch": 0.5272051996285979, + "grad_norm": 1.4111884832382202, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8686888217926025, + "num_tokens": 99673166.0, + "step": 2839 + }, + { + "epoch": 0.5273909006499535, + "grad_norm": 1.4849398136138916, + "learning_rate": 1e-06, + "loss": 0.4366, + "mean_token_accuracy": 0.852852463722229, + "num_tokens": 99709363.0, + "step": 2840 + }, + { + "epoch": 0.5275766016713092, + "grad_norm": 1.5275506973266602, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.8610780239105225, + "num_tokens": 99742811.0, + "step": 2841 + }, + { + "epoch": 0.5277623026926648, + "grad_norm": 1.4935435056686401, + "learning_rate": 1e-06, + "loss": 0.438, + "mean_token_accuracy": 0.8542619943618774, + "num_tokens": 99777771.0, + "step": 2842 + }, + { + "epoch": 0.5279480037140204, + "grad_norm": 1.49073326587677, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.875834584236145, + "num_tokens": 99814246.0, + "step": 2843 + }, + { + "epoch": 0.528133704735376, + "grad_norm": 1.4668922424316406, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8730885982513428, + "num_tokens": 99848962.0, + "step": 2844 + }, + { + "epoch": 0.5283194057567316, + "grad_norm": 1.4952281713485718, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8581607937812805, + "num_tokens": 99883554.0, + "step": 2845 + }, + { + "epoch": 0.5285051067780873, + "grad_norm": 1.4323664903640747, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8578024506568909, + "num_tokens": 99924029.0, + "step": 2846 + }, + { + "epoch": 0.5286908077994429, + "grad_norm": 1.634084939956665, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8588763475418091, + "num_tokens": 99957656.0, + "step": 2847 + }, + { + "epoch": 0.5288765088207985, + "grad_norm": 1.441978096961975, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8661501407623291, + "num_tokens": 99993875.0, + "step": 2848 + }, + { + "epoch": 0.5290622098421541, + "grad_norm": 1.5074182748794556, + "learning_rate": 1e-06, + "loss": 0.4485, + "mean_token_accuracy": 0.852283239364624, + "num_tokens": 100027942.0, + "step": 2849 + }, + { + "epoch": 0.5292479108635098, + "grad_norm": 1.5831571817398071, + "learning_rate": 1e-06, + "loss": 0.4226, + "mean_token_accuracy": 0.8606756925582886, + "num_tokens": 100062423.0, + "step": 2850 + }, + { + "epoch": 0.5294336118848654, + "grad_norm": 1.6221591234207153, + "learning_rate": 1e-06, + "loss": 0.4278, + "mean_token_accuracy": 0.8570481538772583, + "num_tokens": 100094984.0, + "step": 2851 + }, + { + "epoch": 0.529619312906221, + "grad_norm": 1.498324990272522, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.868321418762207, + "num_tokens": 100130754.0, + "step": 2852 + }, + { + "epoch": 0.5298050139275766, + "grad_norm": 1.4627777338027954, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8728448152542114, + "num_tokens": 100166098.0, + "step": 2853 + }, + { + "epoch": 0.5299907149489322, + "grad_norm": 1.3883159160614014, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8638112545013428, + "num_tokens": 100207855.0, + "step": 2854 + }, + { + "epoch": 0.5301764159702879, + "grad_norm": 1.483591914176941, + "learning_rate": 1e-06, + "loss": 0.4187, + "mean_token_accuracy": 0.8583906888961792, + "num_tokens": 100247344.0, + "step": 2855 + }, + { + "epoch": 0.5303621169916435, + "grad_norm": 1.639364242553711, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.866615891456604, + "num_tokens": 100282501.0, + "step": 2856 + }, + { + "epoch": 0.5305478180129991, + "grad_norm": 1.6749221086502075, + "learning_rate": 1e-06, + "loss": 0.4616, + "mean_token_accuracy": 0.8464556932449341, + "num_tokens": 100311274.0, + "step": 2857 + }, + { + "epoch": 0.5307335190343547, + "grad_norm": 1.573528528213501, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8683067560195923, + "num_tokens": 100345225.0, + "step": 2858 + }, + { + "epoch": 0.5309192200557104, + "grad_norm": 1.4170511960983276, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8682756423950195, + "num_tokens": 100385805.0, + "step": 2859 + }, + { + "epoch": 0.531104921077066, + "grad_norm": 1.5879087448120117, + "learning_rate": 1e-06, + "loss": 0.4548, + "mean_token_accuracy": 0.8506075143814087, + "num_tokens": 100416600.0, + "step": 2860 + }, + { + "epoch": 0.5312906220984216, + "grad_norm": 1.4412832260131836, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8685722351074219, + "num_tokens": 100452879.0, + "step": 2861 + }, + { + "epoch": 0.5314763231197771, + "grad_norm": 1.4844046831130981, + "learning_rate": 1e-06, + "loss": 0.4184, + "mean_token_accuracy": 0.8593427538871765, + "num_tokens": 100484406.0, + "step": 2862 + }, + { + "epoch": 0.5316620241411327, + "grad_norm": 1.3466687202453613, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.866580069065094, + "num_tokens": 100523415.0, + "step": 2863 + }, + { + "epoch": 0.5318477251624883, + "grad_norm": 1.551404595375061, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8679574131965637, + "num_tokens": 100557409.0, + "step": 2864 + }, + { + "epoch": 0.532033426183844, + "grad_norm": 1.4982913732528687, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.8644189834594727, + "num_tokens": 100595361.0, + "step": 2865 + }, + { + "epoch": 0.5322191272051996, + "grad_norm": 1.6011934280395508, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8763847351074219, + "num_tokens": 100622620.0, + "step": 2866 + }, + { + "epoch": 0.5324048282265552, + "grad_norm": 1.521285057067871, + "learning_rate": 1e-06, + "loss": 0.416, + "mean_token_accuracy": 0.8557689785957336, + "num_tokens": 100654303.0, + "step": 2867 + }, + { + "epoch": 0.5325905292479108, + "grad_norm": 1.5924023389816284, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8610360026359558, + "num_tokens": 100686199.0, + "step": 2868 + }, + { + "epoch": 0.5327762302692665, + "grad_norm": 1.6350083351135254, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8532691597938538, + "num_tokens": 100719382.0, + "step": 2869 + }, + { + "epoch": 0.5329619312906221, + "grad_norm": 1.525602102279663, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.86668860912323, + "num_tokens": 100755552.0, + "step": 2870 + }, + { + "epoch": 0.5331476323119777, + "grad_norm": 1.5500367879867554, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.8715438842773438, + "num_tokens": 100783931.0, + "step": 2871 + }, + { + "epoch": 0.5333333333333333, + "grad_norm": 1.5113117694854736, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8662967681884766, + "num_tokens": 100819074.0, + "step": 2872 + }, + { + "epoch": 0.533519034354689, + "grad_norm": 1.5819040536880493, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8643145561218262, + "num_tokens": 100848557.0, + "step": 2873 + }, + { + "epoch": 0.5337047353760446, + "grad_norm": 1.446141242980957, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.8649594187736511, + "num_tokens": 100884772.0, + "step": 2874 + }, + { + "epoch": 0.5338904363974002, + "grad_norm": 1.6315473318099976, + "learning_rate": 1e-06, + "loss": 0.4533, + "mean_token_accuracy": 0.8497865200042725, + "num_tokens": 100919849.0, + "step": 2875 + }, + { + "epoch": 0.5340761374187558, + "grad_norm": 1.5226069688796997, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8707926273345947, + "num_tokens": 100954366.0, + "step": 2876 + }, + { + "epoch": 0.5342618384401114, + "grad_norm": 1.5526541471481323, + "learning_rate": 1e-06, + "loss": 0.485, + "mean_token_accuracy": 0.8393268585205078, + "num_tokens": 100991781.0, + "step": 2877 + }, + { + "epoch": 0.5344475394614671, + "grad_norm": 1.618515968322754, + "learning_rate": 1e-06, + "loss": 0.425, + "mean_token_accuracy": 0.8596919178962708, + "num_tokens": 101021804.0, + "step": 2878 + }, + { + "epoch": 0.5346332404828227, + "grad_norm": 1.3958224058151245, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8686140179634094, + "num_tokens": 101059854.0, + "step": 2879 + }, + { + "epoch": 0.5348189415041783, + "grad_norm": 1.3075350522994995, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8802230954170227, + "num_tokens": 101102675.0, + "step": 2880 + }, + { + "epoch": 0.5350046425255339, + "grad_norm": 1.394078016281128, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8686567544937134, + "num_tokens": 101140338.0, + "step": 2881 + }, + { + "epoch": 0.5351903435468895, + "grad_norm": 1.5188260078430176, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.8520197868347168, + "num_tokens": 101177029.0, + "step": 2882 + }, + { + "epoch": 0.5353760445682452, + "grad_norm": 1.6885390281677246, + "learning_rate": 1e-06, + "loss": 0.5145, + "mean_token_accuracy": 0.8330025672912598, + "num_tokens": 101209551.0, + "step": 2883 + }, + { + "epoch": 0.5355617455896008, + "grad_norm": 1.4491057395935059, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8675317168235779, + "num_tokens": 101244423.0, + "step": 2884 + }, + { + "epoch": 0.5357474466109564, + "grad_norm": 1.4866737127304077, + "learning_rate": 1e-06, + "loss": 0.4497, + "mean_token_accuracy": 0.8530135154724121, + "num_tokens": 101282876.0, + "step": 2885 + }, + { + "epoch": 0.5359331476323119, + "grad_norm": 1.4266117811203003, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.854928731918335, + "num_tokens": 101321866.0, + "step": 2886 + }, + { + "epoch": 0.5361188486536675, + "grad_norm": 1.4359008073806763, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8703032732009888, + "num_tokens": 101357787.0, + "step": 2887 + }, + { + "epoch": 0.5363045496750232, + "grad_norm": 1.8201391696929932, + "learning_rate": 1e-06, + "loss": 0.4215, + "mean_token_accuracy": 0.8547918796539307, + "num_tokens": 101388024.0, + "step": 2888 + }, + { + "epoch": 0.5364902506963788, + "grad_norm": 1.5364294052124023, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8571186661720276, + "num_tokens": 101423796.0, + "step": 2889 + }, + { + "epoch": 0.5366759517177344, + "grad_norm": 1.6442158222198486, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.875376284122467, + "num_tokens": 101448909.0, + "step": 2890 + }, + { + "epoch": 0.53686165273909, + "grad_norm": 1.458693504333496, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8715115785598755, + "num_tokens": 101481739.0, + "step": 2891 + }, + { + "epoch": 0.5370473537604457, + "grad_norm": 1.3759371042251587, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8747694492340088, + "num_tokens": 101522773.0, + "step": 2892 + }, + { + "epoch": 0.5372330547818013, + "grad_norm": 1.7137356996536255, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8739070892333984, + "num_tokens": 101554030.0, + "step": 2893 + }, + { + "epoch": 0.5374187558031569, + "grad_norm": 1.5583083629608154, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8581380844116211, + "num_tokens": 101589676.0, + "step": 2894 + }, + { + "epoch": 0.5376044568245125, + "grad_norm": 1.61641263961792, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8616445064544678, + "num_tokens": 101624240.0, + "step": 2895 + }, + { + "epoch": 0.5377901578458681, + "grad_norm": 1.5557539463043213, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8747143149375916, + "num_tokens": 101656246.0, + "step": 2896 + }, + { + "epoch": 0.5379758588672238, + "grad_norm": 1.5054839849472046, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.856602668762207, + "num_tokens": 101690452.0, + "step": 2897 + }, + { + "epoch": 0.5381615598885794, + "grad_norm": 1.49643874168396, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8738270998001099, + "num_tokens": 101727508.0, + "step": 2898 + }, + { + "epoch": 0.538347260909935, + "grad_norm": 1.407155990600586, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8738347291946411, + "num_tokens": 101764369.0, + "step": 2899 + }, + { + "epoch": 0.5385329619312906, + "grad_norm": 1.5745002031326294, + "learning_rate": 1e-06, + "loss": 0.4434, + "mean_token_accuracy": 0.8526073694229126, + "num_tokens": 101798765.0, + "step": 2900 + }, + { + "epoch": 0.5387186629526463, + "grad_norm": 1.6507352590560913, + "learning_rate": 1e-06, + "loss": 0.4833, + "mean_token_accuracy": 0.8476956486701965, + "num_tokens": 101832893.0, + "step": 2901 + }, + { + "epoch": 0.5389043639740019, + "grad_norm": 1.4622783660888672, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8589172959327698, + "num_tokens": 101869481.0, + "step": 2902 + }, + { + "epoch": 0.5390900649953575, + "grad_norm": 1.6664739847183228, + "learning_rate": 1e-06, + "loss": 0.4884, + "mean_token_accuracy": 0.832461953163147, + "num_tokens": 101904408.0, + "step": 2903 + }, + { + "epoch": 0.5392757660167131, + "grad_norm": 1.6136119365692139, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8656132221221924, + "num_tokens": 101934282.0, + "step": 2904 + }, + { + "epoch": 0.5394614670380687, + "grad_norm": 1.5522421598434448, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8633601665496826, + "num_tokens": 101968234.0, + "step": 2905 + }, + { + "epoch": 0.5396471680594244, + "grad_norm": 1.5362095832824707, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8755078315734863, + "num_tokens": 102001652.0, + "step": 2906 + }, + { + "epoch": 0.53983286908078, + "grad_norm": 1.3883994817733765, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8659639358520508, + "num_tokens": 102038711.0, + "step": 2907 + }, + { + "epoch": 0.5400185701021356, + "grad_norm": 1.430698275566101, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8679478168487549, + "num_tokens": 102074631.0, + "step": 2908 + }, + { + "epoch": 0.5402042711234912, + "grad_norm": 1.4490296840667725, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8782161474227905, + "num_tokens": 102109253.0, + "step": 2909 + }, + { + "epoch": 0.5403899721448467, + "grad_norm": 1.3702306747436523, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.876354455947876, + "num_tokens": 102150500.0, + "step": 2910 + }, + { + "epoch": 0.5405756731662024, + "grad_norm": 1.5172958374023438, + "learning_rate": 1e-06, + "loss": 0.3864, + "mean_token_accuracy": 0.868294358253479, + "num_tokens": 102182754.0, + "step": 2911 + }, + { + "epoch": 0.540761374187558, + "grad_norm": 1.5282118320465088, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8646684885025024, + "num_tokens": 102220800.0, + "step": 2912 + }, + { + "epoch": 0.5409470752089136, + "grad_norm": 1.6978614330291748, + "learning_rate": 1e-06, + "loss": 0.4393, + "mean_token_accuracy": 0.8511630296707153, + "num_tokens": 102252553.0, + "step": 2913 + }, + { + "epoch": 0.5411327762302692, + "grad_norm": 1.4553182125091553, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8704811930656433, + "num_tokens": 102289691.0, + "step": 2914 + }, + { + "epoch": 0.5413184772516249, + "grad_norm": 1.5143964290618896, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8620141744613647, + "num_tokens": 102324808.0, + "step": 2915 + }, + { + "epoch": 0.5415041782729805, + "grad_norm": 1.4982508420944214, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8694647550582886, + "num_tokens": 102355629.0, + "step": 2916 + }, + { + "epoch": 0.5416898792943361, + "grad_norm": 1.4597536325454712, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8786814212799072, + "num_tokens": 102390347.0, + "step": 2917 + }, + { + "epoch": 0.5418755803156917, + "grad_norm": 1.5170615911483765, + "learning_rate": 1e-06, + "loss": 0.4443, + "mean_token_accuracy": 0.8530242443084717, + "num_tokens": 102426047.0, + "step": 2918 + }, + { + "epoch": 0.5420612813370473, + "grad_norm": 1.4040249586105347, + "learning_rate": 1e-06, + "loss": 0.4238, + "mean_token_accuracy": 0.8571102023124695, + "num_tokens": 102468186.0, + "step": 2919 + }, + { + "epoch": 0.542246982358403, + "grad_norm": 1.4615836143493652, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8708640336990356, + "num_tokens": 102504454.0, + "step": 2920 + }, + { + "epoch": 0.5424326833797586, + "grad_norm": 1.585071325302124, + "learning_rate": 1e-06, + "loss": 0.4389, + "mean_token_accuracy": 0.8550041913986206, + "num_tokens": 102538942.0, + "step": 2921 + }, + { + "epoch": 0.5426183844011142, + "grad_norm": 1.4064081907272339, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8659105896949768, + "num_tokens": 102575741.0, + "step": 2922 + }, + { + "epoch": 0.5428040854224698, + "grad_norm": 1.5220141410827637, + "learning_rate": 1e-06, + "loss": 0.443, + "mean_token_accuracy": 0.8551903963088989, + "num_tokens": 102608686.0, + "step": 2923 + }, + { + "epoch": 0.5429897864438255, + "grad_norm": 1.4879858493804932, + "learning_rate": 1e-06, + "loss": 0.4201, + "mean_token_accuracy": 0.8561899662017822, + "num_tokens": 102641231.0, + "step": 2924 + }, + { + "epoch": 0.5431754874651811, + "grad_norm": 1.494015097618103, + "learning_rate": 1e-06, + "loss": 0.4159, + "mean_token_accuracy": 0.8600714802742004, + "num_tokens": 102675983.0, + "step": 2925 + }, + { + "epoch": 0.5433611884865367, + "grad_norm": 1.4596208333969116, + "learning_rate": 1e-06, + "loss": 0.4454, + "mean_token_accuracy": 0.8547711968421936, + "num_tokens": 102714274.0, + "step": 2926 + }, + { + "epoch": 0.5435468895078923, + "grad_norm": 1.4297409057617188, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8737314939498901, + "num_tokens": 102749415.0, + "step": 2927 + }, + { + "epoch": 0.5437325905292479, + "grad_norm": 1.4050956964492798, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8735857605934143, + "num_tokens": 102783864.0, + "step": 2928 + }, + { + "epoch": 0.5439182915506036, + "grad_norm": 1.648244023323059, + "learning_rate": 1e-06, + "loss": 0.4235, + "mean_token_accuracy": 0.855492353439331, + "num_tokens": 102816073.0, + "step": 2929 + }, + { + "epoch": 0.5441039925719592, + "grad_norm": 1.6228710412979126, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.8561063408851624, + "num_tokens": 102844643.0, + "step": 2930 + }, + { + "epoch": 0.5442896935933148, + "grad_norm": 1.350818395614624, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8753316402435303, + "num_tokens": 102879956.0, + "step": 2931 + }, + { + "epoch": 0.5444753946146704, + "grad_norm": 1.4556913375854492, + "learning_rate": 1e-06, + "loss": 0.4228, + "mean_token_accuracy": 0.856969952583313, + "num_tokens": 102916053.0, + "step": 2932 + }, + { + "epoch": 0.544661095636026, + "grad_norm": 1.5046484470367432, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.869625449180603, + "num_tokens": 102951004.0, + "step": 2933 + }, + { + "epoch": 0.5448467966573816, + "grad_norm": 1.4327077865600586, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8680179119110107, + "num_tokens": 102988845.0, + "step": 2934 + }, + { + "epoch": 0.5450324976787372, + "grad_norm": 1.4908173084259033, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8620045781135559, + "num_tokens": 103023107.0, + "step": 2935 + }, + { + "epoch": 0.5452181987000928, + "grad_norm": 1.4790799617767334, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.866521954536438, + "num_tokens": 103063047.0, + "step": 2936 + }, + { + "epoch": 0.5454038997214484, + "grad_norm": 1.4830148220062256, + "learning_rate": 1e-06, + "loss": 0.4635, + "mean_token_accuracy": 0.8512356877326965, + "num_tokens": 103101723.0, + "step": 2937 + }, + { + "epoch": 0.545589600742804, + "grad_norm": 1.5191112756729126, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8691739439964294, + "num_tokens": 103132560.0, + "step": 2938 + }, + { + "epoch": 0.5457753017641597, + "grad_norm": 1.4557077884674072, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8719882965087891, + "num_tokens": 103167121.0, + "step": 2939 + }, + { + "epoch": 0.5459610027855153, + "grad_norm": 1.5411877632141113, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8600430488586426, + "num_tokens": 103202544.0, + "step": 2940 + }, + { + "epoch": 0.5461467038068709, + "grad_norm": 1.4254192113876343, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.864432692527771, + "num_tokens": 103242743.0, + "step": 2941 + }, + { + "epoch": 0.5463324048282265, + "grad_norm": 1.4292768239974976, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8549323081970215, + "num_tokens": 103280469.0, + "step": 2942 + }, + { + "epoch": 0.5465181058495822, + "grad_norm": 1.4995061159133911, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8766067028045654, + "num_tokens": 103312194.0, + "step": 2943 + }, + { + "epoch": 0.5467038068709378, + "grad_norm": 1.4818496704101562, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.860321581363678, + "num_tokens": 103347551.0, + "step": 2944 + }, + { + "epoch": 0.5468895078922934, + "grad_norm": 1.529788613319397, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8739409446716309, + "num_tokens": 103378033.0, + "step": 2945 + }, + { + "epoch": 0.547075208913649, + "grad_norm": 1.385335087776184, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8709660768508911, + "num_tokens": 103420616.0, + "step": 2946 + }, + { + "epoch": 0.5472609099350046, + "grad_norm": 1.4117906093597412, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8672977685928345, + "num_tokens": 103461590.0, + "step": 2947 + }, + { + "epoch": 0.5474466109563603, + "grad_norm": 1.543813943862915, + "learning_rate": 1e-06, + "loss": 0.4305, + "mean_token_accuracy": 0.8561193943023682, + "num_tokens": 103495556.0, + "step": 2948 + }, + { + "epoch": 0.5476323119777159, + "grad_norm": 1.4193395376205444, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8598055243492126, + "num_tokens": 103532876.0, + "step": 2949 + }, + { + "epoch": 0.5478180129990715, + "grad_norm": 1.4315154552459717, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.870344340801239, + "num_tokens": 103569019.0, + "step": 2950 + }, + { + "epoch": 0.5480037140204271, + "grad_norm": 1.5025465488433838, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8706357479095459, + "num_tokens": 103599398.0, + "step": 2951 + }, + { + "epoch": 0.5481894150417828, + "grad_norm": 1.4665546417236328, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8695495128631592, + "num_tokens": 103634100.0, + "step": 2952 + }, + { + "epoch": 0.5483751160631384, + "grad_norm": 1.4377033710479736, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8708151578903198, + "num_tokens": 103672957.0, + "step": 2953 + }, + { + "epoch": 0.548560817084494, + "grad_norm": 1.543283462524414, + "learning_rate": 1e-06, + "loss": 0.468, + "mean_token_accuracy": 0.845670223236084, + "num_tokens": 103709220.0, + "step": 2954 + }, + { + "epoch": 0.5487465181058496, + "grad_norm": 1.6172364950180054, + "learning_rate": 1e-06, + "loss": 0.4696, + "mean_token_accuracy": 0.8440426588058472, + "num_tokens": 103741872.0, + "step": 2955 + }, + { + "epoch": 0.5489322191272052, + "grad_norm": 1.4731961488723755, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8597791790962219, + "num_tokens": 103778930.0, + "step": 2956 + }, + { + "epoch": 0.5491179201485609, + "grad_norm": 1.4704670906066895, + "learning_rate": 1e-06, + "loss": 0.4, + "mean_token_accuracy": 0.8674108386039734, + "num_tokens": 103812443.0, + "step": 2957 + }, + { + "epoch": 0.5493036211699165, + "grad_norm": 1.374153733253479, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8724377155303955, + "num_tokens": 103854099.0, + "step": 2958 + }, + { + "epoch": 0.549489322191272, + "grad_norm": 1.775674819946289, + "learning_rate": 1e-06, + "loss": 0.4768, + "mean_token_accuracy": 0.8442089557647705, + "num_tokens": 103891282.0, + "step": 2959 + }, + { + "epoch": 0.5496750232126276, + "grad_norm": 1.4600704908370972, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8634527921676636, + "num_tokens": 103925758.0, + "step": 2960 + }, + { + "epoch": 0.5498607242339832, + "grad_norm": 1.3095074892044067, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8639867901802063, + "num_tokens": 103969675.0, + "step": 2961 + }, + { + "epoch": 0.5500464252553389, + "grad_norm": 1.51960027217865, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.871601939201355, + "num_tokens": 104004068.0, + "step": 2962 + }, + { + "epoch": 0.5502321262766945, + "grad_norm": 1.6722533702850342, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.8612664341926575, + "num_tokens": 104033977.0, + "step": 2963 + }, + { + "epoch": 0.5504178272980501, + "grad_norm": 1.4206244945526123, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8741799592971802, + "num_tokens": 104072636.0, + "step": 2964 + }, + { + "epoch": 0.5506035283194057, + "grad_norm": 1.4295846223831177, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8644875884056091, + "num_tokens": 104108735.0, + "step": 2965 + }, + { + "epoch": 0.5507892293407614, + "grad_norm": 1.401210069656372, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8792670965194702, + "num_tokens": 104143150.0, + "step": 2966 + }, + { + "epoch": 0.550974930362117, + "grad_norm": 1.5401145219802856, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8650367259979248, + "num_tokens": 104176272.0, + "step": 2967 + }, + { + "epoch": 0.5511606313834726, + "grad_norm": 1.5817347764968872, + "learning_rate": 1e-06, + "loss": 0.4681, + "mean_token_accuracy": 0.8464888334274292, + "num_tokens": 104214603.0, + "step": 2968 + }, + { + "epoch": 0.5513463324048282, + "grad_norm": 1.4574850797653198, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8664339780807495, + "num_tokens": 104251686.0, + "step": 2969 + }, + { + "epoch": 0.5515320334261838, + "grad_norm": 1.608241319656372, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8675237894058228, + "num_tokens": 104280743.0, + "step": 2970 + }, + { + "epoch": 0.5517177344475395, + "grad_norm": 1.45661461353302, + "learning_rate": 1e-06, + "loss": 0.454, + "mean_token_accuracy": 0.8489992022514343, + "num_tokens": 104318111.0, + "step": 2971 + }, + { + "epoch": 0.5519034354688951, + "grad_norm": 1.476210594177246, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8571730852127075, + "num_tokens": 104352172.0, + "step": 2972 + }, + { + "epoch": 0.5520891364902507, + "grad_norm": 1.561684250831604, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.863240122795105, + "num_tokens": 104387881.0, + "step": 2973 + }, + { + "epoch": 0.5522748375116063, + "grad_norm": 1.5109004974365234, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8674728870391846, + "num_tokens": 104423315.0, + "step": 2974 + }, + { + "epoch": 0.552460538532962, + "grad_norm": 1.6047048568725586, + "learning_rate": 1e-06, + "loss": 0.4145, + "mean_token_accuracy": 0.8592774271965027, + "num_tokens": 104454572.0, + "step": 2975 + }, + { + "epoch": 0.5526462395543176, + "grad_norm": 1.5843708515167236, + "learning_rate": 1e-06, + "loss": 0.407, + "mean_token_accuracy": 0.8626500368118286, + "num_tokens": 104485690.0, + "step": 2976 + }, + { + "epoch": 0.5528319405756732, + "grad_norm": 1.6224693059921265, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8638267517089844, + "num_tokens": 104518264.0, + "step": 2977 + }, + { + "epoch": 0.5530176415970288, + "grad_norm": 1.4372769594192505, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8660397529602051, + "num_tokens": 104554100.0, + "step": 2978 + }, + { + "epoch": 0.5532033426183844, + "grad_norm": 1.3897743225097656, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8696422576904297, + "num_tokens": 104594313.0, + "step": 2979 + }, + { + "epoch": 0.5533890436397401, + "grad_norm": 1.498752474784851, + "learning_rate": 1e-06, + "loss": 0.4085, + "mean_token_accuracy": 0.8658992052078247, + "num_tokens": 104627995.0, + "step": 2980 + }, + { + "epoch": 0.5535747446610957, + "grad_norm": 1.4572871923446655, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8595683574676514, + "num_tokens": 104665419.0, + "step": 2981 + }, + { + "epoch": 0.5537604456824513, + "grad_norm": 1.4877870082855225, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8689259886741638, + "num_tokens": 104699677.0, + "step": 2982 + }, + { + "epoch": 0.5539461467038068, + "grad_norm": 1.4043986797332764, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8724451065063477, + "num_tokens": 104737080.0, + "step": 2983 + }, + { + "epoch": 0.5541318477251624, + "grad_norm": 1.4036632776260376, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.8689889311790466, + "num_tokens": 104772696.0, + "step": 2984 + }, + { + "epoch": 0.5543175487465181, + "grad_norm": 1.5323729515075684, + "learning_rate": 1e-06, + "loss": 0.4424, + "mean_token_accuracy": 0.8539298176765442, + "num_tokens": 104808599.0, + "step": 2985 + }, + { + "epoch": 0.5545032497678737, + "grad_norm": 1.59994637966156, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.8666243553161621, + "num_tokens": 104839668.0, + "step": 2986 + }, + { + "epoch": 0.5546889507892293, + "grad_norm": 1.7015098333358765, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8712502717971802, + "num_tokens": 104866362.0, + "step": 2987 + }, + { + "epoch": 0.5548746518105849, + "grad_norm": 1.4833942651748657, + "learning_rate": 1e-06, + "loss": 0.447, + "mean_token_accuracy": 0.8510693311691284, + "num_tokens": 104904229.0, + "step": 2988 + }, + { + "epoch": 0.5550603528319406, + "grad_norm": 1.4288042783737183, + "learning_rate": 1e-06, + "loss": 0.4176, + "mean_token_accuracy": 0.8611621856689453, + "num_tokens": 104938503.0, + "step": 2989 + }, + { + "epoch": 0.5552460538532962, + "grad_norm": 1.439116358757019, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.861835241317749, + "num_tokens": 104975154.0, + "step": 2990 + }, + { + "epoch": 0.5554317548746518, + "grad_norm": 1.5104316473007202, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8676828742027283, + "num_tokens": 105007195.0, + "step": 2991 + }, + { + "epoch": 0.5556174558960074, + "grad_norm": 1.41012704372406, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8753647804260254, + "num_tokens": 105042176.0, + "step": 2992 + }, + { + "epoch": 0.555803156917363, + "grad_norm": 1.4187253713607788, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8722844123840332, + "num_tokens": 105080222.0, + "step": 2993 + }, + { + "epoch": 0.5559888579387187, + "grad_norm": 1.3350348472595215, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8798905611038208, + "num_tokens": 105118852.0, + "step": 2994 + }, + { + "epoch": 0.5561745589600743, + "grad_norm": 1.5297187566757202, + "learning_rate": 1e-06, + "loss": 0.4416, + "mean_token_accuracy": 0.8532620668411255, + "num_tokens": 105152988.0, + "step": 2995 + }, + { + "epoch": 0.5563602599814299, + "grad_norm": 1.5432686805725098, + "learning_rate": 1e-06, + "loss": 0.4348, + "mean_token_accuracy": 0.8557921648025513, + "num_tokens": 105187631.0, + "step": 2996 + }, + { + "epoch": 0.5565459610027855, + "grad_norm": 1.584495186805725, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8661261796951294, + "num_tokens": 105218721.0, + "step": 2997 + }, + { + "epoch": 0.5567316620241411, + "grad_norm": 1.4892710447311401, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8667886853218079, + "num_tokens": 105252218.0, + "step": 2998 + }, + { + "epoch": 0.5569173630454968, + "grad_norm": 1.2793121337890625, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8731145858764648, + "num_tokens": 105295997.0, + "step": 2999 + }, + { + "epoch": 0.5571030640668524, + "grad_norm": 1.4520009756088257, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8577324151992798, + "num_tokens": 105330226.0, + "step": 3000 + }, + { + "epoch": 0.557288765088208, + "grad_norm": 1.2986468076705933, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8656777143478394, + "num_tokens": 105370442.0, + "step": 3001 + }, + { + "epoch": 0.5574744661095636, + "grad_norm": 1.438421607017517, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8672515749931335, + "num_tokens": 105406121.0, + "step": 3002 + }, + { + "epoch": 0.5576601671309193, + "grad_norm": 1.5473709106445312, + "learning_rate": 1e-06, + "loss": 0.4504, + "mean_token_accuracy": 0.8522449731826782, + "num_tokens": 105439446.0, + "step": 3003 + }, + { + "epoch": 0.5578458681522749, + "grad_norm": 1.355425477027893, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8774534463882446, + "num_tokens": 105477023.0, + "step": 3004 + }, + { + "epoch": 0.5580315691736305, + "grad_norm": 1.403688907623291, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8753929138183594, + "num_tokens": 105511543.0, + "step": 3005 + }, + { + "epoch": 0.5582172701949861, + "grad_norm": 1.4556854963302612, + "learning_rate": 1e-06, + "loss": 0.3698, + "mean_token_accuracy": 0.8749834299087524, + "num_tokens": 105544411.0, + "step": 3006 + }, + { + "epoch": 0.5584029712163416, + "grad_norm": 1.4714105129241943, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8643149137496948, + "num_tokens": 105582636.0, + "step": 3007 + }, + { + "epoch": 0.5585886722376973, + "grad_norm": 1.4577317237854004, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.865149974822998, + "num_tokens": 105617550.0, + "step": 3008 + }, + { + "epoch": 0.5587743732590529, + "grad_norm": 1.4408491849899292, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8719579577445984, + "num_tokens": 105658229.0, + "step": 3009 + }, + { + "epoch": 0.5589600742804085, + "grad_norm": 1.4893989562988281, + "learning_rate": 1e-06, + "loss": 0.4259, + "mean_token_accuracy": 0.857109785079956, + "num_tokens": 105695484.0, + "step": 3010 + }, + { + "epoch": 0.5591457753017641, + "grad_norm": 1.6226166486740112, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8692530393600464, + "num_tokens": 105727366.0, + "step": 3011 + }, + { + "epoch": 0.5593314763231197, + "grad_norm": 1.6878582239151, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.861607015132904, + "num_tokens": 105754550.0, + "step": 3012 + }, + { + "epoch": 0.5595171773444754, + "grad_norm": 1.6075491905212402, + "learning_rate": 1e-06, + "loss": 0.4232, + "mean_token_accuracy": 0.858980655670166, + "num_tokens": 105785153.0, + "step": 3013 + }, + { + "epoch": 0.559702878365831, + "grad_norm": 1.4229985475540161, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.8670248985290527, + "num_tokens": 105824400.0, + "step": 3014 + }, + { + "epoch": 0.5598885793871866, + "grad_norm": 1.390278697013855, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8714529275894165, + "num_tokens": 105860521.0, + "step": 3015 + }, + { + "epoch": 0.5600742804085422, + "grad_norm": 1.3954746723175049, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8780702352523804, + "num_tokens": 105895864.0, + "step": 3016 + }, + { + "epoch": 0.5602599814298979, + "grad_norm": 1.444397211074829, + "learning_rate": 1e-06, + "loss": 0.4647, + "mean_token_accuracy": 0.8517561554908752, + "num_tokens": 105936339.0, + "step": 3017 + }, + { + "epoch": 0.5604456824512535, + "grad_norm": 1.3601669073104858, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8662182092666626, + "num_tokens": 105977455.0, + "step": 3018 + }, + { + "epoch": 0.5606313834726091, + "grad_norm": 1.5513739585876465, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8644242286682129, + "num_tokens": 106009100.0, + "step": 3019 + }, + { + "epoch": 0.5608170844939647, + "grad_norm": 1.5956498384475708, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8712162375450134, + "num_tokens": 106040630.0, + "step": 3020 + }, + { + "epoch": 0.5610027855153203, + "grad_norm": 1.4796288013458252, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8684936165809631, + "num_tokens": 106071876.0, + "step": 3021 + }, + { + "epoch": 0.561188486536676, + "grad_norm": 1.5616933107376099, + "learning_rate": 1e-06, + "loss": 0.4785, + "mean_token_accuracy": 0.8426852226257324, + "num_tokens": 106103917.0, + "step": 3022 + }, + { + "epoch": 0.5613741875580316, + "grad_norm": 1.5572222471237183, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8571903705596924, + "num_tokens": 106136583.0, + "step": 3023 + }, + { + "epoch": 0.5615598885793872, + "grad_norm": 1.511776089668274, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8681233525276184, + "num_tokens": 106168999.0, + "step": 3024 + }, + { + "epoch": 0.5617455896007428, + "grad_norm": 1.4363925457000732, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8660212159156799, + "num_tokens": 106207442.0, + "step": 3025 + }, + { + "epoch": 0.5619312906220985, + "grad_norm": 1.448846459388733, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.8722308278083801, + "num_tokens": 106244157.0, + "step": 3026 + }, + { + "epoch": 0.5621169916434541, + "grad_norm": 1.395193099975586, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8739328384399414, + "num_tokens": 106282822.0, + "step": 3027 + }, + { + "epoch": 0.5623026926648097, + "grad_norm": 1.7601736783981323, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8668721318244934, + "num_tokens": 106308451.0, + "step": 3028 + }, + { + "epoch": 0.5624883936861653, + "grad_norm": 1.4714858531951904, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.8651747703552246, + "num_tokens": 106346353.0, + "step": 3029 + }, + { + "epoch": 0.5626740947075209, + "grad_norm": 1.4728460311889648, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8596330881118774, + "num_tokens": 106382571.0, + "step": 3030 + }, + { + "epoch": 0.5628597957288765, + "grad_norm": 1.5199453830718994, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.8590928912162781, + "num_tokens": 106421744.0, + "step": 3031 + }, + { + "epoch": 0.5630454967502321, + "grad_norm": 1.3565243482589722, + "learning_rate": 1e-06, + "loss": 0.4089, + "mean_token_accuracy": 0.8612964749336243, + "num_tokens": 106462658.0, + "step": 3032 + }, + { + "epoch": 0.5632311977715877, + "grad_norm": 1.4865511655807495, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8812607526779175, + "num_tokens": 106493281.0, + "step": 3033 + }, + { + "epoch": 0.5634168987929433, + "grad_norm": 1.7233015298843384, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8726844787597656, + "num_tokens": 106527879.0, + "step": 3034 + }, + { + "epoch": 0.5636025998142989, + "grad_norm": 1.4869595766067505, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8567392826080322, + "num_tokens": 106563718.0, + "step": 3035 + }, + { + "epoch": 0.5637883008356546, + "grad_norm": 1.549703598022461, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8678501844406128, + "num_tokens": 106594787.0, + "step": 3036 + }, + { + "epoch": 0.5639740018570102, + "grad_norm": 1.6330188512802124, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8567841053009033, + "num_tokens": 106622849.0, + "step": 3037 + }, + { + "epoch": 0.5641597028783658, + "grad_norm": 1.379385232925415, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8740877509117126, + "num_tokens": 106662510.0, + "step": 3038 + }, + { + "epoch": 0.5643454038997214, + "grad_norm": 1.4018771648406982, + "learning_rate": 1e-06, + "loss": 0.4348, + "mean_token_accuracy": 0.8609975576400757, + "num_tokens": 106699951.0, + "step": 3039 + }, + { + "epoch": 0.564531104921077, + "grad_norm": 1.4662867784500122, + "learning_rate": 1e-06, + "loss": 0.4165, + "mean_token_accuracy": 0.8613410592079163, + "num_tokens": 106735630.0, + "step": 3040 + }, + { + "epoch": 0.5647168059424327, + "grad_norm": 1.35671865940094, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8814457654953003, + "num_tokens": 106773371.0, + "step": 3041 + }, + { + "epoch": 0.5649025069637883, + "grad_norm": 1.5939301252365112, + "learning_rate": 1e-06, + "loss": 0.4498, + "mean_token_accuracy": 0.8490623831748962, + "num_tokens": 106805840.0, + "step": 3042 + }, + { + "epoch": 0.5650882079851439, + "grad_norm": 1.6965878009796143, + "learning_rate": 1e-06, + "loss": 0.4515, + "mean_token_accuracy": 0.8464418649673462, + "num_tokens": 106844074.0, + "step": 3043 + }, + { + "epoch": 0.5652739090064995, + "grad_norm": 1.5074430704116821, + "learning_rate": 1e-06, + "loss": 0.4241, + "mean_token_accuracy": 0.8611171841621399, + "num_tokens": 106880837.0, + "step": 3044 + }, + { + "epoch": 0.5654596100278552, + "grad_norm": 1.4928357601165771, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8669837713241577, + "num_tokens": 106914326.0, + "step": 3045 + }, + { + "epoch": 0.5656453110492108, + "grad_norm": 1.3899592161178589, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8547102212905884, + "num_tokens": 106957312.0, + "step": 3046 + }, + { + "epoch": 0.5658310120705664, + "grad_norm": 1.363591194152832, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8641067147254944, + "num_tokens": 106995148.0, + "step": 3047 + }, + { + "epoch": 0.566016713091922, + "grad_norm": 1.3436930179595947, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8806909322738647, + "num_tokens": 107033150.0, + "step": 3048 + }, + { + "epoch": 0.5662024141132777, + "grad_norm": 1.4803508520126343, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.867904543876648, + "num_tokens": 107069292.0, + "step": 3049 + }, + { + "epoch": 0.5663881151346333, + "grad_norm": 1.554972529411316, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8631044030189514, + "num_tokens": 107103127.0, + "step": 3050 + }, + { + "epoch": 0.5665738161559889, + "grad_norm": 1.4596574306488037, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8772430419921875, + "num_tokens": 107138321.0, + "step": 3051 + }, + { + "epoch": 0.5667595171773445, + "grad_norm": 1.3494549989700317, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8762947916984558, + "num_tokens": 107176900.0, + "step": 3052 + }, + { + "epoch": 0.5669452181987001, + "grad_norm": 1.568415641784668, + "learning_rate": 1e-06, + "loss": 0.429, + "mean_token_accuracy": 0.8576103448867798, + "num_tokens": 107213233.0, + "step": 3053 + }, + { + "epoch": 0.5671309192200558, + "grad_norm": 1.486454963684082, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8670877814292908, + "num_tokens": 107247121.0, + "step": 3054 + }, + { + "epoch": 0.5673166202414113, + "grad_norm": 1.3066719770431519, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8744036555290222, + "num_tokens": 107286748.0, + "step": 3055 + }, + { + "epoch": 0.5675023212627669, + "grad_norm": 1.3989983797073364, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8706868886947632, + "num_tokens": 107322884.0, + "step": 3056 + }, + { + "epoch": 0.5676880222841225, + "grad_norm": 1.4188241958618164, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8761314153671265, + "num_tokens": 107357020.0, + "step": 3057 + }, + { + "epoch": 0.5678737233054781, + "grad_norm": 1.5289162397384644, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8707102537155151, + "num_tokens": 107387599.0, + "step": 3058 + }, + { + "epoch": 0.5680594243268338, + "grad_norm": 1.43552565574646, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8739715218544006, + "num_tokens": 107423750.0, + "step": 3059 + }, + { + "epoch": 0.5682451253481894, + "grad_norm": 1.3539103269577026, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8643208146095276, + "num_tokens": 107464672.0, + "step": 3060 + }, + { + "epoch": 0.568430826369545, + "grad_norm": 1.4975391626358032, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8747420310974121, + "num_tokens": 107498065.0, + "step": 3061 + }, + { + "epoch": 0.5686165273909006, + "grad_norm": 1.6731586456298828, + "learning_rate": 1e-06, + "loss": 0.4182, + "mean_token_accuracy": 0.8575788736343384, + "num_tokens": 107531275.0, + "step": 3062 + }, + { + "epoch": 0.5688022284122562, + "grad_norm": 1.522275686264038, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8712226152420044, + "num_tokens": 107563284.0, + "step": 3063 + }, + { + "epoch": 0.5689879294336119, + "grad_norm": 1.6395456790924072, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.873476505279541, + "num_tokens": 107591036.0, + "step": 3064 + }, + { + "epoch": 0.5691736304549675, + "grad_norm": 1.4832061529159546, + "learning_rate": 1e-06, + "loss": 0.4266, + "mean_token_accuracy": 0.8593735098838806, + "num_tokens": 107625581.0, + "step": 3065 + }, + { + "epoch": 0.5693593314763231, + "grad_norm": 1.4264512062072754, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8853203058242798, + "num_tokens": 107658087.0, + "step": 3066 + }, + { + "epoch": 0.5695450324976787, + "grad_norm": 1.4685232639312744, + "learning_rate": 1e-06, + "loss": 0.423, + "mean_token_accuracy": 0.8552031517028809, + "num_tokens": 107696822.0, + "step": 3067 + }, + { + "epoch": 0.5697307335190344, + "grad_norm": 1.5291192531585693, + "learning_rate": 1e-06, + "loss": 0.4086, + "mean_token_accuracy": 0.8664661645889282, + "num_tokens": 107730345.0, + "step": 3068 + }, + { + "epoch": 0.56991643454039, + "grad_norm": 1.4971389770507812, + "learning_rate": 1e-06, + "loss": 0.4275, + "mean_token_accuracy": 0.8568614721298218, + "num_tokens": 107771610.0, + "step": 3069 + }, + { + "epoch": 0.5701021355617456, + "grad_norm": 1.5551278591156006, + "learning_rate": 1e-06, + "loss": 0.3991, + "mean_token_accuracy": 0.8661575317382812, + "num_tokens": 107805328.0, + "step": 3070 + }, + { + "epoch": 0.5702878365831012, + "grad_norm": 1.5153439044952393, + "learning_rate": 1e-06, + "loss": 0.4276, + "mean_token_accuracy": 0.8573393225669861, + "num_tokens": 107836353.0, + "step": 3071 + }, + { + "epoch": 0.5704735376044568, + "grad_norm": 1.3850057125091553, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8812424540519714, + "num_tokens": 107871753.0, + "step": 3072 + }, + { + "epoch": 0.5706592386258125, + "grad_norm": 1.4145545959472656, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8642995953559875, + "num_tokens": 107907876.0, + "step": 3073 + }, + { + "epoch": 0.5708449396471681, + "grad_norm": 1.481163740158081, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8648856282234192, + "num_tokens": 107942681.0, + "step": 3074 + }, + { + "epoch": 0.5710306406685237, + "grad_norm": 1.5554144382476807, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8570640087127686, + "num_tokens": 107978133.0, + "step": 3075 + }, + { + "epoch": 0.5712163416898793, + "grad_norm": 1.5428204536437988, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8712533712387085, + "num_tokens": 108013423.0, + "step": 3076 + }, + { + "epoch": 0.571402042711235, + "grad_norm": 1.4205265045166016, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8774498701095581, + "num_tokens": 108047502.0, + "step": 3077 + }, + { + "epoch": 0.5715877437325906, + "grad_norm": 1.368078589439392, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8659745454788208, + "num_tokens": 108089889.0, + "step": 3078 + }, + { + "epoch": 0.5717734447539461, + "grad_norm": 1.4261664152145386, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8658143281936646, + "num_tokens": 108126198.0, + "step": 3079 + }, + { + "epoch": 0.5719591457753017, + "grad_norm": 1.4790167808532715, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8720365166664124, + "num_tokens": 108162509.0, + "step": 3080 + }, + { + "epoch": 0.5721448467966573, + "grad_norm": 1.4168696403503418, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8660995960235596, + "num_tokens": 108199582.0, + "step": 3081 + }, + { + "epoch": 0.572330547818013, + "grad_norm": 1.5444387197494507, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8685659170150757, + "num_tokens": 108232024.0, + "step": 3082 + }, + { + "epoch": 0.5725162488393686, + "grad_norm": 1.3740994930267334, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8821556568145752, + "num_tokens": 108271324.0, + "step": 3083 + }, + { + "epoch": 0.5727019498607242, + "grad_norm": 1.5538816452026367, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8590574264526367, + "num_tokens": 108303193.0, + "step": 3084 + }, + { + "epoch": 0.5728876508820798, + "grad_norm": 1.5082861185073853, + "learning_rate": 1e-06, + "loss": 0.426, + "mean_token_accuracy": 0.8620551824569702, + "num_tokens": 108338889.0, + "step": 3085 + }, + { + "epoch": 0.5730733519034354, + "grad_norm": 1.609868049621582, + "learning_rate": 1e-06, + "loss": 0.3993, + "mean_token_accuracy": 0.865456223487854, + "num_tokens": 108370169.0, + "step": 3086 + }, + { + "epoch": 0.5732590529247911, + "grad_norm": 1.7691344022750854, + "learning_rate": 1e-06, + "loss": 0.4527, + "mean_token_accuracy": 0.8576333522796631, + "num_tokens": 108396251.0, + "step": 3087 + }, + { + "epoch": 0.5734447539461467, + "grad_norm": 1.4508963823318481, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8769065141677856, + "num_tokens": 108431942.0, + "step": 3088 + }, + { + "epoch": 0.5736304549675023, + "grad_norm": 1.3998064994812012, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8631671071052551, + "num_tokens": 108469577.0, + "step": 3089 + }, + { + "epoch": 0.5738161559888579, + "grad_norm": 1.525235891342163, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8670732975006104, + "num_tokens": 108499113.0, + "step": 3090 + }, + { + "epoch": 0.5740018570102136, + "grad_norm": 1.560835838317871, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8709467649459839, + "num_tokens": 108532001.0, + "step": 3091 + }, + { + "epoch": 0.5741875580315692, + "grad_norm": 1.4606406688690186, + "learning_rate": 1e-06, + "loss": 0.4305, + "mean_token_accuracy": 0.8547304272651672, + "num_tokens": 108567180.0, + "step": 3092 + }, + { + "epoch": 0.5743732590529248, + "grad_norm": 1.3900785446166992, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8717983365058899, + "num_tokens": 108603933.0, + "step": 3093 + }, + { + "epoch": 0.5745589600742804, + "grad_norm": 1.4903335571289062, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.8706786036491394, + "num_tokens": 108635493.0, + "step": 3094 + }, + { + "epoch": 0.574744661095636, + "grad_norm": 1.421636939048767, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8796199560165405, + "num_tokens": 108671573.0, + "step": 3095 + }, + { + "epoch": 0.5749303621169917, + "grad_norm": 1.5987586975097656, + "learning_rate": 1e-06, + "loss": 0.4298, + "mean_token_accuracy": 0.855233907699585, + "num_tokens": 108704095.0, + "step": 3096 + }, + { + "epoch": 0.5751160631383473, + "grad_norm": 1.549889087677002, + "learning_rate": 1e-06, + "loss": 0.4313, + "mean_token_accuracy": 0.8568845987319946, + "num_tokens": 108740128.0, + "step": 3097 + }, + { + "epoch": 0.5753017641597029, + "grad_norm": 1.4651451110839844, + "learning_rate": 1e-06, + "loss": 0.4563, + "mean_token_accuracy": 0.850060760974884, + "num_tokens": 108782383.0, + "step": 3098 + }, + { + "epoch": 0.5754874651810585, + "grad_norm": 1.640693187713623, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.874483585357666, + "num_tokens": 108810943.0, + "step": 3099 + }, + { + "epoch": 0.5756731662024142, + "grad_norm": 1.6225643157958984, + "learning_rate": 1e-06, + "loss": 0.4811, + "mean_token_accuracy": 0.8388683795928955, + "num_tokens": 108844360.0, + "step": 3100 + }, + { + "epoch": 0.5758588672237698, + "grad_norm": 1.4034610986709595, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.869165301322937, + "num_tokens": 108880369.0, + "step": 3101 + }, + { + "epoch": 0.5760445682451254, + "grad_norm": 1.4405616521835327, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8754898309707642, + "num_tokens": 108914204.0, + "step": 3102 + }, + { + "epoch": 0.576230269266481, + "grad_norm": 1.6074081659317017, + "learning_rate": 1e-06, + "loss": 0.4374, + "mean_token_accuracy": 0.8450477719306946, + "num_tokens": 108943506.0, + "step": 3103 + }, + { + "epoch": 0.5764159702878365, + "grad_norm": 1.5397429466247559, + "learning_rate": 1e-06, + "loss": 0.4212, + "mean_token_accuracy": 0.8584700226783752, + "num_tokens": 108979611.0, + "step": 3104 + }, + { + "epoch": 0.5766016713091922, + "grad_norm": 1.4912196397781372, + "learning_rate": 1e-06, + "loss": 0.4238, + "mean_token_accuracy": 0.8566613793373108, + "num_tokens": 109013499.0, + "step": 3105 + }, + { + "epoch": 0.5767873723305478, + "grad_norm": 1.524368405342102, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8617110848426819, + "num_tokens": 109049096.0, + "step": 3106 + }, + { + "epoch": 0.5769730733519034, + "grad_norm": 1.4535802602767944, + "learning_rate": 1e-06, + "loss": 0.4197, + "mean_token_accuracy": 0.8606884479522705, + "num_tokens": 109084967.0, + "step": 3107 + }, + { + "epoch": 0.577158774373259, + "grad_norm": 1.4500020742416382, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8554012775421143, + "num_tokens": 109124999.0, + "step": 3108 + }, + { + "epoch": 0.5773444753946146, + "grad_norm": 1.4660117626190186, + "learning_rate": 1e-06, + "loss": 0.4199, + "mean_token_accuracy": 0.8642070889472961, + "num_tokens": 109163174.0, + "step": 3109 + }, + { + "epoch": 0.5775301764159703, + "grad_norm": 1.5151410102844238, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8620747327804565, + "num_tokens": 109196205.0, + "step": 3110 + }, + { + "epoch": 0.5777158774373259, + "grad_norm": 1.5305958986282349, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8647947907447815, + "num_tokens": 109230013.0, + "step": 3111 + }, + { + "epoch": 0.5779015784586815, + "grad_norm": 1.419809341430664, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8716652989387512, + "num_tokens": 109270289.0, + "step": 3112 + }, + { + "epoch": 0.5780872794800371, + "grad_norm": 1.3448927402496338, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8673360347747803, + "num_tokens": 109312852.0, + "step": 3113 + }, + { + "epoch": 0.5782729805013928, + "grad_norm": 1.52225661277771, + "learning_rate": 1e-06, + "loss": 0.409, + "mean_token_accuracy": 0.8609833717346191, + "num_tokens": 109347011.0, + "step": 3114 + }, + { + "epoch": 0.5784586815227484, + "grad_norm": 1.525153636932373, + "learning_rate": 1e-06, + "loss": 0.4039, + "mean_token_accuracy": 0.8664754629135132, + "num_tokens": 109378638.0, + "step": 3115 + }, + { + "epoch": 0.578644382544104, + "grad_norm": 1.4406167268753052, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.87962806224823, + "num_tokens": 109411676.0, + "step": 3116 + }, + { + "epoch": 0.5788300835654596, + "grad_norm": 1.4103264808654785, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8788682818412781, + "num_tokens": 109448695.0, + "step": 3117 + }, + { + "epoch": 0.5790157845868152, + "grad_norm": 1.4870034456253052, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.8552806973457336, + "num_tokens": 109482056.0, + "step": 3118 + }, + { + "epoch": 0.5792014856081709, + "grad_norm": 1.464033842086792, + "learning_rate": 1e-06, + "loss": 0.4598, + "mean_token_accuracy": 0.8443112373352051, + "num_tokens": 109520095.0, + "step": 3119 + }, + { + "epoch": 0.5793871866295265, + "grad_norm": 1.5290310382843018, + "learning_rate": 1e-06, + "loss": 0.4049, + "mean_token_accuracy": 0.8653832674026489, + "num_tokens": 109552426.0, + "step": 3120 + }, + { + "epoch": 0.5795728876508821, + "grad_norm": 1.4914597272872925, + "learning_rate": 1e-06, + "loss": 0.4337, + "mean_token_accuracy": 0.85475093126297, + "num_tokens": 109587312.0, + "step": 3121 + }, + { + "epoch": 0.5797585886722377, + "grad_norm": 1.5451617240905762, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8623249530792236, + "num_tokens": 109618522.0, + "step": 3122 + }, + { + "epoch": 0.5799442896935934, + "grad_norm": 1.4898459911346436, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8760234713554382, + "num_tokens": 109652185.0, + "step": 3123 + }, + { + "epoch": 0.580129990714949, + "grad_norm": 1.6231476068496704, + "learning_rate": 1e-06, + "loss": 0.4309, + "mean_token_accuracy": 0.8579404354095459, + "num_tokens": 109683466.0, + "step": 3124 + }, + { + "epoch": 0.5803156917363046, + "grad_norm": 1.48130464553833, + "learning_rate": 1e-06, + "loss": 0.4781, + "mean_token_accuracy": 0.8414827585220337, + "num_tokens": 109722433.0, + "step": 3125 + }, + { + "epoch": 0.5805013927576602, + "grad_norm": 1.4235354661941528, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8697983026504517, + "num_tokens": 109757713.0, + "step": 3126 + }, + { + "epoch": 0.5806870937790158, + "grad_norm": 1.5657274723052979, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8768470287322998, + "num_tokens": 109785721.0, + "step": 3127 + }, + { + "epoch": 0.5808727948003714, + "grad_norm": 1.5774192810058594, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8612363338470459, + "num_tokens": 109820365.0, + "step": 3128 + }, + { + "epoch": 0.581058495821727, + "grad_norm": 1.4797569513320923, + "learning_rate": 1e-06, + "loss": 0.5142, + "mean_token_accuracy": 0.8399315476417542, + "num_tokens": 109858589.0, + "step": 3129 + }, + { + "epoch": 0.5812441968430826, + "grad_norm": 1.5033740997314453, + "learning_rate": 1e-06, + "loss": 0.4156, + "mean_token_accuracy": 0.8604862689971924, + "num_tokens": 109895421.0, + "step": 3130 + }, + { + "epoch": 0.5814298978644382, + "grad_norm": 1.5332050323486328, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8781673908233643, + "num_tokens": 109923261.0, + "step": 3131 + }, + { + "epoch": 0.5816155988857938, + "grad_norm": 1.6173205375671387, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8573958873748779, + "num_tokens": 109952638.0, + "step": 3132 + }, + { + "epoch": 0.5818012999071495, + "grad_norm": 1.4148788452148438, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.865686297416687, + "num_tokens": 109992880.0, + "step": 3133 + }, + { + "epoch": 0.5819870009285051, + "grad_norm": 1.3920183181762695, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.8789078593254089, + "num_tokens": 110029473.0, + "step": 3134 + }, + { + "epoch": 0.5821727019498607, + "grad_norm": 1.4166308641433716, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8678529262542725, + "num_tokens": 110065433.0, + "step": 3135 + }, + { + "epoch": 0.5823584029712163, + "grad_norm": 1.4054216146469116, + "learning_rate": 1e-06, + "loss": 0.4281, + "mean_token_accuracy": 0.8546513915061951, + "num_tokens": 110104718.0, + "step": 3136 + }, + { + "epoch": 0.582544103992572, + "grad_norm": 1.447013258934021, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.867416501045227, + "num_tokens": 110139584.0, + "step": 3137 + }, + { + "epoch": 0.5827298050139276, + "grad_norm": 1.6447603702545166, + "learning_rate": 1e-06, + "loss": 0.442, + "mean_token_accuracy": 0.8507980108261108, + "num_tokens": 110176404.0, + "step": 3138 + }, + { + "epoch": 0.5829155060352832, + "grad_norm": 1.5709857940673828, + "learning_rate": 1e-06, + "loss": 0.4257, + "mean_token_accuracy": 0.8567639589309692, + "num_tokens": 110209009.0, + "step": 3139 + }, + { + "epoch": 0.5831012070566388, + "grad_norm": 1.373815894126892, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8699792623519897, + "num_tokens": 110245681.0, + "step": 3140 + }, + { + "epoch": 0.5832869080779944, + "grad_norm": 1.322949767112732, + "learning_rate": 1e-06, + "loss": 0.4304, + "mean_token_accuracy": 0.8541291952133179, + "num_tokens": 110290726.0, + "step": 3141 + }, + { + "epoch": 0.5834726090993501, + "grad_norm": 1.5514620542526245, + "learning_rate": 1e-06, + "loss": 0.4266, + "mean_token_accuracy": 0.8596206307411194, + "num_tokens": 110324784.0, + "step": 3142 + }, + { + "epoch": 0.5836583101207057, + "grad_norm": 1.4692083597183228, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8755532503128052, + "num_tokens": 110358958.0, + "step": 3143 + }, + { + "epoch": 0.5838440111420613, + "grad_norm": 1.377461552619934, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8711427450180054, + "num_tokens": 110399495.0, + "step": 3144 + }, + { + "epoch": 0.5840297121634169, + "grad_norm": 1.464421272277832, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8690829277038574, + "num_tokens": 110435580.0, + "step": 3145 + }, + { + "epoch": 0.5842154131847725, + "grad_norm": 1.4338542222976685, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8737183809280396, + "num_tokens": 110467650.0, + "step": 3146 + }, + { + "epoch": 0.5844011142061282, + "grad_norm": 1.5484113693237305, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8581846952438354, + "num_tokens": 110499845.0, + "step": 3147 + }, + { + "epoch": 0.5845868152274838, + "grad_norm": 1.4482414722442627, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8752939701080322, + "num_tokens": 110535830.0, + "step": 3148 + }, + { + "epoch": 0.5847725162488394, + "grad_norm": 1.617756962776184, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.874495267868042, + "num_tokens": 110567310.0, + "step": 3149 + }, + { + "epoch": 0.584958217270195, + "grad_norm": 1.5516492128372192, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8649832606315613, + "num_tokens": 110607292.0, + "step": 3150 + }, + { + "epoch": 0.5851439182915507, + "grad_norm": 1.4418771266937256, + "learning_rate": 1e-06, + "loss": 0.4419, + "mean_token_accuracy": 0.8498275876045227, + "num_tokens": 110648348.0, + "step": 3151 + }, + { + "epoch": 0.5853296193129062, + "grad_norm": 1.5093706846237183, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.8573949337005615, + "num_tokens": 110681914.0, + "step": 3152 + }, + { + "epoch": 0.5855153203342618, + "grad_norm": 1.5602184534072876, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8657054901123047, + "num_tokens": 110711854.0, + "step": 3153 + }, + { + "epoch": 0.5857010213556174, + "grad_norm": 1.6595823764801025, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8599889874458313, + "num_tokens": 110741553.0, + "step": 3154 + }, + { + "epoch": 0.585886722376973, + "grad_norm": 1.4130291938781738, + "learning_rate": 1e-06, + "loss": 0.4623, + "mean_token_accuracy": 0.8468543291091919, + "num_tokens": 110783008.0, + "step": 3155 + }, + { + "epoch": 0.5860724233983287, + "grad_norm": 1.4790294170379639, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8601041436195374, + "num_tokens": 110817073.0, + "step": 3156 + }, + { + "epoch": 0.5862581244196843, + "grad_norm": 1.3810762166976929, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8798579573631287, + "num_tokens": 110857560.0, + "step": 3157 + }, + { + "epoch": 0.5864438254410399, + "grad_norm": 1.4265772104263306, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8711577653884888, + "num_tokens": 110892495.0, + "step": 3158 + }, + { + "epoch": 0.5866295264623955, + "grad_norm": 1.3852227926254272, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8669221997261047, + "num_tokens": 110929671.0, + "step": 3159 + }, + { + "epoch": 0.5868152274837511, + "grad_norm": 1.6651861667633057, + "learning_rate": 1e-06, + "loss": 0.4348, + "mean_token_accuracy": 0.8562161326408386, + "num_tokens": 110965154.0, + "step": 3160 + }, + { + "epoch": 0.5870009285051068, + "grad_norm": 1.499557375907898, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8656136393547058, + "num_tokens": 110999885.0, + "step": 3161 + }, + { + "epoch": 0.5871866295264624, + "grad_norm": 1.4803080558776855, + "learning_rate": 1e-06, + "loss": 0.41, + "mean_token_accuracy": 0.8560152053833008, + "num_tokens": 111034938.0, + "step": 3162 + }, + { + "epoch": 0.587372330547818, + "grad_norm": 1.4740253686904907, + "learning_rate": 1e-06, + "loss": 0.4454, + "mean_token_accuracy": 0.856190025806427, + "num_tokens": 111071281.0, + "step": 3163 + }, + { + "epoch": 0.5875580315691736, + "grad_norm": 1.5227088928222656, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8611716032028198, + "num_tokens": 111105262.0, + "step": 3164 + }, + { + "epoch": 0.5877437325905293, + "grad_norm": 1.3774625062942505, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.8626527190208435, + "num_tokens": 111145321.0, + "step": 3165 + }, + { + "epoch": 0.5879294336118849, + "grad_norm": 1.5565879344940186, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8632072806358337, + "num_tokens": 111178613.0, + "step": 3166 + }, + { + "epoch": 0.5881151346332405, + "grad_norm": 1.5819337368011475, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8590534329414368, + "num_tokens": 111213333.0, + "step": 3167 + }, + { + "epoch": 0.5883008356545961, + "grad_norm": 1.3763362169265747, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8743268251419067, + "num_tokens": 111253036.0, + "step": 3168 + }, + { + "epoch": 0.5884865366759517, + "grad_norm": 1.5576261281967163, + "learning_rate": 1e-06, + "loss": 0.4494, + "mean_token_accuracy": 0.8546668887138367, + "num_tokens": 111284288.0, + "step": 3169 + }, + { + "epoch": 0.5886722376973074, + "grad_norm": 1.499884843826294, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8652059435844421, + "num_tokens": 111319132.0, + "step": 3170 + }, + { + "epoch": 0.588857938718663, + "grad_norm": 1.4604874849319458, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.863771915435791, + "num_tokens": 111355408.0, + "step": 3171 + }, + { + "epoch": 0.5890436397400186, + "grad_norm": 1.6374328136444092, + "learning_rate": 1e-06, + "loss": 0.4096, + "mean_token_accuracy": 0.8627380132675171, + "num_tokens": 111385967.0, + "step": 3172 + }, + { + "epoch": 0.5892293407613742, + "grad_norm": 1.6328668594360352, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8662426471710205, + "num_tokens": 111416203.0, + "step": 3173 + }, + { + "epoch": 0.5894150417827299, + "grad_norm": 1.501347541809082, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8651024103164673, + "num_tokens": 111455243.0, + "step": 3174 + }, + { + "epoch": 0.5896007428040855, + "grad_norm": 1.3378512859344482, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8694455027580261, + "num_tokens": 111499617.0, + "step": 3175 + }, + { + "epoch": 0.589786443825441, + "grad_norm": 1.4583499431610107, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8705124258995056, + "num_tokens": 111533489.0, + "step": 3176 + }, + { + "epoch": 0.5899721448467966, + "grad_norm": 1.441172480583191, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8657184839248657, + "num_tokens": 111570713.0, + "step": 3177 + }, + { + "epoch": 0.5901578458681522, + "grad_norm": 1.3775190114974976, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8718806505203247, + "num_tokens": 111609497.0, + "step": 3178 + }, + { + "epoch": 0.5903435468895079, + "grad_norm": 1.5976901054382324, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.872994601726532, + "num_tokens": 111638603.0, + "step": 3179 + }, + { + "epoch": 0.5905292479108635, + "grad_norm": 1.4527390003204346, + "learning_rate": 1e-06, + "loss": 0.4525, + "mean_token_accuracy": 0.8495190739631653, + "num_tokens": 111679056.0, + "step": 3180 + }, + { + "epoch": 0.5907149489322191, + "grad_norm": 1.4764034748077393, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8725529313087463, + "num_tokens": 111712450.0, + "step": 3181 + }, + { + "epoch": 0.5909006499535747, + "grad_norm": 1.412812352180481, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8759621381759644, + "num_tokens": 111747999.0, + "step": 3182 + }, + { + "epoch": 0.5910863509749303, + "grad_norm": 1.3824373483657837, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.870629072189331, + "num_tokens": 111784420.0, + "step": 3183 + }, + { + "epoch": 0.591272051996286, + "grad_norm": 1.52138090133667, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.870978593826294, + "num_tokens": 111817632.0, + "step": 3184 + }, + { + "epoch": 0.5914577530176416, + "grad_norm": 1.5095248222351074, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8695809841156006, + "num_tokens": 111851886.0, + "step": 3185 + }, + { + "epoch": 0.5916434540389972, + "grad_norm": 1.598530650138855, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.861935019493103, + "num_tokens": 111881347.0, + "step": 3186 + }, + { + "epoch": 0.5918291550603528, + "grad_norm": 1.6543142795562744, + "learning_rate": 1e-06, + "loss": 0.4343, + "mean_token_accuracy": 0.8528863191604614, + "num_tokens": 111910799.0, + "step": 3187 + }, + { + "epoch": 0.5920148560817085, + "grad_norm": 1.4415743350982666, + "learning_rate": 1e-06, + "loss": 0.4089, + "mean_token_accuracy": 0.86287522315979, + "num_tokens": 111949261.0, + "step": 3188 + }, + { + "epoch": 0.5922005571030641, + "grad_norm": 1.5138565301895142, + "learning_rate": 1e-06, + "loss": 0.4273, + "mean_token_accuracy": 0.8573057651519775, + "num_tokens": 111985704.0, + "step": 3189 + }, + { + "epoch": 0.5923862581244197, + "grad_norm": 1.5861210823059082, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8686889410018921, + "num_tokens": 112013488.0, + "step": 3190 + }, + { + "epoch": 0.5925719591457753, + "grad_norm": 1.512541651725769, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8654052019119263, + "num_tokens": 112047574.0, + "step": 3191 + }, + { + "epoch": 0.5927576601671309, + "grad_norm": 1.4429088830947876, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8653408288955688, + "num_tokens": 112082340.0, + "step": 3192 + }, + { + "epoch": 0.5929433611884866, + "grad_norm": 1.4900074005126953, + "learning_rate": 1e-06, + "loss": 0.4126, + "mean_token_accuracy": 0.8628304600715637, + "num_tokens": 112118951.0, + "step": 3193 + }, + { + "epoch": 0.5931290622098422, + "grad_norm": 1.531227946281433, + "learning_rate": 1e-06, + "loss": 0.4226, + "mean_token_accuracy": 0.8584507703781128, + "num_tokens": 112154009.0, + "step": 3194 + }, + { + "epoch": 0.5933147632311978, + "grad_norm": 1.508695125579834, + "learning_rate": 1e-06, + "loss": 0.4386, + "mean_token_accuracy": 0.8508501052856445, + "num_tokens": 112188924.0, + "step": 3195 + }, + { + "epoch": 0.5935004642525534, + "grad_norm": 1.5438241958618164, + "learning_rate": 1e-06, + "loss": 0.4596, + "mean_token_accuracy": 0.8502774834632874, + "num_tokens": 112224152.0, + "step": 3196 + }, + { + "epoch": 0.593686165273909, + "grad_norm": 1.5518112182617188, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8675413131713867, + "num_tokens": 112253827.0, + "step": 3197 + }, + { + "epoch": 0.5938718662952647, + "grad_norm": 1.633603811264038, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8679229021072388, + "num_tokens": 112284459.0, + "step": 3198 + }, + { + "epoch": 0.5940575673166203, + "grad_norm": 1.4916467666625977, + "learning_rate": 1e-06, + "loss": 0.4039, + "mean_token_accuracy": 0.8643943071365356, + "num_tokens": 112316729.0, + "step": 3199 + }, + { + "epoch": 0.5942432683379758, + "grad_norm": 1.5497246980667114, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8641015291213989, + "num_tokens": 112354722.0, + "step": 3200 + }, + { + "epoch": 0.5944289693593314, + "grad_norm": 1.4005593061447144, + "learning_rate": 1e-06, + "loss": 0.4372, + "mean_token_accuracy": 0.8567696213722229, + "num_tokens": 112392467.0, + "step": 3201 + }, + { + "epoch": 0.594614670380687, + "grad_norm": 1.4559739828109741, + "learning_rate": 1e-06, + "loss": 0.4167, + "mean_token_accuracy": 0.8589311242103577, + "num_tokens": 112428736.0, + "step": 3202 + }, + { + "epoch": 0.5948003714020427, + "grad_norm": 1.4533278942108154, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8713050484657288, + "num_tokens": 112461351.0, + "step": 3203 + }, + { + "epoch": 0.5949860724233983, + "grad_norm": 1.3876537084579468, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8740916848182678, + "num_tokens": 112493879.0, + "step": 3204 + }, + { + "epoch": 0.5951717734447539, + "grad_norm": 1.4112181663513184, + "learning_rate": 1e-06, + "loss": 0.4031, + "mean_token_accuracy": 0.8615130186080933, + "num_tokens": 112533655.0, + "step": 3205 + }, + { + "epoch": 0.5953574744661095, + "grad_norm": 1.494407057762146, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8719019889831543, + "num_tokens": 112564423.0, + "step": 3206 + }, + { + "epoch": 0.5955431754874652, + "grad_norm": 1.5036299228668213, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8565518260002136, + "num_tokens": 112596177.0, + "step": 3207 + }, + { + "epoch": 0.5957288765088208, + "grad_norm": 1.5950931310653687, + "learning_rate": 1e-06, + "loss": 0.4723, + "mean_token_accuracy": 0.8425227403640747, + "num_tokens": 112627267.0, + "step": 3208 + }, + { + "epoch": 0.5959145775301764, + "grad_norm": 1.771060585975647, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8562183380126953, + "num_tokens": 112654052.0, + "step": 3209 + }, + { + "epoch": 0.596100278551532, + "grad_norm": 1.4705818891525269, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8739781379699707, + "num_tokens": 112687997.0, + "step": 3210 + }, + { + "epoch": 0.5962859795728876, + "grad_norm": 1.5644818544387817, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.8602743148803711, + "num_tokens": 112718707.0, + "step": 3211 + }, + { + "epoch": 0.5964716805942433, + "grad_norm": 1.498134970664978, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8717538714408875, + "num_tokens": 112750941.0, + "step": 3212 + }, + { + "epoch": 0.5966573816155989, + "grad_norm": 1.4816502332687378, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8696750998497009, + "num_tokens": 112785740.0, + "step": 3213 + }, + { + "epoch": 0.5968430826369545, + "grad_norm": 1.5476101636886597, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.863863468170166, + "num_tokens": 112816311.0, + "step": 3214 + }, + { + "epoch": 0.5970287836583101, + "grad_norm": 1.4466289281845093, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8674909472465515, + "num_tokens": 112851735.0, + "step": 3215 + }, + { + "epoch": 0.5972144846796658, + "grad_norm": 1.5193166732788086, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8710466623306274, + "num_tokens": 112884273.0, + "step": 3216 + }, + { + "epoch": 0.5974001857010214, + "grad_norm": 1.4393001794815063, + "learning_rate": 1e-06, + "loss": 0.4311, + "mean_token_accuracy": 0.8577375411987305, + "num_tokens": 112924400.0, + "step": 3217 + }, + { + "epoch": 0.597585886722377, + "grad_norm": 1.5546938180923462, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8655400276184082, + "num_tokens": 112962299.0, + "step": 3218 + }, + { + "epoch": 0.5977715877437326, + "grad_norm": 1.468772530555725, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.861327588558197, + "num_tokens": 112998892.0, + "step": 3219 + }, + { + "epoch": 0.5979572887650882, + "grad_norm": 1.6058841943740845, + "learning_rate": 1e-06, + "loss": 0.4437, + "mean_token_accuracy": 0.8557447791099548, + "num_tokens": 113033214.0, + "step": 3220 + }, + { + "epoch": 0.5981429897864439, + "grad_norm": 1.5326706171035767, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8702585101127625, + "num_tokens": 113067078.0, + "step": 3221 + }, + { + "epoch": 0.5983286908077995, + "grad_norm": 1.6064878702163696, + "learning_rate": 1e-06, + "loss": 0.4089, + "mean_token_accuracy": 0.8658522367477417, + "num_tokens": 113096904.0, + "step": 3222 + }, + { + "epoch": 0.5985143918291551, + "grad_norm": 1.5502352714538574, + "learning_rate": 1e-06, + "loss": 0.4355, + "mean_token_accuracy": 0.8526659607887268, + "num_tokens": 113134603.0, + "step": 3223 + }, + { + "epoch": 0.5987000928505106, + "grad_norm": 1.4623119831085205, + "learning_rate": 1e-06, + "loss": 0.433, + "mean_token_accuracy": 0.8545453548431396, + "num_tokens": 113175723.0, + "step": 3224 + }, + { + "epoch": 0.5988857938718662, + "grad_norm": 1.5832939147949219, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.8574327230453491, + "num_tokens": 113208167.0, + "step": 3225 + }, + { + "epoch": 0.5990714948932219, + "grad_norm": 1.6354286670684814, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.864162266254425, + "num_tokens": 113238455.0, + "step": 3226 + }, + { + "epoch": 0.5992571959145775, + "grad_norm": 1.5657652616500854, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8606683611869812, + "num_tokens": 113277233.0, + "step": 3227 + }, + { + "epoch": 0.5994428969359331, + "grad_norm": 1.4761584997177124, + "learning_rate": 1e-06, + "loss": 0.4174, + "mean_token_accuracy": 0.8647213578224182, + "num_tokens": 113316954.0, + "step": 3228 + }, + { + "epoch": 0.5996285979572887, + "grad_norm": 1.5374826192855835, + "learning_rate": 1e-06, + "loss": 0.4468, + "mean_token_accuracy": 0.8550326824188232, + "num_tokens": 113348661.0, + "step": 3229 + }, + { + "epoch": 0.5998142989786444, + "grad_norm": 1.434332251548767, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8643045425415039, + "num_tokens": 113385481.0, + "step": 3230 + }, + { + "epoch": 0.6, + "grad_norm": 1.4272863864898682, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8612259030342102, + "num_tokens": 113423807.0, + "step": 3231 + }, + { + "epoch": 0.6001857010213556, + "grad_norm": 1.4238497018814087, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8709200620651245, + "num_tokens": 113459787.0, + "step": 3232 + }, + { + "epoch": 0.6003714020427112, + "grad_norm": 1.5351027250289917, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.859269380569458, + "num_tokens": 113492582.0, + "step": 3233 + }, + { + "epoch": 0.6005571030640668, + "grad_norm": 1.5585929155349731, + "learning_rate": 1e-06, + "loss": 0.4102, + "mean_token_accuracy": 0.8654330968856812, + "num_tokens": 113526222.0, + "step": 3234 + }, + { + "epoch": 0.6007428040854225, + "grad_norm": 1.5235259532928467, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8765540719032288, + "num_tokens": 113558997.0, + "step": 3235 + }, + { + "epoch": 0.6009285051067781, + "grad_norm": 1.595720887184143, + "learning_rate": 1e-06, + "loss": 0.4154, + "mean_token_accuracy": 0.8573435544967651, + "num_tokens": 113590003.0, + "step": 3236 + }, + { + "epoch": 0.6011142061281337, + "grad_norm": 1.5411020517349243, + "learning_rate": 1e-06, + "loss": 0.4395, + "mean_token_accuracy": 0.8529524207115173, + "num_tokens": 113627008.0, + "step": 3237 + }, + { + "epoch": 0.6012999071494893, + "grad_norm": 1.4862154722213745, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.8593199253082275, + "num_tokens": 113662409.0, + "step": 3238 + }, + { + "epoch": 0.601485608170845, + "grad_norm": 1.5590896606445312, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.872184157371521, + "num_tokens": 113696336.0, + "step": 3239 + }, + { + "epoch": 0.6016713091922006, + "grad_norm": 1.5255550146102905, + "learning_rate": 1e-06, + "loss": 0.4677, + "mean_token_accuracy": 0.8443647623062134, + "num_tokens": 113735352.0, + "step": 3240 + }, + { + "epoch": 0.6018570102135562, + "grad_norm": 1.4034048318862915, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8745983839035034, + "num_tokens": 113773849.0, + "step": 3241 + }, + { + "epoch": 0.6020427112349118, + "grad_norm": 1.5326367616653442, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8710435628890991, + "num_tokens": 113808596.0, + "step": 3242 + }, + { + "epoch": 0.6022284122562674, + "grad_norm": 1.6505380868911743, + "learning_rate": 1e-06, + "loss": 0.4635, + "mean_token_accuracy": 0.8453598618507385, + "num_tokens": 113841544.0, + "step": 3243 + }, + { + "epoch": 0.6024141132776231, + "grad_norm": 1.5033844709396362, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8730766177177429, + "num_tokens": 113873601.0, + "step": 3244 + }, + { + "epoch": 0.6025998142989787, + "grad_norm": 1.5512354373931885, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8641659021377563, + "num_tokens": 113906478.0, + "step": 3245 + }, + { + "epoch": 0.6027855153203343, + "grad_norm": 1.5038070678710938, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8687379360198975, + "num_tokens": 113942320.0, + "step": 3246 + }, + { + "epoch": 0.6029712163416899, + "grad_norm": 1.4827666282653809, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8725326657295227, + "num_tokens": 113974240.0, + "step": 3247 + }, + { + "epoch": 0.6031569173630454, + "grad_norm": 1.5036993026733398, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.8685422539710999, + "num_tokens": 114007864.0, + "step": 3248 + }, + { + "epoch": 0.6033426183844011, + "grad_norm": 1.5213462114334106, + "learning_rate": 1e-06, + "loss": 0.4207, + "mean_token_accuracy": 0.8581971526145935, + "num_tokens": 114039483.0, + "step": 3249 + }, + { + "epoch": 0.6035283194057567, + "grad_norm": 1.456407904624939, + "learning_rate": 1e-06, + "loss": 0.4808, + "mean_token_accuracy": 0.8398098945617676, + "num_tokens": 114079737.0, + "step": 3250 + }, + { + "epoch": 0.6037140204271123, + "grad_norm": 1.5831539630889893, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8681021332740784, + "num_tokens": 114112474.0, + "step": 3251 + }, + { + "epoch": 0.6038997214484679, + "grad_norm": 1.3623836040496826, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8742401599884033, + "num_tokens": 114150833.0, + "step": 3252 + }, + { + "epoch": 0.6040854224698236, + "grad_norm": 1.4394572973251343, + "learning_rate": 1e-06, + "loss": 0.4598, + "mean_token_accuracy": 0.8474564552307129, + "num_tokens": 114190431.0, + "step": 3253 + }, + { + "epoch": 0.6042711234911792, + "grad_norm": 1.5271532535552979, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8687044382095337, + "num_tokens": 114224373.0, + "step": 3254 + }, + { + "epoch": 0.6044568245125348, + "grad_norm": 1.4068560600280762, + "learning_rate": 1e-06, + "loss": 0.4419, + "mean_token_accuracy": 0.8584907054901123, + "num_tokens": 114266817.0, + "step": 3255 + }, + { + "epoch": 0.6046425255338904, + "grad_norm": 1.4776434898376465, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8677098155021667, + "num_tokens": 114304520.0, + "step": 3256 + }, + { + "epoch": 0.604828226555246, + "grad_norm": 1.5505226850509644, + "learning_rate": 1e-06, + "loss": 0.4648, + "mean_token_accuracy": 0.8505285382270813, + "num_tokens": 114339658.0, + "step": 3257 + }, + { + "epoch": 0.6050139275766017, + "grad_norm": 1.511345386505127, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8709540367126465, + "num_tokens": 114372366.0, + "step": 3258 + }, + { + "epoch": 0.6051996285979573, + "grad_norm": 1.5543071031570435, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8671364784240723, + "num_tokens": 114407162.0, + "step": 3259 + }, + { + "epoch": 0.6053853296193129, + "grad_norm": 1.4924685955047607, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.8614305257797241, + "num_tokens": 114442873.0, + "step": 3260 + }, + { + "epoch": 0.6055710306406685, + "grad_norm": 1.5172810554504395, + "learning_rate": 1e-06, + "loss": 0.3833, + "mean_token_accuracy": 0.8749980926513672, + "num_tokens": 114477720.0, + "step": 3261 + }, + { + "epoch": 0.6057567316620242, + "grad_norm": 1.391034722328186, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8776705861091614, + "num_tokens": 114516055.0, + "step": 3262 + }, + { + "epoch": 0.6059424326833798, + "grad_norm": 1.6224287748336792, + "learning_rate": 1e-06, + "loss": 0.4089, + "mean_token_accuracy": 0.8628395795822144, + "num_tokens": 114543308.0, + "step": 3263 + }, + { + "epoch": 0.6061281337047354, + "grad_norm": 1.4878734350204468, + "learning_rate": 1e-06, + "loss": 0.4206, + "mean_token_accuracy": 0.8613787889480591, + "num_tokens": 114576033.0, + "step": 3264 + }, + { + "epoch": 0.606313834726091, + "grad_norm": 1.4631332159042358, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8646998405456543, + "num_tokens": 114610265.0, + "step": 3265 + }, + { + "epoch": 0.6064995357474466, + "grad_norm": 1.5230646133422852, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8634626269340515, + "num_tokens": 114643889.0, + "step": 3266 + }, + { + "epoch": 0.6066852367688023, + "grad_norm": 1.4793381690979004, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.8674182891845703, + "num_tokens": 114676312.0, + "step": 3267 + }, + { + "epoch": 0.6068709377901579, + "grad_norm": 1.4119576215744019, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8614342212677002, + "num_tokens": 114716743.0, + "step": 3268 + }, + { + "epoch": 0.6070566388115135, + "grad_norm": 1.3482266664505005, + "learning_rate": 1e-06, + "loss": 0.4038, + "mean_token_accuracy": 0.864314615726471, + "num_tokens": 114756486.0, + "step": 3269 + }, + { + "epoch": 0.6072423398328691, + "grad_norm": 1.4939855337142944, + "learning_rate": 1e-06, + "loss": 0.4486, + "mean_token_accuracy": 0.8504355549812317, + "num_tokens": 114797264.0, + "step": 3270 + }, + { + "epoch": 0.6074280408542247, + "grad_norm": 1.6266133785247803, + "learning_rate": 1e-06, + "loss": 0.4357, + "mean_token_accuracy": 0.8578251004219055, + "num_tokens": 114828693.0, + "step": 3271 + }, + { + "epoch": 0.6076137418755804, + "grad_norm": 1.4489192962646484, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8699002265930176, + "num_tokens": 114862673.0, + "step": 3272 + }, + { + "epoch": 0.6077994428969359, + "grad_norm": 1.5531238317489624, + "learning_rate": 1e-06, + "loss": 0.4496, + "mean_token_accuracy": 0.8471637964248657, + "num_tokens": 114896607.0, + "step": 3273 + }, + { + "epoch": 0.6079851439182915, + "grad_norm": 1.5122015476226807, + "learning_rate": 1e-06, + "loss": 0.4072, + "mean_token_accuracy": 0.863632321357727, + "num_tokens": 114929886.0, + "step": 3274 + }, + { + "epoch": 0.6081708449396471, + "grad_norm": 1.4839974641799927, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8631876707077026, + "num_tokens": 114965359.0, + "step": 3275 + }, + { + "epoch": 0.6083565459610027, + "grad_norm": 1.3934317827224731, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.873368501663208, + "num_tokens": 115005043.0, + "step": 3276 + }, + { + "epoch": 0.6085422469823584, + "grad_norm": 1.40770423412323, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8774762153625488, + "num_tokens": 115042374.0, + "step": 3277 + }, + { + "epoch": 0.608727948003714, + "grad_norm": 1.4788001775741577, + "learning_rate": 1e-06, + "loss": 0.4337, + "mean_token_accuracy": 0.8569501638412476, + "num_tokens": 115083281.0, + "step": 3278 + }, + { + "epoch": 0.6089136490250696, + "grad_norm": 1.363682508468628, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8659379482269287, + "num_tokens": 115123071.0, + "step": 3279 + }, + { + "epoch": 0.6090993500464252, + "grad_norm": 1.6858623027801514, + "learning_rate": 1e-06, + "loss": 0.4872, + "mean_token_accuracy": 0.8428884744644165, + "num_tokens": 115152341.0, + "step": 3280 + }, + { + "epoch": 0.6092850510677809, + "grad_norm": 1.4917705059051514, + "learning_rate": 1e-06, + "loss": 0.5087, + "mean_token_accuracy": 0.8325767517089844, + "num_tokens": 115189665.0, + "step": 3281 + }, + { + "epoch": 0.6094707520891365, + "grad_norm": 1.480592966079712, + "learning_rate": 1e-06, + "loss": 0.4165, + "mean_token_accuracy": 0.8649041652679443, + "num_tokens": 115225540.0, + "step": 3282 + }, + { + "epoch": 0.6096564531104921, + "grad_norm": 1.4007264375686646, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8686927556991577, + "num_tokens": 115261223.0, + "step": 3283 + }, + { + "epoch": 0.6098421541318477, + "grad_norm": 1.358003854751587, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8696720600128174, + "num_tokens": 115300306.0, + "step": 3284 + }, + { + "epoch": 0.6100278551532033, + "grad_norm": 1.4577248096466064, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.8677841424942017, + "num_tokens": 115332478.0, + "step": 3285 + }, + { + "epoch": 0.610213556174559, + "grad_norm": 1.6107850074768066, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.879291296005249, + "num_tokens": 115360727.0, + "step": 3286 + }, + { + "epoch": 0.6103992571959146, + "grad_norm": 1.4076032638549805, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8629510402679443, + "num_tokens": 115403110.0, + "step": 3287 + }, + { + "epoch": 0.6105849582172702, + "grad_norm": 1.5119202136993408, + "learning_rate": 1e-06, + "loss": 0.4202, + "mean_token_accuracy": 0.8583099246025085, + "num_tokens": 115436552.0, + "step": 3288 + }, + { + "epoch": 0.6107706592386258, + "grad_norm": 1.5865613222122192, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8569850921630859, + "num_tokens": 115469576.0, + "step": 3289 + }, + { + "epoch": 0.6109563602599815, + "grad_norm": 1.48448646068573, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8677518367767334, + "num_tokens": 115503463.0, + "step": 3290 + }, + { + "epoch": 0.6111420612813371, + "grad_norm": 1.5727214813232422, + "learning_rate": 1e-06, + "loss": 0.4459, + "mean_token_accuracy": 0.8487198352813721, + "num_tokens": 115533843.0, + "step": 3291 + }, + { + "epoch": 0.6113277623026927, + "grad_norm": 1.447072148323059, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8672451376914978, + "num_tokens": 115571592.0, + "step": 3292 + }, + { + "epoch": 0.6115134633240483, + "grad_norm": 1.456089735031128, + "learning_rate": 1e-06, + "loss": 0.4412, + "mean_token_accuracy": 0.8537816405296326, + "num_tokens": 115606922.0, + "step": 3293 + }, + { + "epoch": 0.611699164345404, + "grad_norm": 1.55936861038208, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8663432002067566, + "num_tokens": 115636300.0, + "step": 3294 + }, + { + "epoch": 0.6118848653667596, + "grad_norm": 1.5448771715164185, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8658432960510254, + "num_tokens": 115666577.0, + "step": 3295 + }, + { + "epoch": 0.6120705663881152, + "grad_norm": 1.6056082248687744, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8683902025222778, + "num_tokens": 115696162.0, + "step": 3296 + }, + { + "epoch": 0.6122562674094707, + "grad_norm": 1.5145713090896606, + "learning_rate": 1e-06, + "loss": 0.4334, + "mean_token_accuracy": 0.8556846380233765, + "num_tokens": 115731664.0, + "step": 3297 + }, + { + "epoch": 0.6124419684308263, + "grad_norm": 1.4830145835876465, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8684971928596497, + "num_tokens": 115769236.0, + "step": 3298 + }, + { + "epoch": 0.6126276694521819, + "grad_norm": 1.56150484085083, + "learning_rate": 1e-06, + "loss": 0.4247, + "mean_token_accuracy": 0.8584356904029846, + "num_tokens": 115806675.0, + "step": 3299 + }, + { + "epoch": 0.6128133704735376, + "grad_norm": 1.4118688106536865, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8665117621421814, + "num_tokens": 115844961.0, + "step": 3300 + }, + { + "epoch": 0.6129990714948932, + "grad_norm": 1.5259582996368408, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8601857423782349, + "num_tokens": 115882986.0, + "step": 3301 + }, + { + "epoch": 0.6131847725162488, + "grad_norm": 1.6200780868530273, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8676793575286865, + "num_tokens": 115910420.0, + "step": 3302 + }, + { + "epoch": 0.6133704735376044, + "grad_norm": 1.39448881149292, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8696333765983582, + "num_tokens": 115949679.0, + "step": 3303 + }, + { + "epoch": 0.61355617455896, + "grad_norm": 1.4660594463348389, + "learning_rate": 1e-06, + "loss": 0.4595, + "mean_token_accuracy": 0.8466606736183167, + "num_tokens": 115987040.0, + "step": 3304 + }, + { + "epoch": 0.6137418755803157, + "grad_norm": 1.4613577127456665, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.8586177229881287, + "num_tokens": 116025681.0, + "step": 3305 + }, + { + "epoch": 0.6139275766016713, + "grad_norm": 1.370072841644287, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8722557425498962, + "num_tokens": 116064439.0, + "step": 3306 + }, + { + "epoch": 0.6141132776230269, + "grad_norm": 1.4953114986419678, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8598681092262268, + "num_tokens": 116102998.0, + "step": 3307 + }, + { + "epoch": 0.6142989786443825, + "grad_norm": 1.5240952968597412, + "learning_rate": 1e-06, + "loss": 0.4248, + "mean_token_accuracy": 0.860980212688446, + "num_tokens": 116136046.0, + "step": 3308 + }, + { + "epoch": 0.6144846796657382, + "grad_norm": 1.5557206869125366, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8580477237701416, + "num_tokens": 116172128.0, + "step": 3309 + }, + { + "epoch": 0.6146703806870938, + "grad_norm": 1.500927448272705, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.874000608921051, + "num_tokens": 116206195.0, + "step": 3310 + }, + { + "epoch": 0.6148560817084494, + "grad_norm": 1.553775668144226, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8712667226791382, + "num_tokens": 116237503.0, + "step": 3311 + }, + { + "epoch": 0.615041782729805, + "grad_norm": 1.655001163482666, + "learning_rate": 1e-06, + "loss": 0.4476, + "mean_token_accuracy": 0.8514078855514526, + "num_tokens": 116268175.0, + "step": 3312 + }, + { + "epoch": 0.6152274837511607, + "grad_norm": 1.4763630628585815, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8659589290618896, + "num_tokens": 116301762.0, + "step": 3313 + }, + { + "epoch": 0.6154131847725163, + "grad_norm": 1.4937193393707275, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8619322180747986, + "num_tokens": 116337231.0, + "step": 3314 + }, + { + "epoch": 0.6155988857938719, + "grad_norm": 1.5482982397079468, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8731445670127869, + "num_tokens": 116367914.0, + "step": 3315 + }, + { + "epoch": 0.6157845868152275, + "grad_norm": 1.5653066635131836, + "learning_rate": 1e-06, + "loss": 0.4315, + "mean_token_accuracy": 0.8606778383255005, + "num_tokens": 116400398.0, + "step": 3316 + }, + { + "epoch": 0.6159702878365831, + "grad_norm": 1.4699647426605225, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8680520057678223, + "num_tokens": 116433034.0, + "step": 3317 + }, + { + "epoch": 0.6161559888579388, + "grad_norm": 1.374679446220398, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8742736577987671, + "num_tokens": 116470731.0, + "step": 3318 + }, + { + "epoch": 0.6163416898792944, + "grad_norm": 1.476996898651123, + "learning_rate": 1e-06, + "loss": 0.4292, + "mean_token_accuracy": 0.8553224205970764, + "num_tokens": 116503288.0, + "step": 3319 + }, + { + "epoch": 0.61652739090065, + "grad_norm": 1.5203396081924438, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8599120378494263, + "num_tokens": 116540735.0, + "step": 3320 + }, + { + "epoch": 0.6167130919220055, + "grad_norm": 1.3452203273773193, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8733627200126648, + "num_tokens": 116578734.0, + "step": 3321 + }, + { + "epoch": 0.6168987929433611, + "grad_norm": 1.5373049974441528, + "learning_rate": 1e-06, + "loss": 0.4054, + "mean_token_accuracy": 0.8625974059104919, + "num_tokens": 116615065.0, + "step": 3322 + }, + { + "epoch": 0.6170844939647168, + "grad_norm": 1.5239429473876953, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8615304827690125, + "num_tokens": 116650163.0, + "step": 3323 + }, + { + "epoch": 0.6172701949860724, + "grad_norm": 1.5423027276992798, + "learning_rate": 1e-06, + "loss": 0.4608, + "mean_token_accuracy": 0.8486703634262085, + "num_tokens": 116686904.0, + "step": 3324 + }, + { + "epoch": 0.617455896007428, + "grad_norm": 1.4713525772094727, + "learning_rate": 1e-06, + "loss": 0.4497, + "mean_token_accuracy": 0.8510833978652954, + "num_tokens": 116721844.0, + "step": 3325 + }, + { + "epoch": 0.6176415970287836, + "grad_norm": 1.6496081352233887, + "learning_rate": 1e-06, + "loss": 0.4153, + "mean_token_accuracy": 0.8624802827835083, + "num_tokens": 116750577.0, + "step": 3326 + }, + { + "epoch": 0.6178272980501393, + "grad_norm": 1.532319188117981, + "learning_rate": 1e-06, + "loss": 0.4211, + "mean_token_accuracy": 0.860471785068512, + "num_tokens": 116785310.0, + "step": 3327 + }, + { + "epoch": 0.6180129990714949, + "grad_norm": 1.4850642681121826, + "learning_rate": 1e-06, + "loss": 0.4269, + "mean_token_accuracy": 0.8577606678009033, + "num_tokens": 116823170.0, + "step": 3328 + }, + { + "epoch": 0.6181987000928505, + "grad_norm": 1.4455382823944092, + "learning_rate": 1e-06, + "loss": 0.3623, + "mean_token_accuracy": 0.8760952949523926, + "num_tokens": 116855843.0, + "step": 3329 + }, + { + "epoch": 0.6183844011142061, + "grad_norm": 1.5370330810546875, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8612949848175049, + "num_tokens": 116890014.0, + "step": 3330 + }, + { + "epoch": 0.6185701021355617, + "grad_norm": 1.585254192352295, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.8623337745666504, + "num_tokens": 116921675.0, + "step": 3331 + }, + { + "epoch": 0.6187558031569174, + "grad_norm": 1.4289822578430176, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8550444841384888, + "num_tokens": 116960353.0, + "step": 3332 + }, + { + "epoch": 0.618941504178273, + "grad_norm": 1.4092066287994385, + "learning_rate": 1e-06, + "loss": 0.424, + "mean_token_accuracy": 0.8569517135620117, + "num_tokens": 116998435.0, + "step": 3333 + }, + { + "epoch": 0.6191272051996286, + "grad_norm": 1.6271904706954956, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8687554597854614, + "num_tokens": 117028914.0, + "step": 3334 + }, + { + "epoch": 0.6193129062209842, + "grad_norm": 1.6270374059677124, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8812012076377869, + "num_tokens": 117061697.0, + "step": 3335 + }, + { + "epoch": 0.6194986072423398, + "grad_norm": 1.539696455001831, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.879875898361206, + "num_tokens": 117092001.0, + "step": 3336 + }, + { + "epoch": 0.6196843082636955, + "grad_norm": 1.5636332035064697, + "learning_rate": 1e-06, + "loss": 0.4652, + "mean_token_accuracy": 0.8446124792098999, + "num_tokens": 117124957.0, + "step": 3337 + }, + { + "epoch": 0.6198700092850511, + "grad_norm": 1.5202937126159668, + "learning_rate": 1e-06, + "loss": 0.4378, + "mean_token_accuracy": 0.8545224666595459, + "num_tokens": 117162365.0, + "step": 3338 + }, + { + "epoch": 0.6200557103064067, + "grad_norm": 1.5020166635513306, + "learning_rate": 1e-06, + "loss": 0.4293, + "mean_token_accuracy": 0.8567825555801392, + "num_tokens": 117196828.0, + "step": 3339 + }, + { + "epoch": 0.6202414113277623, + "grad_norm": 1.6709312200546265, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8640246391296387, + "num_tokens": 117224290.0, + "step": 3340 + }, + { + "epoch": 0.620427112349118, + "grad_norm": 1.8083819150924683, + "learning_rate": 1e-06, + "loss": 0.4727, + "mean_token_accuracy": 0.8411673307418823, + "num_tokens": 117260189.0, + "step": 3341 + }, + { + "epoch": 0.6206128133704736, + "grad_norm": 1.4980524778366089, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.867672860622406, + "num_tokens": 117295215.0, + "step": 3342 + }, + { + "epoch": 0.6207985143918292, + "grad_norm": 1.66419517993927, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8658710718154907, + "num_tokens": 117322262.0, + "step": 3343 + }, + { + "epoch": 0.6209842154131848, + "grad_norm": 1.4913663864135742, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.862145185470581, + "num_tokens": 117359252.0, + "step": 3344 + }, + { + "epoch": 0.6211699164345403, + "grad_norm": 1.35677969455719, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8679033517837524, + "num_tokens": 117399865.0, + "step": 3345 + }, + { + "epoch": 0.621355617455896, + "grad_norm": 1.5077166557312012, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8655356764793396, + "num_tokens": 117434249.0, + "step": 3346 + }, + { + "epoch": 0.6215413184772516, + "grad_norm": 1.3199446201324463, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8661308288574219, + "num_tokens": 117474924.0, + "step": 3347 + }, + { + "epoch": 0.6217270194986072, + "grad_norm": 1.5213607549667358, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8815633058547974, + "num_tokens": 117506345.0, + "step": 3348 + }, + { + "epoch": 0.6219127205199628, + "grad_norm": 1.3967753648757935, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8838473558425903, + "num_tokens": 117544239.0, + "step": 3349 + }, + { + "epoch": 0.6220984215413184, + "grad_norm": 1.3143006563186646, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8755311965942383, + "num_tokens": 117584156.0, + "step": 3350 + }, + { + "epoch": 0.6222841225626741, + "grad_norm": 1.495336651802063, + "learning_rate": 1e-06, + "loss": 0.4462, + "mean_token_accuracy": 0.8545538187026978, + "num_tokens": 117620668.0, + "step": 3351 + }, + { + "epoch": 0.6224698235840297, + "grad_norm": 1.5510144233703613, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8722633123397827, + "num_tokens": 117651685.0, + "step": 3352 + }, + { + "epoch": 0.6226555246053853, + "grad_norm": 1.42420494556427, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8664368391036987, + "num_tokens": 117688072.0, + "step": 3353 + }, + { + "epoch": 0.6228412256267409, + "grad_norm": 1.540402889251709, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.86394202709198, + "num_tokens": 117719899.0, + "step": 3354 + }, + { + "epoch": 0.6230269266480966, + "grad_norm": 1.4896262884140015, + "learning_rate": 1e-06, + "loss": 0.4515, + "mean_token_accuracy": 0.8527985215187073, + "num_tokens": 117760178.0, + "step": 3355 + }, + { + "epoch": 0.6232126276694522, + "grad_norm": 1.4189308881759644, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.8759593963623047, + "num_tokens": 117798544.0, + "step": 3356 + }, + { + "epoch": 0.6233983286908078, + "grad_norm": 1.4440065622329712, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8608179092407227, + "num_tokens": 117838305.0, + "step": 3357 + }, + { + "epoch": 0.6235840297121634, + "grad_norm": 1.5469417572021484, + "learning_rate": 1e-06, + "loss": 0.4649, + "mean_token_accuracy": 0.8418357968330383, + "num_tokens": 117876538.0, + "step": 3358 + }, + { + "epoch": 0.623769730733519, + "grad_norm": 1.4385660886764526, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8555572628974915, + "num_tokens": 117915003.0, + "step": 3359 + }, + { + "epoch": 0.6239554317548747, + "grad_norm": 1.3799670934677124, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8697055578231812, + "num_tokens": 117948924.0, + "step": 3360 + }, + { + "epoch": 0.6241411327762303, + "grad_norm": 1.435299038887024, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8668463230133057, + "num_tokens": 117985252.0, + "step": 3361 + }, + { + "epoch": 0.6243268337975859, + "grad_norm": 1.7058682441711426, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8630540370941162, + "num_tokens": 118014133.0, + "step": 3362 + }, + { + "epoch": 0.6245125348189415, + "grad_norm": 1.5682494640350342, + "learning_rate": 1e-06, + "loss": 0.4372, + "mean_token_accuracy": 0.8537818193435669, + "num_tokens": 118052334.0, + "step": 3363 + }, + { + "epoch": 0.6246982358402972, + "grad_norm": 1.5207488536834717, + "learning_rate": 1e-06, + "loss": 0.396, + "mean_token_accuracy": 0.8675475120544434, + "num_tokens": 118087500.0, + "step": 3364 + }, + { + "epoch": 0.6248839368616528, + "grad_norm": 1.3609169721603394, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8644399642944336, + "num_tokens": 118129781.0, + "step": 3365 + }, + { + "epoch": 0.6250696378830084, + "grad_norm": 1.383690357208252, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8564732670783997, + "num_tokens": 118174769.0, + "step": 3366 + }, + { + "epoch": 0.625255338904364, + "grad_norm": 1.5742474794387817, + "learning_rate": 1e-06, + "loss": 0.4331, + "mean_token_accuracy": 0.8528032302856445, + "num_tokens": 118209130.0, + "step": 3367 + }, + { + "epoch": 0.6254410399257196, + "grad_norm": 1.5273804664611816, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8667843341827393, + "num_tokens": 118241744.0, + "step": 3368 + }, + { + "epoch": 0.6256267409470752, + "grad_norm": 1.4591330289840698, + "learning_rate": 1e-06, + "loss": 0.4421, + "mean_token_accuracy": 0.8565310835838318, + "num_tokens": 118281277.0, + "step": 3369 + }, + { + "epoch": 0.6258124419684308, + "grad_norm": 1.4585680961608887, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8700059652328491, + "num_tokens": 118319876.0, + "step": 3370 + }, + { + "epoch": 0.6259981429897864, + "grad_norm": 1.6036531925201416, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.8659709692001343, + "num_tokens": 118352008.0, + "step": 3371 + }, + { + "epoch": 0.626183844011142, + "grad_norm": 1.3914625644683838, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8726078271865845, + "num_tokens": 118390662.0, + "step": 3372 + }, + { + "epoch": 0.6263695450324976, + "grad_norm": 1.5672894716262817, + "learning_rate": 1e-06, + "loss": 0.4211, + "mean_token_accuracy": 0.8603646755218506, + "num_tokens": 118425711.0, + "step": 3373 + }, + { + "epoch": 0.6265552460538533, + "grad_norm": 1.3803887367248535, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8749619126319885, + "num_tokens": 118464226.0, + "step": 3374 + }, + { + "epoch": 0.6267409470752089, + "grad_norm": 1.4689395427703857, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8600353598594666, + "num_tokens": 118500488.0, + "step": 3375 + }, + { + "epoch": 0.6269266480965645, + "grad_norm": 1.3757803440093994, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8640417456626892, + "num_tokens": 118540436.0, + "step": 3376 + }, + { + "epoch": 0.6271123491179201, + "grad_norm": 1.447970986366272, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8775396347045898, + "num_tokens": 118576554.0, + "step": 3377 + }, + { + "epoch": 0.6272980501392758, + "grad_norm": 1.4870994091033936, + "learning_rate": 1e-06, + "loss": 0.4501, + "mean_token_accuracy": 0.8490476608276367, + "num_tokens": 118611888.0, + "step": 3378 + }, + { + "epoch": 0.6274837511606314, + "grad_norm": 1.644615888595581, + "learning_rate": 1e-06, + "loss": 0.4242, + "mean_token_accuracy": 0.8568984270095825, + "num_tokens": 118645092.0, + "step": 3379 + }, + { + "epoch": 0.627669452181987, + "grad_norm": 1.4273521900177002, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.879813551902771, + "num_tokens": 118683036.0, + "step": 3380 + }, + { + "epoch": 0.6278551532033426, + "grad_norm": 1.6648643016815186, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8746498227119446, + "num_tokens": 118714084.0, + "step": 3381 + }, + { + "epoch": 0.6280408542246982, + "grad_norm": 1.4799480438232422, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8750911355018616, + "num_tokens": 118747435.0, + "step": 3382 + }, + { + "epoch": 0.6282265552460539, + "grad_norm": 1.4707250595092773, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8706915378570557, + "num_tokens": 118781975.0, + "step": 3383 + }, + { + "epoch": 0.6284122562674095, + "grad_norm": 1.605220913887024, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8675321340560913, + "num_tokens": 118813999.0, + "step": 3384 + }, + { + "epoch": 0.6285979572887651, + "grad_norm": 1.394412636756897, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8675355911254883, + "num_tokens": 118850782.0, + "step": 3385 + }, + { + "epoch": 0.6287836583101207, + "grad_norm": 1.7578989267349243, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8580573201179504, + "num_tokens": 118881646.0, + "step": 3386 + }, + { + "epoch": 0.6289693593314764, + "grad_norm": 1.5843058824539185, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8601069450378418, + "num_tokens": 118915908.0, + "step": 3387 + }, + { + "epoch": 0.629155060352832, + "grad_norm": 1.4264061450958252, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8759791254997253, + "num_tokens": 118953181.0, + "step": 3388 + }, + { + "epoch": 0.6293407613741876, + "grad_norm": 1.5405988693237305, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8725144863128662, + "num_tokens": 118983894.0, + "step": 3389 + }, + { + "epoch": 0.6295264623955432, + "grad_norm": 1.4522064924240112, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8686338663101196, + "num_tokens": 119019651.0, + "step": 3390 + }, + { + "epoch": 0.6297121634168988, + "grad_norm": 1.4941909313201904, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8652963638305664, + "num_tokens": 119057787.0, + "step": 3391 + }, + { + "epoch": 0.6298978644382545, + "grad_norm": 1.5120387077331543, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.876710057258606, + "num_tokens": 119090861.0, + "step": 3392 + }, + { + "epoch": 0.63008356545961, + "grad_norm": 1.6512126922607422, + "learning_rate": 1e-06, + "loss": 0.4231, + "mean_token_accuracy": 0.8576511740684509, + "num_tokens": 119120407.0, + "step": 3393 + }, + { + "epoch": 0.6302692664809656, + "grad_norm": 1.3429571390151978, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8682154417037964, + "num_tokens": 119164861.0, + "step": 3394 + }, + { + "epoch": 0.6304549675023212, + "grad_norm": 1.5162378549575806, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8698589205741882, + "num_tokens": 119198936.0, + "step": 3395 + }, + { + "epoch": 0.6306406685236768, + "grad_norm": 1.5155389308929443, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.8638168573379517, + "num_tokens": 119234014.0, + "step": 3396 + }, + { + "epoch": 0.6308263695450325, + "grad_norm": 1.513472318649292, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.8596947193145752, + "num_tokens": 119266563.0, + "step": 3397 + }, + { + "epoch": 0.6310120705663881, + "grad_norm": 1.3937904834747314, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8625417947769165, + "num_tokens": 119303340.0, + "step": 3398 + }, + { + "epoch": 0.6311977715877437, + "grad_norm": 1.365425705909729, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8686511516571045, + "num_tokens": 119343261.0, + "step": 3399 + }, + { + "epoch": 0.6313834726090993, + "grad_norm": 1.4413948059082031, + "learning_rate": 1e-06, + "loss": 0.4225, + "mean_token_accuracy": 0.8604139685630798, + "num_tokens": 119382527.0, + "step": 3400 + }, + { + "epoch": 0.631569173630455, + "grad_norm": 1.4612287282943726, + "learning_rate": 1e-06, + "loss": 0.4355, + "mean_token_accuracy": 0.8508779406547546, + "num_tokens": 119421239.0, + "step": 3401 + }, + { + "epoch": 0.6317548746518106, + "grad_norm": 1.7685139179229736, + "learning_rate": 1e-06, + "loss": 0.434, + "mean_token_accuracy": 0.8539875745773315, + "num_tokens": 119450800.0, + "step": 3402 + }, + { + "epoch": 0.6319405756731662, + "grad_norm": 1.4601202011108398, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8587148785591125, + "num_tokens": 119487354.0, + "step": 3403 + }, + { + "epoch": 0.6321262766945218, + "grad_norm": 1.433027982711792, + "learning_rate": 1e-06, + "loss": 0.4219, + "mean_token_accuracy": 0.8592584133148193, + "num_tokens": 119523772.0, + "step": 3404 + }, + { + "epoch": 0.6323119777158774, + "grad_norm": 1.4343726634979248, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8565617799758911, + "num_tokens": 119564296.0, + "step": 3405 + }, + { + "epoch": 0.6324976787372331, + "grad_norm": 1.4877374172210693, + "learning_rate": 1e-06, + "loss": 0.4458, + "mean_token_accuracy": 0.8510506749153137, + "num_tokens": 119602591.0, + "step": 3406 + }, + { + "epoch": 0.6326833797585887, + "grad_norm": 1.4301185607910156, + "learning_rate": 1e-06, + "loss": 0.4163, + "mean_token_accuracy": 0.8604094982147217, + "num_tokens": 119638893.0, + "step": 3407 + }, + { + "epoch": 0.6328690807799443, + "grad_norm": 1.5792220830917358, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.877119779586792, + "num_tokens": 119671354.0, + "step": 3408 + }, + { + "epoch": 0.6330547818012999, + "grad_norm": 1.3031784296035767, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8734943866729736, + "num_tokens": 119713707.0, + "step": 3409 + }, + { + "epoch": 0.6332404828226555, + "grad_norm": 1.5933159589767456, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8671590089797974, + "num_tokens": 119745209.0, + "step": 3410 + }, + { + "epoch": 0.6334261838440112, + "grad_norm": 1.4033955335617065, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8544172048568726, + "num_tokens": 119784245.0, + "step": 3411 + }, + { + "epoch": 0.6336118848653668, + "grad_norm": 1.409753680229187, + "learning_rate": 1e-06, + "loss": 0.4315, + "mean_token_accuracy": 0.8577234745025635, + "num_tokens": 119828716.0, + "step": 3412 + }, + { + "epoch": 0.6337975858867224, + "grad_norm": 1.4407705068588257, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.868416428565979, + "num_tokens": 119865517.0, + "step": 3413 + }, + { + "epoch": 0.633983286908078, + "grad_norm": 1.392762541770935, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8619296550750732, + "num_tokens": 119909585.0, + "step": 3414 + }, + { + "epoch": 0.6341689879294337, + "grad_norm": 1.2918592691421509, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.880720317363739, + "num_tokens": 119949951.0, + "step": 3415 + }, + { + "epoch": 0.6343546889507893, + "grad_norm": 1.6214748620986938, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8680566549301147, + "num_tokens": 119977706.0, + "step": 3416 + }, + { + "epoch": 0.6345403899721448, + "grad_norm": 1.5096808671951294, + "learning_rate": 1e-06, + "loss": 0.4236, + "mean_token_accuracy": 0.8596384525299072, + "num_tokens": 120013182.0, + "step": 3417 + }, + { + "epoch": 0.6347260909935004, + "grad_norm": 1.53545343875885, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8777205944061279, + "num_tokens": 120042867.0, + "step": 3418 + }, + { + "epoch": 0.634911792014856, + "grad_norm": 1.5882351398468018, + "learning_rate": 1e-06, + "loss": 0.4416, + "mean_token_accuracy": 0.8566650748252869, + "num_tokens": 120073197.0, + "step": 3419 + }, + { + "epoch": 0.6350974930362117, + "grad_norm": 1.4230378866195679, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.854845404624939, + "num_tokens": 120116902.0, + "step": 3420 + }, + { + "epoch": 0.6352831940575673, + "grad_norm": 1.506118655204773, + "learning_rate": 1e-06, + "loss": 0.4271, + "mean_token_accuracy": 0.8560730218887329, + "num_tokens": 120151657.0, + "step": 3421 + }, + { + "epoch": 0.6354688950789229, + "grad_norm": 1.4588813781738281, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8600140810012817, + "num_tokens": 120187919.0, + "step": 3422 + }, + { + "epoch": 0.6356545961002785, + "grad_norm": 1.3456590175628662, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8716384172439575, + "num_tokens": 120226336.0, + "step": 3423 + }, + { + "epoch": 0.6358402971216341, + "grad_norm": 1.3531720638275146, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.874639630317688, + "num_tokens": 120263954.0, + "step": 3424 + }, + { + "epoch": 0.6360259981429898, + "grad_norm": 1.3323341608047485, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8797644376754761, + "num_tokens": 120303688.0, + "step": 3425 + }, + { + "epoch": 0.6362116991643454, + "grad_norm": 1.5671271085739136, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.868769645690918, + "num_tokens": 120332436.0, + "step": 3426 + }, + { + "epoch": 0.636397400185701, + "grad_norm": 1.458939552307129, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8638266324996948, + "num_tokens": 120368327.0, + "step": 3427 + }, + { + "epoch": 0.6365831012070566, + "grad_norm": 1.6833401918411255, + "learning_rate": 1e-06, + "loss": 0.4175, + "mean_token_accuracy": 0.8636212348937988, + "num_tokens": 120395412.0, + "step": 3428 + }, + { + "epoch": 0.6367688022284123, + "grad_norm": 1.4609071016311646, + "learning_rate": 1e-06, + "loss": 0.4099, + "mean_token_accuracy": 0.864005982875824, + "num_tokens": 120431644.0, + "step": 3429 + }, + { + "epoch": 0.6369545032497679, + "grad_norm": 1.4426153898239136, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8662862777709961, + "num_tokens": 120468675.0, + "step": 3430 + }, + { + "epoch": 0.6371402042711235, + "grad_norm": 1.5450663566589355, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8652429580688477, + "num_tokens": 120501512.0, + "step": 3431 + }, + { + "epoch": 0.6373259052924791, + "grad_norm": 1.4961857795715332, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8721012473106384, + "num_tokens": 120535040.0, + "step": 3432 + }, + { + "epoch": 0.6375116063138347, + "grad_norm": 1.6501500606536865, + "learning_rate": 1e-06, + "loss": 0.4984, + "mean_token_accuracy": 0.8428474068641663, + "num_tokens": 120570198.0, + "step": 3433 + }, + { + "epoch": 0.6376973073351904, + "grad_norm": 1.676233172416687, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8759784698486328, + "num_tokens": 120602144.0, + "step": 3434 + }, + { + "epoch": 0.637883008356546, + "grad_norm": 1.4556134939193726, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8735976219177246, + "num_tokens": 120640091.0, + "step": 3435 + }, + { + "epoch": 0.6380687093779016, + "grad_norm": 1.3888096809387207, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8729268908500671, + "num_tokens": 120677388.0, + "step": 3436 + }, + { + "epoch": 0.6382544103992572, + "grad_norm": 1.438068151473999, + "learning_rate": 1e-06, + "loss": 0.4474, + "mean_token_accuracy": 0.8559677004814148, + "num_tokens": 120718832.0, + "step": 3437 + }, + { + "epoch": 0.6384401114206129, + "grad_norm": 1.526841402053833, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8742155432701111, + "num_tokens": 120752955.0, + "step": 3438 + }, + { + "epoch": 0.6386258124419685, + "grad_norm": 1.4227341413497925, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8696640133857727, + "num_tokens": 120792448.0, + "step": 3439 + }, + { + "epoch": 0.6388115134633241, + "grad_norm": 1.5491849184036255, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8721966743469238, + "num_tokens": 120825389.0, + "step": 3440 + }, + { + "epoch": 0.6389972144846797, + "grad_norm": 1.4103256464004517, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8674156069755554, + "num_tokens": 120865114.0, + "step": 3441 + }, + { + "epoch": 0.6391829155060352, + "grad_norm": 1.4011282920837402, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8774269223213196, + "num_tokens": 120904130.0, + "step": 3442 + }, + { + "epoch": 0.6393686165273909, + "grad_norm": 1.6065499782562256, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8605390787124634, + "num_tokens": 120937480.0, + "step": 3443 + }, + { + "epoch": 0.6395543175487465, + "grad_norm": 1.4110827445983887, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8726624250411987, + "num_tokens": 120976242.0, + "step": 3444 + }, + { + "epoch": 0.6397400185701021, + "grad_norm": 1.5951935052871704, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8673468828201294, + "num_tokens": 121006908.0, + "step": 3445 + }, + { + "epoch": 0.6399257195914577, + "grad_norm": 1.5731474161148071, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.857722818851471, + "num_tokens": 121038091.0, + "step": 3446 + }, + { + "epoch": 0.6401114206128133, + "grad_norm": 1.4864119291305542, + "learning_rate": 1e-06, + "loss": 0.441, + "mean_token_accuracy": 0.8544328212738037, + "num_tokens": 121072372.0, + "step": 3447 + }, + { + "epoch": 0.640297121634169, + "grad_norm": 1.4365499019622803, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.878575325012207, + "num_tokens": 121104556.0, + "step": 3448 + }, + { + "epoch": 0.6404828226555246, + "grad_norm": 1.6029469966888428, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.8620169162750244, + "num_tokens": 121140061.0, + "step": 3449 + }, + { + "epoch": 0.6406685236768802, + "grad_norm": 1.4644429683685303, + "learning_rate": 1e-06, + "loss": 0.4638, + "mean_token_accuracy": 0.8470888733863831, + "num_tokens": 121179834.0, + "step": 3450 + }, + { + "epoch": 0.6408542246982358, + "grad_norm": 1.5483790636062622, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8616095781326294, + "num_tokens": 121211743.0, + "step": 3451 + }, + { + "epoch": 0.6410399257195915, + "grad_norm": 1.4643994569778442, + "learning_rate": 1e-06, + "loss": 0.417, + "mean_token_accuracy": 0.8626405000686646, + "num_tokens": 121247945.0, + "step": 3452 + }, + { + "epoch": 0.6412256267409471, + "grad_norm": 1.6870503425598145, + "learning_rate": 1e-06, + "loss": 0.4632, + "mean_token_accuracy": 0.8508996963500977, + "num_tokens": 121277753.0, + "step": 3453 + }, + { + "epoch": 0.6414113277623027, + "grad_norm": 1.4872204065322876, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8531184792518616, + "num_tokens": 121312389.0, + "step": 3454 + }, + { + "epoch": 0.6415970287836583, + "grad_norm": 1.563072681427002, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8556421399116516, + "num_tokens": 121345465.0, + "step": 3455 + }, + { + "epoch": 0.6417827298050139, + "grad_norm": 1.5360276699066162, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8710156679153442, + "num_tokens": 121378633.0, + "step": 3456 + }, + { + "epoch": 0.6419684308263696, + "grad_norm": 1.5889387130737305, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8768621683120728, + "num_tokens": 121410059.0, + "step": 3457 + }, + { + "epoch": 0.6421541318477252, + "grad_norm": 1.419764518737793, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.8613795042037964, + "num_tokens": 121447209.0, + "step": 3458 + }, + { + "epoch": 0.6423398328690808, + "grad_norm": 1.4395445585250854, + "learning_rate": 1e-06, + "loss": 0.4459, + "mean_token_accuracy": 0.8514890670776367, + "num_tokens": 121485520.0, + "step": 3459 + }, + { + "epoch": 0.6425255338904364, + "grad_norm": 1.4830501079559326, + "learning_rate": 1e-06, + "loss": 0.4249, + "mean_token_accuracy": 0.853171706199646, + "num_tokens": 121521087.0, + "step": 3460 + }, + { + "epoch": 0.642711234911792, + "grad_norm": 1.389037847518921, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8680200576782227, + "num_tokens": 121560370.0, + "step": 3461 + }, + { + "epoch": 0.6428969359331477, + "grad_norm": 1.5215318202972412, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.868914783000946, + "num_tokens": 121592417.0, + "step": 3462 + }, + { + "epoch": 0.6430826369545033, + "grad_norm": 1.4973244667053223, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8727169036865234, + "num_tokens": 121626840.0, + "step": 3463 + }, + { + "epoch": 0.6432683379758589, + "grad_norm": 1.3385707139968872, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8667813539505005, + "num_tokens": 121666099.0, + "step": 3464 + }, + { + "epoch": 0.6434540389972145, + "grad_norm": 1.4193511009216309, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8616415858268738, + "num_tokens": 121703516.0, + "step": 3465 + }, + { + "epoch": 0.64363974001857, + "grad_norm": 1.4350754022598267, + "learning_rate": 1e-06, + "loss": 0.4097, + "mean_token_accuracy": 0.8615332841873169, + "num_tokens": 121739599.0, + "step": 3466 + }, + { + "epoch": 0.6438254410399257, + "grad_norm": 1.4803812503814697, + "learning_rate": 1e-06, + "loss": 0.4481, + "mean_token_accuracy": 0.8533076643943787, + "num_tokens": 121778454.0, + "step": 3467 + }, + { + "epoch": 0.6440111420612813, + "grad_norm": 1.5934306383132935, + "learning_rate": 1e-06, + "loss": 0.4854, + "mean_token_accuracy": 0.8470322489738464, + "num_tokens": 121812026.0, + "step": 3468 + }, + { + "epoch": 0.6441968430826369, + "grad_norm": 1.3730274438858032, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8720499277114868, + "num_tokens": 121850174.0, + "step": 3469 + }, + { + "epoch": 0.6443825441039925, + "grad_norm": 1.5364654064178467, + "learning_rate": 1e-06, + "loss": 0.4616, + "mean_token_accuracy": 0.8483405709266663, + "num_tokens": 121886455.0, + "step": 3470 + }, + { + "epoch": 0.6445682451253482, + "grad_norm": 1.4636077880859375, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8724023699760437, + "num_tokens": 121918534.0, + "step": 3471 + }, + { + "epoch": 0.6447539461467038, + "grad_norm": 1.3860347270965576, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.871728777885437, + "num_tokens": 121956244.0, + "step": 3472 + }, + { + "epoch": 0.6449396471680594, + "grad_norm": 1.432969570159912, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8716330528259277, + "num_tokens": 121994814.0, + "step": 3473 + }, + { + "epoch": 0.645125348189415, + "grad_norm": 1.5099472999572754, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8732986450195312, + "num_tokens": 122026461.0, + "step": 3474 + }, + { + "epoch": 0.6453110492107706, + "grad_norm": 1.311277151107788, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8798463344573975, + "num_tokens": 122064584.0, + "step": 3475 + }, + { + "epoch": 0.6454967502321263, + "grad_norm": 1.3568270206451416, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8601056337356567, + "num_tokens": 122106212.0, + "step": 3476 + }, + { + "epoch": 0.6456824512534819, + "grad_norm": 1.4836997985839844, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8736217617988586, + "num_tokens": 122139088.0, + "step": 3477 + }, + { + "epoch": 0.6458681522748375, + "grad_norm": 1.3924798965454102, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8670597076416016, + "num_tokens": 122178321.0, + "step": 3478 + }, + { + "epoch": 0.6460538532961931, + "grad_norm": 1.6290395259857178, + "learning_rate": 1e-06, + "loss": 0.4918, + "mean_token_accuracy": 0.8389496207237244, + "num_tokens": 122211170.0, + "step": 3479 + }, + { + "epoch": 0.6462395543175488, + "grad_norm": 1.4588688611984253, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8717994093894958, + "num_tokens": 122247887.0, + "step": 3480 + }, + { + "epoch": 0.6464252553389044, + "grad_norm": 1.5519366264343262, + "learning_rate": 1e-06, + "loss": 0.4254, + "mean_token_accuracy": 0.863244891166687, + "num_tokens": 122278979.0, + "step": 3481 + }, + { + "epoch": 0.64661095636026, + "grad_norm": 1.506765604019165, + "learning_rate": 1e-06, + "loss": 0.4537, + "mean_token_accuracy": 0.8494502305984497, + "num_tokens": 122315962.0, + "step": 3482 + }, + { + "epoch": 0.6467966573816156, + "grad_norm": 1.5985161066055298, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8665615320205688, + "num_tokens": 122349944.0, + "step": 3483 + }, + { + "epoch": 0.6469823584029712, + "grad_norm": 1.4495257139205933, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8716683387756348, + "num_tokens": 122385076.0, + "step": 3484 + }, + { + "epoch": 0.6471680594243269, + "grad_norm": 1.4336283206939697, + "learning_rate": 1e-06, + "loss": 0.407, + "mean_token_accuracy": 0.8652489185333252, + "num_tokens": 122423117.0, + "step": 3485 + }, + { + "epoch": 0.6473537604456825, + "grad_norm": 1.600419044494629, + "learning_rate": 1e-06, + "loss": 0.4596, + "mean_token_accuracy": 0.849053144454956, + "num_tokens": 122456448.0, + "step": 3486 + }, + { + "epoch": 0.6475394614670381, + "grad_norm": 1.5487521886825562, + "learning_rate": 1e-06, + "loss": 0.4206, + "mean_token_accuracy": 0.8575952053070068, + "num_tokens": 122490426.0, + "step": 3487 + }, + { + "epoch": 0.6477251624883937, + "grad_norm": 1.4660006761550903, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.8639377951622009, + "num_tokens": 122523299.0, + "step": 3488 + }, + { + "epoch": 0.6479108635097494, + "grad_norm": 1.5592577457427979, + "learning_rate": 1e-06, + "loss": 0.4297, + "mean_token_accuracy": 0.8564179539680481, + "num_tokens": 122560544.0, + "step": 3489 + }, + { + "epoch": 0.6480965645311049, + "grad_norm": 1.487788200378418, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.871560275554657, + "num_tokens": 122592644.0, + "step": 3490 + }, + { + "epoch": 0.6482822655524605, + "grad_norm": 1.315751314163208, + "learning_rate": 1e-06, + "loss": 0.4033, + "mean_token_accuracy": 0.8639551401138306, + "num_tokens": 122633847.0, + "step": 3491 + }, + { + "epoch": 0.6484679665738161, + "grad_norm": 1.3756792545318604, + "learning_rate": 1e-06, + "loss": 0.4061, + "mean_token_accuracy": 0.8660752773284912, + "num_tokens": 122672958.0, + "step": 3492 + }, + { + "epoch": 0.6486536675951717, + "grad_norm": 1.4597712755203247, + "learning_rate": 1e-06, + "loss": 0.436, + "mean_token_accuracy": 0.8527424931526184, + "num_tokens": 122708196.0, + "step": 3493 + }, + { + "epoch": 0.6488393686165274, + "grad_norm": 1.454405426979065, + "learning_rate": 1e-06, + "loss": 0.4469, + "mean_token_accuracy": 0.8529781103134155, + "num_tokens": 122750072.0, + "step": 3494 + }, + { + "epoch": 0.649025069637883, + "grad_norm": 1.5172123908996582, + "learning_rate": 1e-06, + "loss": 0.3873, + "mean_token_accuracy": 0.8686325550079346, + "num_tokens": 122782777.0, + "step": 3495 + }, + { + "epoch": 0.6492107706592386, + "grad_norm": 1.4534528255462646, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8842573761940002, + "num_tokens": 122816562.0, + "step": 3496 + }, + { + "epoch": 0.6493964716805942, + "grad_norm": 1.5225552320480347, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8751149773597717, + "num_tokens": 122849447.0, + "step": 3497 + }, + { + "epoch": 0.6495821727019498, + "grad_norm": 1.6210992336273193, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.877913236618042, + "num_tokens": 122876721.0, + "step": 3498 + }, + { + "epoch": 0.6497678737233055, + "grad_norm": 1.323075532913208, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.870604395866394, + "num_tokens": 122919552.0, + "step": 3499 + }, + { + "epoch": 0.6499535747446611, + "grad_norm": 1.4622682332992554, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8650006651878357, + "num_tokens": 122952329.0, + "step": 3500 + }, + { + "epoch": 0.6501392757660167, + "grad_norm": 1.356992483139038, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8718604445457458, + "num_tokens": 122991120.0, + "step": 3501 + }, + { + "epoch": 0.6503249767873723, + "grad_norm": 1.4968699216842651, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8746169805526733, + "num_tokens": 123025248.0, + "step": 3502 + }, + { + "epoch": 0.650510677808728, + "grad_norm": 1.419919490814209, + "learning_rate": 1e-06, + "loss": 0.3833, + "mean_token_accuracy": 0.8712900280952454, + "num_tokens": 123060944.0, + "step": 3503 + }, + { + "epoch": 0.6506963788300836, + "grad_norm": 1.6729872226715088, + "learning_rate": 1e-06, + "loss": 0.4545, + "mean_token_accuracy": 0.8470711708068848, + "num_tokens": 123091689.0, + "step": 3504 + }, + { + "epoch": 0.6508820798514392, + "grad_norm": 1.4426854848861694, + "learning_rate": 1e-06, + "loss": 0.4504, + "mean_token_accuracy": 0.8481158018112183, + "num_tokens": 123129632.0, + "step": 3505 + }, + { + "epoch": 0.6510677808727948, + "grad_norm": 1.2977392673492432, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8737001419067383, + "num_tokens": 123168698.0, + "step": 3506 + }, + { + "epoch": 0.6512534818941504, + "grad_norm": 1.3935943841934204, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.8658835887908936, + "num_tokens": 123206929.0, + "step": 3507 + }, + { + "epoch": 0.6514391829155061, + "grad_norm": 1.2938530445098877, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8731182813644409, + "num_tokens": 123249156.0, + "step": 3508 + }, + { + "epoch": 0.6516248839368617, + "grad_norm": 1.4187253713607788, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8768045902252197, + "num_tokens": 123285641.0, + "step": 3509 + }, + { + "epoch": 0.6518105849582173, + "grad_norm": 1.382126808166504, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.864206075668335, + "num_tokens": 123325036.0, + "step": 3510 + }, + { + "epoch": 0.6519962859795729, + "grad_norm": 1.4946391582489014, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.8661736845970154, + "num_tokens": 123359007.0, + "step": 3511 + }, + { + "epoch": 0.6521819870009286, + "grad_norm": 1.5585765838623047, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8681473731994629, + "num_tokens": 123391821.0, + "step": 3512 + }, + { + "epoch": 0.6523676880222842, + "grad_norm": 1.349889874458313, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8818002939224243, + "num_tokens": 123429777.0, + "step": 3513 + }, + { + "epoch": 0.6525533890436397, + "grad_norm": 1.5435409545898438, + "learning_rate": 1e-06, + "loss": 0.457, + "mean_token_accuracy": 0.8487993478775024, + "num_tokens": 123462512.0, + "step": 3514 + }, + { + "epoch": 0.6527390900649953, + "grad_norm": 1.480671763420105, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8651825785636902, + "num_tokens": 123496278.0, + "step": 3515 + }, + { + "epoch": 0.6529247910863509, + "grad_norm": 1.4716545343399048, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8789281845092773, + "num_tokens": 123526417.0, + "step": 3516 + }, + { + "epoch": 0.6531104921077066, + "grad_norm": 1.5647096633911133, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8677746057510376, + "num_tokens": 123557363.0, + "step": 3517 + }, + { + "epoch": 0.6532961931290622, + "grad_norm": 1.5920780897140503, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.8613243103027344, + "num_tokens": 123588009.0, + "step": 3518 + }, + { + "epoch": 0.6534818941504178, + "grad_norm": 1.4081063270568848, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8658419847488403, + "num_tokens": 123628576.0, + "step": 3519 + }, + { + "epoch": 0.6536675951717734, + "grad_norm": 1.4095784425735474, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8752228021621704, + "num_tokens": 123668265.0, + "step": 3520 + }, + { + "epoch": 0.653853296193129, + "grad_norm": 1.8104534149169922, + "learning_rate": 1e-06, + "loss": 0.3878, + "mean_token_accuracy": 0.8683562278747559, + "num_tokens": 123699366.0, + "step": 3521 + }, + { + "epoch": 0.6540389972144847, + "grad_norm": 1.614820122718811, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8699030876159668, + "num_tokens": 123726647.0, + "step": 3522 + }, + { + "epoch": 0.6542246982358403, + "grad_norm": 1.5197556018829346, + "learning_rate": 1e-06, + "loss": 0.4186, + "mean_token_accuracy": 0.859410285949707, + "num_tokens": 123762564.0, + "step": 3523 + }, + { + "epoch": 0.6544103992571959, + "grad_norm": 1.3908615112304688, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.868587076663971, + "num_tokens": 123800533.0, + "step": 3524 + }, + { + "epoch": 0.6545961002785515, + "grad_norm": 1.6120741367340088, + "learning_rate": 1e-06, + "loss": 0.4156, + "mean_token_accuracy": 0.8609925508499146, + "num_tokens": 123829700.0, + "step": 3525 + }, + { + "epoch": 0.6547818012999072, + "grad_norm": 1.5162302255630493, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8774611353874207, + "num_tokens": 123865767.0, + "step": 3526 + }, + { + "epoch": 0.6549675023212628, + "grad_norm": 1.3759667873382568, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8644248247146606, + "num_tokens": 123902654.0, + "step": 3527 + }, + { + "epoch": 0.6551532033426184, + "grad_norm": 1.4480904340744019, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.874488353729248, + "num_tokens": 123934942.0, + "step": 3528 + }, + { + "epoch": 0.655338904363974, + "grad_norm": 1.5212076902389526, + "learning_rate": 1e-06, + "loss": 0.4022, + "mean_token_accuracy": 0.864362359046936, + "num_tokens": 123970195.0, + "step": 3529 + }, + { + "epoch": 0.6555246053853296, + "grad_norm": 1.439089059829712, + "learning_rate": 1e-06, + "loss": 0.4081, + "mean_token_accuracy": 0.8630516529083252, + "num_tokens": 124003813.0, + "step": 3530 + }, + { + "epoch": 0.6557103064066853, + "grad_norm": 1.4173535108566284, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8668116927146912, + "num_tokens": 124041414.0, + "step": 3531 + }, + { + "epoch": 0.6558960074280409, + "grad_norm": 1.2987792491912842, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8700346946716309, + "num_tokens": 124083036.0, + "step": 3532 + }, + { + "epoch": 0.6560817084493965, + "grad_norm": 1.4931716918945312, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8756943941116333, + "num_tokens": 124115088.0, + "step": 3533 + }, + { + "epoch": 0.6562674094707521, + "grad_norm": 1.329939365386963, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8769961595535278, + "num_tokens": 124153707.0, + "step": 3534 + }, + { + "epoch": 0.6564531104921077, + "grad_norm": 1.5720163583755493, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8733170032501221, + "num_tokens": 124184598.0, + "step": 3535 + }, + { + "epoch": 0.6566388115134634, + "grad_norm": 1.4900262355804443, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8858870267868042, + "num_tokens": 124216901.0, + "step": 3536 + }, + { + "epoch": 0.656824512534819, + "grad_norm": 1.5224215984344482, + "learning_rate": 1e-06, + "loss": 0.4296, + "mean_token_accuracy": 0.8579608798027039, + "num_tokens": 124256482.0, + "step": 3537 + }, + { + "epoch": 0.6570102135561745, + "grad_norm": 1.5932568311691284, + "learning_rate": 1e-06, + "loss": 0.4, + "mean_token_accuracy": 0.8640252947807312, + "num_tokens": 124284760.0, + "step": 3538 + }, + { + "epoch": 0.6571959145775301, + "grad_norm": 1.578421711921692, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8659862279891968, + "num_tokens": 124314057.0, + "step": 3539 + }, + { + "epoch": 0.6573816155988857, + "grad_norm": 1.4332784414291382, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8719245195388794, + "num_tokens": 124349922.0, + "step": 3540 + }, + { + "epoch": 0.6575673166202414, + "grad_norm": 1.4866704940795898, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8709627389907837, + "num_tokens": 124383015.0, + "step": 3541 + }, + { + "epoch": 0.657753017641597, + "grad_norm": 1.5032989978790283, + "learning_rate": 1e-06, + "loss": 0.4126, + "mean_token_accuracy": 0.8630664348602295, + "num_tokens": 124421673.0, + "step": 3542 + }, + { + "epoch": 0.6579387186629526, + "grad_norm": 1.5089237689971924, + "learning_rate": 1e-06, + "loss": 0.4127, + "mean_token_accuracy": 0.8600289821624756, + "num_tokens": 124455374.0, + "step": 3543 + }, + { + "epoch": 0.6581244196843082, + "grad_norm": 1.427810549736023, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8706825375556946, + "num_tokens": 124491525.0, + "step": 3544 + }, + { + "epoch": 0.6583101207056639, + "grad_norm": 1.519370675086975, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.866650402545929, + "num_tokens": 124522072.0, + "step": 3545 + }, + { + "epoch": 0.6584958217270195, + "grad_norm": 1.7534229755401611, + "learning_rate": 1e-06, + "loss": 0.446, + "mean_token_accuracy": 0.8583096861839294, + "num_tokens": 124549018.0, + "step": 3546 + }, + { + "epoch": 0.6586815227483751, + "grad_norm": 1.498935580253601, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8551800847053528, + "num_tokens": 124585010.0, + "step": 3547 + }, + { + "epoch": 0.6588672237697307, + "grad_norm": 1.4704910516738892, + "learning_rate": 1e-06, + "loss": 0.4234, + "mean_token_accuracy": 0.8615171909332275, + "num_tokens": 124624884.0, + "step": 3548 + }, + { + "epoch": 0.6590529247910863, + "grad_norm": 1.3277842998504639, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8740249872207642, + "num_tokens": 124664092.0, + "step": 3549 + }, + { + "epoch": 0.659238625812442, + "grad_norm": 1.4778648614883423, + "learning_rate": 1e-06, + "loss": 0.4031, + "mean_token_accuracy": 0.8661471605300903, + "num_tokens": 124698229.0, + "step": 3550 + }, + { + "epoch": 0.6594243268337976, + "grad_norm": 1.4985723495483398, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8752144575119019, + "num_tokens": 124729749.0, + "step": 3551 + }, + { + "epoch": 0.6596100278551532, + "grad_norm": 1.5363504886627197, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8656867146492004, + "num_tokens": 124761771.0, + "step": 3552 + }, + { + "epoch": 0.6597957288765088, + "grad_norm": 1.5785458087921143, + "learning_rate": 1e-06, + "loss": 0.4341, + "mean_token_accuracy": 0.8524996638298035, + "num_tokens": 124794251.0, + "step": 3553 + }, + { + "epoch": 0.6599814298978645, + "grad_norm": 1.4237942695617676, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.867865800857544, + "num_tokens": 124834680.0, + "step": 3554 + }, + { + "epoch": 0.6601671309192201, + "grad_norm": 1.4836318492889404, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8741595149040222, + "num_tokens": 124873160.0, + "step": 3555 + }, + { + "epoch": 0.6603528319405757, + "grad_norm": 1.4228283166885376, + "learning_rate": 1e-06, + "loss": 0.4667, + "mean_token_accuracy": 0.8461063504219055, + "num_tokens": 124914817.0, + "step": 3556 + }, + { + "epoch": 0.6605385329619313, + "grad_norm": 1.399776816368103, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8728688955307007, + "num_tokens": 124950284.0, + "step": 3557 + }, + { + "epoch": 0.660724233983287, + "grad_norm": 1.4853081703186035, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8659873008728027, + "num_tokens": 124984260.0, + "step": 3558 + }, + { + "epoch": 0.6609099350046426, + "grad_norm": 1.4803690910339355, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8682709336280823, + "num_tokens": 125017878.0, + "step": 3559 + }, + { + "epoch": 0.6610956360259982, + "grad_norm": 1.4255428314208984, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8731871247291565, + "num_tokens": 125056495.0, + "step": 3560 + }, + { + "epoch": 0.6612813370473538, + "grad_norm": 1.5011262893676758, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8521174192428589, + "num_tokens": 125089409.0, + "step": 3561 + }, + { + "epoch": 0.6614670380687093, + "grad_norm": 1.293039321899414, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8645948767662048, + "num_tokens": 125131914.0, + "step": 3562 + }, + { + "epoch": 0.661652739090065, + "grad_norm": 1.607352614402771, + "learning_rate": 1e-06, + "loss": 0.4712, + "mean_token_accuracy": 0.8466709852218628, + "num_tokens": 125166042.0, + "step": 3563 + }, + { + "epoch": 0.6618384401114206, + "grad_norm": 1.4816415309906006, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8636524677276611, + "num_tokens": 125203307.0, + "step": 3564 + }, + { + "epoch": 0.6620241411327762, + "grad_norm": 1.4021567106246948, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8687418699264526, + "num_tokens": 125238645.0, + "step": 3565 + }, + { + "epoch": 0.6622098421541318, + "grad_norm": 1.4568054676055908, + "learning_rate": 1e-06, + "loss": 0.4477, + "mean_token_accuracy": 0.8504116535186768, + "num_tokens": 125275584.0, + "step": 3566 + }, + { + "epoch": 0.6623955431754874, + "grad_norm": 1.400840163230896, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8799753189086914, + "num_tokens": 125308408.0, + "step": 3567 + }, + { + "epoch": 0.662581244196843, + "grad_norm": 1.409216046333313, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8702272176742554, + "num_tokens": 125345975.0, + "step": 3568 + }, + { + "epoch": 0.6627669452181987, + "grad_norm": 1.5720239877700806, + "learning_rate": 1e-06, + "loss": 0.4577, + "mean_token_accuracy": 0.8510939478874207, + "num_tokens": 125381103.0, + "step": 3569 + }, + { + "epoch": 0.6629526462395543, + "grad_norm": 1.5117661952972412, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.8731316924095154, + "num_tokens": 125412203.0, + "step": 3570 + }, + { + "epoch": 0.6631383472609099, + "grad_norm": 1.558646321296692, + "learning_rate": 1e-06, + "loss": 0.4085, + "mean_token_accuracy": 0.8638054728507996, + "num_tokens": 125440510.0, + "step": 3571 + }, + { + "epoch": 0.6633240482822655, + "grad_norm": 1.3932219743728638, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8548464775085449, + "num_tokens": 125475637.0, + "step": 3572 + }, + { + "epoch": 0.6635097493036212, + "grad_norm": 1.4229822158813477, + "learning_rate": 1e-06, + "loss": 0.4171, + "mean_token_accuracy": 0.8576468229293823, + "num_tokens": 125513827.0, + "step": 3573 + }, + { + "epoch": 0.6636954503249768, + "grad_norm": 1.6073660850524902, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8845610618591309, + "num_tokens": 125543081.0, + "step": 3574 + }, + { + "epoch": 0.6638811513463324, + "grad_norm": 1.4665939807891846, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8672977089881897, + "num_tokens": 125576526.0, + "step": 3575 + }, + { + "epoch": 0.664066852367688, + "grad_norm": 1.3113385438919067, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8917263746261597, + "num_tokens": 125614300.0, + "step": 3576 + }, + { + "epoch": 0.6642525533890437, + "grad_norm": 1.5269041061401367, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8557575345039368, + "num_tokens": 125653396.0, + "step": 3577 + }, + { + "epoch": 0.6644382544103993, + "grad_norm": 1.4467653036117554, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8794653415679932, + "num_tokens": 125687583.0, + "step": 3578 + }, + { + "epoch": 0.6646239554317549, + "grad_norm": 1.583001732826233, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8652593493461609, + "num_tokens": 125715830.0, + "step": 3579 + }, + { + "epoch": 0.6648096564531105, + "grad_norm": 1.4150781631469727, + "learning_rate": 1e-06, + "loss": 0.4235, + "mean_token_accuracy": 0.8593720197677612, + "num_tokens": 125755777.0, + "step": 3580 + }, + { + "epoch": 0.6649953574744661, + "grad_norm": 1.382315754890442, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8627327084541321, + "num_tokens": 125795479.0, + "step": 3581 + }, + { + "epoch": 0.6651810584958218, + "grad_norm": 1.4311498403549194, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8731065988540649, + "num_tokens": 125832307.0, + "step": 3582 + }, + { + "epoch": 0.6653667595171774, + "grad_norm": 1.446431279182434, + "learning_rate": 1e-06, + "loss": 0.4121, + "mean_token_accuracy": 0.8559281826019287, + "num_tokens": 125874400.0, + "step": 3583 + }, + { + "epoch": 0.665552460538533, + "grad_norm": 1.3862159252166748, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8754266500473022, + "num_tokens": 125911589.0, + "step": 3584 + }, + { + "epoch": 0.6657381615598886, + "grad_norm": 1.5968492031097412, + "learning_rate": 1e-06, + "loss": 0.437, + "mean_token_accuracy": 0.8558709621429443, + "num_tokens": 125942473.0, + "step": 3585 + }, + { + "epoch": 0.6659238625812441, + "grad_norm": 1.387170672416687, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8651899695396423, + "num_tokens": 125982859.0, + "step": 3586 + }, + { + "epoch": 0.6661095636025998, + "grad_norm": 1.4611200094223022, + "learning_rate": 1e-06, + "loss": 0.4266, + "mean_token_accuracy": 0.8595246076583862, + "num_tokens": 126017592.0, + "step": 3587 + }, + { + "epoch": 0.6662952646239554, + "grad_norm": 1.676871657371521, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.8577502965927124, + "num_tokens": 126049102.0, + "step": 3588 + }, + { + "epoch": 0.666480965645311, + "grad_norm": 1.4352295398712158, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8754088878631592, + "num_tokens": 126081048.0, + "step": 3589 + }, + { + "epoch": 0.6666666666666666, + "grad_norm": 1.555621862411499, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8742213249206543, + "num_tokens": 126109145.0, + "step": 3590 + }, + { + "epoch": 0.6668523676880223, + "grad_norm": 1.488595724105835, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.8673287630081177, + "num_tokens": 126144498.0, + "step": 3591 + }, + { + "epoch": 0.6670380687093779, + "grad_norm": 1.4024947881698608, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8571251630783081, + "num_tokens": 126181919.0, + "step": 3592 + }, + { + "epoch": 0.6672237697307335, + "grad_norm": 1.5525712966918945, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.869093656539917, + "num_tokens": 126209571.0, + "step": 3593 + }, + { + "epoch": 0.6674094707520891, + "grad_norm": 1.500781536102295, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8685953617095947, + "num_tokens": 126244492.0, + "step": 3594 + }, + { + "epoch": 0.6675951717734447, + "grad_norm": 1.4899219274520874, + "learning_rate": 1e-06, + "loss": 0.4247, + "mean_token_accuracy": 0.8561009764671326, + "num_tokens": 126280296.0, + "step": 3595 + }, + { + "epoch": 0.6677808727948004, + "grad_norm": 1.5203547477722168, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8737329244613647, + "num_tokens": 126312723.0, + "step": 3596 + }, + { + "epoch": 0.667966573816156, + "grad_norm": 1.6698758602142334, + "learning_rate": 1e-06, + "loss": 0.4485, + "mean_token_accuracy": 0.8579388856887817, + "num_tokens": 126341256.0, + "step": 3597 + }, + { + "epoch": 0.6681522748375116, + "grad_norm": 1.4965602159500122, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8659828305244446, + "num_tokens": 126379816.0, + "step": 3598 + }, + { + "epoch": 0.6683379758588672, + "grad_norm": 1.4762672185897827, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8617702722549438, + "num_tokens": 126416173.0, + "step": 3599 + }, + { + "epoch": 0.6685236768802229, + "grad_norm": 1.4790377616882324, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8614950180053711, + "num_tokens": 126449606.0, + "step": 3600 + }, + { + "epoch": 0.6687093779015785, + "grad_norm": 1.4733290672302246, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.8724240660667419, + "num_tokens": 126483587.0, + "step": 3601 + }, + { + "epoch": 0.6688950789229341, + "grad_norm": 1.40427565574646, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.8655798435211182, + "num_tokens": 126525190.0, + "step": 3602 + }, + { + "epoch": 0.6690807799442897, + "grad_norm": 1.437436580657959, + "learning_rate": 1e-06, + "loss": 0.4167, + "mean_token_accuracy": 0.8587230443954468, + "num_tokens": 126559171.0, + "step": 3603 + }, + { + "epoch": 0.6692664809656453, + "grad_norm": 1.4710276126861572, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.865025520324707, + "num_tokens": 126600151.0, + "step": 3604 + }, + { + "epoch": 0.669452181987001, + "grad_norm": 1.485258936882019, + "learning_rate": 1e-06, + "loss": 0.4285, + "mean_token_accuracy": 0.8567299842834473, + "num_tokens": 126637680.0, + "step": 3605 + }, + { + "epoch": 0.6696378830083566, + "grad_norm": 1.437630295753479, + "learning_rate": 1e-06, + "loss": 0.4327, + "mean_token_accuracy": 0.8567132949829102, + "num_tokens": 126677983.0, + "step": 3606 + }, + { + "epoch": 0.6698235840297122, + "grad_norm": 1.6182643175125122, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8675620555877686, + "num_tokens": 126706437.0, + "step": 3607 + }, + { + "epoch": 0.6700092850510678, + "grad_norm": 1.4641917943954468, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8717255592346191, + "num_tokens": 126739644.0, + "step": 3608 + }, + { + "epoch": 0.6701949860724234, + "grad_norm": 1.7194126844406128, + "learning_rate": 1e-06, + "loss": 0.4154, + "mean_token_accuracy": 0.864026665687561, + "num_tokens": 126766325.0, + "step": 3609 + }, + { + "epoch": 0.6703806870937791, + "grad_norm": 1.6549147367477417, + "learning_rate": 1e-06, + "loss": 0.4249, + "mean_token_accuracy": 0.8612757325172424, + "num_tokens": 126799544.0, + "step": 3610 + }, + { + "epoch": 0.6705663881151346, + "grad_norm": 1.7988234758377075, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.866706371307373, + "num_tokens": 126835552.0, + "step": 3611 + }, + { + "epoch": 0.6707520891364902, + "grad_norm": 1.8865524530410767, + "learning_rate": 1e-06, + "loss": 0.4478, + "mean_token_accuracy": 0.8532551527023315, + "num_tokens": 126872280.0, + "step": 3612 + }, + { + "epoch": 0.6709377901578458, + "grad_norm": 1.4082086086273193, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8756135106086731, + "num_tokens": 126910112.0, + "step": 3613 + }, + { + "epoch": 0.6711234911792014, + "grad_norm": 1.6039234399795532, + "learning_rate": 1e-06, + "loss": 0.4377, + "mean_token_accuracy": 0.8540182113647461, + "num_tokens": 126950162.0, + "step": 3614 + }, + { + "epoch": 0.6713091922005571, + "grad_norm": 1.5369300842285156, + "learning_rate": 1e-06, + "loss": 0.4047, + "mean_token_accuracy": 0.8636519908905029, + "num_tokens": 126984178.0, + "step": 3615 + }, + { + "epoch": 0.6714948932219127, + "grad_norm": 1.506710171699524, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8680063486099243, + "num_tokens": 127020715.0, + "step": 3616 + }, + { + "epoch": 0.6716805942432683, + "grad_norm": 1.4080816507339478, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8682851791381836, + "num_tokens": 127060584.0, + "step": 3617 + }, + { + "epoch": 0.6718662952646239, + "grad_norm": 1.8359962701797485, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8692066073417664, + "num_tokens": 127092550.0, + "step": 3618 + }, + { + "epoch": 0.6720519962859796, + "grad_norm": 1.7407690286636353, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8647322058677673, + "num_tokens": 127116914.0, + "step": 3619 + }, + { + "epoch": 0.6722376973073352, + "grad_norm": 1.421657681465149, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8757038712501526, + "num_tokens": 127153789.0, + "step": 3620 + }, + { + "epoch": 0.6724233983286908, + "grad_norm": 1.5153756141662598, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8667312264442444, + "num_tokens": 127190628.0, + "step": 3621 + }, + { + "epoch": 0.6726090993500464, + "grad_norm": 1.3309487104415894, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8812462091445923, + "num_tokens": 127232115.0, + "step": 3622 + }, + { + "epoch": 0.672794800371402, + "grad_norm": 1.508371114730835, + "learning_rate": 1e-06, + "loss": 0.4326, + "mean_token_accuracy": 0.8565011024475098, + "num_tokens": 127269152.0, + "step": 3623 + }, + { + "epoch": 0.6729805013927577, + "grad_norm": 1.5311516523361206, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8797270059585571, + "num_tokens": 127303645.0, + "step": 3624 + }, + { + "epoch": 0.6731662024141133, + "grad_norm": 1.7166353464126587, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8620785474777222, + "num_tokens": 127333655.0, + "step": 3625 + }, + { + "epoch": 0.6733519034354689, + "grad_norm": 1.554591178894043, + "learning_rate": 1e-06, + "loss": 0.4205, + "mean_token_accuracy": 0.8587316870689392, + "num_tokens": 127368045.0, + "step": 3626 + }, + { + "epoch": 0.6735376044568245, + "grad_norm": 1.4388527870178223, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8771381974220276, + "num_tokens": 127404055.0, + "step": 3627 + }, + { + "epoch": 0.6737233054781802, + "grad_norm": 1.626989483833313, + "learning_rate": 1e-06, + "loss": 0.4327, + "mean_token_accuracy": 0.8563379645347595, + "num_tokens": 127432009.0, + "step": 3628 + }, + { + "epoch": 0.6739090064995358, + "grad_norm": 1.523188829421997, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8673606514930725, + "num_tokens": 127464899.0, + "step": 3629 + }, + { + "epoch": 0.6740947075208914, + "grad_norm": 1.5112154483795166, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8801466822624207, + "num_tokens": 127502988.0, + "step": 3630 + }, + { + "epoch": 0.674280408542247, + "grad_norm": 1.6225415468215942, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8603847026824951, + "num_tokens": 127535454.0, + "step": 3631 + }, + { + "epoch": 0.6744661095636026, + "grad_norm": 1.5536667108535767, + "learning_rate": 1e-06, + "loss": 0.473, + "mean_token_accuracy": 0.8431335687637329, + "num_tokens": 127574128.0, + "step": 3632 + }, + { + "epoch": 0.6746518105849583, + "grad_norm": 1.5720067024230957, + "learning_rate": 1e-06, + "loss": 0.4514, + "mean_token_accuracy": 0.8546853065490723, + "num_tokens": 127609526.0, + "step": 3633 + }, + { + "epoch": 0.6748375116063139, + "grad_norm": 1.4772948026657104, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8766822814941406, + "num_tokens": 127641324.0, + "step": 3634 + }, + { + "epoch": 0.6750232126276694, + "grad_norm": 1.4878208637237549, + "learning_rate": 1e-06, + "loss": 0.413, + "mean_token_accuracy": 0.8669679164886475, + "num_tokens": 127680516.0, + "step": 3635 + }, + { + "epoch": 0.675208913649025, + "grad_norm": 1.4649150371551514, + "learning_rate": 1e-06, + "loss": 0.4289, + "mean_token_accuracy": 0.8564008474349976, + "num_tokens": 127720371.0, + "step": 3636 + }, + { + "epoch": 0.6753946146703806, + "grad_norm": 1.5223356485366821, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8696171045303345, + "num_tokens": 127752871.0, + "step": 3637 + }, + { + "epoch": 0.6755803156917363, + "grad_norm": 1.3661595582962036, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8707165718078613, + "num_tokens": 127789760.0, + "step": 3638 + }, + { + "epoch": 0.6757660167130919, + "grad_norm": 1.4374091625213623, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.866641640663147, + "num_tokens": 127826341.0, + "step": 3639 + }, + { + "epoch": 0.6759517177344475, + "grad_norm": 1.4834625720977783, + "learning_rate": 1e-06, + "loss": 0.4238, + "mean_token_accuracy": 0.8615736365318298, + "num_tokens": 127861097.0, + "step": 3640 + }, + { + "epoch": 0.6761374187558031, + "grad_norm": 1.493571162223816, + "learning_rate": 1e-06, + "loss": 0.442, + "mean_token_accuracy": 0.852660596370697, + "num_tokens": 127896338.0, + "step": 3641 + }, + { + "epoch": 0.6763231197771588, + "grad_norm": 1.5180460214614868, + "learning_rate": 1e-06, + "loss": 0.4535, + "mean_token_accuracy": 0.8524094820022583, + "num_tokens": 127939482.0, + "step": 3642 + }, + { + "epoch": 0.6765088207985144, + "grad_norm": 1.4095085859298706, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8756992220878601, + "num_tokens": 127973559.0, + "step": 3643 + }, + { + "epoch": 0.67669452181987, + "grad_norm": 1.4169652462005615, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.859915554523468, + "num_tokens": 128009966.0, + "step": 3644 + }, + { + "epoch": 0.6768802228412256, + "grad_norm": 1.614936351776123, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8784012794494629, + "num_tokens": 128037700.0, + "step": 3645 + }, + { + "epoch": 0.6770659238625812, + "grad_norm": 1.5117790699005127, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8602393865585327, + "num_tokens": 128072794.0, + "step": 3646 + }, + { + "epoch": 0.6772516248839369, + "grad_norm": 1.5611310005187988, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8563827276229858, + "num_tokens": 128106581.0, + "step": 3647 + }, + { + "epoch": 0.6774373259052925, + "grad_norm": 1.4437613487243652, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8661157488822937, + "num_tokens": 128143149.0, + "step": 3648 + }, + { + "epoch": 0.6776230269266481, + "grad_norm": 1.5057600736618042, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8738724589347839, + "num_tokens": 128177033.0, + "step": 3649 + }, + { + "epoch": 0.6778087279480037, + "grad_norm": 1.36944580078125, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8712078928947449, + "num_tokens": 128219794.0, + "step": 3650 + }, + { + "epoch": 0.6779944289693594, + "grad_norm": 1.571436882019043, + "learning_rate": 1e-06, + "loss": 0.4711, + "mean_token_accuracy": 0.8433568477630615, + "num_tokens": 128254798.0, + "step": 3651 + }, + { + "epoch": 0.678180129990715, + "grad_norm": 1.4582360982894897, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8698675632476807, + "num_tokens": 128288450.0, + "step": 3652 + }, + { + "epoch": 0.6783658310120706, + "grad_norm": 1.4380335807800293, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8560930490493774, + "num_tokens": 128324807.0, + "step": 3653 + }, + { + "epoch": 0.6785515320334262, + "grad_norm": 1.4789390563964844, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8618615865707397, + "num_tokens": 128358157.0, + "step": 3654 + }, + { + "epoch": 0.6787372330547818, + "grad_norm": 1.494370937347412, + "learning_rate": 1e-06, + "loss": 0.4461, + "mean_token_accuracy": 0.8504229784011841, + "num_tokens": 128395222.0, + "step": 3655 + }, + { + "epoch": 0.6789229340761375, + "grad_norm": 1.4849611520767212, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8808667659759521, + "num_tokens": 128427881.0, + "step": 3656 + }, + { + "epoch": 0.6791086350974931, + "grad_norm": 1.4157402515411377, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8700555562973022, + "num_tokens": 128462133.0, + "step": 3657 + }, + { + "epoch": 0.6792943361188487, + "grad_norm": 1.457863688468933, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8667833805084229, + "num_tokens": 128499226.0, + "step": 3658 + }, + { + "epoch": 0.6794800371402042, + "grad_norm": 1.3795852661132812, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8656404614448547, + "num_tokens": 128537849.0, + "step": 3659 + }, + { + "epoch": 0.6796657381615598, + "grad_norm": 1.3778300285339355, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8635112047195435, + "num_tokens": 128577256.0, + "step": 3660 + }, + { + "epoch": 0.6798514391829155, + "grad_norm": 1.5413389205932617, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8661732077598572, + "num_tokens": 128607967.0, + "step": 3661 + }, + { + "epoch": 0.6800371402042711, + "grad_norm": 1.3865389823913574, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8728144764900208, + "num_tokens": 128647085.0, + "step": 3662 + }, + { + "epoch": 0.6802228412256267, + "grad_norm": 1.4379998445510864, + "learning_rate": 1e-06, + "loss": 0.427, + "mean_token_accuracy": 0.8553211092948914, + "num_tokens": 128683854.0, + "step": 3663 + }, + { + "epoch": 0.6804085422469823, + "grad_norm": 1.4514516592025757, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8686730265617371, + "num_tokens": 128717559.0, + "step": 3664 + }, + { + "epoch": 0.680594243268338, + "grad_norm": 1.5276955366134644, + "learning_rate": 1e-06, + "loss": 0.4359, + "mean_token_accuracy": 0.8567806482315063, + "num_tokens": 128753391.0, + "step": 3665 + }, + { + "epoch": 0.6807799442896936, + "grad_norm": 1.6360331773757935, + "learning_rate": 1e-06, + "loss": 0.4483, + "mean_token_accuracy": 0.8573865294456482, + "num_tokens": 128785822.0, + "step": 3666 + }, + { + "epoch": 0.6809656453110492, + "grad_norm": 1.4138431549072266, + "learning_rate": 1e-06, + "loss": 0.3966, + "mean_token_accuracy": 0.8678653240203857, + "num_tokens": 128822445.0, + "step": 3667 + }, + { + "epoch": 0.6811513463324048, + "grad_norm": 1.3497908115386963, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8776307106018066, + "num_tokens": 128861562.0, + "step": 3668 + }, + { + "epoch": 0.6813370473537604, + "grad_norm": 1.6968997716903687, + "learning_rate": 1e-06, + "loss": 0.4145, + "mean_token_accuracy": 0.8623155355453491, + "num_tokens": 128893293.0, + "step": 3669 + }, + { + "epoch": 0.6815227483751161, + "grad_norm": 1.3812429904937744, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8674536347389221, + "num_tokens": 128930968.0, + "step": 3670 + }, + { + "epoch": 0.6817084493964717, + "grad_norm": 1.3281291723251343, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.875716507434845, + "num_tokens": 128971358.0, + "step": 3671 + }, + { + "epoch": 0.6818941504178273, + "grad_norm": 1.4553301334381104, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8680291771888733, + "num_tokens": 129006734.0, + "step": 3672 + }, + { + "epoch": 0.6820798514391829, + "grad_norm": 1.564658284187317, + "learning_rate": 1e-06, + "loss": 0.4247, + "mean_token_accuracy": 0.8580313324928284, + "num_tokens": 129041789.0, + "step": 3673 + }, + { + "epoch": 0.6822655524605385, + "grad_norm": 1.4423373937606812, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8749476671218872, + "num_tokens": 129078662.0, + "step": 3674 + }, + { + "epoch": 0.6824512534818942, + "grad_norm": 1.4742977619171143, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8742094039916992, + "num_tokens": 129110919.0, + "step": 3675 + }, + { + "epoch": 0.6826369545032498, + "grad_norm": 1.402459979057312, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8712834715843201, + "num_tokens": 129151555.0, + "step": 3676 + }, + { + "epoch": 0.6828226555246054, + "grad_norm": 1.4285868406295776, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8728432655334473, + "num_tokens": 129187060.0, + "step": 3677 + }, + { + "epoch": 0.683008356545961, + "grad_norm": 1.5812454223632812, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.875599205493927, + "num_tokens": 129217144.0, + "step": 3678 + }, + { + "epoch": 0.6831940575673167, + "grad_norm": 1.466542363166809, + "learning_rate": 1e-06, + "loss": 0.427, + "mean_token_accuracy": 0.8607890605926514, + "num_tokens": 129256616.0, + "step": 3679 + }, + { + "epoch": 0.6833797585886723, + "grad_norm": 1.3212037086486816, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8807903528213501, + "num_tokens": 129296668.0, + "step": 3680 + }, + { + "epoch": 0.6835654596100279, + "grad_norm": 1.6237080097198486, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8558815717697144, + "num_tokens": 129326165.0, + "step": 3681 + }, + { + "epoch": 0.6837511606313835, + "grad_norm": 1.435660719871521, + "learning_rate": 1e-06, + "loss": 0.4644, + "mean_token_accuracy": 0.8461858034133911, + "num_tokens": 129362517.0, + "step": 3682 + }, + { + "epoch": 0.683936861652739, + "grad_norm": 1.436526894569397, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8756453990936279, + "num_tokens": 129395893.0, + "step": 3683 + }, + { + "epoch": 0.6841225626740947, + "grad_norm": 1.4003007411956787, + "learning_rate": 1e-06, + "loss": 0.4211, + "mean_token_accuracy": 0.8600224852561951, + "num_tokens": 129436430.0, + "step": 3684 + }, + { + "epoch": 0.6843082636954503, + "grad_norm": 1.365220308303833, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8687750101089478, + "num_tokens": 129477385.0, + "step": 3685 + }, + { + "epoch": 0.6844939647168059, + "grad_norm": 1.4061388969421387, + "learning_rate": 1e-06, + "loss": 0.3967, + "mean_token_accuracy": 0.8666470646858215, + "num_tokens": 129518738.0, + "step": 3686 + }, + { + "epoch": 0.6846796657381615, + "grad_norm": 1.5059670209884644, + "learning_rate": 1e-06, + "loss": 0.4433, + "mean_token_accuracy": 0.8550407886505127, + "num_tokens": 129554462.0, + "step": 3687 + }, + { + "epoch": 0.6848653667595171, + "grad_norm": 1.548668384552002, + "learning_rate": 1e-06, + "loss": 0.4412, + "mean_token_accuracy": 0.854602038860321, + "num_tokens": 129590513.0, + "step": 3688 + }, + { + "epoch": 0.6850510677808728, + "grad_norm": 1.4062385559082031, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.8622093796730042, + "num_tokens": 129628851.0, + "step": 3689 + }, + { + "epoch": 0.6852367688022284, + "grad_norm": 1.4330389499664307, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8726223111152649, + "num_tokens": 129662472.0, + "step": 3690 + }, + { + "epoch": 0.685422469823584, + "grad_norm": 1.5581560134887695, + "learning_rate": 1e-06, + "loss": 0.4339, + "mean_token_accuracy": 0.8573954701423645, + "num_tokens": 129694087.0, + "step": 3691 + }, + { + "epoch": 0.6856081708449396, + "grad_norm": 1.3838677406311035, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8795661926269531, + "num_tokens": 129729435.0, + "step": 3692 + }, + { + "epoch": 0.6857938718662953, + "grad_norm": 1.4106866121292114, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8727688789367676, + "num_tokens": 129769045.0, + "step": 3693 + }, + { + "epoch": 0.6859795728876509, + "grad_norm": 1.5422788858413696, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8666688799858093, + "num_tokens": 129801259.0, + "step": 3694 + }, + { + "epoch": 0.6861652739090065, + "grad_norm": 1.4443645477294922, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8583523631095886, + "num_tokens": 129836812.0, + "step": 3695 + }, + { + "epoch": 0.6863509749303621, + "grad_norm": 1.3375905752182007, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8711287975311279, + "num_tokens": 129877058.0, + "step": 3696 + }, + { + "epoch": 0.6865366759517177, + "grad_norm": 1.4609683752059937, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8750194907188416, + "num_tokens": 129914907.0, + "step": 3697 + }, + { + "epoch": 0.6867223769730734, + "grad_norm": 1.4180389642715454, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.875105082988739, + "num_tokens": 129951277.0, + "step": 3698 + }, + { + "epoch": 0.686908077994429, + "grad_norm": 1.5408389568328857, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8516712188720703, + "num_tokens": 129983988.0, + "step": 3699 + }, + { + "epoch": 0.6870937790157846, + "grad_norm": 1.399574875831604, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8718807101249695, + "num_tokens": 130025105.0, + "step": 3700 + }, + { + "epoch": 0.6872794800371402, + "grad_norm": 1.5268394947052002, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8589845299720764, + "num_tokens": 130056461.0, + "step": 3701 + }, + { + "epoch": 0.6874651810584959, + "grad_norm": 1.4693282842636108, + "learning_rate": 1e-06, + "loss": 0.4562, + "mean_token_accuracy": 0.8511512279510498, + "num_tokens": 130093952.0, + "step": 3702 + }, + { + "epoch": 0.6876508820798515, + "grad_norm": 1.3181426525115967, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8737012147903442, + "num_tokens": 130134937.0, + "step": 3703 + }, + { + "epoch": 0.6878365831012071, + "grad_norm": 1.3508113622665405, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8646184206008911, + "num_tokens": 130173915.0, + "step": 3704 + }, + { + "epoch": 0.6880222841225627, + "grad_norm": 1.4789105653762817, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8715777397155762, + "num_tokens": 130207481.0, + "step": 3705 + }, + { + "epoch": 0.6882079851439183, + "grad_norm": 1.51768958568573, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.8634151816368103, + "num_tokens": 130240909.0, + "step": 3706 + }, + { + "epoch": 0.6883936861652739, + "grad_norm": 1.4780641794204712, + "learning_rate": 1e-06, + "loss": 0.4072, + "mean_token_accuracy": 0.8641551733016968, + "num_tokens": 130278027.0, + "step": 3707 + }, + { + "epoch": 0.6885793871866295, + "grad_norm": 1.3798496723175049, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.864722728729248, + "num_tokens": 130319614.0, + "step": 3708 + }, + { + "epoch": 0.6887650882079851, + "grad_norm": 1.4481276273727417, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.870132327079773, + "num_tokens": 130356011.0, + "step": 3709 + }, + { + "epoch": 0.6889507892293407, + "grad_norm": 1.4838941097259521, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8706007599830627, + "num_tokens": 130391841.0, + "step": 3710 + }, + { + "epoch": 0.6891364902506963, + "grad_norm": 1.5400627851486206, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.8677898049354553, + "num_tokens": 130423555.0, + "step": 3711 + }, + { + "epoch": 0.689322191272052, + "grad_norm": 1.5441021919250488, + "learning_rate": 1e-06, + "loss": 0.4245, + "mean_token_accuracy": 0.8551099896430969, + "num_tokens": 130461673.0, + "step": 3712 + }, + { + "epoch": 0.6895078922934076, + "grad_norm": 1.6532683372497559, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8712887763977051, + "num_tokens": 130486335.0, + "step": 3713 + }, + { + "epoch": 0.6896935933147632, + "grad_norm": 1.6064226627349854, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8575350046157837, + "num_tokens": 130517620.0, + "step": 3714 + }, + { + "epoch": 0.6898792943361188, + "grad_norm": 1.5344635248184204, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8682268857955933, + "num_tokens": 130549076.0, + "step": 3715 + }, + { + "epoch": 0.6900649953574745, + "grad_norm": 1.6464142799377441, + "learning_rate": 1e-06, + "loss": 0.4603, + "mean_token_accuracy": 0.8485170602798462, + "num_tokens": 130580801.0, + "step": 3716 + }, + { + "epoch": 0.6902506963788301, + "grad_norm": 1.5347850322723389, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.865787923336029, + "num_tokens": 130611071.0, + "step": 3717 + }, + { + "epoch": 0.6904363974001857, + "grad_norm": 1.5122227668762207, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8755646347999573, + "num_tokens": 130641511.0, + "step": 3718 + }, + { + "epoch": 0.6906220984215413, + "grad_norm": 1.4201678037643433, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8802661299705505, + "num_tokens": 130671777.0, + "step": 3719 + }, + { + "epoch": 0.6908077994428969, + "grad_norm": 1.4491339921951294, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8706096410751343, + "num_tokens": 130705818.0, + "step": 3720 + }, + { + "epoch": 0.6909935004642526, + "grad_norm": 1.4161303043365479, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8771395683288574, + "num_tokens": 130739954.0, + "step": 3721 + }, + { + "epoch": 0.6911792014856082, + "grad_norm": 1.636866569519043, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8648054599761963, + "num_tokens": 130772862.0, + "step": 3722 + }, + { + "epoch": 0.6913649025069638, + "grad_norm": 1.4011846780776978, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8764070272445679, + "num_tokens": 130809025.0, + "step": 3723 + }, + { + "epoch": 0.6915506035283194, + "grad_norm": 1.4188629388809204, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.862945556640625, + "num_tokens": 130844583.0, + "step": 3724 + }, + { + "epoch": 0.691736304549675, + "grad_norm": 1.5798959732055664, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8639330863952637, + "num_tokens": 130872211.0, + "step": 3725 + }, + { + "epoch": 0.6919220055710307, + "grad_norm": 1.5294156074523926, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8645330667495728, + "num_tokens": 130901421.0, + "step": 3726 + }, + { + "epoch": 0.6921077065923863, + "grad_norm": 1.510923981666565, + "learning_rate": 1e-06, + "loss": 0.4422, + "mean_token_accuracy": 0.8512291312217712, + "num_tokens": 130938053.0, + "step": 3727 + }, + { + "epoch": 0.6922934076137419, + "grad_norm": 1.5057588815689087, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8713706731796265, + "num_tokens": 130965446.0, + "step": 3728 + }, + { + "epoch": 0.6924791086350975, + "grad_norm": 1.5261675119400024, + "learning_rate": 1e-06, + "loss": 0.441, + "mean_token_accuracy": 0.8546329736709595, + "num_tokens": 130999385.0, + "step": 3729 + }, + { + "epoch": 0.6926648096564532, + "grad_norm": 1.602838397026062, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8629463911056519, + "num_tokens": 131033905.0, + "step": 3730 + }, + { + "epoch": 0.6928505106778087, + "grad_norm": 1.5089802742004395, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.861944317817688, + "num_tokens": 131064487.0, + "step": 3731 + }, + { + "epoch": 0.6930362116991643, + "grad_norm": 1.6349081993103027, + "learning_rate": 1e-06, + "loss": 0.4066, + "mean_token_accuracy": 0.8647894859313965, + "num_tokens": 131094591.0, + "step": 3732 + }, + { + "epoch": 0.6932219127205199, + "grad_norm": 1.5282173156738281, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8627017736434937, + "num_tokens": 131128774.0, + "step": 3733 + }, + { + "epoch": 0.6934076137418755, + "grad_norm": 1.5716298818588257, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8607927560806274, + "num_tokens": 131163428.0, + "step": 3734 + }, + { + "epoch": 0.6935933147632312, + "grad_norm": 1.5605531930923462, + "learning_rate": 1e-06, + "loss": 0.4251, + "mean_token_accuracy": 0.8574244976043701, + "num_tokens": 131196507.0, + "step": 3735 + }, + { + "epoch": 0.6937790157845868, + "grad_norm": 1.4327856302261353, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8638442158699036, + "num_tokens": 131238312.0, + "step": 3736 + }, + { + "epoch": 0.6939647168059424, + "grad_norm": 1.518307089805603, + "learning_rate": 1e-06, + "loss": 0.4707, + "mean_token_accuracy": 0.8481897115707397, + "num_tokens": 131275913.0, + "step": 3737 + }, + { + "epoch": 0.694150417827298, + "grad_norm": 1.5755281448364258, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8632879257202148, + "num_tokens": 131312446.0, + "step": 3738 + }, + { + "epoch": 0.6943361188486536, + "grad_norm": 1.4391202926635742, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8745514154434204, + "num_tokens": 131348240.0, + "step": 3739 + }, + { + "epoch": 0.6945218198700093, + "grad_norm": 1.3881598711013794, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8779680132865906, + "num_tokens": 131383936.0, + "step": 3740 + }, + { + "epoch": 0.6947075208913649, + "grad_norm": 1.6509426832199097, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8629846572875977, + "num_tokens": 131416002.0, + "step": 3741 + }, + { + "epoch": 0.6948932219127205, + "grad_norm": 1.453762412071228, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.866954505443573, + "num_tokens": 131452496.0, + "step": 3742 + }, + { + "epoch": 0.6950789229340761, + "grad_norm": 1.476640224456787, + "learning_rate": 1e-06, + "loss": 0.4144, + "mean_token_accuracy": 0.8586719036102295, + "num_tokens": 131486954.0, + "step": 3743 + }, + { + "epoch": 0.6952646239554318, + "grad_norm": 1.4044326543807983, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8650226593017578, + "num_tokens": 131522277.0, + "step": 3744 + }, + { + "epoch": 0.6954503249767874, + "grad_norm": 1.436407208442688, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8747355341911316, + "num_tokens": 131558209.0, + "step": 3745 + }, + { + "epoch": 0.695636025998143, + "grad_norm": 1.5987842082977295, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8749812841415405, + "num_tokens": 131587917.0, + "step": 3746 + }, + { + "epoch": 0.6958217270194986, + "grad_norm": 1.4088101387023926, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8598212003707886, + "num_tokens": 131626472.0, + "step": 3747 + }, + { + "epoch": 0.6960074280408542, + "grad_norm": 1.4971153736114502, + "learning_rate": 1e-06, + "loss": 0.414, + "mean_token_accuracy": 0.8624442219734192, + "num_tokens": 131661650.0, + "step": 3748 + }, + { + "epoch": 0.6961931290622099, + "grad_norm": 1.4295456409454346, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8766024708747864, + "num_tokens": 131696241.0, + "step": 3749 + }, + { + "epoch": 0.6963788300835655, + "grad_norm": 1.347917914390564, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8664827942848206, + "num_tokens": 131735673.0, + "step": 3750 + }, + { + "epoch": 0.6965645311049211, + "grad_norm": 1.3917235136032104, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8773648738861084, + "num_tokens": 131772854.0, + "step": 3751 + }, + { + "epoch": 0.6967502321262767, + "grad_norm": 1.5669541358947754, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8619586825370789, + "num_tokens": 131807713.0, + "step": 3752 + }, + { + "epoch": 0.6969359331476324, + "grad_norm": 1.4715739488601685, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8782349824905396, + "num_tokens": 131841052.0, + "step": 3753 + }, + { + "epoch": 0.697121634168988, + "grad_norm": 1.4843339920043945, + "learning_rate": 1e-06, + "loss": 0.4234, + "mean_token_accuracy": 0.8561804294586182, + "num_tokens": 131878830.0, + "step": 3754 + }, + { + "epoch": 0.6973073351903436, + "grad_norm": 1.4701433181762695, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8582860231399536, + "num_tokens": 131912467.0, + "step": 3755 + }, + { + "epoch": 0.6974930362116991, + "grad_norm": 1.4631186723709106, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8633040189743042, + "num_tokens": 131949608.0, + "step": 3756 + }, + { + "epoch": 0.6976787372330547, + "grad_norm": 1.5008355379104614, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8630252480506897, + "num_tokens": 131984138.0, + "step": 3757 + }, + { + "epoch": 0.6978644382544104, + "grad_norm": 1.336832880973816, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8789117336273193, + "num_tokens": 132023855.0, + "step": 3758 + }, + { + "epoch": 0.698050139275766, + "grad_norm": 1.427697777748108, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8720798492431641, + "num_tokens": 132057818.0, + "step": 3759 + }, + { + "epoch": 0.6982358402971216, + "grad_norm": 1.5020824670791626, + "learning_rate": 1e-06, + "loss": 0.4224, + "mean_token_accuracy": 0.8559032678604126, + "num_tokens": 132092150.0, + "step": 3760 + }, + { + "epoch": 0.6984215413184772, + "grad_norm": 1.4944571256637573, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8712884783744812, + "num_tokens": 132125615.0, + "step": 3761 + }, + { + "epoch": 0.6986072423398328, + "grad_norm": 1.4215986728668213, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8640590906143188, + "num_tokens": 132162276.0, + "step": 3762 + }, + { + "epoch": 0.6987929433611885, + "grad_norm": 1.4320576190948486, + "learning_rate": 1e-06, + "loss": 0.4643, + "mean_token_accuracy": 0.8495853543281555, + "num_tokens": 132203825.0, + "step": 3763 + }, + { + "epoch": 0.6989786443825441, + "grad_norm": 1.5130822658538818, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8654618859291077, + "num_tokens": 132239174.0, + "step": 3764 + }, + { + "epoch": 0.6991643454038997, + "grad_norm": 1.3609380722045898, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.863849401473999, + "num_tokens": 132278945.0, + "step": 3765 + }, + { + "epoch": 0.6993500464252553, + "grad_norm": 1.4044383764266968, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8727526664733887, + "num_tokens": 132318537.0, + "step": 3766 + }, + { + "epoch": 0.699535747446611, + "grad_norm": 1.53383207321167, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8647242784500122, + "num_tokens": 132351707.0, + "step": 3767 + }, + { + "epoch": 0.6997214484679666, + "grad_norm": 1.3889251947402954, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8747433423995972, + "num_tokens": 132390270.0, + "step": 3768 + }, + { + "epoch": 0.6999071494893222, + "grad_norm": 1.457321286201477, + "learning_rate": 1e-06, + "loss": 0.4336, + "mean_token_accuracy": 0.8555033802986145, + "num_tokens": 132430401.0, + "step": 3769 + }, + { + "epoch": 0.7000928505106778, + "grad_norm": 1.3929965496063232, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.874815821647644, + "num_tokens": 132468863.0, + "step": 3770 + }, + { + "epoch": 0.7002785515320334, + "grad_norm": 1.5507481098175049, + "learning_rate": 1e-06, + "loss": 0.411, + "mean_token_accuracy": 0.8628186583518982, + "num_tokens": 132502597.0, + "step": 3771 + }, + { + "epoch": 0.7004642525533891, + "grad_norm": 1.6376774311065674, + "learning_rate": 1e-06, + "loss": 0.3864, + "mean_token_accuracy": 0.8705576658248901, + "num_tokens": 132530395.0, + "step": 3772 + }, + { + "epoch": 0.7006499535747447, + "grad_norm": 1.4425361156463623, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8691045641899109, + "num_tokens": 132566391.0, + "step": 3773 + }, + { + "epoch": 0.7008356545961003, + "grad_norm": 1.3935115337371826, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8795140981674194, + "num_tokens": 132604023.0, + "step": 3774 + }, + { + "epoch": 0.7010213556174559, + "grad_norm": 1.575622797012329, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8719546794891357, + "num_tokens": 132634493.0, + "step": 3775 + }, + { + "epoch": 0.7012070566388116, + "grad_norm": 1.4607521295547485, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.872177004814148, + "num_tokens": 132671710.0, + "step": 3776 + }, + { + "epoch": 0.7013927576601672, + "grad_norm": 1.4959913492202759, + "learning_rate": 1e-06, + "loss": 0.4231, + "mean_token_accuracy": 0.8607534170150757, + "num_tokens": 132708494.0, + "step": 3777 + }, + { + "epoch": 0.7015784586815228, + "grad_norm": 1.3701624870300293, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8667423725128174, + "num_tokens": 132746991.0, + "step": 3778 + }, + { + "epoch": 0.7017641597028784, + "grad_norm": 1.7321975231170654, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8747094869613647, + "num_tokens": 132774944.0, + "step": 3779 + }, + { + "epoch": 0.7019498607242339, + "grad_norm": 1.6722800731658936, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8694274425506592, + "num_tokens": 132802339.0, + "step": 3780 + }, + { + "epoch": 0.7021355617455896, + "grad_norm": 1.3538076877593994, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8660314679145813, + "num_tokens": 132842060.0, + "step": 3781 + }, + { + "epoch": 0.7023212627669452, + "grad_norm": 1.4777109622955322, + "learning_rate": 1e-06, + "loss": 0.4768, + "mean_token_accuracy": 0.8424034118652344, + "num_tokens": 132880463.0, + "step": 3782 + }, + { + "epoch": 0.7025069637883008, + "grad_norm": 1.5746090412139893, + "learning_rate": 1e-06, + "loss": 0.4075, + "mean_token_accuracy": 0.8662357330322266, + "num_tokens": 132915253.0, + "step": 3783 + }, + { + "epoch": 0.7026926648096564, + "grad_norm": 1.4894860982894897, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.867484450340271, + "num_tokens": 132949876.0, + "step": 3784 + }, + { + "epoch": 0.702878365831012, + "grad_norm": 1.6088107824325562, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8670032024383545, + "num_tokens": 132981919.0, + "step": 3785 + }, + { + "epoch": 0.7030640668523677, + "grad_norm": 1.5138713121414185, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8730055093765259, + "num_tokens": 133013908.0, + "step": 3786 + }, + { + "epoch": 0.7032497678737233, + "grad_norm": 1.5542083978652954, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8779813051223755, + "num_tokens": 133045733.0, + "step": 3787 + }, + { + "epoch": 0.7034354688950789, + "grad_norm": 1.4562629461288452, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.881493330001831, + "num_tokens": 133081263.0, + "step": 3788 + }, + { + "epoch": 0.7036211699164345, + "grad_norm": 1.5139195919036865, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8660410046577454, + "num_tokens": 133122510.0, + "step": 3789 + }, + { + "epoch": 0.7038068709377902, + "grad_norm": 1.4560184478759766, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8655869364738464, + "num_tokens": 133157114.0, + "step": 3790 + }, + { + "epoch": 0.7039925719591458, + "grad_norm": 1.4838560819625854, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8634426593780518, + "num_tokens": 133192776.0, + "step": 3791 + }, + { + "epoch": 0.7041782729805014, + "grad_norm": 1.361987590789795, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8579250574111938, + "num_tokens": 133234275.0, + "step": 3792 + }, + { + "epoch": 0.704363974001857, + "grad_norm": 1.5272274017333984, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.863811194896698, + "num_tokens": 133270712.0, + "step": 3793 + }, + { + "epoch": 0.7045496750232126, + "grad_norm": 1.5285142660140991, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8606115579605103, + "num_tokens": 133305202.0, + "step": 3794 + }, + { + "epoch": 0.7047353760445683, + "grad_norm": 1.692376732826233, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8634047508239746, + "num_tokens": 133331942.0, + "step": 3795 + }, + { + "epoch": 0.7049210770659239, + "grad_norm": 1.581238865852356, + "learning_rate": 1e-06, + "loss": 0.4417, + "mean_token_accuracy": 0.8547376990318298, + "num_tokens": 133366336.0, + "step": 3796 + }, + { + "epoch": 0.7051067780872795, + "grad_norm": 1.6047625541687012, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.879555344581604, + "num_tokens": 133393568.0, + "step": 3797 + }, + { + "epoch": 0.7052924791086351, + "grad_norm": 1.4025059938430786, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8826014995574951, + "num_tokens": 133432306.0, + "step": 3798 + }, + { + "epoch": 0.7054781801299908, + "grad_norm": 1.494681715965271, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8668789267539978, + "num_tokens": 133467790.0, + "step": 3799 + }, + { + "epoch": 0.7056638811513464, + "grad_norm": 1.2782235145568848, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8778796195983887, + "num_tokens": 133507884.0, + "step": 3800 + }, + { + "epoch": 0.705849582172702, + "grad_norm": 1.4965509176254272, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.8645918369293213, + "num_tokens": 133542456.0, + "step": 3801 + }, + { + "epoch": 0.7060352831940576, + "grad_norm": 1.5148537158966064, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8649543523788452, + "num_tokens": 133580935.0, + "step": 3802 + }, + { + "epoch": 0.7062209842154132, + "grad_norm": 1.5462678670883179, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.8575243949890137, + "num_tokens": 133616703.0, + "step": 3803 + }, + { + "epoch": 0.7064066852367687, + "grad_norm": 1.5539889335632324, + "learning_rate": 1e-06, + "loss": 0.4243, + "mean_token_accuracy": 0.8611229658126831, + "num_tokens": 133651738.0, + "step": 3804 + }, + { + "epoch": 0.7065923862581244, + "grad_norm": 1.4287232160568237, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8700911402702332, + "num_tokens": 133685977.0, + "step": 3805 + }, + { + "epoch": 0.70677808727948, + "grad_norm": 1.3500571250915527, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8696606159210205, + "num_tokens": 133725784.0, + "step": 3806 + }, + { + "epoch": 0.7069637883008356, + "grad_norm": 1.722334623336792, + "learning_rate": 1e-06, + "loss": 0.4252, + "mean_token_accuracy": 0.8560072183609009, + "num_tokens": 133754089.0, + "step": 3807 + }, + { + "epoch": 0.7071494893221912, + "grad_norm": 1.6208925247192383, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8770612478256226, + "num_tokens": 133782378.0, + "step": 3808 + }, + { + "epoch": 0.7073351903435469, + "grad_norm": 1.5035008192062378, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8723472952842712, + "num_tokens": 133815892.0, + "step": 3809 + }, + { + "epoch": 0.7075208913649025, + "grad_norm": 1.4750925302505493, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8798144459724426, + "num_tokens": 133850631.0, + "step": 3810 + }, + { + "epoch": 0.7077065923862581, + "grad_norm": 1.537314772605896, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8731700778007507, + "num_tokens": 133881977.0, + "step": 3811 + }, + { + "epoch": 0.7078922934076137, + "grad_norm": 1.7742799520492554, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8619943261146545, + "num_tokens": 133910474.0, + "step": 3812 + }, + { + "epoch": 0.7080779944289693, + "grad_norm": 1.4870436191558838, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8723715543746948, + "num_tokens": 133943347.0, + "step": 3813 + }, + { + "epoch": 0.708263695450325, + "grad_norm": 1.557968258857727, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.860358715057373, + "num_tokens": 133977758.0, + "step": 3814 + }, + { + "epoch": 0.7084493964716806, + "grad_norm": 1.4394382238388062, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8648302555084229, + "num_tokens": 134017061.0, + "step": 3815 + }, + { + "epoch": 0.7086350974930362, + "grad_norm": 1.5358517169952393, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8602652549743652, + "num_tokens": 134054551.0, + "step": 3816 + }, + { + "epoch": 0.7088207985143918, + "grad_norm": 1.7484354972839355, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8813542723655701, + "num_tokens": 134087891.0, + "step": 3817 + }, + { + "epoch": 0.7090064995357475, + "grad_norm": 1.3908195495605469, + "learning_rate": 1e-06, + "loss": 0.4151, + "mean_token_accuracy": 0.8605867624282837, + "num_tokens": 134125993.0, + "step": 3818 + }, + { + "epoch": 0.7091922005571031, + "grad_norm": 1.5009191036224365, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8605002164840698, + "num_tokens": 134160737.0, + "step": 3819 + }, + { + "epoch": 0.7093779015784587, + "grad_norm": 1.4415109157562256, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8626144528388977, + "num_tokens": 134202225.0, + "step": 3820 + }, + { + "epoch": 0.7095636025998143, + "grad_norm": 1.455657720565796, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8754655122756958, + "num_tokens": 134235642.0, + "step": 3821 + }, + { + "epoch": 0.70974930362117, + "grad_norm": 1.5950623750686646, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.8654167056083679, + "num_tokens": 134267242.0, + "step": 3822 + }, + { + "epoch": 0.7099350046425256, + "grad_norm": 1.4194951057434082, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8542160391807556, + "num_tokens": 134302847.0, + "step": 3823 + }, + { + "epoch": 0.7101207056638812, + "grad_norm": 1.427626609802246, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8757723569869995, + "num_tokens": 134343551.0, + "step": 3824 + }, + { + "epoch": 0.7103064066852368, + "grad_norm": 1.4052298069000244, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8722927570343018, + "num_tokens": 134378463.0, + "step": 3825 + }, + { + "epoch": 0.7104921077065924, + "grad_norm": 1.744377613067627, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.865544319152832, + "num_tokens": 134404498.0, + "step": 3826 + }, + { + "epoch": 0.7106778087279481, + "grad_norm": 1.5828144550323486, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.8609398603439331, + "num_tokens": 134436601.0, + "step": 3827 + }, + { + "epoch": 0.7108635097493036, + "grad_norm": 1.617460012435913, + "learning_rate": 1e-06, + "loss": 0.4288, + "mean_token_accuracy": 0.8594568967819214, + "num_tokens": 134469516.0, + "step": 3828 + }, + { + "epoch": 0.7110492107706592, + "grad_norm": 1.6045385599136353, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8702163100242615, + "num_tokens": 134501978.0, + "step": 3829 + }, + { + "epoch": 0.7112349117920148, + "grad_norm": 1.5524650812149048, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8623298406600952, + "num_tokens": 134537552.0, + "step": 3830 + }, + { + "epoch": 0.7114206128133704, + "grad_norm": 1.482947826385498, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8667672872543335, + "num_tokens": 134569653.0, + "step": 3831 + }, + { + "epoch": 0.7116063138347261, + "grad_norm": 1.517768383026123, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8654186725616455, + "num_tokens": 134600853.0, + "step": 3832 + }, + { + "epoch": 0.7117920148560817, + "grad_norm": 1.2882442474365234, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8714131712913513, + "num_tokens": 134644030.0, + "step": 3833 + }, + { + "epoch": 0.7119777158774373, + "grad_norm": 1.6284452676773071, + "learning_rate": 1e-06, + "loss": 0.4357, + "mean_token_accuracy": 0.8537391424179077, + "num_tokens": 134674983.0, + "step": 3834 + }, + { + "epoch": 0.7121634168987929, + "grad_norm": 1.5666183233261108, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8642847537994385, + "num_tokens": 134705509.0, + "step": 3835 + }, + { + "epoch": 0.7123491179201485, + "grad_norm": 1.4085811376571655, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8764443397521973, + "num_tokens": 134741937.0, + "step": 3836 + }, + { + "epoch": 0.7125348189415042, + "grad_norm": 1.5358798503875732, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8721710443496704, + "num_tokens": 134775314.0, + "step": 3837 + }, + { + "epoch": 0.7127205199628598, + "grad_norm": 1.5098615884780884, + "learning_rate": 1e-06, + "loss": 0.4459, + "mean_token_accuracy": 0.8504490852355957, + "num_tokens": 134809018.0, + "step": 3838 + }, + { + "epoch": 0.7129062209842154, + "grad_norm": 1.581382393836975, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.8623500466346741, + "num_tokens": 134842713.0, + "step": 3839 + }, + { + "epoch": 0.713091922005571, + "grad_norm": 1.6263253688812256, + "learning_rate": 1e-06, + "loss": 0.4483, + "mean_token_accuracy": 0.8500321507453918, + "num_tokens": 134875535.0, + "step": 3840 + }, + { + "epoch": 0.7132776230269267, + "grad_norm": 1.3519140481948853, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8759486079216003, + "num_tokens": 134916226.0, + "step": 3841 + }, + { + "epoch": 0.7134633240482823, + "grad_norm": 1.6003081798553467, + "learning_rate": 1e-06, + "loss": 0.4948, + "mean_token_accuracy": 0.8348633646965027, + "num_tokens": 134946321.0, + "step": 3842 + }, + { + "epoch": 0.7136490250696379, + "grad_norm": 1.506001353263855, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8661918640136719, + "num_tokens": 134980596.0, + "step": 3843 + }, + { + "epoch": 0.7138347260909935, + "grad_norm": 1.5211434364318848, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8648523688316345, + "num_tokens": 135012425.0, + "step": 3844 + }, + { + "epoch": 0.7140204271123491, + "grad_norm": 1.4657444953918457, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8749302625656128, + "num_tokens": 135046345.0, + "step": 3845 + }, + { + "epoch": 0.7142061281337048, + "grad_norm": 1.4258453845977783, + "learning_rate": 1e-06, + "loss": 0.4113, + "mean_token_accuracy": 0.8619562387466431, + "num_tokens": 135084679.0, + "step": 3846 + }, + { + "epoch": 0.7143918291550604, + "grad_norm": 1.3997060060501099, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8753824234008789, + "num_tokens": 135119285.0, + "step": 3847 + }, + { + "epoch": 0.714577530176416, + "grad_norm": 1.5245455503463745, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.874697744846344, + "num_tokens": 135150595.0, + "step": 3848 + }, + { + "epoch": 0.7147632311977716, + "grad_norm": 1.420576572418213, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8666455745697021, + "num_tokens": 135186016.0, + "step": 3849 + }, + { + "epoch": 0.7149489322191273, + "grad_norm": 1.5928590297698975, + "learning_rate": 1e-06, + "loss": 0.3987, + "mean_token_accuracy": 0.865896463394165, + "num_tokens": 135220285.0, + "step": 3850 + }, + { + "epoch": 0.7151346332404829, + "grad_norm": 1.6040410995483398, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8682854175567627, + "num_tokens": 135250926.0, + "step": 3851 + }, + { + "epoch": 0.7153203342618384, + "grad_norm": 1.5030878782272339, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8697625398635864, + "num_tokens": 135281464.0, + "step": 3852 + }, + { + "epoch": 0.715506035283194, + "grad_norm": 1.5504683256149292, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8647266626358032, + "num_tokens": 135315531.0, + "step": 3853 + }, + { + "epoch": 0.7156917363045496, + "grad_norm": 1.4662630558013916, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8785541653633118, + "num_tokens": 135352817.0, + "step": 3854 + }, + { + "epoch": 0.7158774373259053, + "grad_norm": 1.5433986186981201, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8685632944107056, + "num_tokens": 135384265.0, + "step": 3855 + }, + { + "epoch": 0.7160631383472609, + "grad_norm": 1.4935641288757324, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8734679222106934, + "num_tokens": 135417993.0, + "step": 3856 + }, + { + "epoch": 0.7162488393686165, + "grad_norm": 1.494093656539917, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.8683609962463379, + "num_tokens": 135450656.0, + "step": 3857 + }, + { + "epoch": 0.7164345403899721, + "grad_norm": 1.4333956241607666, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8779165744781494, + "num_tokens": 135486553.0, + "step": 3858 + }, + { + "epoch": 0.7166202414113277, + "grad_norm": 1.5697402954101562, + "learning_rate": 1e-06, + "loss": 0.4568, + "mean_token_accuracy": 0.844976544380188, + "num_tokens": 135525090.0, + "step": 3859 + }, + { + "epoch": 0.7168059424326834, + "grad_norm": 1.3949546813964844, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8650090098381042, + "num_tokens": 135566757.0, + "step": 3860 + }, + { + "epoch": 0.716991643454039, + "grad_norm": 1.476684808731079, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8609600067138672, + "num_tokens": 135602845.0, + "step": 3861 + }, + { + "epoch": 0.7171773444753946, + "grad_norm": 1.4899483919143677, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8746464252471924, + "num_tokens": 135634104.0, + "step": 3862 + }, + { + "epoch": 0.7173630454967502, + "grad_norm": 1.5280652046203613, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8717108964920044, + "num_tokens": 135668013.0, + "step": 3863 + }, + { + "epoch": 0.7175487465181059, + "grad_norm": 1.4673248529434204, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8671373128890991, + "num_tokens": 135700803.0, + "step": 3864 + }, + { + "epoch": 0.7177344475394615, + "grad_norm": 1.5038232803344727, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.866202712059021, + "num_tokens": 135735755.0, + "step": 3865 + }, + { + "epoch": 0.7179201485608171, + "grad_norm": 1.3875365257263184, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8596001863479614, + "num_tokens": 135775760.0, + "step": 3866 + }, + { + "epoch": 0.7181058495821727, + "grad_norm": 1.604933738708496, + "learning_rate": 1e-06, + "loss": 0.4457, + "mean_token_accuracy": 0.8572461605072021, + "num_tokens": 135807702.0, + "step": 3867 + }, + { + "epoch": 0.7182915506035283, + "grad_norm": 1.5666059255599976, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.8612861633300781, + "num_tokens": 135837959.0, + "step": 3868 + }, + { + "epoch": 0.718477251624884, + "grad_norm": 1.5593842267990112, + "learning_rate": 1e-06, + "loss": 0.4399, + "mean_token_accuracy": 0.8514494299888611, + "num_tokens": 135871050.0, + "step": 3869 + }, + { + "epoch": 0.7186629526462396, + "grad_norm": 1.4758301973342896, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8774250745773315, + "num_tokens": 135903446.0, + "step": 3870 + }, + { + "epoch": 0.7188486536675952, + "grad_norm": 1.4823659658432007, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8820937275886536, + "num_tokens": 135934047.0, + "step": 3871 + }, + { + "epoch": 0.7190343546889508, + "grad_norm": 1.4607487916946411, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8732775449752808, + "num_tokens": 135967624.0, + "step": 3872 + }, + { + "epoch": 0.7192200557103064, + "grad_norm": 1.5810904502868652, + "learning_rate": 1e-06, + "loss": 0.4096, + "mean_token_accuracy": 0.8643584251403809, + "num_tokens": 135998830.0, + "step": 3873 + }, + { + "epoch": 0.7194057567316621, + "grad_norm": 1.4494516849517822, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8701114058494568, + "num_tokens": 136034906.0, + "step": 3874 + }, + { + "epoch": 0.7195914577530177, + "grad_norm": 1.3385769128799438, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8664237260818481, + "num_tokens": 136075250.0, + "step": 3875 + }, + { + "epoch": 0.7197771587743732, + "grad_norm": 1.4823001623153687, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8821091651916504, + "num_tokens": 136109102.0, + "step": 3876 + }, + { + "epoch": 0.7199628597957288, + "grad_norm": 1.5447473526000977, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8679956197738647, + "num_tokens": 136142185.0, + "step": 3877 + }, + { + "epoch": 0.7201485608170844, + "grad_norm": 1.5678097009658813, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.8569774031639099, + "num_tokens": 136173878.0, + "step": 3878 + }, + { + "epoch": 0.7203342618384401, + "grad_norm": 1.423660397529602, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.872641384601593, + "num_tokens": 136210043.0, + "step": 3879 + }, + { + "epoch": 0.7205199628597957, + "grad_norm": 1.4478460550308228, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8750026226043701, + "num_tokens": 136242574.0, + "step": 3880 + }, + { + "epoch": 0.7207056638811513, + "grad_norm": 1.6004878282546997, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8603547811508179, + "num_tokens": 136273460.0, + "step": 3881 + }, + { + "epoch": 0.7208913649025069, + "grad_norm": 1.4794094562530518, + "learning_rate": 1e-06, + "loss": 0.4372, + "mean_token_accuracy": 0.8542205095291138, + "num_tokens": 136312879.0, + "step": 3882 + }, + { + "epoch": 0.7210770659238626, + "grad_norm": 1.6978392601013184, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8768853545188904, + "num_tokens": 136339699.0, + "step": 3883 + }, + { + "epoch": 0.7212627669452182, + "grad_norm": 1.3174759149551392, + "learning_rate": 1e-06, + "loss": 0.4236, + "mean_token_accuracy": 0.8565586805343628, + "num_tokens": 136386348.0, + "step": 3884 + }, + { + "epoch": 0.7214484679665738, + "grad_norm": 1.5969734191894531, + "learning_rate": 1e-06, + "loss": 0.4814, + "mean_token_accuracy": 0.8394978046417236, + "num_tokens": 136422552.0, + "step": 3885 + }, + { + "epoch": 0.7216341689879294, + "grad_norm": 1.476819634437561, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8637270927429199, + "num_tokens": 136459043.0, + "step": 3886 + }, + { + "epoch": 0.721819870009285, + "grad_norm": 1.499699592590332, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8645059466362, + "num_tokens": 136491509.0, + "step": 3887 + }, + { + "epoch": 0.7220055710306407, + "grad_norm": 1.4364712238311768, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.8638747334480286, + "num_tokens": 136526361.0, + "step": 3888 + }, + { + "epoch": 0.7221912720519963, + "grad_norm": 1.413198709487915, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8748520016670227, + "num_tokens": 136563047.0, + "step": 3889 + }, + { + "epoch": 0.7223769730733519, + "grad_norm": 1.5165067911148071, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.883773922920227, + "num_tokens": 136591859.0, + "step": 3890 + }, + { + "epoch": 0.7225626740947075, + "grad_norm": 1.373978614807129, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.873531699180603, + "num_tokens": 136630297.0, + "step": 3891 + }, + { + "epoch": 0.7227483751160632, + "grad_norm": 1.4197849035263062, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.865969181060791, + "num_tokens": 136665171.0, + "step": 3892 + }, + { + "epoch": 0.7229340761374188, + "grad_norm": 1.4880610704421997, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8614929914474487, + "num_tokens": 136703223.0, + "step": 3893 + }, + { + "epoch": 0.7231197771587744, + "grad_norm": 1.505262017250061, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8650166988372803, + "num_tokens": 136741314.0, + "step": 3894 + }, + { + "epoch": 0.72330547818013, + "grad_norm": 1.5270296335220337, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8728243708610535, + "num_tokens": 136777182.0, + "step": 3895 + }, + { + "epoch": 0.7234911792014856, + "grad_norm": 1.472633957862854, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.8609429001808167, + "num_tokens": 136814285.0, + "step": 3896 + }, + { + "epoch": 0.7236768802228413, + "grad_norm": 1.3734854459762573, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.8649901151657104, + "num_tokens": 136857653.0, + "step": 3897 + }, + { + "epoch": 0.7238625812441969, + "grad_norm": 1.4874118566513062, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8628761768341064, + "num_tokens": 136894227.0, + "step": 3898 + }, + { + "epoch": 0.7240482822655525, + "grad_norm": 1.6652601957321167, + "learning_rate": 1e-06, + "loss": 0.4362, + "mean_token_accuracy": 0.8560175895690918, + "num_tokens": 136923811.0, + "step": 3899 + }, + { + "epoch": 0.724233983286908, + "grad_norm": 1.541709303855896, + "learning_rate": 1e-06, + "loss": 0.426, + "mean_token_accuracy": 0.8655896782875061, + "num_tokens": 136962224.0, + "step": 3900 + }, + { + "epoch": 0.7244196843082636, + "grad_norm": 1.5751936435699463, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8584206700325012, + "num_tokens": 136994110.0, + "step": 3901 + }, + { + "epoch": 0.7246053853296193, + "grad_norm": 1.4272263050079346, + "learning_rate": 1e-06, + "loss": 0.4546, + "mean_token_accuracy": 0.8517013192176819, + "num_tokens": 137032271.0, + "step": 3902 + }, + { + "epoch": 0.7247910863509749, + "grad_norm": 1.5510753393173218, + "learning_rate": 1e-06, + "loss": 0.4391, + "mean_token_accuracy": 0.8526506423950195, + "num_tokens": 137067746.0, + "step": 3903 + }, + { + "epoch": 0.7249767873723305, + "grad_norm": 1.412095546722412, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.861789882183075, + "num_tokens": 137106691.0, + "step": 3904 + }, + { + "epoch": 0.7251624883936861, + "grad_norm": 1.5393460988998413, + "learning_rate": 1e-06, + "loss": 0.4375, + "mean_token_accuracy": 0.8519428968429565, + "num_tokens": 137139066.0, + "step": 3905 + }, + { + "epoch": 0.7253481894150418, + "grad_norm": 1.3994150161743164, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8700485229492188, + "num_tokens": 137177793.0, + "step": 3906 + }, + { + "epoch": 0.7255338904363974, + "grad_norm": 1.3369803428649902, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8808174729347229, + "num_tokens": 137216003.0, + "step": 3907 + }, + { + "epoch": 0.725719591457753, + "grad_norm": 1.3032225370407104, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8752363920211792, + "num_tokens": 137260650.0, + "step": 3908 + }, + { + "epoch": 0.7259052924791086, + "grad_norm": 1.4523752927780151, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8694571256637573, + "num_tokens": 137299042.0, + "step": 3909 + }, + { + "epoch": 0.7260909935004642, + "grad_norm": 1.4506025314331055, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8636667132377625, + "num_tokens": 137337355.0, + "step": 3910 + }, + { + "epoch": 0.7262766945218199, + "grad_norm": 1.5917950868606567, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8814862966537476, + "num_tokens": 137364668.0, + "step": 3911 + }, + { + "epoch": 0.7264623955431755, + "grad_norm": 1.6422215700149536, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8702133297920227, + "num_tokens": 137392812.0, + "step": 3912 + }, + { + "epoch": 0.7266480965645311, + "grad_norm": 1.3853366374969482, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8624571561813354, + "num_tokens": 137432075.0, + "step": 3913 + }, + { + "epoch": 0.7268337975858867, + "grad_norm": 1.430823564529419, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8642017841339111, + "num_tokens": 137467264.0, + "step": 3914 + }, + { + "epoch": 0.7270194986072424, + "grad_norm": 1.4403899908065796, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8775723576545715, + "num_tokens": 137505227.0, + "step": 3915 + }, + { + "epoch": 0.727205199628598, + "grad_norm": 1.4480979442596436, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8655813336372375, + "num_tokens": 137539252.0, + "step": 3916 + }, + { + "epoch": 0.7273909006499536, + "grad_norm": 1.545469880104065, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8595898151397705, + "num_tokens": 137573488.0, + "step": 3917 + }, + { + "epoch": 0.7275766016713092, + "grad_norm": 1.6348960399627686, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8675207495689392, + "num_tokens": 137603725.0, + "step": 3918 + }, + { + "epoch": 0.7277623026926648, + "grad_norm": 1.4124634265899658, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8783499002456665, + "num_tokens": 137639669.0, + "step": 3919 + }, + { + "epoch": 0.7279480037140205, + "grad_norm": 1.4802875518798828, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8774101138114929, + "num_tokens": 137671647.0, + "step": 3920 + }, + { + "epoch": 0.7281337047353761, + "grad_norm": 1.3572421073913574, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8606544137001038, + "num_tokens": 137718167.0, + "step": 3921 + }, + { + "epoch": 0.7283194057567317, + "grad_norm": 1.6061408519744873, + "learning_rate": 1e-06, + "loss": 0.4615, + "mean_token_accuracy": 0.8461273312568665, + "num_tokens": 137751866.0, + "step": 3922 + }, + { + "epoch": 0.7285051067780873, + "grad_norm": 1.3444242477416992, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8788166046142578, + "num_tokens": 137792750.0, + "step": 3923 + }, + { + "epoch": 0.728690807799443, + "grad_norm": 1.4624261856079102, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8668270111083984, + "num_tokens": 137827459.0, + "step": 3924 + }, + { + "epoch": 0.7288765088207985, + "grad_norm": 1.4755502939224243, + "learning_rate": 1e-06, + "loss": 0.4107, + "mean_token_accuracy": 0.8605727553367615, + "num_tokens": 137865474.0, + "step": 3925 + }, + { + "epoch": 0.7290622098421541, + "grad_norm": 1.4387707710266113, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8783211708068848, + "num_tokens": 137901148.0, + "step": 3926 + }, + { + "epoch": 0.7292479108635097, + "grad_norm": 1.7439398765563965, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8632806539535522, + "num_tokens": 137940969.0, + "step": 3927 + }, + { + "epoch": 0.7294336118848653, + "grad_norm": 1.447418212890625, + "learning_rate": 1e-06, + "loss": 0.4359, + "mean_token_accuracy": 0.8559394478797913, + "num_tokens": 137980145.0, + "step": 3928 + }, + { + "epoch": 0.729619312906221, + "grad_norm": 1.4463369846343994, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8734780550003052, + "num_tokens": 138013681.0, + "step": 3929 + }, + { + "epoch": 0.7298050139275766, + "grad_norm": 1.6808239221572876, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8649353981018066, + "num_tokens": 138039415.0, + "step": 3930 + }, + { + "epoch": 0.7299907149489322, + "grad_norm": 1.389749526977539, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8736277222633362, + "num_tokens": 138076067.0, + "step": 3931 + }, + { + "epoch": 0.7301764159702878, + "grad_norm": 1.337321400642395, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8716802000999451, + "num_tokens": 138113699.0, + "step": 3932 + }, + { + "epoch": 0.7303621169916434, + "grad_norm": 1.3498204946517944, + "learning_rate": 1e-06, + "loss": 0.4289, + "mean_token_accuracy": 0.847374439239502, + "num_tokens": 138154720.0, + "step": 3933 + }, + { + "epoch": 0.7305478180129991, + "grad_norm": 1.5530552864074707, + "learning_rate": 1e-06, + "loss": 0.4894, + "mean_token_accuracy": 0.8370401859283447, + "num_tokens": 138190618.0, + "step": 3934 + }, + { + "epoch": 0.7307335190343547, + "grad_norm": 1.3882640600204468, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8721013069152832, + "num_tokens": 138229659.0, + "step": 3935 + }, + { + "epoch": 0.7309192200557103, + "grad_norm": 1.5362696647644043, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8699935674667358, + "num_tokens": 138260828.0, + "step": 3936 + }, + { + "epoch": 0.7311049210770659, + "grad_norm": 1.5890601873397827, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8732076287269592, + "num_tokens": 138290775.0, + "step": 3937 + }, + { + "epoch": 0.7312906220984215, + "grad_norm": 1.4565653800964355, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8710335493087769, + "num_tokens": 138321779.0, + "step": 3938 + }, + { + "epoch": 0.7314763231197772, + "grad_norm": 1.5434203147888184, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8701229691505432, + "num_tokens": 138353745.0, + "step": 3939 + }, + { + "epoch": 0.7316620241411328, + "grad_norm": 1.4929139614105225, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.863792359828949, + "num_tokens": 138390092.0, + "step": 3940 + }, + { + "epoch": 0.7318477251624884, + "grad_norm": 1.3596365451812744, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8775947093963623, + "num_tokens": 138429072.0, + "step": 3941 + }, + { + "epoch": 0.732033426183844, + "grad_norm": 1.645926594734192, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8690534830093384, + "num_tokens": 138461078.0, + "step": 3942 + }, + { + "epoch": 0.7322191272051997, + "grad_norm": 1.4702104330062866, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8726372718811035, + "num_tokens": 138490714.0, + "step": 3943 + }, + { + "epoch": 0.7324048282265553, + "grad_norm": 1.4497511386871338, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8683154582977295, + "num_tokens": 138525383.0, + "step": 3944 + }, + { + "epoch": 0.7325905292479109, + "grad_norm": 1.3875086307525635, + "learning_rate": 1e-06, + "loss": 0.4369, + "mean_token_accuracy": 0.8568674325942993, + "num_tokens": 138567601.0, + "step": 3945 + }, + { + "epoch": 0.7327762302692665, + "grad_norm": 1.6225422620773315, + "learning_rate": 1e-06, + "loss": 0.4231, + "mean_token_accuracy": 0.8622781038284302, + "num_tokens": 138597975.0, + "step": 3946 + }, + { + "epoch": 0.7329619312906221, + "grad_norm": 1.4041988849639893, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8670946955680847, + "num_tokens": 138632743.0, + "step": 3947 + }, + { + "epoch": 0.7331476323119778, + "grad_norm": 1.3315479755401611, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8704754114151001, + "num_tokens": 138676624.0, + "step": 3948 + }, + { + "epoch": 0.7333333333333333, + "grad_norm": 1.3784537315368652, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8731364011764526, + "num_tokens": 138713128.0, + "step": 3949 + }, + { + "epoch": 0.7335190343546889, + "grad_norm": 1.4534589052200317, + "learning_rate": 1e-06, + "loss": 0.4275, + "mean_token_accuracy": 0.8611701726913452, + "num_tokens": 138750263.0, + "step": 3950 + }, + { + "epoch": 0.7337047353760445, + "grad_norm": 1.5840903520584106, + "learning_rate": 1e-06, + "loss": 0.431, + "mean_token_accuracy": 0.8531301021575928, + "num_tokens": 138784272.0, + "step": 3951 + }, + { + "epoch": 0.7338904363974001, + "grad_norm": 1.558203935623169, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8731005787849426, + "num_tokens": 138818575.0, + "step": 3952 + }, + { + "epoch": 0.7340761374187558, + "grad_norm": 1.3917254209518433, + "learning_rate": 1e-06, + "loss": 0.4097, + "mean_token_accuracy": 0.8638235926628113, + "num_tokens": 138856751.0, + "step": 3953 + }, + { + "epoch": 0.7342618384401114, + "grad_norm": 1.5487836599349976, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8707695007324219, + "num_tokens": 138887877.0, + "step": 3954 + }, + { + "epoch": 0.734447539461467, + "grad_norm": 1.5034667253494263, + "learning_rate": 1e-06, + "loss": 0.4061, + "mean_token_accuracy": 0.8628677129745483, + "num_tokens": 138924340.0, + "step": 3955 + }, + { + "epoch": 0.7346332404828226, + "grad_norm": 1.4322164058685303, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8699337244033813, + "num_tokens": 138961842.0, + "step": 3956 + }, + { + "epoch": 0.7348189415041783, + "grad_norm": 1.3675110340118408, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8673449158668518, + "num_tokens": 139002277.0, + "step": 3957 + }, + { + "epoch": 0.7350046425255339, + "grad_norm": 1.4499478340148926, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8603479862213135, + "num_tokens": 139039666.0, + "step": 3958 + }, + { + "epoch": 0.7351903435468895, + "grad_norm": 1.4822862148284912, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8753523230552673, + "num_tokens": 139072694.0, + "step": 3959 + }, + { + "epoch": 0.7353760445682451, + "grad_norm": 1.6857563257217407, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.858048677444458, + "num_tokens": 139102294.0, + "step": 3960 + }, + { + "epoch": 0.7355617455896007, + "grad_norm": 1.4605499505996704, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8734868168830872, + "num_tokens": 139137063.0, + "step": 3961 + }, + { + "epoch": 0.7357474466109564, + "grad_norm": 1.4257879257202148, + "learning_rate": 1e-06, + "loss": 0.4453, + "mean_token_accuracy": 0.8466150164604187, + "num_tokens": 139182031.0, + "step": 3962 + }, + { + "epoch": 0.735933147632312, + "grad_norm": 1.4546544551849365, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.857035756111145, + "num_tokens": 139218386.0, + "step": 3963 + }, + { + "epoch": 0.7361188486536676, + "grad_norm": 1.4838662147521973, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8638260364532471, + "num_tokens": 139253596.0, + "step": 3964 + }, + { + "epoch": 0.7363045496750232, + "grad_norm": 1.4119813442230225, + "learning_rate": 1e-06, + "loss": 0.4318, + "mean_token_accuracy": 0.8550409078598022, + "num_tokens": 139291662.0, + "step": 3965 + }, + { + "epoch": 0.7364902506963789, + "grad_norm": 1.4835197925567627, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8752501010894775, + "num_tokens": 139323560.0, + "step": 3966 + }, + { + "epoch": 0.7366759517177345, + "grad_norm": 1.520742416381836, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8646643757820129, + "num_tokens": 139354367.0, + "step": 3967 + }, + { + "epoch": 0.7368616527390901, + "grad_norm": 1.4637649059295654, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8638601303100586, + "num_tokens": 139386229.0, + "step": 3968 + }, + { + "epoch": 0.7370473537604457, + "grad_norm": 1.469151496887207, + "learning_rate": 1e-06, + "loss": 0.4042, + "mean_token_accuracy": 0.8611970543861389, + "num_tokens": 139423613.0, + "step": 3969 + }, + { + "epoch": 0.7372330547818013, + "grad_norm": 1.6235041618347168, + "learning_rate": 1e-06, + "loss": 0.413, + "mean_token_accuracy": 0.8606391549110413, + "num_tokens": 139454691.0, + "step": 3970 + }, + { + "epoch": 0.737418755803157, + "grad_norm": 1.618180274963379, + "learning_rate": 1e-06, + "loss": 0.4544, + "mean_token_accuracy": 0.8487458229064941, + "num_tokens": 139484530.0, + "step": 3971 + }, + { + "epoch": 0.7376044568245126, + "grad_norm": 1.3145794868469238, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8728348016738892, + "num_tokens": 139525627.0, + "step": 3972 + }, + { + "epoch": 0.7377901578458681, + "grad_norm": 1.5595550537109375, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8751474022865295, + "num_tokens": 139556118.0, + "step": 3973 + }, + { + "epoch": 0.7379758588672237, + "grad_norm": 1.52383553981781, + "learning_rate": 1e-06, + "loss": 0.4488, + "mean_token_accuracy": 0.8557536602020264, + "num_tokens": 139588227.0, + "step": 3974 + }, + { + "epoch": 0.7381615598885793, + "grad_norm": 1.4977585077285767, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8832539916038513, + "num_tokens": 139621401.0, + "step": 3975 + }, + { + "epoch": 0.738347260909935, + "grad_norm": 1.423345923423767, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8693244457244873, + "num_tokens": 139657561.0, + "step": 3976 + }, + { + "epoch": 0.7385329619312906, + "grad_norm": 1.5130460262298584, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8688253164291382, + "num_tokens": 139691492.0, + "step": 3977 + }, + { + "epoch": 0.7387186629526462, + "grad_norm": 1.4339282512664795, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8842450380325317, + "num_tokens": 139728775.0, + "step": 3978 + }, + { + "epoch": 0.7389043639740018, + "grad_norm": 1.589447021484375, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.8619915246963501, + "num_tokens": 139759919.0, + "step": 3979 + }, + { + "epoch": 0.7390900649953575, + "grad_norm": 1.4007375240325928, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8654740452766418, + "num_tokens": 139796821.0, + "step": 3980 + }, + { + "epoch": 0.7392757660167131, + "grad_norm": 1.2837954759597778, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8745406866073608, + "num_tokens": 139839989.0, + "step": 3981 + }, + { + "epoch": 0.7394614670380687, + "grad_norm": 1.3308420181274414, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8668801784515381, + "num_tokens": 139879702.0, + "step": 3982 + }, + { + "epoch": 0.7396471680594243, + "grad_norm": 1.3755462169647217, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8638036251068115, + "num_tokens": 139919171.0, + "step": 3983 + }, + { + "epoch": 0.7398328690807799, + "grad_norm": 1.4370067119598389, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.8661304116249084, + "num_tokens": 139951484.0, + "step": 3984 + }, + { + "epoch": 0.7400185701021356, + "grad_norm": 1.501537561416626, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8726012110710144, + "num_tokens": 139981821.0, + "step": 3985 + }, + { + "epoch": 0.7402042711234912, + "grad_norm": 1.4698476791381836, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8593464493751526, + "num_tokens": 140018336.0, + "step": 3986 + }, + { + "epoch": 0.7403899721448468, + "grad_norm": 1.481258749961853, + "learning_rate": 1e-06, + "loss": 0.4054, + "mean_token_accuracy": 0.8662081956863403, + "num_tokens": 140053899.0, + "step": 3987 + }, + { + "epoch": 0.7405756731662024, + "grad_norm": 1.5530635118484497, + "learning_rate": 1e-06, + "loss": 0.4468, + "mean_token_accuracy": 0.8504177927970886, + "num_tokens": 140090228.0, + "step": 3988 + }, + { + "epoch": 0.740761374187558, + "grad_norm": 1.3915873765945435, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8577654361724854, + "num_tokens": 140130851.0, + "step": 3989 + }, + { + "epoch": 0.7409470752089137, + "grad_norm": 1.4361339807510376, + "learning_rate": 1e-06, + "loss": 0.4137, + "mean_token_accuracy": 0.8601821660995483, + "num_tokens": 140171694.0, + "step": 3990 + }, + { + "epoch": 0.7411327762302693, + "grad_norm": 1.568824291229248, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8651225566864014, + "num_tokens": 140203043.0, + "step": 3991 + }, + { + "epoch": 0.7413184772516249, + "grad_norm": 1.400780200958252, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8676513433456421, + "num_tokens": 140243552.0, + "step": 3992 + }, + { + "epoch": 0.7415041782729805, + "grad_norm": 1.4744782447814941, + "learning_rate": 1e-06, + "loss": 0.444, + "mean_token_accuracy": 0.8527308702468872, + "num_tokens": 140284018.0, + "step": 3993 + }, + { + "epoch": 0.7416898792943362, + "grad_norm": 1.4856677055358887, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8721513152122498, + "num_tokens": 140316818.0, + "step": 3994 + }, + { + "epoch": 0.7418755803156918, + "grad_norm": 1.6107226610183716, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8746885061264038, + "num_tokens": 140345592.0, + "step": 3995 + }, + { + "epoch": 0.7420612813370474, + "grad_norm": 1.5348727703094482, + "learning_rate": 1e-06, + "loss": 0.4324, + "mean_token_accuracy": 0.8561911582946777, + "num_tokens": 140379810.0, + "step": 3996 + }, + { + "epoch": 0.7422469823584029, + "grad_norm": 1.3502726554870605, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.86814284324646, + "num_tokens": 140419343.0, + "step": 3997 + }, + { + "epoch": 0.7424326833797585, + "grad_norm": 1.2454912662506104, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8865362405776978, + "num_tokens": 140459318.0, + "step": 3998 + }, + { + "epoch": 0.7426183844011142, + "grad_norm": 1.509879231452942, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8693286776542664, + "num_tokens": 140499723.0, + "step": 3999 + }, + { + "epoch": 0.7428040854224698, + "grad_norm": 1.3971409797668457, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8739365339279175, + "num_tokens": 140545445.0, + "step": 4000 + }, + { + "epoch": 0.7429897864438254, + "grad_norm": 1.454973578453064, + "learning_rate": 1e-06, + "loss": 0.4271, + "mean_token_accuracy": 0.8539445996284485, + "num_tokens": 140583824.0, + "step": 4001 + }, + { + "epoch": 0.743175487465181, + "grad_norm": 1.428399920463562, + "learning_rate": 1e-06, + "loss": 0.4427, + "mean_token_accuracy": 0.8522265553474426, + "num_tokens": 140621596.0, + "step": 4002 + }, + { + "epoch": 0.7433611884865367, + "grad_norm": 1.4388662576675415, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8701059818267822, + "num_tokens": 140655780.0, + "step": 4003 + }, + { + "epoch": 0.7435468895078923, + "grad_norm": 1.3869484663009644, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8594268560409546, + "num_tokens": 140692908.0, + "step": 4004 + }, + { + "epoch": 0.7437325905292479, + "grad_norm": 1.4729251861572266, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.876118540763855, + "num_tokens": 140727897.0, + "step": 4005 + }, + { + "epoch": 0.7439182915506035, + "grad_norm": 1.4518170356750488, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8615155220031738, + "num_tokens": 140764879.0, + "step": 4006 + }, + { + "epoch": 0.7441039925719591, + "grad_norm": 1.4158216714859009, + "learning_rate": 1e-06, + "loss": 0.4378, + "mean_token_accuracy": 0.8551397323608398, + "num_tokens": 140807809.0, + "step": 4007 + }, + { + "epoch": 0.7442896935933148, + "grad_norm": 1.689464807510376, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8782101273536682, + "num_tokens": 140837326.0, + "step": 4008 + }, + { + "epoch": 0.7444753946146704, + "grad_norm": 1.369179368019104, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.871493399143219, + "num_tokens": 140875291.0, + "step": 4009 + }, + { + "epoch": 0.744661095636026, + "grad_norm": 1.437311053276062, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8817027807235718, + "num_tokens": 140908825.0, + "step": 4010 + }, + { + "epoch": 0.7448467966573816, + "grad_norm": 1.4502110481262207, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.884742259979248, + "num_tokens": 140940294.0, + "step": 4011 + }, + { + "epoch": 0.7450324976787372, + "grad_norm": 1.6045429706573486, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8675312399864197, + "num_tokens": 140973559.0, + "step": 4012 + }, + { + "epoch": 0.7452181987000929, + "grad_norm": 1.3764450550079346, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8639158606529236, + "num_tokens": 141009950.0, + "step": 4013 + }, + { + "epoch": 0.7454038997214485, + "grad_norm": 1.4074655771255493, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.863377571105957, + "num_tokens": 141048764.0, + "step": 4014 + }, + { + "epoch": 0.7455896007428041, + "grad_norm": 1.4618040323257446, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8896653056144714, + "num_tokens": 141079307.0, + "step": 4015 + }, + { + "epoch": 0.7457753017641597, + "grad_norm": 1.45021390914917, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8717293739318848, + "num_tokens": 141114030.0, + "step": 4016 + }, + { + "epoch": 0.7459610027855154, + "grad_norm": 1.4078514575958252, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8570642471313477, + "num_tokens": 141152809.0, + "step": 4017 + }, + { + "epoch": 0.746146703806871, + "grad_norm": 1.5862425565719604, + "learning_rate": 1e-06, + "loss": 0.4308, + "mean_token_accuracy": 0.852861762046814, + "num_tokens": 141184986.0, + "step": 4018 + }, + { + "epoch": 0.7463324048282266, + "grad_norm": 1.5146398544311523, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8606436252593994, + "num_tokens": 141219487.0, + "step": 4019 + }, + { + "epoch": 0.7465181058495822, + "grad_norm": 1.4263248443603516, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.8684922456741333, + "num_tokens": 141253834.0, + "step": 4020 + }, + { + "epoch": 0.7467038068709377, + "grad_norm": 1.3846601247787476, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8702013492584229, + "num_tokens": 141290795.0, + "step": 4021 + }, + { + "epoch": 0.7468895078922934, + "grad_norm": 1.4620188474655151, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8671576976776123, + "num_tokens": 141328605.0, + "step": 4022 + }, + { + "epoch": 0.747075208913649, + "grad_norm": 1.330392837524414, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8728325963020325, + "num_tokens": 141367056.0, + "step": 4023 + }, + { + "epoch": 0.7472609099350046, + "grad_norm": 1.3226498365402222, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8902323246002197, + "num_tokens": 141405816.0, + "step": 4024 + }, + { + "epoch": 0.7474466109563602, + "grad_norm": 1.6325633525848389, + "learning_rate": 1e-06, + "loss": 0.3967, + "mean_token_accuracy": 0.8662066459655762, + "num_tokens": 141434905.0, + "step": 4025 + }, + { + "epoch": 0.7476323119777158, + "grad_norm": 1.5440006256103516, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8783234357833862, + "num_tokens": 141467796.0, + "step": 4026 + }, + { + "epoch": 0.7478180129990715, + "grad_norm": 1.37333083152771, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8768007755279541, + "num_tokens": 141503438.0, + "step": 4027 + }, + { + "epoch": 0.7480037140204271, + "grad_norm": 1.3076322078704834, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.880521297454834, + "num_tokens": 141544016.0, + "step": 4028 + }, + { + "epoch": 0.7481894150417827, + "grad_norm": 1.4454200267791748, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8811613321304321, + "num_tokens": 141578549.0, + "step": 4029 + }, + { + "epoch": 0.7483751160631383, + "grad_norm": 1.5329076051712036, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8767404556274414, + "num_tokens": 141609046.0, + "step": 4030 + }, + { + "epoch": 0.748560817084494, + "grad_norm": 1.4274425506591797, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.871198296546936, + "num_tokens": 141646875.0, + "step": 4031 + }, + { + "epoch": 0.7487465181058496, + "grad_norm": 1.3963466882705688, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8794934153556824, + "num_tokens": 141684203.0, + "step": 4032 + }, + { + "epoch": 0.7489322191272052, + "grad_norm": 1.4657254219055176, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8644809722900391, + "num_tokens": 141720201.0, + "step": 4033 + }, + { + "epoch": 0.7491179201485608, + "grad_norm": 1.363416075706482, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8725953102111816, + "num_tokens": 141757560.0, + "step": 4034 + }, + { + "epoch": 0.7493036211699164, + "grad_norm": 1.4846432209014893, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8613755702972412, + "num_tokens": 141793870.0, + "step": 4035 + }, + { + "epoch": 0.7494893221912721, + "grad_norm": 1.303788661956787, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8764728307723999, + "num_tokens": 141832625.0, + "step": 4036 + }, + { + "epoch": 0.7496750232126277, + "grad_norm": 1.4994158744812012, + "learning_rate": 1e-06, + "loss": 0.4409, + "mean_token_accuracy": 0.857335090637207, + "num_tokens": 141867008.0, + "step": 4037 + }, + { + "epoch": 0.7498607242339833, + "grad_norm": 1.3645870685577393, + "learning_rate": 1e-06, + "loss": 0.4187, + "mean_token_accuracy": 0.8567726612091064, + "num_tokens": 141907433.0, + "step": 4038 + }, + { + "epoch": 0.7500464252553389, + "grad_norm": 1.499672293663025, + "learning_rate": 1e-06, + "loss": 0.4134, + "mean_token_accuracy": 0.8632603287696838, + "num_tokens": 141941428.0, + "step": 4039 + }, + { + "epoch": 0.7502321262766946, + "grad_norm": 1.4511404037475586, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8742157220840454, + "num_tokens": 141977601.0, + "step": 4040 + }, + { + "epoch": 0.7504178272980502, + "grad_norm": 1.353268027305603, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8777867555618286, + "num_tokens": 142015171.0, + "step": 4041 + }, + { + "epoch": 0.7506035283194058, + "grad_norm": 1.470913052558899, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8698869347572327, + "num_tokens": 142052852.0, + "step": 4042 + }, + { + "epoch": 0.7507892293407614, + "grad_norm": 1.5613999366760254, + "learning_rate": 1e-06, + "loss": 0.4327, + "mean_token_accuracy": 0.8549935221672058, + "num_tokens": 142090269.0, + "step": 4043 + }, + { + "epoch": 0.750974930362117, + "grad_norm": 1.5514891147613525, + "learning_rate": 1e-06, + "loss": 0.4315, + "mean_token_accuracy": 0.8554372191429138, + "num_tokens": 142123280.0, + "step": 4044 + }, + { + "epoch": 0.7511606313834726, + "grad_norm": 1.500539779663086, + "learning_rate": 1e-06, + "loss": 0.4244, + "mean_token_accuracy": 0.8558436632156372, + "num_tokens": 142161692.0, + "step": 4045 + }, + { + "epoch": 0.7513463324048282, + "grad_norm": 1.4871597290039062, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.880423903465271, + "num_tokens": 142195103.0, + "step": 4046 + }, + { + "epoch": 0.7515320334261838, + "grad_norm": 1.5626229047775269, + "learning_rate": 1e-06, + "loss": 0.4054, + "mean_token_accuracy": 0.8676470518112183, + "num_tokens": 142229192.0, + "step": 4047 + }, + { + "epoch": 0.7517177344475394, + "grad_norm": 1.4287948608398438, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8764421939849854, + "num_tokens": 142268798.0, + "step": 4048 + }, + { + "epoch": 0.751903435468895, + "grad_norm": 1.4062221050262451, + "learning_rate": 1e-06, + "loss": 0.4335, + "mean_token_accuracy": 0.8565168380737305, + "num_tokens": 142308100.0, + "step": 4049 + }, + { + "epoch": 0.7520891364902507, + "grad_norm": 1.5094982385635376, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8656038045883179, + "num_tokens": 142343214.0, + "step": 4050 + }, + { + "epoch": 0.7522748375116063, + "grad_norm": 1.5466749668121338, + "learning_rate": 1e-06, + "loss": 0.4239, + "mean_token_accuracy": 0.8611805438995361, + "num_tokens": 142374092.0, + "step": 4051 + }, + { + "epoch": 0.7524605385329619, + "grad_norm": 1.6866523027420044, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8749682307243347, + "num_tokens": 142402887.0, + "step": 4052 + }, + { + "epoch": 0.7526462395543175, + "grad_norm": 1.4311987161636353, + "learning_rate": 1e-06, + "loss": 0.4511, + "mean_token_accuracy": 0.8516175746917725, + "num_tokens": 142439528.0, + "step": 4053 + }, + { + "epoch": 0.7528319405756732, + "grad_norm": 1.4385106563568115, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8715450763702393, + "num_tokens": 142476247.0, + "step": 4054 + }, + { + "epoch": 0.7530176415970288, + "grad_norm": 1.5256284475326538, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8675440549850464, + "num_tokens": 142515353.0, + "step": 4055 + }, + { + "epoch": 0.7532033426183844, + "grad_norm": 1.436012625694275, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8622541427612305, + "num_tokens": 142551376.0, + "step": 4056 + }, + { + "epoch": 0.75338904363974, + "grad_norm": 1.5123887062072754, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8634037971496582, + "num_tokens": 142586569.0, + "step": 4057 + }, + { + "epoch": 0.7535747446610956, + "grad_norm": 1.4286258220672607, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8596839308738708, + "num_tokens": 142624923.0, + "step": 4058 + }, + { + "epoch": 0.7537604456824513, + "grad_norm": 1.5290627479553223, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8605283498764038, + "num_tokens": 142661158.0, + "step": 4059 + }, + { + "epoch": 0.7539461467038069, + "grad_norm": 1.506537914276123, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.8686702847480774, + "num_tokens": 142696851.0, + "step": 4060 + }, + { + "epoch": 0.7541318477251625, + "grad_norm": 1.375619649887085, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8783921599388123, + "num_tokens": 142734451.0, + "step": 4061 + }, + { + "epoch": 0.7543175487465181, + "grad_norm": 1.5889705419540405, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8707635402679443, + "num_tokens": 142763976.0, + "step": 4062 + }, + { + "epoch": 0.7545032497678738, + "grad_norm": 1.3771426677703857, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.871483325958252, + "num_tokens": 142800307.0, + "step": 4063 + }, + { + "epoch": 0.7546889507892294, + "grad_norm": 1.5676559209823608, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8657637238502502, + "num_tokens": 142834894.0, + "step": 4064 + }, + { + "epoch": 0.754874651810585, + "grad_norm": 1.5590784549713135, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8629356026649475, + "num_tokens": 142871015.0, + "step": 4065 + }, + { + "epoch": 0.7550603528319406, + "grad_norm": 1.6839569807052612, + "learning_rate": 1e-06, + "loss": 0.4098, + "mean_token_accuracy": 0.8635315895080566, + "num_tokens": 142899086.0, + "step": 4066 + }, + { + "epoch": 0.7552460538532962, + "grad_norm": 1.6402510404586792, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8721622228622437, + "num_tokens": 142927150.0, + "step": 4067 + }, + { + "epoch": 0.7554317548746519, + "grad_norm": 1.5042409896850586, + "learning_rate": 1e-06, + "loss": 0.4232, + "mean_token_accuracy": 0.8578730225563049, + "num_tokens": 142960975.0, + "step": 4068 + }, + { + "epoch": 0.7556174558960074, + "grad_norm": 1.6647766828536987, + "learning_rate": 1e-06, + "loss": 0.4688, + "mean_token_accuracy": 0.8447742462158203, + "num_tokens": 142994045.0, + "step": 4069 + }, + { + "epoch": 0.755803156917363, + "grad_norm": 1.4347819089889526, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.875644326210022, + "num_tokens": 143028636.0, + "step": 4070 + }, + { + "epoch": 0.7559888579387186, + "grad_norm": 1.3635599613189697, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8820711374282837, + "num_tokens": 143064968.0, + "step": 4071 + }, + { + "epoch": 0.7561745589600742, + "grad_norm": 1.5716586112976074, + "learning_rate": 1e-06, + "loss": 0.442, + "mean_token_accuracy": 0.8554710149765015, + "num_tokens": 143100010.0, + "step": 4072 + }, + { + "epoch": 0.7563602599814299, + "grad_norm": 1.5307793617248535, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.8612303733825684, + "num_tokens": 143134447.0, + "step": 4073 + }, + { + "epoch": 0.7565459610027855, + "grad_norm": 1.5207263231277466, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8695960640907288, + "num_tokens": 143166729.0, + "step": 4074 + }, + { + "epoch": 0.7567316620241411, + "grad_norm": 1.564591407775879, + "learning_rate": 1e-06, + "loss": 0.4302, + "mean_token_accuracy": 0.8571587800979614, + "num_tokens": 143198786.0, + "step": 4075 + }, + { + "epoch": 0.7569173630454967, + "grad_norm": 1.4931414127349854, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8835130929946899, + "num_tokens": 143227819.0, + "step": 4076 + }, + { + "epoch": 0.7571030640668523, + "grad_norm": 1.3435479402542114, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.874548614025116, + "num_tokens": 143266762.0, + "step": 4077 + }, + { + "epoch": 0.757288765088208, + "grad_norm": 1.3844950199127197, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8705599308013916, + "num_tokens": 143302504.0, + "step": 4078 + }, + { + "epoch": 0.7574744661095636, + "grad_norm": 1.5546321868896484, + "learning_rate": 1e-06, + "loss": 0.4213, + "mean_token_accuracy": 0.857141375541687, + "num_tokens": 143335214.0, + "step": 4079 + }, + { + "epoch": 0.7576601671309192, + "grad_norm": 1.6031972169876099, + "learning_rate": 1e-06, + "loss": 0.4617, + "mean_token_accuracy": 0.8461491465568542, + "num_tokens": 143366478.0, + "step": 4080 + }, + { + "epoch": 0.7578458681522748, + "grad_norm": 1.459606409072876, + "learning_rate": 1e-06, + "loss": 0.4079, + "mean_token_accuracy": 0.8626506924629211, + "num_tokens": 143403908.0, + "step": 4081 + }, + { + "epoch": 0.7580315691736305, + "grad_norm": 1.498223900794983, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8711974620819092, + "num_tokens": 143436957.0, + "step": 4082 + }, + { + "epoch": 0.7582172701949861, + "grad_norm": 1.529679536819458, + "learning_rate": 1e-06, + "loss": 0.4453, + "mean_token_accuracy": 0.8532432317733765, + "num_tokens": 143472028.0, + "step": 4083 + }, + { + "epoch": 0.7584029712163417, + "grad_norm": 1.4287805557250977, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8687053322792053, + "num_tokens": 143507996.0, + "step": 4084 + }, + { + "epoch": 0.7585886722376973, + "grad_norm": 1.551557183265686, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.872012734413147, + "num_tokens": 143542376.0, + "step": 4085 + }, + { + "epoch": 0.758774373259053, + "grad_norm": 1.4538391828536987, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8730669021606445, + "num_tokens": 143579086.0, + "step": 4086 + }, + { + "epoch": 0.7589600742804086, + "grad_norm": 1.4493223428726196, + "learning_rate": 1e-06, + "loss": 0.4197, + "mean_token_accuracy": 0.8602200150489807, + "num_tokens": 143616703.0, + "step": 4087 + }, + { + "epoch": 0.7591457753017642, + "grad_norm": 1.4957590103149414, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8741206526756287, + "num_tokens": 143646161.0, + "step": 4088 + }, + { + "epoch": 0.7593314763231198, + "grad_norm": 1.4717369079589844, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8659254312515259, + "num_tokens": 143680219.0, + "step": 4089 + }, + { + "epoch": 0.7595171773444754, + "grad_norm": 1.4219547510147095, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8715538382530212, + "num_tokens": 143719415.0, + "step": 4090 + }, + { + "epoch": 0.7597028783658311, + "grad_norm": 1.446815848350525, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.8618425130844116, + "num_tokens": 143757916.0, + "step": 4091 + }, + { + "epoch": 0.7598885793871867, + "grad_norm": 1.6108577251434326, + "learning_rate": 1e-06, + "loss": 0.4269, + "mean_token_accuracy": 0.8547021150588989, + "num_tokens": 143789880.0, + "step": 4092 + }, + { + "epoch": 0.7600742804085423, + "grad_norm": 1.4639431238174438, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8720701336860657, + "num_tokens": 143826318.0, + "step": 4093 + }, + { + "epoch": 0.7602599814298978, + "grad_norm": 1.3325234651565552, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8786025047302246, + "num_tokens": 143862911.0, + "step": 4094 + }, + { + "epoch": 0.7604456824512534, + "grad_norm": 1.45928156375885, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8663491606712341, + "num_tokens": 143899630.0, + "step": 4095 + }, + { + "epoch": 0.7606313834726091, + "grad_norm": 1.442471981048584, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8713316917419434, + "num_tokens": 143933120.0, + "step": 4096 + }, + { + "epoch": 0.7608170844939647, + "grad_norm": 1.3968385457992554, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.871623158454895, + "num_tokens": 143970338.0, + "step": 4097 + }, + { + "epoch": 0.7610027855153203, + "grad_norm": 1.4516605138778687, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.877955436706543, + "num_tokens": 144007562.0, + "step": 4098 + }, + { + "epoch": 0.7611884865366759, + "grad_norm": 1.5758692026138306, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8736022114753723, + "num_tokens": 144036546.0, + "step": 4099 + }, + { + "epoch": 0.7613741875580315, + "grad_norm": 1.40431547164917, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.8621724843978882, + "num_tokens": 144074151.0, + "step": 4100 + }, + { + "epoch": 0.7615598885793872, + "grad_norm": 1.5626883506774902, + "learning_rate": 1e-06, + "loss": 0.4596, + "mean_token_accuracy": 0.8468915224075317, + "num_tokens": 144107431.0, + "step": 4101 + }, + { + "epoch": 0.7617455896007428, + "grad_norm": 1.4040606021881104, + "learning_rate": 1e-06, + "loss": 0.4318, + "mean_token_accuracy": 0.8575447201728821, + "num_tokens": 144147379.0, + "step": 4102 + }, + { + "epoch": 0.7619312906220984, + "grad_norm": 1.3639013767242432, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8685154318809509, + "num_tokens": 144185418.0, + "step": 4103 + }, + { + "epoch": 0.762116991643454, + "grad_norm": 1.5802844762802124, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8724039793014526, + "num_tokens": 144213443.0, + "step": 4104 + }, + { + "epoch": 0.7623026926648097, + "grad_norm": 1.4445830583572388, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8665984272956848, + "num_tokens": 144247669.0, + "step": 4105 + }, + { + "epoch": 0.7624883936861653, + "grad_norm": 1.4993983507156372, + "learning_rate": 1e-06, + "loss": 0.455, + "mean_token_accuracy": 0.852267861366272, + "num_tokens": 144281633.0, + "step": 4106 + }, + { + "epoch": 0.7626740947075209, + "grad_norm": 1.4481804370880127, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8745157718658447, + "num_tokens": 144316640.0, + "step": 4107 + }, + { + "epoch": 0.7628597957288765, + "grad_norm": 1.5033037662506104, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8573806285858154, + "num_tokens": 144350605.0, + "step": 4108 + }, + { + "epoch": 0.7630454967502321, + "grad_norm": 1.4326051473617554, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.863282322883606, + "num_tokens": 144389344.0, + "step": 4109 + }, + { + "epoch": 0.7632311977715878, + "grad_norm": 1.4288976192474365, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8685510158538818, + "num_tokens": 144424056.0, + "step": 4110 + }, + { + "epoch": 0.7634168987929434, + "grad_norm": 1.4261703491210938, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.8628720045089722, + "num_tokens": 144462533.0, + "step": 4111 + }, + { + "epoch": 0.763602599814299, + "grad_norm": 1.4486216306686401, + "learning_rate": 1e-06, + "loss": 0.4054, + "mean_token_accuracy": 0.863290548324585, + "num_tokens": 144497100.0, + "step": 4112 + }, + { + "epoch": 0.7637883008356546, + "grad_norm": 1.5104750394821167, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8721722364425659, + "num_tokens": 144529490.0, + "step": 4113 + }, + { + "epoch": 0.7639740018570103, + "grad_norm": 1.3784526586532593, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8755133748054504, + "num_tokens": 144562318.0, + "step": 4114 + }, + { + "epoch": 0.7641597028783659, + "grad_norm": 1.4052021503448486, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8800268173217773, + "num_tokens": 144594477.0, + "step": 4115 + }, + { + "epoch": 0.7643454038997215, + "grad_norm": 1.5154272317886353, + "learning_rate": 1e-06, + "loss": 0.4234, + "mean_token_accuracy": 0.8609412908554077, + "num_tokens": 144627002.0, + "step": 4116 + }, + { + "epoch": 0.7645311049210771, + "grad_norm": 1.3682135343551636, + "learning_rate": 1e-06, + "loss": 0.4303, + "mean_token_accuracy": 0.8555614352226257, + "num_tokens": 144668935.0, + "step": 4117 + }, + { + "epoch": 0.7647168059424326, + "grad_norm": 1.46824049949646, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8769677877426147, + "num_tokens": 144701760.0, + "step": 4118 + }, + { + "epoch": 0.7649025069637883, + "grad_norm": 1.4834530353546143, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8740501403808594, + "num_tokens": 144734648.0, + "step": 4119 + }, + { + "epoch": 0.7650882079851439, + "grad_norm": 1.589632511138916, + "learning_rate": 1e-06, + "loss": 0.4265, + "mean_token_accuracy": 0.8573746681213379, + "num_tokens": 144766375.0, + "step": 4120 + }, + { + "epoch": 0.7652739090064995, + "grad_norm": 1.4141861200332642, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8641924858093262, + "num_tokens": 144805029.0, + "step": 4121 + }, + { + "epoch": 0.7654596100278551, + "grad_norm": 1.4889650344848633, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8643198013305664, + "num_tokens": 144840033.0, + "step": 4122 + }, + { + "epoch": 0.7656453110492107, + "grad_norm": 1.4549658298492432, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8749579191207886, + "num_tokens": 144875278.0, + "step": 4123 + }, + { + "epoch": 0.7658310120705664, + "grad_norm": 1.5499796867370605, + "learning_rate": 1e-06, + "loss": 0.4158, + "mean_token_accuracy": 0.8597497344017029, + "num_tokens": 144911319.0, + "step": 4124 + }, + { + "epoch": 0.766016713091922, + "grad_norm": 1.6158835887908936, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8722172379493713, + "num_tokens": 144941176.0, + "step": 4125 + }, + { + "epoch": 0.7662024141132776, + "grad_norm": 1.4839427471160889, + "learning_rate": 1e-06, + "loss": 0.3983, + "mean_token_accuracy": 0.8647838830947876, + "num_tokens": 144975168.0, + "step": 4126 + }, + { + "epoch": 0.7663881151346332, + "grad_norm": 1.544854760169983, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8717043995857239, + "num_tokens": 145006382.0, + "step": 4127 + }, + { + "epoch": 0.7665738161559889, + "grad_norm": 1.4261822700500488, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8814923167228699, + "num_tokens": 145041270.0, + "step": 4128 + }, + { + "epoch": 0.7667595171773445, + "grad_norm": 1.3519858121871948, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8819169998168945, + "num_tokens": 145079334.0, + "step": 4129 + }, + { + "epoch": 0.7669452181987001, + "grad_norm": 1.4771913290023804, + "learning_rate": 1e-06, + "loss": 0.4163, + "mean_token_accuracy": 0.8595727682113647, + "num_tokens": 145115983.0, + "step": 4130 + }, + { + "epoch": 0.7671309192200557, + "grad_norm": 1.5290547609329224, + "learning_rate": 1e-06, + "loss": 0.4205, + "mean_token_accuracy": 0.8642396330833435, + "num_tokens": 145150756.0, + "step": 4131 + }, + { + "epoch": 0.7673166202414113, + "grad_norm": 1.5090166330337524, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8695691227912903, + "num_tokens": 145183369.0, + "step": 4132 + }, + { + "epoch": 0.767502321262767, + "grad_norm": 1.6872401237487793, + "learning_rate": 1e-06, + "loss": 0.426, + "mean_token_accuracy": 0.8577898740768433, + "num_tokens": 145211597.0, + "step": 4133 + }, + { + "epoch": 0.7676880222841226, + "grad_norm": 1.3990304470062256, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.880316436290741, + "num_tokens": 145243643.0, + "step": 4134 + }, + { + "epoch": 0.7678737233054782, + "grad_norm": 1.5593931674957275, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8726903200149536, + "num_tokens": 145275160.0, + "step": 4135 + }, + { + "epoch": 0.7680594243268338, + "grad_norm": 1.5671638250350952, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8719234466552734, + "num_tokens": 145305492.0, + "step": 4136 + }, + { + "epoch": 0.7682451253481895, + "grad_norm": 1.5326138734817505, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8775267601013184, + "num_tokens": 145338651.0, + "step": 4137 + }, + { + "epoch": 0.7684308263695451, + "grad_norm": 1.6265113353729248, + "learning_rate": 1e-06, + "loss": 0.4218, + "mean_token_accuracy": 0.8594217300415039, + "num_tokens": 145370362.0, + "step": 4138 + }, + { + "epoch": 0.7686165273909007, + "grad_norm": 1.4896290302276611, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8646162748336792, + "num_tokens": 145407776.0, + "step": 4139 + }, + { + "epoch": 0.7688022284122563, + "grad_norm": 1.527699589729309, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8661541938781738, + "num_tokens": 145441522.0, + "step": 4140 + }, + { + "epoch": 0.7689879294336119, + "grad_norm": 1.4466886520385742, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.867782711982727, + "num_tokens": 145477300.0, + "step": 4141 + }, + { + "epoch": 0.7691736304549674, + "grad_norm": 1.4819495677947998, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8777168989181519, + "num_tokens": 145511829.0, + "step": 4142 + }, + { + "epoch": 0.7693593314763231, + "grad_norm": 1.55104660987854, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8724972009658813, + "num_tokens": 145543529.0, + "step": 4143 + }, + { + "epoch": 0.7695450324976787, + "grad_norm": 1.5557701587677002, + "learning_rate": 1e-06, + "loss": 0.442, + "mean_token_accuracy": 0.8540785908699036, + "num_tokens": 145578239.0, + "step": 4144 + }, + { + "epoch": 0.7697307335190343, + "grad_norm": 1.4880280494689941, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8752203583717346, + "num_tokens": 145608841.0, + "step": 4145 + }, + { + "epoch": 0.7699164345403899, + "grad_norm": 1.9141278266906738, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8638476133346558, + "num_tokens": 145639323.0, + "step": 4146 + }, + { + "epoch": 0.7701021355617456, + "grad_norm": 1.5151454210281372, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8579588532447815, + "num_tokens": 145672855.0, + "step": 4147 + }, + { + "epoch": 0.7702878365831012, + "grad_norm": 1.5379528999328613, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8696743845939636, + "num_tokens": 145704958.0, + "step": 4148 + }, + { + "epoch": 0.7704735376044568, + "grad_norm": 1.4620075225830078, + "learning_rate": 1e-06, + "loss": 0.4285, + "mean_token_accuracy": 0.8571958541870117, + "num_tokens": 145742355.0, + "step": 4149 + }, + { + "epoch": 0.7706592386258124, + "grad_norm": 1.390338659286499, + "learning_rate": 1e-06, + "loss": 0.4233, + "mean_token_accuracy": 0.8576906323432922, + "num_tokens": 145785380.0, + "step": 4150 + }, + { + "epoch": 0.770844939647168, + "grad_norm": 1.49502432346344, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8689720630645752, + "num_tokens": 145818195.0, + "step": 4151 + }, + { + "epoch": 0.7710306406685237, + "grad_norm": 1.466523289680481, + "learning_rate": 1e-06, + "loss": 0.4202, + "mean_token_accuracy": 0.859764575958252, + "num_tokens": 145852854.0, + "step": 4152 + }, + { + "epoch": 0.7712163416898793, + "grad_norm": 1.5601590871810913, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8587005734443665, + "num_tokens": 145882809.0, + "step": 4153 + }, + { + "epoch": 0.7714020427112349, + "grad_norm": 1.4548614025115967, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8687023520469666, + "num_tokens": 145919051.0, + "step": 4154 + }, + { + "epoch": 0.7715877437325905, + "grad_norm": 1.367629051208496, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8577137589454651, + "num_tokens": 145961288.0, + "step": 4155 + }, + { + "epoch": 0.7717734447539462, + "grad_norm": 1.5485608577728271, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.8634119033813477, + "num_tokens": 145993624.0, + "step": 4156 + }, + { + "epoch": 0.7719591457753018, + "grad_norm": 1.4653816223144531, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8602052330970764, + "num_tokens": 146028654.0, + "step": 4157 + }, + { + "epoch": 0.7721448467966574, + "grad_norm": 1.3920972347259521, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8663740158081055, + "num_tokens": 146071507.0, + "step": 4158 + }, + { + "epoch": 0.772330547818013, + "grad_norm": 1.5000957250595093, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8657957315444946, + "num_tokens": 146105879.0, + "step": 4159 + }, + { + "epoch": 0.7725162488393686, + "grad_norm": 1.5050967931747437, + "learning_rate": 1e-06, + "loss": 0.449, + "mean_token_accuracy": 0.8522063493728638, + "num_tokens": 146140914.0, + "step": 4160 + }, + { + "epoch": 0.7727019498607243, + "grad_norm": 1.3179221153259277, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8617433309555054, + "num_tokens": 146186179.0, + "step": 4161 + }, + { + "epoch": 0.7728876508820799, + "grad_norm": 1.4616003036499023, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8741273880004883, + "num_tokens": 146224327.0, + "step": 4162 + }, + { + "epoch": 0.7730733519034355, + "grad_norm": 1.495237946510315, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.86658775806427, + "num_tokens": 146255341.0, + "step": 4163 + }, + { + "epoch": 0.7732590529247911, + "grad_norm": 1.604143500328064, + "learning_rate": 1e-06, + "loss": 0.4388, + "mean_token_accuracy": 0.8538568019866943, + "num_tokens": 146288776.0, + "step": 4164 + }, + { + "epoch": 0.7734447539461468, + "grad_norm": 1.365248680114746, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8751352429389954, + "num_tokens": 146325367.0, + "step": 4165 + }, + { + "epoch": 0.7736304549675023, + "grad_norm": 1.4724067449569702, + "learning_rate": 1e-06, + "loss": 0.4049, + "mean_token_accuracy": 0.862970232963562, + "num_tokens": 146359936.0, + "step": 4166 + }, + { + "epoch": 0.7738161559888579, + "grad_norm": 1.5519442558288574, + "learning_rate": 1e-06, + "loss": 0.44, + "mean_token_accuracy": 0.8550089597702026, + "num_tokens": 146391249.0, + "step": 4167 + }, + { + "epoch": 0.7740018570102135, + "grad_norm": 1.6309616565704346, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8704659938812256, + "num_tokens": 146417544.0, + "step": 4168 + }, + { + "epoch": 0.7741875580315691, + "grad_norm": 1.576401710510254, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8766748309135437, + "num_tokens": 146444329.0, + "step": 4169 + }, + { + "epoch": 0.7743732590529248, + "grad_norm": 1.5781574249267578, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.863267719745636, + "num_tokens": 146475279.0, + "step": 4170 + }, + { + "epoch": 0.7745589600742804, + "grad_norm": 1.4882652759552002, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8716592788696289, + "num_tokens": 146510618.0, + "step": 4171 + }, + { + "epoch": 0.774744661095636, + "grad_norm": 1.6165937185287476, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8736490607261658, + "num_tokens": 146541693.0, + "step": 4172 + }, + { + "epoch": 0.7749303621169916, + "grad_norm": 1.5165568590164185, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8665549755096436, + "num_tokens": 146574148.0, + "step": 4173 + }, + { + "epoch": 0.7751160631383472, + "grad_norm": 1.5452793836593628, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.858726978302002, + "num_tokens": 146609611.0, + "step": 4174 + }, + { + "epoch": 0.7753017641597029, + "grad_norm": 1.4317131042480469, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8699764013290405, + "num_tokens": 146650754.0, + "step": 4175 + }, + { + "epoch": 0.7754874651810585, + "grad_norm": 1.442805290222168, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.8654168844223022, + "num_tokens": 146686899.0, + "step": 4176 + }, + { + "epoch": 0.7756731662024141, + "grad_norm": 1.4681105613708496, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8638691902160645, + "num_tokens": 146725065.0, + "step": 4177 + }, + { + "epoch": 0.7758588672237697, + "grad_norm": 1.5366711616516113, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8678560853004456, + "num_tokens": 146755138.0, + "step": 4178 + }, + { + "epoch": 0.7760445682451254, + "grad_norm": 1.4657820463180542, + "learning_rate": 1e-06, + "loss": 0.4168, + "mean_token_accuracy": 0.8604849576950073, + "num_tokens": 146796009.0, + "step": 4179 + }, + { + "epoch": 0.776230269266481, + "grad_norm": 1.4386783838272095, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8763004541397095, + "num_tokens": 146832087.0, + "step": 4180 + }, + { + "epoch": 0.7764159702878366, + "grad_norm": 1.435674786567688, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8695617914199829, + "num_tokens": 146865931.0, + "step": 4181 + }, + { + "epoch": 0.7766016713091922, + "grad_norm": 1.5937334299087524, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8595536947250366, + "num_tokens": 146897833.0, + "step": 4182 + }, + { + "epoch": 0.7767873723305478, + "grad_norm": 1.5663915872573853, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.872056245803833, + "num_tokens": 146932384.0, + "step": 4183 + }, + { + "epoch": 0.7769730733519035, + "grad_norm": 1.5725809335708618, + "learning_rate": 1e-06, + "loss": 0.4412, + "mean_token_accuracy": 0.8551104068756104, + "num_tokens": 146970803.0, + "step": 4184 + }, + { + "epoch": 0.7771587743732591, + "grad_norm": 1.590898036956787, + "learning_rate": 1e-06, + "loss": 0.4725, + "mean_token_accuracy": 0.840880274772644, + "num_tokens": 147007359.0, + "step": 4185 + }, + { + "epoch": 0.7773444753946147, + "grad_norm": 1.5135911703109741, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8525117635726929, + "num_tokens": 147044337.0, + "step": 4186 + }, + { + "epoch": 0.7775301764159703, + "grad_norm": 1.4752568006515503, + "learning_rate": 1e-06, + "loss": 0.4333, + "mean_token_accuracy": 0.8528724312782288, + "num_tokens": 147083832.0, + "step": 4187 + }, + { + "epoch": 0.777715877437326, + "grad_norm": 1.4616018533706665, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8671640157699585, + "num_tokens": 147122380.0, + "step": 4188 + }, + { + "epoch": 0.7779015784586816, + "grad_norm": 1.65715754032135, + "learning_rate": 1e-06, + "loss": 0.4449, + "mean_token_accuracy": 0.8517516255378723, + "num_tokens": 147156714.0, + "step": 4189 + }, + { + "epoch": 0.7780872794800371, + "grad_norm": 1.4259957075119019, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8644031286239624, + "num_tokens": 147196000.0, + "step": 4190 + }, + { + "epoch": 0.7782729805013927, + "grad_norm": 1.6325730085372925, + "learning_rate": 1e-06, + "loss": 0.4139, + "mean_token_accuracy": 0.8606816530227661, + "num_tokens": 147222175.0, + "step": 4191 + }, + { + "epoch": 0.7784586815227483, + "grad_norm": 1.5384958982467651, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8686597347259521, + "num_tokens": 147254563.0, + "step": 4192 + }, + { + "epoch": 0.778644382544104, + "grad_norm": 1.7318123579025269, + "learning_rate": 1e-06, + "loss": 0.4593, + "mean_token_accuracy": 0.8486735224723816, + "num_tokens": 147288861.0, + "step": 4193 + }, + { + "epoch": 0.7788300835654596, + "grad_norm": 1.4072659015655518, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8755215406417847, + "num_tokens": 147325065.0, + "step": 4194 + }, + { + "epoch": 0.7790157845868152, + "grad_norm": 1.4131436347961426, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.8606914281845093, + "num_tokens": 147362931.0, + "step": 4195 + }, + { + "epoch": 0.7792014856081708, + "grad_norm": 1.4822407960891724, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8676052093505859, + "num_tokens": 147398093.0, + "step": 4196 + }, + { + "epoch": 0.7793871866295264, + "grad_norm": 1.5218257904052734, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8813974261283875, + "num_tokens": 147426422.0, + "step": 4197 + }, + { + "epoch": 0.7795728876508821, + "grad_norm": 1.5298410654067993, + "learning_rate": 1e-06, + "loss": 0.4481, + "mean_token_accuracy": 0.8497194051742554, + "num_tokens": 147465938.0, + "step": 4198 + }, + { + "epoch": 0.7797585886722377, + "grad_norm": 1.5251444578170776, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8607409000396729, + "num_tokens": 147499409.0, + "step": 4199 + }, + { + "epoch": 0.7799442896935933, + "grad_norm": 1.5935992002487183, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8714412450790405, + "num_tokens": 147530916.0, + "step": 4200 + }, + { + "epoch": 0.7801299907149489, + "grad_norm": 1.653610110282898, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8561683297157288, + "num_tokens": 147562349.0, + "step": 4201 + }, + { + "epoch": 0.7803156917363046, + "grad_norm": 1.5919255018234253, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.8596392273902893, + "num_tokens": 147597052.0, + "step": 4202 + }, + { + "epoch": 0.7805013927576602, + "grad_norm": 1.465988039970398, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8731950521469116, + "num_tokens": 147631071.0, + "step": 4203 + }, + { + "epoch": 0.7806870937790158, + "grad_norm": 1.5586113929748535, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8753013610839844, + "num_tokens": 147662926.0, + "step": 4204 + }, + { + "epoch": 0.7808727948003714, + "grad_norm": 1.575979232788086, + "learning_rate": 1e-06, + "loss": 0.4339, + "mean_token_accuracy": 0.8556935787200928, + "num_tokens": 147696704.0, + "step": 4205 + }, + { + "epoch": 0.781058495821727, + "grad_norm": 1.4476546049118042, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8666895031929016, + "num_tokens": 147732072.0, + "step": 4206 + }, + { + "epoch": 0.7812441968430827, + "grad_norm": 1.5496113300323486, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8593514561653137, + "num_tokens": 147764394.0, + "step": 4207 + }, + { + "epoch": 0.7814298978644383, + "grad_norm": 1.4962223768234253, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.865453839302063, + "num_tokens": 147801275.0, + "step": 4208 + }, + { + "epoch": 0.7816155988857939, + "grad_norm": 1.4510663747787476, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8687009811401367, + "num_tokens": 147838035.0, + "step": 4209 + }, + { + "epoch": 0.7818012999071495, + "grad_norm": 1.4560338258743286, + "learning_rate": 1e-06, + "loss": 0.4489, + "mean_token_accuracy": 0.847292959690094, + "num_tokens": 147877951.0, + "step": 4210 + }, + { + "epoch": 0.7819870009285051, + "grad_norm": 1.5716911554336548, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.8622308969497681, + "num_tokens": 147908439.0, + "step": 4211 + }, + { + "epoch": 0.7821727019498608, + "grad_norm": 1.3883329629898071, + "learning_rate": 1e-06, + "loss": 0.4079, + "mean_token_accuracy": 0.861322283744812, + "num_tokens": 147950503.0, + "step": 4212 + }, + { + "epoch": 0.7823584029712164, + "grad_norm": 1.3452774286270142, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8524519205093384, + "num_tokens": 147992015.0, + "step": 4213 + }, + { + "epoch": 0.7825441039925719, + "grad_norm": 1.4209775924682617, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.875014066696167, + "num_tokens": 148027057.0, + "step": 4214 + }, + { + "epoch": 0.7827298050139275, + "grad_norm": 1.4389175176620483, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.873266339302063, + "num_tokens": 148062784.0, + "step": 4215 + }, + { + "epoch": 0.7829155060352831, + "grad_norm": 1.4621373414993286, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8610584139823914, + "num_tokens": 148097954.0, + "step": 4216 + }, + { + "epoch": 0.7831012070566388, + "grad_norm": 1.4231047630310059, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8699688911437988, + "num_tokens": 148135096.0, + "step": 4217 + }, + { + "epoch": 0.7832869080779944, + "grad_norm": 1.5210607051849365, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8595042824745178, + "num_tokens": 148169164.0, + "step": 4218 + }, + { + "epoch": 0.78347260909935, + "grad_norm": 1.5397859811782837, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.86714768409729, + "num_tokens": 148202143.0, + "step": 4219 + }, + { + "epoch": 0.7836583101207056, + "grad_norm": 1.4026427268981934, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8775376081466675, + "num_tokens": 148237017.0, + "step": 4220 + }, + { + "epoch": 0.7838440111420613, + "grad_norm": 1.375595211982727, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8794983625411987, + "num_tokens": 148273272.0, + "step": 4221 + }, + { + "epoch": 0.7840297121634169, + "grad_norm": 1.5244090557098389, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8647890090942383, + "num_tokens": 148312714.0, + "step": 4222 + }, + { + "epoch": 0.7842154131847725, + "grad_norm": 1.5576132535934448, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8706291913986206, + "num_tokens": 148345969.0, + "step": 4223 + }, + { + "epoch": 0.7844011142061281, + "grad_norm": 1.454367756843567, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8659919500350952, + "num_tokens": 148384672.0, + "step": 4224 + }, + { + "epoch": 0.7845868152274837, + "grad_norm": 1.4456523656845093, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8762226700782776, + "num_tokens": 148417219.0, + "step": 4225 + }, + { + "epoch": 0.7847725162488394, + "grad_norm": 1.412618637084961, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8652851581573486, + "num_tokens": 148454047.0, + "step": 4226 + }, + { + "epoch": 0.784958217270195, + "grad_norm": 1.5322399139404297, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8810688853263855, + "num_tokens": 148487656.0, + "step": 4227 + }, + { + "epoch": 0.7851439182915506, + "grad_norm": 1.4132332801818848, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.8646718859672546, + "num_tokens": 148524277.0, + "step": 4228 + }, + { + "epoch": 0.7853296193129062, + "grad_norm": 1.471097707748413, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.874051034450531, + "num_tokens": 148557498.0, + "step": 4229 + }, + { + "epoch": 0.7855153203342619, + "grad_norm": 1.322424054145813, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.873909056186676, + "num_tokens": 148597609.0, + "step": 4230 + }, + { + "epoch": 0.7857010213556175, + "grad_norm": 1.4254149198532104, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8685673475265503, + "num_tokens": 148632959.0, + "step": 4231 + }, + { + "epoch": 0.7858867223769731, + "grad_norm": 1.4320836067199707, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8682749271392822, + "num_tokens": 148670310.0, + "step": 4232 + }, + { + "epoch": 0.7860724233983287, + "grad_norm": 1.4221692085266113, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8767656683921814, + "num_tokens": 148707895.0, + "step": 4233 + }, + { + "epoch": 0.7862581244196843, + "grad_norm": 1.4995546340942383, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8571673631668091, + "num_tokens": 148743094.0, + "step": 4234 + }, + { + "epoch": 0.78644382544104, + "grad_norm": 1.4471261501312256, + "learning_rate": 1e-06, + "loss": 0.4377, + "mean_token_accuracy": 0.8560522794723511, + "num_tokens": 148780807.0, + "step": 4235 + }, + { + "epoch": 0.7866295264623956, + "grad_norm": 1.5245636701583862, + "learning_rate": 1e-06, + "loss": 0.4298, + "mean_token_accuracy": 0.8583990931510925, + "num_tokens": 148815932.0, + "step": 4236 + }, + { + "epoch": 0.7868152274837512, + "grad_norm": 1.7218577861785889, + "learning_rate": 1e-06, + "loss": 0.4402, + "mean_token_accuracy": 0.854134202003479, + "num_tokens": 148845523.0, + "step": 4237 + }, + { + "epoch": 0.7870009285051067, + "grad_norm": 1.6097722053527832, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8595160245895386, + "num_tokens": 148877434.0, + "step": 4238 + }, + { + "epoch": 0.7871866295264623, + "grad_norm": 1.4103868007659912, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8672453761100769, + "num_tokens": 148914318.0, + "step": 4239 + }, + { + "epoch": 0.787372330547818, + "grad_norm": 1.486951470375061, + "learning_rate": 1e-06, + "loss": 0.4941, + "mean_token_accuracy": 0.8404355049133301, + "num_tokens": 148952304.0, + "step": 4240 + }, + { + "epoch": 0.7875580315691736, + "grad_norm": 1.3994165658950806, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8750385046005249, + "num_tokens": 148990085.0, + "step": 4241 + }, + { + "epoch": 0.7877437325905292, + "grad_norm": 1.4277371168136597, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8750104904174805, + "num_tokens": 149023832.0, + "step": 4242 + }, + { + "epoch": 0.7879294336118848, + "grad_norm": 1.4950346946716309, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8711057901382446, + "num_tokens": 149057465.0, + "step": 4243 + }, + { + "epoch": 0.7881151346332405, + "grad_norm": 1.3624327182769775, + "learning_rate": 1e-06, + "loss": 0.4195, + "mean_token_accuracy": 0.8590812683105469, + "num_tokens": 149097852.0, + "step": 4244 + }, + { + "epoch": 0.7883008356545961, + "grad_norm": 1.4269671440124512, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8651672601699829, + "num_tokens": 149133243.0, + "step": 4245 + }, + { + "epoch": 0.7884865366759517, + "grad_norm": 1.2931993007659912, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8692319989204407, + "num_tokens": 149177667.0, + "step": 4246 + }, + { + "epoch": 0.7886722376973073, + "grad_norm": 1.493039608001709, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8714919090270996, + "num_tokens": 149211828.0, + "step": 4247 + }, + { + "epoch": 0.7888579387186629, + "grad_norm": 1.5582765340805054, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.8645790815353394, + "num_tokens": 149244468.0, + "step": 4248 + }, + { + "epoch": 0.7890436397400186, + "grad_norm": 1.5889732837677002, + "learning_rate": 1e-06, + "loss": 0.4462, + "mean_token_accuracy": 0.8536362648010254, + "num_tokens": 149278527.0, + "step": 4249 + }, + { + "epoch": 0.7892293407613742, + "grad_norm": 1.3959262371063232, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8747129440307617, + "num_tokens": 149316120.0, + "step": 4250 + }, + { + "epoch": 0.7894150417827298, + "grad_norm": 1.40336275100708, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8826848268508911, + "num_tokens": 149349460.0, + "step": 4251 + }, + { + "epoch": 0.7896007428040854, + "grad_norm": 1.461996078491211, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8675516843795776, + "num_tokens": 149383484.0, + "step": 4252 + }, + { + "epoch": 0.789786443825441, + "grad_norm": 1.546278953552246, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8608030676841736, + "num_tokens": 149416021.0, + "step": 4253 + }, + { + "epoch": 0.7899721448467967, + "grad_norm": 1.5269464254379272, + "learning_rate": 1e-06, + "loss": 0.4183, + "mean_token_accuracy": 0.8569588661193848, + "num_tokens": 149451660.0, + "step": 4254 + }, + { + "epoch": 0.7901578458681523, + "grad_norm": 1.4613367319107056, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.866973340511322, + "num_tokens": 149486252.0, + "step": 4255 + }, + { + "epoch": 0.7903435468895079, + "grad_norm": 1.4708762168884277, + "learning_rate": 1e-06, + "loss": 0.4455, + "mean_token_accuracy": 0.8501659631729126, + "num_tokens": 149522177.0, + "step": 4256 + }, + { + "epoch": 0.7905292479108635, + "grad_norm": 1.3292856216430664, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8739930391311646, + "num_tokens": 149562612.0, + "step": 4257 + }, + { + "epoch": 0.7907149489322192, + "grad_norm": 1.4420833587646484, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.867602527141571, + "num_tokens": 149596526.0, + "step": 4258 + }, + { + "epoch": 0.7909006499535748, + "grad_norm": 1.5592859983444214, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.8582231998443604, + "num_tokens": 149630652.0, + "step": 4259 + }, + { + "epoch": 0.7910863509749304, + "grad_norm": 1.3631010055541992, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8664659261703491, + "num_tokens": 149671999.0, + "step": 4260 + }, + { + "epoch": 0.791272051996286, + "grad_norm": 1.4305006265640259, + "learning_rate": 1e-06, + "loss": 0.4181, + "mean_token_accuracy": 0.8574156761169434, + "num_tokens": 149707050.0, + "step": 4261 + }, + { + "epoch": 0.7914577530176417, + "grad_norm": 1.4089146852493286, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8690640330314636, + "num_tokens": 149742088.0, + "step": 4262 + }, + { + "epoch": 0.7916434540389972, + "grad_norm": 1.412907600402832, + "learning_rate": 1e-06, + "loss": 0.4282, + "mean_token_accuracy": 0.85957932472229, + "num_tokens": 149781935.0, + "step": 4263 + }, + { + "epoch": 0.7918291550603528, + "grad_norm": 1.4421324729919434, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8614884614944458, + "num_tokens": 149821389.0, + "step": 4264 + }, + { + "epoch": 0.7920148560817084, + "grad_norm": 1.5296943187713623, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8849063515663147, + "num_tokens": 149850366.0, + "step": 4265 + }, + { + "epoch": 0.792200557103064, + "grad_norm": 1.497775673866272, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8699319362640381, + "num_tokens": 149884942.0, + "step": 4266 + }, + { + "epoch": 0.7923862581244197, + "grad_norm": 1.5665595531463623, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8624739646911621, + "num_tokens": 149919395.0, + "step": 4267 + }, + { + "epoch": 0.7925719591457753, + "grad_norm": 1.5013768672943115, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8846033811569214, + "num_tokens": 149951747.0, + "step": 4268 + }, + { + "epoch": 0.7927576601671309, + "grad_norm": 1.4042670726776123, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8671258687973022, + "num_tokens": 149988791.0, + "step": 4269 + }, + { + "epoch": 0.7929433611884865, + "grad_norm": 1.4390236139297485, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.878669261932373, + "num_tokens": 150024458.0, + "step": 4270 + }, + { + "epoch": 0.7931290622098421, + "grad_norm": 1.479337453842163, + "learning_rate": 1e-06, + "loss": 0.4183, + "mean_token_accuracy": 0.8617490530014038, + "num_tokens": 150061149.0, + "step": 4271 + }, + { + "epoch": 0.7933147632311978, + "grad_norm": 1.3605835437774658, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8809747099876404, + "num_tokens": 150097459.0, + "step": 4272 + }, + { + "epoch": 0.7935004642525534, + "grad_norm": 1.5300770998001099, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.871261715888977, + "num_tokens": 150129707.0, + "step": 4273 + }, + { + "epoch": 0.793686165273909, + "grad_norm": 1.6184635162353516, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8635837435722351, + "num_tokens": 150159348.0, + "step": 4274 + }, + { + "epoch": 0.7938718662952646, + "grad_norm": 1.490491509437561, + "learning_rate": 1e-06, + "loss": 0.4344, + "mean_token_accuracy": 0.8553652763366699, + "num_tokens": 150193515.0, + "step": 4275 + }, + { + "epoch": 0.7940575673166202, + "grad_norm": 1.3718907833099365, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8742977380752563, + "num_tokens": 150230654.0, + "step": 4276 + }, + { + "epoch": 0.7942432683379759, + "grad_norm": 1.4333224296569824, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8681797385215759, + "num_tokens": 150266450.0, + "step": 4277 + }, + { + "epoch": 0.7944289693593315, + "grad_norm": 1.3958390951156616, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8705391883850098, + "num_tokens": 150300484.0, + "step": 4278 + }, + { + "epoch": 0.7946146703806871, + "grad_norm": 1.5912790298461914, + "learning_rate": 1e-06, + "loss": 0.4469, + "mean_token_accuracy": 0.8472079634666443, + "num_tokens": 150332462.0, + "step": 4279 + }, + { + "epoch": 0.7948003714020427, + "grad_norm": 1.4320653676986694, + "learning_rate": 1e-06, + "loss": 0.4229, + "mean_token_accuracy": 0.857153058052063, + "num_tokens": 150371320.0, + "step": 4280 + }, + { + "epoch": 0.7949860724233984, + "grad_norm": 1.452396035194397, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8694531321525574, + "num_tokens": 150407049.0, + "step": 4281 + }, + { + "epoch": 0.795171773444754, + "grad_norm": 1.3932514190673828, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8738869428634644, + "num_tokens": 150443333.0, + "step": 4282 + }, + { + "epoch": 0.7953574744661096, + "grad_norm": 1.4034405946731567, + "learning_rate": 1e-06, + "loss": 0.4201, + "mean_token_accuracy": 0.8610149025917053, + "num_tokens": 150481563.0, + "step": 4283 + }, + { + "epoch": 0.7955431754874652, + "grad_norm": 1.4656116962432861, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8715781569480896, + "num_tokens": 150510824.0, + "step": 4284 + }, + { + "epoch": 0.7957288765088208, + "grad_norm": 1.566403865814209, + "learning_rate": 1e-06, + "loss": 0.4216, + "mean_token_accuracy": 0.8565119504928589, + "num_tokens": 150544281.0, + "step": 4285 + }, + { + "epoch": 0.7959145775301765, + "grad_norm": 1.5258041620254517, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8623394966125488, + "num_tokens": 150579381.0, + "step": 4286 + }, + { + "epoch": 0.796100278551532, + "grad_norm": 1.3732550144195557, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8682514429092407, + "num_tokens": 150618730.0, + "step": 4287 + }, + { + "epoch": 0.7962859795728876, + "grad_norm": 1.488387107849121, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.875205397605896, + "num_tokens": 150652258.0, + "step": 4288 + }, + { + "epoch": 0.7964716805942432, + "grad_norm": 1.4072778224945068, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.879876971244812, + "num_tokens": 150685605.0, + "step": 4289 + }, + { + "epoch": 0.7966573816155988, + "grad_norm": 1.5145899057388306, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8796318769454956, + "num_tokens": 150721664.0, + "step": 4290 + }, + { + "epoch": 0.7968430826369545, + "grad_norm": 1.4034969806671143, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8660592436790466, + "num_tokens": 150759127.0, + "step": 4291 + }, + { + "epoch": 0.7970287836583101, + "grad_norm": 1.5200541019439697, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8798890113830566, + "num_tokens": 150789033.0, + "step": 4292 + }, + { + "epoch": 0.7972144846796657, + "grad_norm": 1.6076432466506958, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8708123564720154, + "num_tokens": 150822750.0, + "step": 4293 + }, + { + "epoch": 0.7974001857010213, + "grad_norm": 1.3078045845031738, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8867218494415283, + "num_tokens": 150859505.0, + "step": 4294 + }, + { + "epoch": 0.797585886722377, + "grad_norm": 1.4260274171829224, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8655760884284973, + "num_tokens": 150895980.0, + "step": 4295 + }, + { + "epoch": 0.7977715877437326, + "grad_norm": 1.5943689346313477, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8777768015861511, + "num_tokens": 150923774.0, + "step": 4296 + }, + { + "epoch": 0.7979572887650882, + "grad_norm": 1.4515695571899414, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8610270023345947, + "num_tokens": 150958864.0, + "step": 4297 + }, + { + "epoch": 0.7981429897864438, + "grad_norm": 1.3812119960784912, + "learning_rate": 1e-06, + "loss": 0.4586, + "mean_token_accuracy": 0.8502476215362549, + "num_tokens": 151002391.0, + "step": 4298 + }, + { + "epoch": 0.7983286908077994, + "grad_norm": 1.4590173959732056, + "learning_rate": 1e-06, + "loss": 0.4338, + "mean_token_accuracy": 0.8545418977737427, + "num_tokens": 151040397.0, + "step": 4299 + }, + { + "epoch": 0.7985143918291551, + "grad_norm": 1.5066908597946167, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8655508756637573, + "num_tokens": 151077187.0, + "step": 4300 + }, + { + "epoch": 0.7987000928505107, + "grad_norm": 1.5705468654632568, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8612083196640015, + "num_tokens": 151110752.0, + "step": 4301 + }, + { + "epoch": 0.7988857938718663, + "grad_norm": 1.4213358163833618, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8610806465148926, + "num_tokens": 151153031.0, + "step": 4302 + }, + { + "epoch": 0.7990714948932219, + "grad_norm": 1.6377543210983276, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.8646152019500732, + "num_tokens": 151181259.0, + "step": 4303 + }, + { + "epoch": 0.7992571959145776, + "grad_norm": 1.429118275642395, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8725091218948364, + "num_tokens": 151218770.0, + "step": 4304 + }, + { + "epoch": 0.7994428969359332, + "grad_norm": 1.4718334674835205, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8565081357955933, + "num_tokens": 151254041.0, + "step": 4305 + }, + { + "epoch": 0.7996285979572888, + "grad_norm": 1.5443674325942993, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.875834584236145, + "num_tokens": 151284384.0, + "step": 4306 + }, + { + "epoch": 0.7998142989786444, + "grad_norm": 1.4097338914871216, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8672487735748291, + "num_tokens": 151323053.0, + "step": 4307 + }, + { + "epoch": 0.8, + "grad_norm": 1.485573172569275, + "learning_rate": 1e-06, + "loss": 0.4549, + "mean_token_accuracy": 0.8474547266960144, + "num_tokens": 151361245.0, + "step": 4308 + }, + { + "epoch": 0.8001857010213557, + "grad_norm": 1.3740034103393555, + "learning_rate": 1e-06, + "loss": 0.4066, + "mean_token_accuracy": 0.8622073531150818, + "num_tokens": 151399928.0, + "step": 4309 + }, + { + "epoch": 0.8003714020427113, + "grad_norm": 1.5283981561660767, + "learning_rate": 1e-06, + "loss": 0.4442, + "mean_token_accuracy": 0.851109504699707, + "num_tokens": 151434958.0, + "step": 4310 + }, + { + "epoch": 0.8005571030640668, + "grad_norm": 1.4276573657989502, + "learning_rate": 1e-06, + "loss": 0.4, + "mean_token_accuracy": 0.864879846572876, + "num_tokens": 151473277.0, + "step": 4311 + }, + { + "epoch": 0.8007428040854224, + "grad_norm": 1.3368289470672607, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8712689876556396, + "num_tokens": 151510983.0, + "step": 4312 + }, + { + "epoch": 0.800928505106778, + "grad_norm": 1.4312189817428589, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8650440573692322, + "num_tokens": 151545684.0, + "step": 4313 + }, + { + "epoch": 0.8011142061281337, + "grad_norm": 1.4777593612670898, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8844903707504272, + "num_tokens": 151579237.0, + "step": 4314 + }, + { + "epoch": 0.8012999071494893, + "grad_norm": 1.4550683498382568, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8864401578903198, + "num_tokens": 151609802.0, + "step": 4315 + }, + { + "epoch": 0.8014856081708449, + "grad_norm": 1.4971927404403687, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8738589286804199, + "num_tokens": 151640436.0, + "step": 4316 + }, + { + "epoch": 0.8016713091922005, + "grad_norm": 1.415717363357544, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8797393441200256, + "num_tokens": 151677443.0, + "step": 4317 + }, + { + "epoch": 0.8018570102135562, + "grad_norm": 1.5025962591171265, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8746228218078613, + "num_tokens": 151710504.0, + "step": 4318 + }, + { + "epoch": 0.8020427112349118, + "grad_norm": 1.470097541809082, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8751316070556641, + "num_tokens": 151743077.0, + "step": 4319 + }, + { + "epoch": 0.8022284122562674, + "grad_norm": 1.6058012247085571, + "learning_rate": 1e-06, + "loss": 0.4513, + "mean_token_accuracy": 0.8565194606781006, + "num_tokens": 151775058.0, + "step": 4320 + }, + { + "epoch": 0.802414113277623, + "grad_norm": 1.4595121145248413, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8760969638824463, + "num_tokens": 151809331.0, + "step": 4321 + }, + { + "epoch": 0.8025998142989786, + "grad_norm": 1.4585543870925903, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8762280941009521, + "num_tokens": 151842936.0, + "step": 4322 + }, + { + "epoch": 0.8027855153203343, + "grad_norm": 1.407335638999939, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8792284727096558, + "num_tokens": 151882512.0, + "step": 4323 + }, + { + "epoch": 0.8029712163416899, + "grad_norm": 1.3724335432052612, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8627673387527466, + "num_tokens": 151922374.0, + "step": 4324 + }, + { + "epoch": 0.8031569173630455, + "grad_norm": 1.5551069974899292, + "learning_rate": 1e-06, + "loss": 0.4097, + "mean_token_accuracy": 0.8626559376716614, + "num_tokens": 151955336.0, + "step": 4325 + }, + { + "epoch": 0.8033426183844011, + "grad_norm": 1.4234095811843872, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.875186562538147, + "num_tokens": 151989402.0, + "step": 4326 + }, + { + "epoch": 0.8035283194057568, + "grad_norm": 1.3889168500900269, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8637504577636719, + "num_tokens": 152027129.0, + "step": 4327 + }, + { + "epoch": 0.8037140204271124, + "grad_norm": 1.6398203372955322, + "learning_rate": 1e-06, + "loss": 0.4502, + "mean_token_accuracy": 0.8514752388000488, + "num_tokens": 152057635.0, + "step": 4328 + }, + { + "epoch": 0.803899721448468, + "grad_norm": 1.6062514781951904, + "learning_rate": 1e-06, + "loss": 0.4397, + "mean_token_accuracy": 0.8550019264221191, + "num_tokens": 152087417.0, + "step": 4329 + }, + { + "epoch": 0.8040854224698236, + "grad_norm": 1.4880225658416748, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8804962038993835, + "num_tokens": 152115119.0, + "step": 4330 + }, + { + "epoch": 0.8042711234911792, + "grad_norm": 1.616533875465393, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8622199296951294, + "num_tokens": 152141321.0, + "step": 4331 + }, + { + "epoch": 0.8044568245125349, + "grad_norm": 1.6225712299346924, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8741847276687622, + "num_tokens": 152169630.0, + "step": 4332 + }, + { + "epoch": 0.8046425255338905, + "grad_norm": 1.3970199823379517, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8692842721939087, + "num_tokens": 152207875.0, + "step": 4333 + }, + { + "epoch": 0.8048282265552461, + "grad_norm": 1.5027581453323364, + "learning_rate": 1e-06, + "loss": 0.4137, + "mean_token_accuracy": 0.8627045750617981, + "num_tokens": 152241228.0, + "step": 4334 + }, + { + "epoch": 0.8050139275766016, + "grad_norm": 1.4709192514419556, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8753564953804016, + "num_tokens": 152273456.0, + "step": 4335 + }, + { + "epoch": 0.8051996285979572, + "grad_norm": 1.451826810836792, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8674672842025757, + "num_tokens": 152308048.0, + "step": 4336 + }, + { + "epoch": 0.8053853296193129, + "grad_norm": 1.393862009048462, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.87123703956604, + "num_tokens": 152347517.0, + "step": 4337 + }, + { + "epoch": 0.8055710306406685, + "grad_norm": 1.3574243783950806, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8744686841964722, + "num_tokens": 152380957.0, + "step": 4338 + }, + { + "epoch": 0.8057567316620241, + "grad_norm": 1.412564992904663, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8714169263839722, + "num_tokens": 152414904.0, + "step": 4339 + }, + { + "epoch": 0.8059424326833797, + "grad_norm": 1.3896642923355103, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8748428821563721, + "num_tokens": 152453753.0, + "step": 4340 + }, + { + "epoch": 0.8061281337047354, + "grad_norm": 1.3358200788497925, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8684599995613098, + "num_tokens": 152493617.0, + "step": 4341 + }, + { + "epoch": 0.806313834726091, + "grad_norm": 1.4020943641662598, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8794793486595154, + "num_tokens": 152531486.0, + "step": 4342 + }, + { + "epoch": 0.8064995357474466, + "grad_norm": 1.561964988708496, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8667652606964111, + "num_tokens": 152560370.0, + "step": 4343 + }, + { + "epoch": 0.8066852367688022, + "grad_norm": 1.3729629516601562, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8715054988861084, + "num_tokens": 152599595.0, + "step": 4344 + }, + { + "epoch": 0.8068709377901578, + "grad_norm": 1.3277689218521118, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.878117561340332, + "num_tokens": 152634587.0, + "step": 4345 + }, + { + "epoch": 0.8070566388115135, + "grad_norm": 1.3243887424468994, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8803708553314209, + "num_tokens": 152675604.0, + "step": 4346 + }, + { + "epoch": 0.8072423398328691, + "grad_norm": 1.4466077089309692, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8836171627044678, + "num_tokens": 152709380.0, + "step": 4347 + }, + { + "epoch": 0.8074280408542247, + "grad_norm": 1.505454421043396, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8721410036087036, + "num_tokens": 152741394.0, + "step": 4348 + }, + { + "epoch": 0.8076137418755803, + "grad_norm": 1.7128647565841675, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8670928478240967, + "num_tokens": 152768671.0, + "step": 4349 + }, + { + "epoch": 0.807799442896936, + "grad_norm": 1.4864674806594849, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8654875755310059, + "num_tokens": 152800925.0, + "step": 4350 + }, + { + "epoch": 0.8079851439182916, + "grad_norm": 1.6240577697753906, + "learning_rate": 1e-06, + "loss": 0.411, + "mean_token_accuracy": 0.8614637851715088, + "num_tokens": 152832030.0, + "step": 4351 + }, + { + "epoch": 0.8081708449396472, + "grad_norm": 1.3833715915679932, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8809746503829956, + "num_tokens": 152866862.0, + "step": 4352 + }, + { + "epoch": 0.8083565459610028, + "grad_norm": 1.3419673442840576, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.875788688659668, + "num_tokens": 152907785.0, + "step": 4353 + }, + { + "epoch": 0.8085422469823584, + "grad_norm": 1.3435289859771729, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8735365271568298, + "num_tokens": 152949860.0, + "step": 4354 + }, + { + "epoch": 0.8087279480037141, + "grad_norm": 1.394057273864746, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8614050149917603, + "num_tokens": 152986601.0, + "step": 4355 + }, + { + "epoch": 0.8089136490250697, + "grad_norm": 1.552368402481079, + "learning_rate": 1e-06, + "loss": 0.4348, + "mean_token_accuracy": 0.8514739274978638, + "num_tokens": 153016062.0, + "step": 4356 + }, + { + "epoch": 0.8090993500464253, + "grad_norm": 1.397349238395691, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8731481432914734, + "num_tokens": 153052543.0, + "step": 4357 + }, + { + "epoch": 0.8092850510677809, + "grad_norm": 1.5142101049423218, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8740532398223877, + "num_tokens": 153085312.0, + "step": 4358 + }, + { + "epoch": 0.8094707520891364, + "grad_norm": 1.6623742580413818, + "learning_rate": 1e-06, + "loss": 0.4283, + "mean_token_accuracy": 0.8580923676490784, + "num_tokens": 153114256.0, + "step": 4359 + }, + { + "epoch": 0.8096564531104921, + "grad_norm": 1.627278447151184, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8615342974662781, + "num_tokens": 153147245.0, + "step": 4360 + }, + { + "epoch": 0.8098421541318477, + "grad_norm": 1.6328399181365967, + "learning_rate": 1e-06, + "loss": 0.4344, + "mean_token_accuracy": 0.852800726890564, + "num_tokens": 153178916.0, + "step": 4361 + }, + { + "epoch": 0.8100278551532033, + "grad_norm": 1.4879647493362427, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8705496191978455, + "num_tokens": 153212659.0, + "step": 4362 + }, + { + "epoch": 0.8102135561745589, + "grad_norm": 1.419058084487915, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8868318796157837, + "num_tokens": 153249055.0, + "step": 4363 + }, + { + "epoch": 0.8103992571959145, + "grad_norm": 1.3796379566192627, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8631755709648132, + "num_tokens": 153290160.0, + "step": 4364 + }, + { + "epoch": 0.8105849582172702, + "grad_norm": 1.607505202293396, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8588657975196838, + "num_tokens": 153320808.0, + "step": 4365 + }, + { + "epoch": 0.8107706592386258, + "grad_norm": 1.4521433115005493, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.869568943977356, + "num_tokens": 153357548.0, + "step": 4366 + }, + { + "epoch": 0.8109563602599814, + "grad_norm": 1.3724608421325684, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8682172894477844, + "num_tokens": 153394796.0, + "step": 4367 + }, + { + "epoch": 0.811142061281337, + "grad_norm": 1.5173132419586182, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8808436393737793, + "num_tokens": 153430062.0, + "step": 4368 + }, + { + "epoch": 0.8113277623026927, + "grad_norm": 1.293441653251648, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8657259941101074, + "num_tokens": 153472448.0, + "step": 4369 + }, + { + "epoch": 0.8115134633240483, + "grad_norm": 1.6800075769424438, + "learning_rate": 1e-06, + "loss": 0.4554, + "mean_token_accuracy": 0.8639504909515381, + "num_tokens": 153505381.0, + "step": 4370 + }, + { + "epoch": 0.8116991643454039, + "grad_norm": 1.3446844816207886, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8892313241958618, + "num_tokens": 153543544.0, + "step": 4371 + }, + { + "epoch": 0.8118848653667595, + "grad_norm": 1.4924812316894531, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8753167390823364, + "num_tokens": 153575345.0, + "step": 4372 + }, + { + "epoch": 0.8120705663881151, + "grad_norm": 1.5450081825256348, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.8626903295516968, + "num_tokens": 153611713.0, + "step": 4373 + }, + { + "epoch": 0.8122562674094708, + "grad_norm": 1.5324996709823608, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8717682361602783, + "num_tokens": 153643445.0, + "step": 4374 + }, + { + "epoch": 0.8124419684308264, + "grad_norm": 1.5823615789413452, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8707560300827026, + "num_tokens": 153673485.0, + "step": 4375 + }, + { + "epoch": 0.812627669452182, + "grad_norm": 1.5514333248138428, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.8621442317962646, + "num_tokens": 153703903.0, + "step": 4376 + }, + { + "epoch": 0.8128133704735376, + "grad_norm": 1.4870507717132568, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8671998977661133, + "num_tokens": 153741450.0, + "step": 4377 + }, + { + "epoch": 0.8129990714948933, + "grad_norm": 1.4169799089431763, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8726785182952881, + "num_tokens": 153774868.0, + "step": 4378 + }, + { + "epoch": 0.8131847725162489, + "grad_norm": 1.277401328086853, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.8665664196014404, + "num_tokens": 153817159.0, + "step": 4379 + }, + { + "epoch": 0.8133704735376045, + "grad_norm": 1.4877066612243652, + "learning_rate": 1e-06, + "loss": 0.4456, + "mean_token_accuracy": 0.8512812256813049, + "num_tokens": 153853133.0, + "step": 4380 + }, + { + "epoch": 0.8135561745589601, + "grad_norm": 1.3208434581756592, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8684566617012024, + "num_tokens": 153893402.0, + "step": 4381 + }, + { + "epoch": 0.8137418755803157, + "grad_norm": 1.4539005756378174, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.867049515247345, + "num_tokens": 153929388.0, + "step": 4382 + }, + { + "epoch": 0.8139275766016713, + "grad_norm": 1.288206934928894, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8669750094413757, + "num_tokens": 153969508.0, + "step": 4383 + }, + { + "epoch": 0.8141132776230269, + "grad_norm": 1.4411375522613525, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8650621175765991, + "num_tokens": 154005767.0, + "step": 4384 + }, + { + "epoch": 0.8142989786443825, + "grad_norm": 1.4540032148361206, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8756599426269531, + "num_tokens": 154043743.0, + "step": 4385 + }, + { + "epoch": 0.8144846796657381, + "grad_norm": 1.5941944122314453, + "learning_rate": 1e-06, + "loss": 0.4693, + "mean_token_accuracy": 0.843764066696167, + "num_tokens": 154081616.0, + "step": 4386 + }, + { + "epoch": 0.8146703806870937, + "grad_norm": 1.4228116273880005, + "learning_rate": 1e-06, + "loss": 0.4247, + "mean_token_accuracy": 0.8573015928268433, + "num_tokens": 154121442.0, + "step": 4387 + }, + { + "epoch": 0.8148560817084494, + "grad_norm": 1.431827425956726, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8651183843612671, + "num_tokens": 154155815.0, + "step": 4388 + }, + { + "epoch": 0.815041782729805, + "grad_norm": 1.5363149642944336, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.874544084072113, + "num_tokens": 154189302.0, + "step": 4389 + }, + { + "epoch": 0.8152274837511606, + "grad_norm": 1.5110701322555542, + "learning_rate": 1e-06, + "loss": 0.3993, + "mean_token_accuracy": 0.8577661514282227, + "num_tokens": 154225641.0, + "step": 4390 + }, + { + "epoch": 0.8154131847725162, + "grad_norm": 1.4252729415893555, + "learning_rate": 1e-06, + "loss": 0.4213, + "mean_token_accuracy": 0.858937680721283, + "num_tokens": 154265352.0, + "step": 4391 + }, + { + "epoch": 0.8155988857938719, + "grad_norm": 1.5255905389785767, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8679010272026062, + "num_tokens": 154298170.0, + "step": 4392 + }, + { + "epoch": 0.8157845868152275, + "grad_norm": 1.3369367122650146, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8689461946487427, + "num_tokens": 154337931.0, + "step": 4393 + }, + { + "epoch": 0.8159702878365831, + "grad_norm": 1.4967868328094482, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8657969832420349, + "num_tokens": 154370388.0, + "step": 4394 + }, + { + "epoch": 0.8161559888579387, + "grad_norm": 1.6002041101455688, + "learning_rate": 1e-06, + "loss": 0.4837, + "mean_token_accuracy": 0.8427270650863647, + "num_tokens": 154402206.0, + "step": 4395 + }, + { + "epoch": 0.8163416898792943, + "grad_norm": 1.5218523740768433, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8706046342849731, + "num_tokens": 154432207.0, + "step": 4396 + }, + { + "epoch": 0.81652739090065, + "grad_norm": 1.4678552150726318, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8706960678100586, + "num_tokens": 154464213.0, + "step": 4397 + }, + { + "epoch": 0.8167130919220056, + "grad_norm": 1.4957754611968994, + "learning_rate": 1e-06, + "loss": 0.4206, + "mean_token_accuracy": 0.8586704730987549, + "num_tokens": 154500090.0, + "step": 4398 + }, + { + "epoch": 0.8168987929433612, + "grad_norm": 1.4626102447509766, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8620758056640625, + "num_tokens": 154534951.0, + "step": 4399 + }, + { + "epoch": 0.8170844939647168, + "grad_norm": 1.5735684633255005, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8748660087585449, + "num_tokens": 154562383.0, + "step": 4400 + }, + { + "epoch": 0.8172701949860725, + "grad_norm": 1.447258710861206, + "learning_rate": 1e-06, + "loss": 0.4669, + "mean_token_accuracy": 0.845693051815033, + "num_tokens": 154602631.0, + "step": 4401 + }, + { + "epoch": 0.8174558960074281, + "grad_norm": 1.4489589929580688, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8786540031433105, + "num_tokens": 154635453.0, + "step": 4402 + }, + { + "epoch": 0.8176415970287837, + "grad_norm": 1.4560799598693848, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8828036785125732, + "num_tokens": 154667723.0, + "step": 4403 + }, + { + "epoch": 0.8178272980501393, + "grad_norm": 1.5701237916946411, + "learning_rate": 1e-06, + "loss": 0.424, + "mean_token_accuracy": 0.8628985285758972, + "num_tokens": 154698842.0, + "step": 4404 + }, + { + "epoch": 0.8180129990714949, + "grad_norm": 1.4248387813568115, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8652523756027222, + "num_tokens": 154736463.0, + "step": 4405 + }, + { + "epoch": 0.8181987000928506, + "grad_norm": 1.5275706052780151, + "learning_rate": 1e-06, + "loss": 0.3864, + "mean_token_accuracy": 0.8697653412818909, + "num_tokens": 154766585.0, + "step": 4406 + }, + { + "epoch": 0.8183844011142062, + "grad_norm": 1.3741343021392822, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8714292645454407, + "num_tokens": 154802481.0, + "step": 4407 + }, + { + "epoch": 0.8185701021355617, + "grad_norm": 1.544641375541687, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.862708330154419, + "num_tokens": 154836602.0, + "step": 4408 + }, + { + "epoch": 0.8187558031569173, + "grad_norm": 1.526307225227356, + "learning_rate": 1e-06, + "loss": 0.4078, + "mean_token_accuracy": 0.8619241118431091, + "num_tokens": 154870220.0, + "step": 4409 + }, + { + "epoch": 0.8189415041782729, + "grad_norm": 1.5287383794784546, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.8614821434020996, + "num_tokens": 154905998.0, + "step": 4410 + }, + { + "epoch": 0.8191272051996286, + "grad_norm": 1.44206702709198, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.8576103448867798, + "num_tokens": 154945643.0, + "step": 4411 + }, + { + "epoch": 0.8193129062209842, + "grad_norm": 1.4855003356933594, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.864578127861023, + "num_tokens": 154979300.0, + "step": 4412 + }, + { + "epoch": 0.8194986072423398, + "grad_norm": 1.3432319164276123, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8630848526954651, + "num_tokens": 155024237.0, + "step": 4413 + }, + { + "epoch": 0.8196843082636954, + "grad_norm": 1.6167142391204834, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.859952449798584, + "num_tokens": 155057292.0, + "step": 4414 + }, + { + "epoch": 0.819870009285051, + "grad_norm": 1.5260231494903564, + "learning_rate": 1e-06, + "loss": 0.427, + "mean_token_accuracy": 0.8577095866203308, + "num_tokens": 155090146.0, + "step": 4415 + }, + { + "epoch": 0.8200557103064067, + "grad_norm": 1.4017609357833862, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8710254430770874, + "num_tokens": 155125961.0, + "step": 4416 + }, + { + "epoch": 0.8202414113277623, + "grad_norm": 1.5468297004699707, + "learning_rate": 1e-06, + "loss": 0.4267, + "mean_token_accuracy": 0.8580654859542847, + "num_tokens": 155160752.0, + "step": 4417 + }, + { + "epoch": 0.8204271123491179, + "grad_norm": 1.453808069229126, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8666590452194214, + "num_tokens": 155196253.0, + "step": 4418 + }, + { + "epoch": 0.8206128133704735, + "grad_norm": 1.6326110363006592, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8717930912971497, + "num_tokens": 155227903.0, + "step": 4419 + }, + { + "epoch": 0.8207985143918292, + "grad_norm": 1.5944634675979614, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8674827814102173, + "num_tokens": 155254679.0, + "step": 4420 + }, + { + "epoch": 0.8209842154131848, + "grad_norm": 1.43600332736969, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.869617223739624, + "num_tokens": 155294967.0, + "step": 4421 + }, + { + "epoch": 0.8211699164345404, + "grad_norm": 1.4403305053710938, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8743588924407959, + "num_tokens": 155330433.0, + "step": 4422 + }, + { + "epoch": 0.821355617455896, + "grad_norm": 1.4365590810775757, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8637329339981079, + "num_tokens": 155367736.0, + "step": 4423 + }, + { + "epoch": 0.8215413184772516, + "grad_norm": 1.3739736080169678, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8749344944953918, + "num_tokens": 155402437.0, + "step": 4424 + }, + { + "epoch": 0.8217270194986073, + "grad_norm": 1.6152186393737793, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.869560182094574, + "num_tokens": 155432658.0, + "step": 4425 + }, + { + "epoch": 0.8219127205199629, + "grad_norm": 1.5058693885803223, + "learning_rate": 1e-06, + "loss": 0.4212, + "mean_token_accuracy": 0.8538002967834473, + "num_tokens": 155469228.0, + "step": 4426 + }, + { + "epoch": 0.8220984215413185, + "grad_norm": 1.4416718482971191, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.8683409094810486, + "num_tokens": 155508704.0, + "step": 4427 + }, + { + "epoch": 0.8222841225626741, + "grad_norm": 1.5121874809265137, + "learning_rate": 1e-06, + "loss": 0.4294, + "mean_token_accuracy": 0.8585560321807861, + "num_tokens": 155542189.0, + "step": 4428 + }, + { + "epoch": 0.8224698235840298, + "grad_norm": 1.5389883518218994, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8683013319969177, + "num_tokens": 155574121.0, + "step": 4429 + }, + { + "epoch": 0.8226555246053854, + "grad_norm": 1.5094890594482422, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8841402530670166, + "num_tokens": 155605200.0, + "step": 4430 + }, + { + "epoch": 0.822841225626741, + "grad_norm": 1.461625576019287, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8757894039154053, + "num_tokens": 155641308.0, + "step": 4431 + }, + { + "epoch": 0.8230269266480965, + "grad_norm": 1.6995973587036133, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.8549230098724365, + "num_tokens": 155672188.0, + "step": 4432 + }, + { + "epoch": 0.8232126276694521, + "grad_norm": 1.718695044517517, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8633071184158325, + "num_tokens": 155698914.0, + "step": 4433 + }, + { + "epoch": 0.8233983286908078, + "grad_norm": 1.3423948287963867, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8758576512336731, + "num_tokens": 155734100.0, + "step": 4434 + }, + { + "epoch": 0.8235840297121634, + "grad_norm": 1.4712629318237305, + "learning_rate": 1e-06, + "loss": 0.4047, + "mean_token_accuracy": 0.8653088808059692, + "num_tokens": 155767159.0, + "step": 4435 + }, + { + "epoch": 0.823769730733519, + "grad_norm": 1.5875213146209717, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.885766327381134, + "num_tokens": 155796825.0, + "step": 4436 + }, + { + "epoch": 0.8239554317548746, + "grad_norm": 1.3969022035598755, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8756624460220337, + "num_tokens": 155836523.0, + "step": 4437 + }, + { + "epoch": 0.8241411327762302, + "grad_norm": 1.3642879724502563, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8727135062217712, + "num_tokens": 155873854.0, + "step": 4438 + }, + { + "epoch": 0.8243268337975859, + "grad_norm": 1.4270684719085693, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8885639309883118, + "num_tokens": 155904745.0, + "step": 4439 + }, + { + "epoch": 0.8245125348189415, + "grad_norm": 1.4901049137115479, + "learning_rate": 1e-06, + "loss": 0.4476, + "mean_token_accuracy": 0.8537468910217285, + "num_tokens": 155941838.0, + "step": 4440 + }, + { + "epoch": 0.8246982358402971, + "grad_norm": 1.5328911542892456, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8710111975669861, + "num_tokens": 155968989.0, + "step": 4441 + }, + { + "epoch": 0.8248839368616527, + "grad_norm": 1.5827488899230957, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8599827289581299, + "num_tokens": 155999345.0, + "step": 4442 + }, + { + "epoch": 0.8250696378830084, + "grad_norm": 1.6069616079330444, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8669947385787964, + "num_tokens": 156031862.0, + "step": 4443 + }, + { + "epoch": 0.825255338904364, + "grad_norm": 1.5009801387786865, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8697458505630493, + "num_tokens": 156069511.0, + "step": 4444 + }, + { + "epoch": 0.8254410399257196, + "grad_norm": 1.4351730346679688, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.86841881275177, + "num_tokens": 156103804.0, + "step": 4445 + }, + { + "epoch": 0.8256267409470752, + "grad_norm": 1.4439034461975098, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8671405911445618, + "num_tokens": 156138524.0, + "step": 4446 + }, + { + "epoch": 0.8258124419684308, + "grad_norm": 1.6674543619155884, + "learning_rate": 1e-06, + "loss": 0.4377, + "mean_token_accuracy": 0.8542786836624146, + "num_tokens": 156167570.0, + "step": 4447 + }, + { + "epoch": 0.8259981429897865, + "grad_norm": 1.4622719287872314, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8762276768684387, + "num_tokens": 156203350.0, + "step": 4448 + }, + { + "epoch": 0.8261838440111421, + "grad_norm": 1.4267054796218872, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8719260692596436, + "num_tokens": 156239342.0, + "step": 4449 + }, + { + "epoch": 0.8263695450324977, + "grad_norm": 1.4559719562530518, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8719119429588318, + "num_tokens": 156274572.0, + "step": 4450 + }, + { + "epoch": 0.8265552460538533, + "grad_norm": 1.4995181560516357, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.869949460029602, + "num_tokens": 156305916.0, + "step": 4451 + }, + { + "epoch": 0.826740947075209, + "grad_norm": 1.3166331052780151, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.881239116191864, + "num_tokens": 156346285.0, + "step": 4452 + }, + { + "epoch": 0.8269266480965646, + "grad_norm": 1.8411407470703125, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8755558729171753, + "num_tokens": 156384992.0, + "step": 4453 + }, + { + "epoch": 0.8271123491179202, + "grad_norm": 1.4200505018234253, + "learning_rate": 1e-06, + "loss": 0.4148, + "mean_token_accuracy": 0.8615482449531555, + "num_tokens": 156423038.0, + "step": 4454 + }, + { + "epoch": 0.8272980501392758, + "grad_norm": 1.4448150396347046, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8744724988937378, + "num_tokens": 156459100.0, + "step": 4455 + }, + { + "epoch": 0.8274837511606313, + "grad_norm": 1.526253342628479, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8725409507751465, + "num_tokens": 156490378.0, + "step": 4456 + }, + { + "epoch": 0.827669452181987, + "grad_norm": 1.4188024997711182, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.882121205329895, + "num_tokens": 156527112.0, + "step": 4457 + }, + { + "epoch": 0.8278551532033426, + "grad_norm": 1.378009557723999, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8679887056350708, + "num_tokens": 156564505.0, + "step": 4458 + }, + { + "epoch": 0.8280408542246982, + "grad_norm": 1.4424364566802979, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8678427934646606, + "num_tokens": 156601265.0, + "step": 4459 + }, + { + "epoch": 0.8282265552460538, + "grad_norm": 1.6520835161209106, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8770948648452759, + "num_tokens": 156633598.0, + "step": 4460 + }, + { + "epoch": 0.8284122562674094, + "grad_norm": 1.3972450494766235, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8726855516433716, + "num_tokens": 156671326.0, + "step": 4461 + }, + { + "epoch": 0.8285979572887651, + "grad_norm": 1.398919701576233, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8751446008682251, + "num_tokens": 156709137.0, + "step": 4462 + }, + { + "epoch": 0.8287836583101207, + "grad_norm": 1.560891032218933, + "learning_rate": 1e-06, + "loss": 0.4482, + "mean_token_accuracy": 0.8512319922447205, + "num_tokens": 156745790.0, + "step": 4463 + }, + { + "epoch": 0.8289693593314763, + "grad_norm": 1.3857157230377197, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8827484846115112, + "num_tokens": 156783851.0, + "step": 4464 + }, + { + "epoch": 0.8291550603528319, + "grad_norm": 1.627467393875122, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8633069396018982, + "num_tokens": 156813077.0, + "step": 4465 + }, + { + "epoch": 0.8293407613741876, + "grad_norm": 1.5589418411254883, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8649924397468567, + "num_tokens": 156844142.0, + "step": 4466 + }, + { + "epoch": 0.8295264623955432, + "grad_norm": 1.616617202758789, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8648035526275635, + "num_tokens": 156873947.0, + "step": 4467 + }, + { + "epoch": 0.8297121634168988, + "grad_norm": 1.421315312385559, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8684825897216797, + "num_tokens": 156908546.0, + "step": 4468 + }, + { + "epoch": 0.8298978644382544, + "grad_norm": 1.4066431522369385, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8652012944221497, + "num_tokens": 156946839.0, + "step": 4469 + }, + { + "epoch": 0.83008356545961, + "grad_norm": 1.5337244272232056, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8544573187828064, + "num_tokens": 156982380.0, + "step": 4470 + }, + { + "epoch": 0.8302692664809657, + "grad_norm": 1.5265249013900757, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8567918539047241, + "num_tokens": 157016455.0, + "step": 4471 + }, + { + "epoch": 0.8304549675023213, + "grad_norm": 1.5265729427337646, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8674758076667786, + "num_tokens": 157051450.0, + "step": 4472 + }, + { + "epoch": 0.8306406685236769, + "grad_norm": 1.3831541538238525, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8836504220962524, + "num_tokens": 157084690.0, + "step": 4473 + }, + { + "epoch": 0.8308263695450325, + "grad_norm": 1.3684251308441162, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8779478669166565, + "num_tokens": 157121722.0, + "step": 4474 + }, + { + "epoch": 0.8310120705663882, + "grad_norm": 1.4725998640060425, + "learning_rate": 1e-06, + "loss": 0.403, + "mean_token_accuracy": 0.8647059202194214, + "num_tokens": 157157788.0, + "step": 4475 + }, + { + "epoch": 0.8311977715877438, + "grad_norm": 1.5325605869293213, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8669177293777466, + "num_tokens": 157193555.0, + "step": 4476 + }, + { + "epoch": 0.8313834726090994, + "grad_norm": 1.418468713760376, + "learning_rate": 1e-06, + "loss": 0.4351, + "mean_token_accuracy": 0.8577579259872437, + "num_tokens": 157234636.0, + "step": 4477 + }, + { + "epoch": 0.831569173630455, + "grad_norm": 1.526427984237671, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8713339567184448, + "num_tokens": 157267212.0, + "step": 4478 + }, + { + "epoch": 0.8317548746518106, + "grad_norm": 1.666540265083313, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8617404103279114, + "num_tokens": 157293918.0, + "step": 4479 + }, + { + "epoch": 0.8319405756731661, + "grad_norm": 1.4959540367126465, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8788197040557861, + "num_tokens": 157326942.0, + "step": 4480 + }, + { + "epoch": 0.8321262766945218, + "grad_norm": 1.449615240097046, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8717511892318726, + "num_tokens": 157362123.0, + "step": 4481 + }, + { + "epoch": 0.8323119777158774, + "grad_norm": 1.3619229793548584, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8724526762962341, + "num_tokens": 157401976.0, + "step": 4482 + }, + { + "epoch": 0.832497678737233, + "grad_norm": 1.4341689348220825, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8755013942718506, + "num_tokens": 157434769.0, + "step": 4483 + }, + { + "epoch": 0.8326833797585886, + "grad_norm": 1.380246639251709, + "learning_rate": 1e-06, + "loss": 0.4007, + "mean_token_accuracy": 0.866276204586029, + "num_tokens": 157472566.0, + "step": 4484 + }, + { + "epoch": 0.8328690807799443, + "grad_norm": 1.5730334520339966, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8612958788871765, + "num_tokens": 157503714.0, + "step": 4485 + }, + { + "epoch": 0.8330547818012999, + "grad_norm": 1.4714293479919434, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8817507028579712, + "num_tokens": 157534379.0, + "step": 4486 + }, + { + "epoch": 0.8332404828226555, + "grad_norm": 1.5364527702331543, + "learning_rate": 1e-06, + "loss": 0.4045, + "mean_token_accuracy": 0.8633732795715332, + "num_tokens": 157563934.0, + "step": 4487 + }, + { + "epoch": 0.8334261838440111, + "grad_norm": 1.4518767595291138, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8798855543136597, + "num_tokens": 157602398.0, + "step": 4488 + }, + { + "epoch": 0.8336118848653667, + "grad_norm": 1.3763625621795654, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8676410913467407, + "num_tokens": 157641820.0, + "step": 4489 + }, + { + "epoch": 0.8337975858867224, + "grad_norm": 1.3853039741516113, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8717555999755859, + "num_tokens": 157678319.0, + "step": 4490 + }, + { + "epoch": 0.833983286908078, + "grad_norm": 1.614864468574524, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8634036779403687, + "num_tokens": 157708692.0, + "step": 4491 + }, + { + "epoch": 0.8341689879294336, + "grad_norm": 1.5226351022720337, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8637144565582275, + "num_tokens": 157744538.0, + "step": 4492 + }, + { + "epoch": 0.8343546889507892, + "grad_norm": 1.49471116065979, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8744279146194458, + "num_tokens": 157778181.0, + "step": 4493 + }, + { + "epoch": 0.8345403899721449, + "grad_norm": 1.4432836771011353, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8714994788169861, + "num_tokens": 157809295.0, + "step": 4494 + }, + { + "epoch": 0.8347260909935005, + "grad_norm": 1.4594136476516724, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8665615320205688, + "num_tokens": 157844923.0, + "step": 4495 + }, + { + "epoch": 0.8349117920148561, + "grad_norm": 1.4632163047790527, + "learning_rate": 1e-06, + "loss": 0.4464, + "mean_token_accuracy": 0.8492079973220825, + "num_tokens": 157883270.0, + "step": 4496 + }, + { + "epoch": 0.8350974930362117, + "grad_norm": 1.4643279314041138, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8739925622940063, + "num_tokens": 157917125.0, + "step": 4497 + }, + { + "epoch": 0.8352831940575673, + "grad_norm": 1.407478928565979, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8773593306541443, + "num_tokens": 157952610.0, + "step": 4498 + }, + { + "epoch": 0.835468895078923, + "grad_norm": 1.3518853187561035, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.8699768781661987, + "num_tokens": 157991414.0, + "step": 4499 + }, + { + "epoch": 0.8356545961002786, + "grad_norm": 1.3536381721496582, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8636759519577026, + "num_tokens": 158028565.0, + "step": 4500 + }, + { + "epoch": 0.8358402971216342, + "grad_norm": 1.5582728385925293, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8660720586776733, + "num_tokens": 158058056.0, + "step": 4501 + }, + { + "epoch": 0.8360259981429898, + "grad_norm": 1.6323301792144775, + "learning_rate": 1e-06, + "loss": 0.4374, + "mean_token_accuracy": 0.8541072607040405, + "num_tokens": 158088109.0, + "step": 4502 + }, + { + "epoch": 0.8362116991643455, + "grad_norm": 1.4913753271102905, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8679664134979248, + "num_tokens": 158121427.0, + "step": 4503 + }, + { + "epoch": 0.836397400185701, + "grad_norm": 1.494083285331726, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8785080909729004, + "num_tokens": 158153658.0, + "step": 4504 + }, + { + "epoch": 0.8365831012070566, + "grad_norm": 1.469000220298767, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8834683895111084, + "num_tokens": 158183711.0, + "step": 4505 + }, + { + "epoch": 0.8367688022284122, + "grad_norm": 1.4741368293762207, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8682658672332764, + "num_tokens": 158217476.0, + "step": 4506 + }, + { + "epoch": 0.8369545032497678, + "grad_norm": 1.3581620454788208, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.88100665807724, + "num_tokens": 158254831.0, + "step": 4507 + }, + { + "epoch": 0.8371402042711235, + "grad_norm": 1.5117757320404053, + "learning_rate": 1e-06, + "loss": 0.4083, + "mean_token_accuracy": 0.8612488508224487, + "num_tokens": 158289926.0, + "step": 4508 + }, + { + "epoch": 0.8373259052924791, + "grad_norm": 1.485046625137329, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8544722199440002, + "num_tokens": 158325001.0, + "step": 4509 + }, + { + "epoch": 0.8375116063138347, + "grad_norm": 1.5113520622253418, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8718620538711548, + "num_tokens": 158354670.0, + "step": 4510 + }, + { + "epoch": 0.8376973073351903, + "grad_norm": 1.480517029762268, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8767070174217224, + "num_tokens": 158385966.0, + "step": 4511 + }, + { + "epoch": 0.8378830083565459, + "grad_norm": 1.5826910734176636, + "learning_rate": 1e-06, + "loss": 0.4172, + "mean_token_accuracy": 0.8648614883422852, + "num_tokens": 158416928.0, + "step": 4512 + }, + { + "epoch": 0.8380687093779016, + "grad_norm": 1.543745994567871, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8788245916366577, + "num_tokens": 158446176.0, + "step": 4513 + }, + { + "epoch": 0.8382544103992572, + "grad_norm": 1.5889966487884521, + "learning_rate": 1e-06, + "loss": 0.4421, + "mean_token_accuracy": 0.8497097492218018, + "num_tokens": 158477559.0, + "step": 4514 + }, + { + "epoch": 0.8384401114206128, + "grad_norm": 1.5167278051376343, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8641489148139954, + "num_tokens": 158514955.0, + "step": 4515 + }, + { + "epoch": 0.8386258124419684, + "grad_norm": 1.4923981428146362, + "learning_rate": 1e-06, + "loss": 0.4322, + "mean_token_accuracy": 0.8545559644699097, + "num_tokens": 158551389.0, + "step": 4516 + }, + { + "epoch": 0.838811513463324, + "grad_norm": 1.4395414590835571, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8654856085777283, + "num_tokens": 158586555.0, + "step": 4517 + }, + { + "epoch": 0.8389972144846797, + "grad_norm": 1.6199893951416016, + "learning_rate": 1e-06, + "loss": 0.4273, + "mean_token_accuracy": 0.8598426580429077, + "num_tokens": 158616268.0, + "step": 4518 + }, + { + "epoch": 0.8391829155060353, + "grad_norm": 1.3687435388565063, + "learning_rate": 1e-06, + "loss": 0.4096, + "mean_token_accuracy": 0.865814208984375, + "num_tokens": 158657396.0, + "step": 4519 + }, + { + "epoch": 0.8393686165273909, + "grad_norm": 1.4568806886672974, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8756235837936401, + "num_tokens": 158690960.0, + "step": 4520 + }, + { + "epoch": 0.8395543175487465, + "grad_norm": 1.4460453987121582, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8764055967330933, + "num_tokens": 158720813.0, + "step": 4521 + }, + { + "epoch": 0.8397400185701022, + "grad_norm": 1.6879342794418335, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8773020505905151, + "num_tokens": 158745611.0, + "step": 4522 + }, + { + "epoch": 0.8399257195914578, + "grad_norm": 1.4924153089523315, + "learning_rate": 1e-06, + "loss": 0.4305, + "mean_token_accuracy": 0.855461835861206, + "num_tokens": 158781425.0, + "step": 4523 + }, + { + "epoch": 0.8401114206128134, + "grad_norm": 1.3956114053726196, + "learning_rate": 1e-06, + "loss": 0.3964, + "mean_token_accuracy": 0.8620362877845764, + "num_tokens": 158818313.0, + "step": 4524 + }, + { + "epoch": 0.840297121634169, + "grad_norm": 1.4108129739761353, + "learning_rate": 1e-06, + "loss": 0.4102, + "mean_token_accuracy": 0.8630297183990479, + "num_tokens": 158860772.0, + "step": 4525 + }, + { + "epoch": 0.8404828226555247, + "grad_norm": 1.4277069568634033, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8707178831100464, + "num_tokens": 158897296.0, + "step": 4526 + }, + { + "epoch": 0.8406685236768803, + "grad_norm": 1.3997611999511719, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8852469325065613, + "num_tokens": 158932351.0, + "step": 4527 + }, + { + "epoch": 0.8408542246982358, + "grad_norm": 1.4636938571929932, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8765361905097961, + "num_tokens": 158964003.0, + "step": 4528 + }, + { + "epoch": 0.8410399257195914, + "grad_norm": 1.6341553926467896, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8670846223831177, + "num_tokens": 159002398.0, + "step": 4529 + }, + { + "epoch": 0.841225626740947, + "grad_norm": 1.6019278764724731, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8702957630157471, + "num_tokens": 159032194.0, + "step": 4530 + }, + { + "epoch": 0.8414113277623027, + "grad_norm": 1.6639155149459839, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8830106854438782, + "num_tokens": 159054696.0, + "step": 4531 + }, + { + "epoch": 0.8415970287836583, + "grad_norm": 1.4409172534942627, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.873690128326416, + "num_tokens": 159095258.0, + "step": 4532 + }, + { + "epoch": 0.8417827298050139, + "grad_norm": 1.5326249599456787, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8758399486541748, + "num_tokens": 159125907.0, + "step": 4533 + }, + { + "epoch": 0.8419684308263695, + "grad_norm": 1.307153344154358, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8745685815811157, + "num_tokens": 159168063.0, + "step": 4534 + }, + { + "epoch": 0.8421541318477251, + "grad_norm": 1.4632019996643066, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.870932936668396, + "num_tokens": 159198241.0, + "step": 4535 + }, + { + "epoch": 0.8423398328690808, + "grad_norm": 1.4829119443893433, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8711150884628296, + "num_tokens": 159234516.0, + "step": 4536 + }, + { + "epoch": 0.8425255338904364, + "grad_norm": 1.3787031173706055, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8699474334716797, + "num_tokens": 159273976.0, + "step": 4537 + }, + { + "epoch": 0.842711234911792, + "grad_norm": 1.4953564405441284, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8713232278823853, + "num_tokens": 159307099.0, + "step": 4538 + }, + { + "epoch": 0.8428969359331476, + "grad_norm": 1.6144226789474487, + "learning_rate": 1e-06, + "loss": 0.4403, + "mean_token_accuracy": 0.8535959720611572, + "num_tokens": 159342525.0, + "step": 4539 + }, + { + "epoch": 0.8430826369545033, + "grad_norm": 1.337264895439148, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8797381520271301, + "num_tokens": 159379816.0, + "step": 4540 + }, + { + "epoch": 0.8432683379758589, + "grad_norm": 1.5324466228485107, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.853455662727356, + "num_tokens": 159415753.0, + "step": 4541 + }, + { + "epoch": 0.8434540389972145, + "grad_norm": 1.5350819826126099, + "learning_rate": 1e-06, + "loss": 0.4621, + "mean_token_accuracy": 0.8439083099365234, + "num_tokens": 159449696.0, + "step": 4542 + }, + { + "epoch": 0.8436397400185701, + "grad_norm": 1.5126943588256836, + "learning_rate": 1e-06, + "loss": 0.4271, + "mean_token_accuracy": 0.857315182685852, + "num_tokens": 159487399.0, + "step": 4543 + }, + { + "epoch": 0.8438254410399257, + "grad_norm": 1.4668048620224, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8797028064727783, + "num_tokens": 159521078.0, + "step": 4544 + }, + { + "epoch": 0.8440111420612814, + "grad_norm": 1.6477264165878296, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8722490668296814, + "num_tokens": 159553759.0, + "step": 4545 + }, + { + "epoch": 0.844196843082637, + "grad_norm": 1.5819084644317627, + "learning_rate": 1e-06, + "loss": 0.4982, + "mean_token_accuracy": 0.8399105072021484, + "num_tokens": 159594028.0, + "step": 4546 + }, + { + "epoch": 0.8443825441039926, + "grad_norm": 1.3539354801177979, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8668508529663086, + "num_tokens": 159633644.0, + "step": 4547 + }, + { + "epoch": 0.8445682451253482, + "grad_norm": 1.5415846109390259, + "learning_rate": 1e-06, + "loss": 0.4283, + "mean_token_accuracy": 0.8601201772689819, + "num_tokens": 159665545.0, + "step": 4548 + }, + { + "epoch": 0.8447539461467038, + "grad_norm": 1.4939143657684326, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8825566172599792, + "num_tokens": 159693452.0, + "step": 4549 + }, + { + "epoch": 0.8449396471680595, + "grad_norm": 1.5426764488220215, + "learning_rate": 1e-06, + "loss": 0.439, + "mean_token_accuracy": 0.8542369604110718, + "num_tokens": 159728400.0, + "step": 4550 + }, + { + "epoch": 0.8451253481894151, + "grad_norm": 1.4105688333511353, + "learning_rate": 1e-06, + "loss": 0.4275, + "mean_token_accuracy": 0.8552511930465698, + "num_tokens": 159768628.0, + "step": 4551 + }, + { + "epoch": 0.8453110492107706, + "grad_norm": 1.3912590742111206, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8732092380523682, + "num_tokens": 159806104.0, + "step": 4552 + }, + { + "epoch": 0.8454967502321262, + "grad_norm": 1.4167841672897339, + "learning_rate": 1e-06, + "loss": 0.4258, + "mean_token_accuracy": 0.8611109256744385, + "num_tokens": 159844779.0, + "step": 4553 + }, + { + "epoch": 0.8456824512534818, + "grad_norm": 1.6141247749328613, + "learning_rate": 1e-06, + "loss": 0.4185, + "mean_token_accuracy": 0.8601219058036804, + "num_tokens": 159874675.0, + "step": 4554 + }, + { + "epoch": 0.8458681522748375, + "grad_norm": 1.4602335691452026, + "learning_rate": 1e-06, + "loss": 0.3966, + "mean_token_accuracy": 0.8619397878646851, + "num_tokens": 159909278.0, + "step": 4555 + }, + { + "epoch": 0.8460538532961931, + "grad_norm": 1.4599730968475342, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8651276230812073, + "num_tokens": 159945181.0, + "step": 4556 + }, + { + "epoch": 0.8462395543175487, + "grad_norm": 1.5010286569595337, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8717650175094604, + "num_tokens": 159975498.0, + "step": 4557 + }, + { + "epoch": 0.8464252553389043, + "grad_norm": 1.4528677463531494, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8805110454559326, + "num_tokens": 160007586.0, + "step": 4558 + }, + { + "epoch": 0.84661095636026, + "grad_norm": 1.5430302619934082, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.858748733997345, + "num_tokens": 160045592.0, + "step": 4559 + }, + { + "epoch": 0.8467966573816156, + "grad_norm": 1.464764952659607, + "learning_rate": 1e-06, + "loss": 0.4537, + "mean_token_accuracy": 0.8501055240631104, + "num_tokens": 160084708.0, + "step": 4560 + }, + { + "epoch": 0.8469823584029712, + "grad_norm": 1.6101571321487427, + "learning_rate": 1e-06, + "loss": 0.4622, + "mean_token_accuracy": 0.8443412780761719, + "num_tokens": 160117584.0, + "step": 4561 + }, + { + "epoch": 0.8471680594243268, + "grad_norm": 1.395214557647705, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8688726425170898, + "num_tokens": 160157119.0, + "step": 4562 + }, + { + "epoch": 0.8473537604456824, + "grad_norm": 1.3471145629882812, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8805049657821655, + "num_tokens": 160193832.0, + "step": 4563 + }, + { + "epoch": 0.8475394614670381, + "grad_norm": 1.3504571914672852, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8691055774688721, + "num_tokens": 160235298.0, + "step": 4564 + }, + { + "epoch": 0.8477251624883937, + "grad_norm": 1.5788815021514893, + "learning_rate": 1e-06, + "loss": 0.4121, + "mean_token_accuracy": 0.8631076812744141, + "num_tokens": 160264770.0, + "step": 4565 + }, + { + "epoch": 0.8479108635097493, + "grad_norm": 1.4539830684661865, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8653757572174072, + "num_tokens": 160298651.0, + "step": 4566 + }, + { + "epoch": 0.8480965645311049, + "grad_norm": 1.5406886339187622, + "learning_rate": 1e-06, + "loss": 0.4352, + "mean_token_accuracy": 0.8566300868988037, + "num_tokens": 160334933.0, + "step": 4567 + }, + { + "epoch": 0.8482822655524606, + "grad_norm": 1.5147124528884888, + "learning_rate": 1e-06, + "loss": 0.4805, + "mean_token_accuracy": 0.8409689664840698, + "num_tokens": 160371790.0, + "step": 4568 + }, + { + "epoch": 0.8484679665738162, + "grad_norm": 1.455092191696167, + "learning_rate": 1e-06, + "loss": 0.4219, + "mean_token_accuracy": 0.8592526316642761, + "num_tokens": 160407719.0, + "step": 4569 + }, + { + "epoch": 0.8486536675951718, + "grad_norm": 1.4611634016036987, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8819177150726318, + "num_tokens": 160438797.0, + "step": 4570 + }, + { + "epoch": 0.8488393686165274, + "grad_norm": 1.4527398347854614, + "learning_rate": 1e-06, + "loss": 0.4453, + "mean_token_accuracy": 0.8548319935798645, + "num_tokens": 160477822.0, + "step": 4571 + }, + { + "epoch": 0.849025069637883, + "grad_norm": 1.352391242980957, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.8658139705657959, + "num_tokens": 160516753.0, + "step": 4572 + }, + { + "epoch": 0.8492107706592387, + "grad_norm": 1.389193058013916, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8787838816642761, + "num_tokens": 160550360.0, + "step": 4573 + }, + { + "epoch": 0.8493964716805943, + "grad_norm": 1.4735143184661865, + "learning_rate": 1e-06, + "loss": 0.4219, + "mean_token_accuracy": 0.8611718416213989, + "num_tokens": 160586361.0, + "step": 4574 + }, + { + "epoch": 0.8495821727019499, + "grad_norm": 1.5756409168243408, + "learning_rate": 1e-06, + "loss": 0.4401, + "mean_token_accuracy": 0.851931631565094, + "num_tokens": 160619049.0, + "step": 4575 + }, + { + "epoch": 0.8497678737233055, + "grad_norm": 1.3197124004364014, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8827530741691589, + "num_tokens": 160655510.0, + "step": 4576 + }, + { + "epoch": 0.849953574744661, + "grad_norm": 1.6237316131591797, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8675010204315186, + "num_tokens": 160687644.0, + "step": 4577 + }, + { + "epoch": 0.8501392757660167, + "grad_norm": 1.3387608528137207, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8878269195556641, + "num_tokens": 160724866.0, + "step": 4578 + }, + { + "epoch": 0.8503249767873723, + "grad_norm": 1.5750001668930054, + "learning_rate": 1e-06, + "loss": 0.4539, + "mean_token_accuracy": 0.8507932424545288, + "num_tokens": 160755818.0, + "step": 4579 + }, + { + "epoch": 0.8505106778087279, + "grad_norm": 1.4190791845321655, + "learning_rate": 1e-06, + "loss": 0.4349, + "mean_token_accuracy": 0.8565270900726318, + "num_tokens": 160795146.0, + "step": 4580 + }, + { + "epoch": 0.8506963788300835, + "grad_norm": 1.4273021221160889, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8662580251693726, + "num_tokens": 160832379.0, + "step": 4581 + }, + { + "epoch": 0.8508820798514392, + "grad_norm": 1.7824926376342773, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8664267063140869, + "num_tokens": 160865458.0, + "step": 4582 + }, + { + "epoch": 0.8510677808727948, + "grad_norm": 1.799908995628357, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8628247380256653, + "num_tokens": 160889005.0, + "step": 4583 + }, + { + "epoch": 0.8512534818941504, + "grad_norm": 1.4120633602142334, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8666867017745972, + "num_tokens": 160930254.0, + "step": 4584 + }, + { + "epoch": 0.851439182915506, + "grad_norm": 1.3652197122573853, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8804128170013428, + "num_tokens": 160965088.0, + "step": 4585 + }, + { + "epoch": 0.8516248839368616, + "grad_norm": 1.3934446573257446, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8616628646850586, + "num_tokens": 161005429.0, + "step": 4586 + }, + { + "epoch": 0.8518105849582173, + "grad_norm": 1.3706709146499634, + "learning_rate": 1e-06, + "loss": 0.4115, + "mean_token_accuracy": 0.861229658126831, + "num_tokens": 161046304.0, + "step": 4587 + }, + { + "epoch": 0.8519962859795729, + "grad_norm": 1.3976126909255981, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8769164085388184, + "num_tokens": 161081322.0, + "step": 4588 + }, + { + "epoch": 0.8521819870009285, + "grad_norm": 1.35893976688385, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8728957176208496, + "num_tokens": 161118013.0, + "step": 4589 + }, + { + "epoch": 0.8523676880222841, + "grad_norm": 1.5687769651412964, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8778938055038452, + "num_tokens": 161144241.0, + "step": 4590 + }, + { + "epoch": 0.8525533890436398, + "grad_norm": 1.4127737283706665, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8686709403991699, + "num_tokens": 161184154.0, + "step": 4591 + }, + { + "epoch": 0.8527390900649954, + "grad_norm": 1.4521516561508179, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.8590152263641357, + "num_tokens": 161221019.0, + "step": 4592 + }, + { + "epoch": 0.852924791086351, + "grad_norm": 1.512271523475647, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8737636804580688, + "num_tokens": 161253824.0, + "step": 4593 + }, + { + "epoch": 0.8531104921077066, + "grad_norm": 1.5006262063980103, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8714276552200317, + "num_tokens": 161286299.0, + "step": 4594 + }, + { + "epoch": 0.8532961931290622, + "grad_norm": 1.5408155918121338, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8673433065414429, + "num_tokens": 161317177.0, + "step": 4595 + }, + { + "epoch": 0.8534818941504179, + "grad_norm": 1.3815244436264038, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8776899576187134, + "num_tokens": 161354110.0, + "step": 4596 + }, + { + "epoch": 0.8536675951717735, + "grad_norm": 1.5443236827850342, + "learning_rate": 1e-06, + "loss": 0.3958, + "mean_token_accuracy": 0.8674078583717346, + "num_tokens": 161384202.0, + "step": 4597 + }, + { + "epoch": 0.8538532961931291, + "grad_norm": 1.323679804801941, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8727337121963501, + "num_tokens": 161421932.0, + "step": 4598 + }, + { + "epoch": 0.8540389972144847, + "grad_norm": 1.485488772392273, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.879284679889679, + "num_tokens": 161453084.0, + "step": 4599 + }, + { + "epoch": 0.8542246982358404, + "grad_norm": 1.4379385709762573, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8745196461677551, + "num_tokens": 161487189.0, + "step": 4600 + }, + { + "epoch": 0.8544103992571959, + "grad_norm": 1.497646689414978, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8744251728057861, + "num_tokens": 161520881.0, + "step": 4601 + }, + { + "epoch": 0.8545961002785515, + "grad_norm": 1.5874923467636108, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8581312894821167, + "num_tokens": 161552564.0, + "step": 4602 + }, + { + "epoch": 0.8547818012999071, + "grad_norm": 1.452324390411377, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8704233765602112, + "num_tokens": 161584301.0, + "step": 4603 + }, + { + "epoch": 0.8549675023212627, + "grad_norm": 1.4259809255599976, + "learning_rate": 1e-06, + "loss": 0.4417, + "mean_token_accuracy": 0.8545448184013367, + "num_tokens": 161624158.0, + "step": 4604 + }, + { + "epoch": 0.8551532033426184, + "grad_norm": 1.3391262292861938, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8673586845397949, + "num_tokens": 161664004.0, + "step": 4605 + }, + { + "epoch": 0.855338904363974, + "grad_norm": 1.4450627565383911, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.866050660610199, + "num_tokens": 161702555.0, + "step": 4606 + }, + { + "epoch": 0.8555246053853296, + "grad_norm": 1.4556912183761597, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8627572059631348, + "num_tokens": 161738023.0, + "step": 4607 + }, + { + "epoch": 0.8557103064066852, + "grad_norm": 1.53220534324646, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8743526935577393, + "num_tokens": 161768625.0, + "step": 4608 + }, + { + "epoch": 0.8558960074280408, + "grad_norm": 1.554979681968689, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8852477073669434, + "num_tokens": 161799014.0, + "step": 4609 + }, + { + "epoch": 0.8560817084493965, + "grad_norm": 1.4057393074035645, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8804247975349426, + "num_tokens": 161835515.0, + "step": 4610 + }, + { + "epoch": 0.8562674094707521, + "grad_norm": 1.4171171188354492, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8687861561775208, + "num_tokens": 161872747.0, + "step": 4611 + }, + { + "epoch": 0.8564531104921077, + "grad_norm": 1.3670642375946045, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8645421862602234, + "num_tokens": 161910524.0, + "step": 4612 + }, + { + "epoch": 0.8566388115134633, + "grad_norm": 1.457801103591919, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8814681768417358, + "num_tokens": 161942434.0, + "step": 4613 + }, + { + "epoch": 0.856824512534819, + "grad_norm": 1.543992042541504, + "learning_rate": 1e-06, + "loss": 0.4303, + "mean_token_accuracy": 0.8541833162307739, + "num_tokens": 161979529.0, + "step": 4614 + }, + { + "epoch": 0.8570102135561746, + "grad_norm": 1.41824209690094, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.870790421962738, + "num_tokens": 162018560.0, + "step": 4615 + }, + { + "epoch": 0.8571959145775302, + "grad_norm": 1.5447732210159302, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.8708246946334839, + "num_tokens": 162050643.0, + "step": 4616 + }, + { + "epoch": 0.8573816155988858, + "grad_norm": 1.3935770988464355, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8804188370704651, + "num_tokens": 162087054.0, + "step": 4617 + }, + { + "epoch": 0.8575673166202414, + "grad_norm": 1.3774648904800415, + "learning_rate": 1e-06, + "loss": 0.4049, + "mean_token_accuracy": 0.865247368812561, + "num_tokens": 162127489.0, + "step": 4618 + }, + { + "epoch": 0.8577530176415971, + "grad_norm": 1.6237210035324097, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8667001128196716, + "num_tokens": 162161497.0, + "step": 4619 + }, + { + "epoch": 0.8579387186629527, + "grad_norm": 1.5002851486206055, + "learning_rate": 1e-06, + "loss": 0.4712, + "mean_token_accuracy": 0.8396521210670471, + "num_tokens": 162199924.0, + "step": 4620 + }, + { + "epoch": 0.8581244196843083, + "grad_norm": 1.4192684888839722, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8745238780975342, + "num_tokens": 162235677.0, + "step": 4621 + }, + { + "epoch": 0.8583101207056639, + "grad_norm": 1.3723878860473633, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8712165355682373, + "num_tokens": 162274202.0, + "step": 4622 + }, + { + "epoch": 0.8584958217270195, + "grad_norm": 1.407332420349121, + "learning_rate": 1e-06, + "loss": 0.4129, + "mean_token_accuracy": 0.8637993335723877, + "num_tokens": 162314575.0, + "step": 4623 + }, + { + "epoch": 0.8586815227483752, + "grad_norm": 1.65924072265625, + "learning_rate": 1e-06, + "loss": 0.4394, + "mean_token_accuracy": 0.8521211743354797, + "num_tokens": 162344415.0, + "step": 4624 + }, + { + "epoch": 0.8588672237697307, + "grad_norm": 1.4931679964065552, + "learning_rate": 1e-06, + "loss": 0.4302, + "mean_token_accuracy": 0.8538749814033508, + "num_tokens": 162381559.0, + "step": 4625 + }, + { + "epoch": 0.8590529247910863, + "grad_norm": 1.4874709844589233, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8754981756210327, + "num_tokens": 162416583.0, + "step": 4626 + }, + { + "epoch": 0.8592386258124419, + "grad_norm": 1.4193682670593262, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8693976998329163, + "num_tokens": 162453536.0, + "step": 4627 + }, + { + "epoch": 0.8594243268337975, + "grad_norm": 1.3942186832427979, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.8674589395523071, + "num_tokens": 162492572.0, + "step": 4628 + }, + { + "epoch": 0.8596100278551532, + "grad_norm": 1.6407077312469482, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.865161120891571, + "num_tokens": 162521519.0, + "step": 4629 + }, + { + "epoch": 0.8597957288765088, + "grad_norm": 1.394739031791687, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8692684769630432, + "num_tokens": 162556360.0, + "step": 4630 + }, + { + "epoch": 0.8599814298978644, + "grad_norm": 1.3771427869796753, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8740612268447876, + "num_tokens": 162591475.0, + "step": 4631 + }, + { + "epoch": 0.86016713091922, + "grad_norm": 1.4669253826141357, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8683004975318909, + "num_tokens": 162626317.0, + "step": 4632 + }, + { + "epoch": 0.8603528319405757, + "grad_norm": 1.4282212257385254, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8704911470413208, + "num_tokens": 162662090.0, + "step": 4633 + }, + { + "epoch": 0.8605385329619313, + "grad_norm": 1.396040678024292, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8846660852432251, + "num_tokens": 162694460.0, + "step": 4634 + }, + { + "epoch": 0.8607242339832869, + "grad_norm": 1.4090055227279663, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8589483499526978, + "num_tokens": 162731207.0, + "step": 4635 + }, + { + "epoch": 0.8609099350046425, + "grad_norm": 1.485068678855896, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8637884855270386, + "num_tokens": 162767782.0, + "step": 4636 + }, + { + "epoch": 0.8610956360259981, + "grad_norm": 1.6268999576568604, + "learning_rate": 1e-06, + "loss": 0.4462, + "mean_token_accuracy": 0.8523966670036316, + "num_tokens": 162803055.0, + "step": 4637 + }, + { + "epoch": 0.8612813370473538, + "grad_norm": 1.3662657737731934, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8712368011474609, + "num_tokens": 162842209.0, + "step": 4638 + }, + { + "epoch": 0.8614670380687094, + "grad_norm": 1.3802168369293213, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8760459423065186, + "num_tokens": 162880504.0, + "step": 4639 + }, + { + "epoch": 0.861652739090065, + "grad_norm": 1.3987066745758057, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8907032012939453, + "num_tokens": 162915932.0, + "step": 4640 + }, + { + "epoch": 0.8618384401114206, + "grad_norm": 1.3871556520462036, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8721531629562378, + "num_tokens": 162954855.0, + "step": 4641 + }, + { + "epoch": 0.8620241411327763, + "grad_norm": 1.509688377380371, + "learning_rate": 1e-06, + "loss": 0.4434, + "mean_token_accuracy": 0.8563363552093506, + "num_tokens": 162993051.0, + "step": 4642 + }, + { + "epoch": 0.8622098421541319, + "grad_norm": 1.5733307600021362, + "learning_rate": 1e-06, + "loss": 0.4497, + "mean_token_accuracy": 0.855318009853363, + "num_tokens": 163030492.0, + "step": 4643 + }, + { + "epoch": 0.8623955431754875, + "grad_norm": 1.5571379661560059, + "learning_rate": 1e-06, + "loss": 0.4518, + "mean_token_accuracy": 0.850037157535553, + "num_tokens": 163068073.0, + "step": 4644 + }, + { + "epoch": 0.8625812441968431, + "grad_norm": 1.5312501192092896, + "learning_rate": 1e-06, + "loss": 0.4244, + "mean_token_accuracy": 0.8562010526657104, + "num_tokens": 163103078.0, + "step": 4645 + }, + { + "epoch": 0.8627669452181987, + "grad_norm": 1.3221379518508911, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.875573456287384, + "num_tokens": 163138600.0, + "step": 4646 + }, + { + "epoch": 0.8629526462395544, + "grad_norm": 1.3167279958724976, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8736100196838379, + "num_tokens": 163178273.0, + "step": 4647 + }, + { + "epoch": 0.86313834726091, + "grad_norm": 1.4752777814865112, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8717114925384521, + "num_tokens": 163208655.0, + "step": 4648 + }, + { + "epoch": 0.8633240482822655, + "grad_norm": 1.371476411819458, + "learning_rate": 1e-06, + "loss": 0.4186, + "mean_token_accuracy": 0.8571078181266785, + "num_tokens": 163250406.0, + "step": 4649 + }, + { + "epoch": 0.8635097493036211, + "grad_norm": 1.4045450687408447, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.873879075050354, + "num_tokens": 163286893.0, + "step": 4650 + }, + { + "epoch": 0.8636954503249767, + "grad_norm": 1.470354437828064, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8665114045143127, + "num_tokens": 163320994.0, + "step": 4651 + }, + { + "epoch": 0.8638811513463324, + "grad_norm": 1.5960065126419067, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8724935054779053, + "num_tokens": 163348760.0, + "step": 4652 + }, + { + "epoch": 0.864066852367688, + "grad_norm": 1.4446054697036743, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8618456125259399, + "num_tokens": 163385066.0, + "step": 4653 + }, + { + "epoch": 0.8642525533890436, + "grad_norm": 1.5004115104675293, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8660337924957275, + "num_tokens": 163416573.0, + "step": 4654 + }, + { + "epoch": 0.8644382544103992, + "grad_norm": 1.4185731410980225, + "learning_rate": 1e-06, + "loss": 0.4358, + "mean_token_accuracy": 0.8560993671417236, + "num_tokens": 163459357.0, + "step": 4655 + }, + { + "epoch": 0.8646239554317549, + "grad_norm": 1.5384517908096313, + "learning_rate": 1e-06, + "loss": 0.41, + "mean_token_accuracy": 0.8619202375411987, + "num_tokens": 163491432.0, + "step": 4656 + }, + { + "epoch": 0.8648096564531105, + "grad_norm": 1.3935630321502686, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.864203155040741, + "num_tokens": 163530853.0, + "step": 4657 + }, + { + "epoch": 0.8649953574744661, + "grad_norm": 1.379319190979004, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8845236301422119, + "num_tokens": 163568516.0, + "step": 4658 + }, + { + "epoch": 0.8651810584958217, + "grad_norm": 1.4649686813354492, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8728392720222473, + "num_tokens": 163602309.0, + "step": 4659 + }, + { + "epoch": 0.8653667595171773, + "grad_norm": 1.3839136362075806, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8615486025810242, + "num_tokens": 163642826.0, + "step": 4660 + }, + { + "epoch": 0.865552460538533, + "grad_norm": 1.5597801208496094, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8824589252471924, + "num_tokens": 163670382.0, + "step": 4661 + }, + { + "epoch": 0.8657381615598886, + "grad_norm": 1.5733343362808228, + "learning_rate": 1e-06, + "loss": 0.4344, + "mean_token_accuracy": 0.8573125600814819, + "num_tokens": 163704404.0, + "step": 4662 + }, + { + "epoch": 0.8659238625812442, + "grad_norm": 1.2704135179519653, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8747692108154297, + "num_tokens": 163744568.0, + "step": 4663 + }, + { + "epoch": 0.8661095636025998, + "grad_norm": 1.3673055171966553, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.8606101274490356, + "num_tokens": 163783510.0, + "step": 4664 + }, + { + "epoch": 0.8662952646239555, + "grad_norm": 1.556307315826416, + "learning_rate": 1e-06, + "loss": 0.3929, + "mean_token_accuracy": 0.8668880462646484, + "num_tokens": 163814174.0, + "step": 4665 + }, + { + "epoch": 0.8664809656453111, + "grad_norm": 1.4140784740447998, + "learning_rate": 1e-06, + "loss": 0.3964, + "mean_token_accuracy": 0.8657386302947998, + "num_tokens": 163851885.0, + "step": 4666 + }, + { + "epoch": 0.8666666666666667, + "grad_norm": 1.4373959302902222, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8839154839515686, + "num_tokens": 163886047.0, + "step": 4667 + }, + { + "epoch": 0.8668523676880223, + "grad_norm": 1.4470748901367188, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8720129728317261, + "num_tokens": 163917359.0, + "step": 4668 + }, + { + "epoch": 0.8670380687093779, + "grad_norm": 1.5106267929077148, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8804579973220825, + "num_tokens": 163947205.0, + "step": 4669 + }, + { + "epoch": 0.8672237697307336, + "grad_norm": 1.4346033334732056, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8782879114151001, + "num_tokens": 163983744.0, + "step": 4670 + }, + { + "epoch": 0.8674094707520892, + "grad_norm": 1.449357271194458, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8637053370475769, + "num_tokens": 164019502.0, + "step": 4671 + }, + { + "epoch": 0.8675951717734448, + "grad_norm": 1.4192389249801636, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8808640837669373, + "num_tokens": 164058289.0, + "step": 4672 + }, + { + "epoch": 0.8677808727948003, + "grad_norm": 1.3864712715148926, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8757816553115845, + "num_tokens": 164098708.0, + "step": 4673 + }, + { + "epoch": 0.8679665738161559, + "grad_norm": 1.3689754009246826, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8738721609115601, + "num_tokens": 164136890.0, + "step": 4674 + }, + { + "epoch": 0.8681522748375116, + "grad_norm": 1.5420829057693481, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8763160705566406, + "num_tokens": 164166671.0, + "step": 4675 + }, + { + "epoch": 0.8683379758588672, + "grad_norm": 1.3927741050720215, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8630715608596802, + "num_tokens": 164205427.0, + "step": 4676 + }, + { + "epoch": 0.8685236768802228, + "grad_norm": 1.6039432287216187, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8786047697067261, + "num_tokens": 164237312.0, + "step": 4677 + }, + { + "epoch": 0.8687093779015784, + "grad_norm": 1.486438512802124, + "learning_rate": 1e-06, + "loss": 0.4233, + "mean_token_accuracy": 0.858014702796936, + "num_tokens": 164274036.0, + "step": 4678 + }, + { + "epoch": 0.868895078922934, + "grad_norm": 1.4361462593078613, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.876993715763092, + "num_tokens": 164307667.0, + "step": 4679 + }, + { + "epoch": 0.8690807799442897, + "grad_norm": 1.4014275074005127, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8631022572517395, + "num_tokens": 164346077.0, + "step": 4680 + }, + { + "epoch": 0.8692664809656453, + "grad_norm": 1.3726836442947388, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8658239245414734, + "num_tokens": 164384797.0, + "step": 4681 + }, + { + "epoch": 0.8694521819870009, + "grad_norm": 1.4374611377716064, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8638213276863098, + "num_tokens": 164421713.0, + "step": 4682 + }, + { + "epoch": 0.8696378830083565, + "grad_norm": 1.302642583847046, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8684841990470886, + "num_tokens": 164463658.0, + "step": 4683 + }, + { + "epoch": 0.8698235840297122, + "grad_norm": 1.4523574113845825, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8665491342544556, + "num_tokens": 164501664.0, + "step": 4684 + }, + { + "epoch": 0.8700092850510678, + "grad_norm": 1.5833059549331665, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8745133876800537, + "num_tokens": 164533837.0, + "step": 4685 + }, + { + "epoch": 0.8701949860724234, + "grad_norm": 1.5713971853256226, + "learning_rate": 1e-06, + "loss": 0.4384, + "mean_token_accuracy": 0.8580344319343567, + "num_tokens": 164565690.0, + "step": 4686 + }, + { + "epoch": 0.870380687093779, + "grad_norm": 1.3398813009262085, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8610277771949768, + "num_tokens": 164607866.0, + "step": 4687 + }, + { + "epoch": 0.8705663881151346, + "grad_norm": 1.6267999410629272, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.860284686088562, + "num_tokens": 164637881.0, + "step": 4688 + }, + { + "epoch": 0.8707520891364903, + "grad_norm": 1.4934731721878052, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8824485540390015, + "num_tokens": 164671074.0, + "step": 4689 + }, + { + "epoch": 0.8709377901578459, + "grad_norm": 1.4654719829559326, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8707285523414612, + "num_tokens": 164705744.0, + "step": 4690 + }, + { + "epoch": 0.8711234911792015, + "grad_norm": 1.5631684064865112, + "learning_rate": 1e-06, + "loss": 0.4215, + "mean_token_accuracy": 0.85630202293396, + "num_tokens": 164740290.0, + "step": 4691 + }, + { + "epoch": 0.8713091922005571, + "grad_norm": 1.4089643955230713, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8804112672805786, + "num_tokens": 164774736.0, + "step": 4692 + }, + { + "epoch": 0.8714948932219128, + "grad_norm": 1.4154577255249023, + "learning_rate": 1e-06, + "loss": 0.4194, + "mean_token_accuracy": 0.8569228053092957, + "num_tokens": 164815772.0, + "step": 4693 + }, + { + "epoch": 0.8716805942432684, + "grad_norm": 1.484598994255066, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8766342997550964, + "num_tokens": 164848926.0, + "step": 4694 + }, + { + "epoch": 0.871866295264624, + "grad_norm": 1.5153272151947021, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.8663976192474365, + "num_tokens": 164882131.0, + "step": 4695 + }, + { + "epoch": 0.8720519962859796, + "grad_norm": 1.679890513420105, + "learning_rate": 1e-06, + "loss": 0.4297, + "mean_token_accuracy": 0.8566983938217163, + "num_tokens": 164909118.0, + "step": 4696 + }, + { + "epoch": 0.8722376973073351, + "grad_norm": 1.5592858791351318, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.867792010307312, + "num_tokens": 164945147.0, + "step": 4697 + }, + { + "epoch": 0.8724233983286908, + "grad_norm": 1.7291542291641235, + "learning_rate": 1e-06, + "loss": 0.4058, + "mean_token_accuracy": 0.8629997968673706, + "num_tokens": 164972609.0, + "step": 4698 + }, + { + "epoch": 0.8726090993500464, + "grad_norm": 1.4783315658569336, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8697714805603027, + "num_tokens": 165004934.0, + "step": 4699 + }, + { + "epoch": 0.872794800371402, + "grad_norm": 1.5433036088943481, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8735402822494507, + "num_tokens": 165036398.0, + "step": 4700 + }, + { + "epoch": 0.8729805013927576, + "grad_norm": 1.5129070281982422, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.8690729737281799, + "num_tokens": 165069504.0, + "step": 4701 + }, + { + "epoch": 0.8731662024141132, + "grad_norm": 1.3804221153259277, + "learning_rate": 1e-06, + "loss": 0.4289, + "mean_token_accuracy": 0.8576400279998779, + "num_tokens": 165113092.0, + "step": 4702 + }, + { + "epoch": 0.8733519034354689, + "grad_norm": 1.320908546447754, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8671631813049316, + "num_tokens": 165156467.0, + "step": 4703 + }, + { + "epoch": 0.8735376044568245, + "grad_norm": 1.4529876708984375, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8614965081214905, + "num_tokens": 165197139.0, + "step": 4704 + }, + { + "epoch": 0.8737233054781801, + "grad_norm": 1.580573320388794, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8671744465827942, + "num_tokens": 165227630.0, + "step": 4705 + }, + { + "epoch": 0.8739090064995357, + "grad_norm": 1.4473369121551514, + "learning_rate": 1e-06, + "loss": 0.4295, + "mean_token_accuracy": 0.8565014600753784, + "num_tokens": 165266827.0, + "step": 4706 + }, + { + "epoch": 0.8740947075208914, + "grad_norm": 1.5507575273513794, + "learning_rate": 1e-06, + "loss": 0.432, + "mean_token_accuracy": 0.8585197925567627, + "num_tokens": 165302912.0, + "step": 4707 + }, + { + "epoch": 0.874280408542247, + "grad_norm": 1.303067684173584, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8804718852043152, + "num_tokens": 165342418.0, + "step": 4708 + }, + { + "epoch": 0.8744661095636026, + "grad_norm": 1.430674433708191, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.8711003065109253, + "num_tokens": 165378512.0, + "step": 4709 + }, + { + "epoch": 0.8746518105849582, + "grad_norm": 1.577439308166504, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.8597962260246277, + "num_tokens": 165406646.0, + "step": 4710 + }, + { + "epoch": 0.8748375116063138, + "grad_norm": 1.4386979341506958, + "learning_rate": 1e-06, + "loss": 0.4295, + "mean_token_accuracy": 0.856781542301178, + "num_tokens": 165443894.0, + "step": 4711 + }, + { + "epoch": 0.8750232126276695, + "grad_norm": 1.3835164308547974, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8684084415435791, + "num_tokens": 165481915.0, + "step": 4712 + }, + { + "epoch": 0.8752089136490251, + "grad_norm": 1.506327748298645, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8747967481613159, + "num_tokens": 165510814.0, + "step": 4713 + }, + { + "epoch": 0.8753946146703807, + "grad_norm": 1.4095839262008667, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8734213709831238, + "num_tokens": 165548422.0, + "step": 4714 + }, + { + "epoch": 0.8755803156917363, + "grad_norm": 1.480114459991455, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.8671435117721558, + "num_tokens": 165579449.0, + "step": 4715 + }, + { + "epoch": 0.875766016713092, + "grad_norm": 1.3926326036453247, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8665111064910889, + "num_tokens": 165616765.0, + "step": 4716 + }, + { + "epoch": 0.8759517177344476, + "grad_norm": 1.3037586212158203, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8695001602172852, + "num_tokens": 165656688.0, + "step": 4717 + }, + { + "epoch": 0.8761374187558032, + "grad_norm": 1.433032512664795, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8806073665618896, + "num_tokens": 165688650.0, + "step": 4718 + }, + { + "epoch": 0.8763231197771588, + "grad_norm": 1.457322359085083, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.8575977087020874, + "num_tokens": 165724346.0, + "step": 4719 + }, + { + "epoch": 0.8765088207985144, + "grad_norm": 1.283337950706482, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8811851739883423, + "num_tokens": 165765664.0, + "step": 4720 + }, + { + "epoch": 0.87669452181987, + "grad_norm": 1.4478893280029297, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8682763576507568, + "num_tokens": 165801915.0, + "step": 4721 + }, + { + "epoch": 0.8768802228412256, + "grad_norm": 1.4071882963180542, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8784762024879456, + "num_tokens": 165837493.0, + "step": 4722 + }, + { + "epoch": 0.8770659238625812, + "grad_norm": 1.6483100652694702, + "learning_rate": 1e-06, + "loss": 0.4624, + "mean_token_accuracy": 0.8507354259490967, + "num_tokens": 165867820.0, + "step": 4723 + }, + { + "epoch": 0.8772516248839368, + "grad_norm": 1.4164499044418335, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8699512481689453, + "num_tokens": 165904966.0, + "step": 4724 + }, + { + "epoch": 0.8774373259052924, + "grad_norm": 1.3338335752487183, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8676736950874329, + "num_tokens": 165944760.0, + "step": 4725 + }, + { + "epoch": 0.8776230269266481, + "grad_norm": 1.3865785598754883, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8704304695129395, + "num_tokens": 165980913.0, + "step": 4726 + }, + { + "epoch": 0.8778087279480037, + "grad_norm": 1.5044463872909546, + "learning_rate": 1e-06, + "loss": 0.4184, + "mean_token_accuracy": 0.8617592453956604, + "num_tokens": 166013229.0, + "step": 4727 + }, + { + "epoch": 0.8779944289693593, + "grad_norm": 1.2367948293685913, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.876186728477478, + "num_tokens": 166057600.0, + "step": 4728 + }, + { + "epoch": 0.8781801299907149, + "grad_norm": 1.63838529586792, + "learning_rate": 1e-06, + "loss": 0.4724, + "mean_token_accuracy": 0.8494746685028076, + "num_tokens": 166089100.0, + "step": 4729 + }, + { + "epoch": 0.8783658310120706, + "grad_norm": 1.4929320812225342, + "learning_rate": 1e-06, + "loss": 0.4195, + "mean_token_accuracy": 0.8587639927864075, + "num_tokens": 166127546.0, + "step": 4730 + }, + { + "epoch": 0.8785515320334262, + "grad_norm": 1.5515108108520508, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8644477725028992, + "num_tokens": 166158302.0, + "step": 4731 + }, + { + "epoch": 0.8787372330547818, + "grad_norm": 1.4742071628570557, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8663758635520935, + "num_tokens": 166192256.0, + "step": 4732 + }, + { + "epoch": 0.8789229340761374, + "grad_norm": 1.5011193752288818, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8658961057662964, + "num_tokens": 166225442.0, + "step": 4733 + }, + { + "epoch": 0.879108635097493, + "grad_norm": 1.30973219871521, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8769066333770752, + "num_tokens": 166265279.0, + "step": 4734 + }, + { + "epoch": 0.8792943361188487, + "grad_norm": 1.4526395797729492, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8650327324867249, + "num_tokens": 166301554.0, + "step": 4735 + }, + { + "epoch": 0.8794800371402043, + "grad_norm": 1.4715992212295532, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8698975443840027, + "num_tokens": 166334216.0, + "step": 4736 + }, + { + "epoch": 0.8796657381615599, + "grad_norm": 1.714213252067566, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8648502826690674, + "num_tokens": 166362709.0, + "step": 4737 + }, + { + "epoch": 0.8798514391829155, + "grad_norm": 1.5321038961410522, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8697283267974854, + "num_tokens": 166393240.0, + "step": 4738 + }, + { + "epoch": 0.8800371402042712, + "grad_norm": 1.5247644186019897, + "learning_rate": 1e-06, + "loss": 0.4185, + "mean_token_accuracy": 0.8584001064300537, + "num_tokens": 166428541.0, + "step": 4739 + }, + { + "epoch": 0.8802228412256268, + "grad_norm": 1.4320658445358276, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.873427152633667, + "num_tokens": 166464069.0, + "step": 4740 + }, + { + "epoch": 0.8804085422469824, + "grad_norm": 1.3723911046981812, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8723419904708862, + "num_tokens": 166502339.0, + "step": 4741 + }, + { + "epoch": 0.880594243268338, + "grad_norm": 1.4939966201782227, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.8643843531608582, + "num_tokens": 166537565.0, + "step": 4742 + }, + { + "epoch": 0.8807799442896936, + "grad_norm": 1.2999095916748047, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8788682222366333, + "num_tokens": 166577104.0, + "step": 4743 + }, + { + "epoch": 0.8809656453110493, + "grad_norm": 1.4994903802871704, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8697794675827026, + "num_tokens": 166610029.0, + "step": 4744 + }, + { + "epoch": 0.8811513463324049, + "grad_norm": 1.383527398109436, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8768170475959778, + "num_tokens": 166645444.0, + "step": 4745 + }, + { + "epoch": 0.8813370473537604, + "grad_norm": 1.7176185846328735, + "learning_rate": 1e-06, + "loss": 0.4536, + "mean_token_accuracy": 0.8539929389953613, + "num_tokens": 166676207.0, + "step": 4746 + }, + { + "epoch": 0.881522748375116, + "grad_norm": 1.337660312652588, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8707806468009949, + "num_tokens": 166716245.0, + "step": 4747 + }, + { + "epoch": 0.8817084493964716, + "grad_norm": 1.5887600183486938, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8589895963668823, + "num_tokens": 166746038.0, + "step": 4748 + }, + { + "epoch": 0.8818941504178273, + "grad_norm": 1.2663205862045288, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.877395510673523, + "num_tokens": 166786905.0, + "step": 4749 + }, + { + "epoch": 0.8820798514391829, + "grad_norm": 1.3958103656768799, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8760128021240234, + "num_tokens": 166822429.0, + "step": 4750 + }, + { + "epoch": 0.8822655524605385, + "grad_norm": 1.3809247016906738, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8735183477401733, + "num_tokens": 166863833.0, + "step": 4751 + }, + { + "epoch": 0.8824512534818941, + "grad_norm": 1.3722243309020996, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8775662779808044, + "num_tokens": 166899953.0, + "step": 4752 + }, + { + "epoch": 0.8826369545032497, + "grad_norm": 1.453750491142273, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8683773279190063, + "num_tokens": 166940104.0, + "step": 4753 + }, + { + "epoch": 0.8828226555246054, + "grad_norm": 1.7137939929962158, + "learning_rate": 1e-06, + "loss": 0.4716, + "mean_token_accuracy": 0.8517782688140869, + "num_tokens": 166969465.0, + "step": 4754 + }, + { + "epoch": 0.883008356545961, + "grad_norm": 1.4958187341690063, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8821313977241516, + "num_tokens": 166999836.0, + "step": 4755 + }, + { + "epoch": 0.8831940575673166, + "grad_norm": 1.4942256212234497, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8798807859420776, + "num_tokens": 167031236.0, + "step": 4756 + }, + { + "epoch": 0.8833797585886722, + "grad_norm": 1.551899790763855, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8570290207862854, + "num_tokens": 167063967.0, + "step": 4757 + }, + { + "epoch": 0.8835654596100279, + "grad_norm": 1.4211686849594116, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8837780356407166, + "num_tokens": 167095439.0, + "step": 4758 + }, + { + "epoch": 0.8837511606313835, + "grad_norm": 1.4081777334213257, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8616202473640442, + "num_tokens": 167133330.0, + "step": 4759 + }, + { + "epoch": 0.8839368616527391, + "grad_norm": 1.3324488401412964, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8839216828346252, + "num_tokens": 167169490.0, + "step": 4760 + }, + { + "epoch": 0.8841225626740947, + "grad_norm": 1.461962342262268, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8695148825645447, + "num_tokens": 167204928.0, + "step": 4761 + }, + { + "epoch": 0.8843082636954503, + "grad_norm": 1.4832723140716553, + "learning_rate": 1e-06, + "loss": 0.4259, + "mean_token_accuracy": 0.8636230230331421, + "num_tokens": 167242484.0, + "step": 4762 + }, + { + "epoch": 0.884493964716806, + "grad_norm": 1.4893382787704468, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8766541481018066, + "num_tokens": 167273255.0, + "step": 4763 + }, + { + "epoch": 0.8846796657381616, + "grad_norm": 1.51304030418396, + "learning_rate": 1e-06, + "loss": 0.4607, + "mean_token_accuracy": 0.8485662937164307, + "num_tokens": 167308915.0, + "step": 4764 + }, + { + "epoch": 0.8848653667595172, + "grad_norm": 1.6096441745758057, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8772429823875427, + "num_tokens": 167339005.0, + "step": 4765 + }, + { + "epoch": 0.8850510677808728, + "grad_norm": 1.4896047115325928, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8622957468032837, + "num_tokens": 167371566.0, + "step": 4766 + }, + { + "epoch": 0.8852367688022285, + "grad_norm": 1.4513428211212158, + "learning_rate": 1e-06, + "loss": 0.4403, + "mean_token_accuracy": 0.8570611476898193, + "num_tokens": 167411266.0, + "step": 4767 + }, + { + "epoch": 0.8854224698235841, + "grad_norm": 1.3574861288070679, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8735045790672302, + "num_tokens": 167450363.0, + "step": 4768 + }, + { + "epoch": 0.8856081708449397, + "grad_norm": 1.559830904006958, + "learning_rate": 1e-06, + "loss": 0.4261, + "mean_token_accuracy": 0.8589088916778564, + "num_tokens": 167484442.0, + "step": 4769 + }, + { + "epoch": 0.8857938718662952, + "grad_norm": 1.2893877029418945, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8753557801246643, + "num_tokens": 167525637.0, + "step": 4770 + }, + { + "epoch": 0.8859795728876508, + "grad_norm": 1.3091543912887573, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8824005126953125, + "num_tokens": 167563564.0, + "step": 4771 + }, + { + "epoch": 0.8861652739090065, + "grad_norm": 1.3172024488449097, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8785938620567322, + "num_tokens": 167601840.0, + "step": 4772 + }, + { + "epoch": 0.8863509749303621, + "grad_norm": 1.3146016597747803, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8684375882148743, + "num_tokens": 167645600.0, + "step": 4773 + }, + { + "epoch": 0.8865366759517177, + "grad_norm": 1.5788748264312744, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8672696352005005, + "num_tokens": 167675412.0, + "step": 4774 + }, + { + "epoch": 0.8867223769730733, + "grad_norm": 1.5887383222579956, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8687704801559448, + "num_tokens": 167706836.0, + "step": 4775 + }, + { + "epoch": 0.886908077994429, + "grad_norm": 1.4574283361434937, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8723205327987671, + "num_tokens": 167740221.0, + "step": 4776 + }, + { + "epoch": 0.8870937790157846, + "grad_norm": 1.475712537765503, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.88404780626297, + "num_tokens": 167771800.0, + "step": 4777 + }, + { + "epoch": 0.8872794800371402, + "grad_norm": 1.3688421249389648, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.8680053949356079, + "num_tokens": 167809829.0, + "step": 4778 + }, + { + "epoch": 0.8874651810584958, + "grad_norm": 1.323329210281372, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8620633482933044, + "num_tokens": 167854185.0, + "step": 4779 + }, + { + "epoch": 0.8876508820798514, + "grad_norm": 1.5164426565170288, + "learning_rate": 1e-06, + "loss": 0.4148, + "mean_token_accuracy": 0.8606853485107422, + "num_tokens": 167889379.0, + "step": 4780 + }, + { + "epoch": 0.887836583101207, + "grad_norm": 1.5045469999313354, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8729966878890991, + "num_tokens": 167919162.0, + "step": 4781 + }, + { + "epoch": 0.8880222841225627, + "grad_norm": 1.3624621629714966, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.8617735505104065, + "num_tokens": 167958501.0, + "step": 4782 + }, + { + "epoch": 0.8882079851439183, + "grad_norm": 1.3645117282867432, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8804284930229187, + "num_tokens": 167997308.0, + "step": 4783 + }, + { + "epoch": 0.8883936861652739, + "grad_norm": 1.4861869812011719, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8558561205863953, + "num_tokens": 168034528.0, + "step": 4784 + }, + { + "epoch": 0.8885793871866295, + "grad_norm": 1.3111021518707275, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8860563635826111, + "num_tokens": 168069567.0, + "step": 4785 + }, + { + "epoch": 0.8887650882079852, + "grad_norm": 1.4131253957748413, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8711543083190918, + "num_tokens": 168109381.0, + "step": 4786 + }, + { + "epoch": 0.8889507892293408, + "grad_norm": 1.463075876235962, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8698536157608032, + "num_tokens": 168144903.0, + "step": 4787 + }, + { + "epoch": 0.8891364902506964, + "grad_norm": 1.6005080938339233, + "learning_rate": 1e-06, + "loss": 0.4534, + "mean_token_accuracy": 0.8503831624984741, + "num_tokens": 168177834.0, + "step": 4788 + }, + { + "epoch": 0.889322191272052, + "grad_norm": 1.4411081075668335, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8768720626831055, + "num_tokens": 168210300.0, + "step": 4789 + }, + { + "epoch": 0.8895078922934077, + "grad_norm": 1.4864771366119385, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8853874206542969, + "num_tokens": 168241705.0, + "step": 4790 + }, + { + "epoch": 0.8896935933147633, + "grad_norm": 1.5265151262283325, + "learning_rate": 1e-06, + "loss": 0.4332, + "mean_token_accuracy": 0.8542643189430237, + "num_tokens": 168277149.0, + "step": 4791 + }, + { + "epoch": 0.8898792943361189, + "grad_norm": 1.4357343912124634, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.8621202707290649, + "num_tokens": 168315197.0, + "step": 4792 + }, + { + "epoch": 0.8900649953574745, + "grad_norm": 1.2830016613006592, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8733253479003906, + "num_tokens": 168358076.0, + "step": 4793 + }, + { + "epoch": 0.89025069637883, + "grad_norm": 1.5281662940979004, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8722198605537415, + "num_tokens": 168392344.0, + "step": 4794 + }, + { + "epoch": 0.8904363974001857, + "grad_norm": 1.372573733329773, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8583742380142212, + "num_tokens": 168434550.0, + "step": 4795 + }, + { + "epoch": 0.8906220984215413, + "grad_norm": 1.603914737701416, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8622859716415405, + "num_tokens": 168463897.0, + "step": 4796 + }, + { + "epoch": 0.8908077994428969, + "grad_norm": 1.4537458419799805, + "learning_rate": 1e-06, + "loss": 0.4225, + "mean_token_accuracy": 0.8573909997940063, + "num_tokens": 168501385.0, + "step": 4797 + }, + { + "epoch": 0.8909935004642525, + "grad_norm": 1.548998236656189, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8661316633224487, + "num_tokens": 168539216.0, + "step": 4798 + }, + { + "epoch": 0.8911792014856081, + "grad_norm": 1.5533180236816406, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8699958920478821, + "num_tokens": 168569303.0, + "step": 4799 + }, + { + "epoch": 0.8913649025069638, + "grad_norm": 1.459061861038208, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8663872480392456, + "num_tokens": 168603300.0, + "step": 4800 + }, + { + "epoch": 0.8915506035283194, + "grad_norm": 1.4884679317474365, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8690770864486694, + "num_tokens": 168640326.0, + "step": 4801 + }, + { + "epoch": 0.891736304549675, + "grad_norm": 1.514736533164978, + "learning_rate": 1e-06, + "loss": 0.4803, + "mean_token_accuracy": 0.8444130420684814, + "num_tokens": 168678993.0, + "step": 4802 + }, + { + "epoch": 0.8919220055710306, + "grad_norm": 1.409411907196045, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8812041282653809, + "num_tokens": 168714599.0, + "step": 4803 + }, + { + "epoch": 0.8921077065923863, + "grad_norm": 1.4132565259933472, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.8625444173812866, + "num_tokens": 168753744.0, + "step": 4804 + }, + { + "epoch": 0.8922934076137419, + "grad_norm": 1.4690792560577393, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.8673104047775269, + "num_tokens": 168788723.0, + "step": 4805 + }, + { + "epoch": 0.8924791086350975, + "grad_norm": 1.4266304969787598, + "learning_rate": 1e-06, + "loss": 0.4144, + "mean_token_accuracy": 0.8608930110931396, + "num_tokens": 168828884.0, + "step": 4806 + }, + { + "epoch": 0.8926648096564531, + "grad_norm": 1.4710478782653809, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8724551200866699, + "num_tokens": 168860515.0, + "step": 4807 + }, + { + "epoch": 0.8928505106778087, + "grad_norm": 1.4167784452438354, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.868039608001709, + "num_tokens": 168898508.0, + "step": 4808 + }, + { + "epoch": 0.8930362116991644, + "grad_norm": 1.3813748359680176, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8724155426025391, + "num_tokens": 168937397.0, + "step": 4809 + }, + { + "epoch": 0.89322191272052, + "grad_norm": 1.3632550239562988, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8735392093658447, + "num_tokens": 168975549.0, + "step": 4810 + }, + { + "epoch": 0.8934076137418756, + "grad_norm": 1.4705214500427246, + "learning_rate": 1e-06, + "loss": 0.4411, + "mean_token_accuracy": 0.8533387184143066, + "num_tokens": 169015865.0, + "step": 4811 + }, + { + "epoch": 0.8935933147632312, + "grad_norm": 1.3649429082870483, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8741934895515442, + "num_tokens": 169052151.0, + "step": 4812 + }, + { + "epoch": 0.8937790157845868, + "grad_norm": 1.4283168315887451, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8833653926849365, + "num_tokens": 169084534.0, + "step": 4813 + }, + { + "epoch": 0.8939647168059425, + "grad_norm": 1.3573201894760132, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8729093670845032, + "num_tokens": 169121214.0, + "step": 4814 + }, + { + "epoch": 0.8941504178272981, + "grad_norm": 1.4069993495941162, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8696565628051758, + "num_tokens": 169158177.0, + "step": 4815 + }, + { + "epoch": 0.8943361188486537, + "grad_norm": 1.485360860824585, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8711416125297546, + "num_tokens": 169187866.0, + "step": 4816 + }, + { + "epoch": 0.8945218198700093, + "grad_norm": 1.3980947732925415, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8806086182594299, + "num_tokens": 169225269.0, + "step": 4817 + }, + { + "epoch": 0.8947075208913648, + "grad_norm": 1.5271106958389282, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.874969482421875, + "num_tokens": 169258008.0, + "step": 4818 + }, + { + "epoch": 0.8948932219127205, + "grad_norm": 1.708999514579773, + "learning_rate": 1e-06, + "loss": 0.4574, + "mean_token_accuracy": 0.8532171249389648, + "num_tokens": 169289354.0, + "step": 4819 + }, + { + "epoch": 0.8950789229340761, + "grad_norm": 1.4460091590881348, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8733060359954834, + "num_tokens": 169320253.0, + "step": 4820 + }, + { + "epoch": 0.8952646239554317, + "grad_norm": 1.4254916906356812, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8795843124389648, + "num_tokens": 169355490.0, + "step": 4821 + }, + { + "epoch": 0.8954503249767873, + "grad_norm": 1.5178848505020142, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8634144067764282, + "num_tokens": 169392552.0, + "step": 4822 + }, + { + "epoch": 0.895636025998143, + "grad_norm": 1.446341633796692, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8693130016326904, + "num_tokens": 169425785.0, + "step": 4823 + }, + { + "epoch": 0.8958217270194986, + "grad_norm": 1.5423320531845093, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8778427839279175, + "num_tokens": 169454929.0, + "step": 4824 + }, + { + "epoch": 0.8960074280408542, + "grad_norm": 1.3516356945037842, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8821628093719482, + "num_tokens": 169497950.0, + "step": 4825 + }, + { + "epoch": 0.8961931290622098, + "grad_norm": 1.4517943859100342, + "learning_rate": 1e-06, + "loss": 0.3958, + "mean_token_accuracy": 0.8677703142166138, + "num_tokens": 169531842.0, + "step": 4826 + }, + { + "epoch": 0.8963788300835654, + "grad_norm": 1.4418280124664307, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8736256957054138, + "num_tokens": 169571178.0, + "step": 4827 + }, + { + "epoch": 0.8965645311049211, + "grad_norm": 1.5892360210418701, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8684179782867432, + "num_tokens": 169604264.0, + "step": 4828 + }, + { + "epoch": 0.8967502321262767, + "grad_norm": 1.4704978466033936, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8798755407333374, + "num_tokens": 169633731.0, + "step": 4829 + }, + { + "epoch": 0.8969359331476323, + "grad_norm": 1.444591999053955, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8663825988769531, + "num_tokens": 169667977.0, + "step": 4830 + }, + { + "epoch": 0.8971216341689879, + "grad_norm": 1.527289867401123, + "learning_rate": 1e-06, + "loss": 0.4446, + "mean_token_accuracy": 0.8487565517425537, + "num_tokens": 169702161.0, + "step": 4831 + }, + { + "epoch": 0.8973073351903436, + "grad_norm": 1.4940521717071533, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8679206371307373, + "num_tokens": 169735214.0, + "step": 4832 + }, + { + "epoch": 0.8974930362116992, + "grad_norm": 1.4661051034927368, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8774145841598511, + "num_tokens": 169767006.0, + "step": 4833 + }, + { + "epoch": 0.8976787372330548, + "grad_norm": 1.5929129123687744, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8691810965538025, + "num_tokens": 169798343.0, + "step": 4834 + }, + { + "epoch": 0.8978644382544104, + "grad_norm": 1.5332680940628052, + "learning_rate": 1e-06, + "loss": 0.4295, + "mean_token_accuracy": 0.8544508218765259, + "num_tokens": 169832567.0, + "step": 4835 + }, + { + "epoch": 0.898050139275766, + "grad_norm": 1.572872281074524, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8645522594451904, + "num_tokens": 169867905.0, + "step": 4836 + }, + { + "epoch": 0.8982358402971217, + "grad_norm": 1.4350284337997437, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8663449287414551, + "num_tokens": 169906528.0, + "step": 4837 + }, + { + "epoch": 0.8984215413184773, + "grad_norm": 1.4487658739089966, + "learning_rate": 1e-06, + "loss": 0.4075, + "mean_token_accuracy": 0.860902726650238, + "num_tokens": 169942080.0, + "step": 4838 + }, + { + "epoch": 0.8986072423398329, + "grad_norm": 1.4335943460464478, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8674382567405701, + "num_tokens": 169983144.0, + "step": 4839 + }, + { + "epoch": 0.8987929433611885, + "grad_norm": 1.5057040452957153, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8698930740356445, + "num_tokens": 170014869.0, + "step": 4840 + }, + { + "epoch": 0.8989786443825442, + "grad_norm": 1.3054742813110352, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8813691735267639, + "num_tokens": 170051505.0, + "step": 4841 + }, + { + "epoch": 0.8991643454038997, + "grad_norm": 1.3505420684814453, + "learning_rate": 1e-06, + "loss": 0.4518, + "mean_token_accuracy": 0.8500977754592896, + "num_tokens": 170093664.0, + "step": 4842 + }, + { + "epoch": 0.8993500464252553, + "grad_norm": 1.5844030380249023, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.8659168481826782, + "num_tokens": 170124854.0, + "step": 4843 + }, + { + "epoch": 0.8995357474466109, + "grad_norm": 1.2859748601913452, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8682684898376465, + "num_tokens": 170167083.0, + "step": 4844 + }, + { + "epoch": 0.8997214484679665, + "grad_norm": 1.489041805267334, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.866168737411499, + "num_tokens": 170200286.0, + "step": 4845 + }, + { + "epoch": 0.8999071494893222, + "grad_norm": 1.4228613376617432, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8675075769424438, + "num_tokens": 170236970.0, + "step": 4846 + }, + { + "epoch": 0.9000928505106778, + "grad_norm": 1.4648237228393555, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8738669753074646, + "num_tokens": 170272469.0, + "step": 4847 + }, + { + "epoch": 0.9002785515320334, + "grad_norm": 1.4701011180877686, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8824779987335205, + "num_tokens": 170304823.0, + "step": 4848 + }, + { + "epoch": 0.900464252553389, + "grad_norm": 1.374903678894043, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8764537572860718, + "num_tokens": 170343503.0, + "step": 4849 + }, + { + "epoch": 0.9006499535747446, + "grad_norm": 1.4265447854995728, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.865265429019928, + "num_tokens": 170380398.0, + "step": 4850 + }, + { + "epoch": 0.9008356545961003, + "grad_norm": 1.4016180038452148, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8678222298622131, + "num_tokens": 170422172.0, + "step": 4851 + }, + { + "epoch": 0.9010213556174559, + "grad_norm": 1.4206066131591797, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8717818856239319, + "num_tokens": 170460188.0, + "step": 4852 + }, + { + "epoch": 0.9012070566388115, + "grad_norm": 1.5417306423187256, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8700105547904968, + "num_tokens": 170490410.0, + "step": 4853 + }, + { + "epoch": 0.9013927576601671, + "grad_norm": 1.3709027767181396, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8781417608261108, + "num_tokens": 170528623.0, + "step": 4854 + }, + { + "epoch": 0.9015784586815228, + "grad_norm": 1.422966480255127, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8658878803253174, + "num_tokens": 170564455.0, + "step": 4855 + }, + { + "epoch": 0.9017641597028784, + "grad_norm": 1.4528048038482666, + "learning_rate": 1e-06, + "loss": 0.4251, + "mean_token_accuracy": 0.8575242757797241, + "num_tokens": 170603679.0, + "step": 4856 + }, + { + "epoch": 0.901949860724234, + "grad_norm": 1.4444149732589722, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8681344985961914, + "num_tokens": 170640800.0, + "step": 4857 + }, + { + "epoch": 0.9021355617455896, + "grad_norm": 1.509055495262146, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8783034086227417, + "num_tokens": 170673944.0, + "step": 4858 + }, + { + "epoch": 0.9023212627669452, + "grad_norm": 1.5147134065628052, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8776580095291138, + "num_tokens": 170706687.0, + "step": 4859 + }, + { + "epoch": 0.9025069637883009, + "grad_norm": 1.5241199731826782, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8673285841941833, + "num_tokens": 170738128.0, + "step": 4860 + }, + { + "epoch": 0.9026926648096565, + "grad_norm": 1.4526337385177612, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8634917140007019, + "num_tokens": 170774507.0, + "step": 4861 + }, + { + "epoch": 0.9028783658310121, + "grad_norm": 1.2852548360824585, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8691818714141846, + "num_tokens": 170814340.0, + "step": 4862 + }, + { + "epoch": 0.9030640668523677, + "grad_norm": 1.371621012687683, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8773421049118042, + "num_tokens": 170854431.0, + "step": 4863 + }, + { + "epoch": 0.9032497678737234, + "grad_norm": 1.3251229524612427, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8735581040382385, + "num_tokens": 170896740.0, + "step": 4864 + }, + { + "epoch": 0.903435468895079, + "grad_norm": 1.3634812831878662, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8609936833381653, + "num_tokens": 170936419.0, + "step": 4865 + }, + { + "epoch": 0.9036211699164345, + "grad_norm": 1.4202923774719238, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.873184323310852, + "num_tokens": 170972396.0, + "step": 4866 + }, + { + "epoch": 0.9038068709377901, + "grad_norm": 1.4480165243148804, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.879142701625824, + "num_tokens": 171004363.0, + "step": 4867 + }, + { + "epoch": 0.9039925719591457, + "grad_norm": 1.4329602718353271, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8728072643280029, + "num_tokens": 171045391.0, + "step": 4868 + }, + { + "epoch": 0.9041782729805014, + "grad_norm": 1.5470954179763794, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8677529096603394, + "num_tokens": 171077710.0, + "step": 4869 + }, + { + "epoch": 0.904363974001857, + "grad_norm": 1.4471385478973389, + "learning_rate": 1e-06, + "loss": 0.4236, + "mean_token_accuracy": 0.8530632853507996, + "num_tokens": 171113731.0, + "step": 4870 + }, + { + "epoch": 0.9045496750232126, + "grad_norm": 1.4027183055877686, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8677995800971985, + "num_tokens": 171149441.0, + "step": 4871 + }, + { + "epoch": 0.9047353760445682, + "grad_norm": 1.5259089469909668, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8667317032814026, + "num_tokens": 171179449.0, + "step": 4872 + }, + { + "epoch": 0.9049210770659238, + "grad_norm": 1.3800219297409058, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8671566247940063, + "num_tokens": 171218649.0, + "step": 4873 + }, + { + "epoch": 0.9051067780872795, + "grad_norm": 1.3970222473144531, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8749163150787354, + "num_tokens": 171253436.0, + "step": 4874 + }, + { + "epoch": 0.9052924791086351, + "grad_norm": 1.5126162767410278, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8755490779876709, + "num_tokens": 171283837.0, + "step": 4875 + }, + { + "epoch": 0.9054781801299907, + "grad_norm": 1.2998212575912476, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8758929371833801, + "num_tokens": 171323726.0, + "step": 4876 + }, + { + "epoch": 0.9056638811513463, + "grad_norm": 1.3506639003753662, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.875390887260437, + "num_tokens": 171361238.0, + "step": 4877 + }, + { + "epoch": 0.905849582172702, + "grad_norm": 1.4153084754943848, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.8652917146682739, + "num_tokens": 171397561.0, + "step": 4878 + }, + { + "epoch": 0.9060352831940576, + "grad_norm": 1.4390089511871338, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8666024804115295, + "num_tokens": 171432475.0, + "step": 4879 + }, + { + "epoch": 0.9062209842154132, + "grad_norm": 1.5535187721252441, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8716068863868713, + "num_tokens": 171463214.0, + "step": 4880 + }, + { + "epoch": 0.9064066852367688, + "grad_norm": 1.5702781677246094, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8664436340332031, + "num_tokens": 171495789.0, + "step": 4881 + }, + { + "epoch": 0.9065923862581244, + "grad_norm": 1.6236470937728882, + "learning_rate": 1e-06, + "loss": 0.4427, + "mean_token_accuracy": 0.8504034280776978, + "num_tokens": 171528415.0, + "step": 4882 + }, + { + "epoch": 0.9067780872794801, + "grad_norm": 1.5620888471603394, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8723427057266235, + "num_tokens": 171562343.0, + "step": 4883 + }, + { + "epoch": 0.9069637883008357, + "grad_norm": 1.407409906387329, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8726062178611755, + "num_tokens": 171602249.0, + "step": 4884 + }, + { + "epoch": 0.9071494893221913, + "grad_norm": 1.5084774494171143, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.8608411550521851, + "num_tokens": 171637608.0, + "step": 4885 + }, + { + "epoch": 0.9073351903435469, + "grad_norm": 1.497722864151001, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.8604280948638916, + "num_tokens": 171672840.0, + "step": 4886 + }, + { + "epoch": 0.9075208913649025, + "grad_norm": 1.3737741708755493, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8566344380378723, + "num_tokens": 171713282.0, + "step": 4887 + }, + { + "epoch": 0.9077065923862582, + "grad_norm": 1.533540964126587, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8711971640586853, + "num_tokens": 171744839.0, + "step": 4888 + }, + { + "epoch": 0.9078922934076138, + "grad_norm": 1.5322277545928955, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8663643598556519, + "num_tokens": 171784804.0, + "step": 4889 + }, + { + "epoch": 0.9080779944289693, + "grad_norm": 1.476155400276184, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.863680362701416, + "num_tokens": 171822586.0, + "step": 4890 + }, + { + "epoch": 0.9082636954503249, + "grad_norm": 1.5374091863632202, + "learning_rate": 1e-06, + "loss": 0.4277, + "mean_token_accuracy": 0.860169529914856, + "num_tokens": 171857361.0, + "step": 4891 + }, + { + "epoch": 0.9084493964716805, + "grad_norm": 1.4357964992523193, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8827040195465088, + "num_tokens": 171892517.0, + "step": 4892 + }, + { + "epoch": 0.9086350974930362, + "grad_norm": 1.3179033994674683, + "learning_rate": 1e-06, + "loss": 0.4258, + "mean_token_accuracy": 0.856715202331543, + "num_tokens": 171938540.0, + "step": 4893 + }, + { + "epoch": 0.9088207985143918, + "grad_norm": 1.4388278722763062, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8620123863220215, + "num_tokens": 171973326.0, + "step": 4894 + }, + { + "epoch": 0.9090064995357474, + "grad_norm": 1.5022753477096558, + "learning_rate": 1e-06, + "loss": 0.4265, + "mean_token_accuracy": 0.8597049713134766, + "num_tokens": 172009050.0, + "step": 4895 + }, + { + "epoch": 0.909192200557103, + "grad_norm": 1.6904491186141968, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8845095634460449, + "num_tokens": 172035772.0, + "step": 4896 + }, + { + "epoch": 0.9093779015784587, + "grad_norm": 1.3793361186981201, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8806976079940796, + "num_tokens": 172074814.0, + "step": 4897 + }, + { + "epoch": 0.9095636025998143, + "grad_norm": 1.529566764831543, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8797377347946167, + "num_tokens": 172107465.0, + "step": 4898 + }, + { + "epoch": 0.9097493036211699, + "grad_norm": 1.5969371795654297, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8810225129127502, + "num_tokens": 172137228.0, + "step": 4899 + }, + { + "epoch": 0.9099350046425255, + "grad_norm": 1.4021788835525513, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.871243953704834, + "num_tokens": 172174940.0, + "step": 4900 + }, + { + "epoch": 0.9101207056638811, + "grad_norm": 1.3424148559570312, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8738458752632141, + "num_tokens": 172216605.0, + "step": 4901 + }, + { + "epoch": 0.9103064066852368, + "grad_norm": 1.5247079133987427, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8687790632247925, + "num_tokens": 172253057.0, + "step": 4902 + }, + { + "epoch": 0.9104921077065924, + "grad_norm": 1.377903938293457, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8676984310150146, + "num_tokens": 172291834.0, + "step": 4903 + }, + { + "epoch": 0.910677808727948, + "grad_norm": 1.699110746383667, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.85873943567276, + "num_tokens": 172319592.0, + "step": 4904 + }, + { + "epoch": 0.9108635097493036, + "grad_norm": 1.3387407064437866, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8799559473991394, + "num_tokens": 172359855.0, + "step": 4905 + }, + { + "epoch": 0.9110492107706593, + "grad_norm": 1.354806900024414, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8753020763397217, + "num_tokens": 172396650.0, + "step": 4906 + }, + { + "epoch": 0.9112349117920149, + "grad_norm": 1.587409496307373, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8664718270301819, + "num_tokens": 172427645.0, + "step": 4907 + }, + { + "epoch": 0.9114206128133705, + "grad_norm": 1.6093664169311523, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8762493133544922, + "num_tokens": 172457757.0, + "step": 4908 + }, + { + "epoch": 0.9116063138347261, + "grad_norm": 1.4720269441604614, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8686755299568176, + "num_tokens": 172494658.0, + "step": 4909 + }, + { + "epoch": 0.9117920148560817, + "grad_norm": 1.3815683126449585, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.882219672203064, + "num_tokens": 172533156.0, + "step": 4910 + }, + { + "epoch": 0.9119777158774374, + "grad_norm": 1.5968654155731201, + "learning_rate": 1e-06, + "loss": 0.4359, + "mean_token_accuracy": 0.8510522842407227, + "num_tokens": 172564595.0, + "step": 4911 + }, + { + "epoch": 0.912163416898793, + "grad_norm": 1.3785046339035034, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8814458847045898, + "num_tokens": 172600011.0, + "step": 4912 + }, + { + "epoch": 0.9123491179201486, + "grad_norm": 1.5415107011795044, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8676857948303223, + "num_tokens": 172631137.0, + "step": 4913 + }, + { + "epoch": 0.9125348189415042, + "grad_norm": 1.492526650428772, + "learning_rate": 1e-06, + "loss": 0.4597, + "mean_token_accuracy": 0.8436508178710938, + "num_tokens": 172669961.0, + "step": 4914 + }, + { + "epoch": 0.9127205199628597, + "grad_norm": 1.6231521368026733, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8711570501327515, + "num_tokens": 172704926.0, + "step": 4915 + }, + { + "epoch": 0.9129062209842154, + "grad_norm": 1.3496627807617188, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.8653949499130249, + "num_tokens": 172745538.0, + "step": 4916 + }, + { + "epoch": 0.913091922005571, + "grad_norm": 1.463630199432373, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8728243708610535, + "num_tokens": 172780582.0, + "step": 4917 + }, + { + "epoch": 0.9132776230269266, + "grad_norm": 1.5236728191375732, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.870898962020874, + "num_tokens": 172815088.0, + "step": 4918 + }, + { + "epoch": 0.9134633240482822, + "grad_norm": 1.472212314605713, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8749830722808838, + "num_tokens": 172848133.0, + "step": 4919 + }, + { + "epoch": 0.9136490250696379, + "grad_norm": 1.4643909931182861, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.864338755607605, + "num_tokens": 172882601.0, + "step": 4920 + }, + { + "epoch": 0.9138347260909935, + "grad_norm": 1.4286330938339233, + "learning_rate": 1e-06, + "loss": 0.4183, + "mean_token_accuracy": 0.8585737347602844, + "num_tokens": 172923185.0, + "step": 4921 + }, + { + "epoch": 0.9140204271123491, + "grad_norm": 1.7016936540603638, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.8626382350921631, + "num_tokens": 172952640.0, + "step": 4922 + }, + { + "epoch": 0.9142061281337047, + "grad_norm": 1.450466275215149, + "learning_rate": 1e-06, + "loss": 0.4395, + "mean_token_accuracy": 0.856513261795044, + "num_tokens": 172989007.0, + "step": 4923 + }, + { + "epoch": 0.9143918291550603, + "grad_norm": 1.4297959804534912, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8769185543060303, + "num_tokens": 173023462.0, + "step": 4924 + }, + { + "epoch": 0.914577530176416, + "grad_norm": 1.4257795810699463, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8707365393638611, + "num_tokens": 173059718.0, + "step": 4925 + }, + { + "epoch": 0.9147632311977716, + "grad_norm": 1.427337408065796, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8790769577026367, + "num_tokens": 173093440.0, + "step": 4926 + }, + { + "epoch": 0.9149489322191272, + "grad_norm": 1.4462897777557373, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.8652009963989258, + "num_tokens": 173129465.0, + "step": 4927 + }, + { + "epoch": 0.9151346332404828, + "grad_norm": 1.3787059783935547, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8734291791915894, + "num_tokens": 173169016.0, + "step": 4928 + }, + { + "epoch": 0.9153203342618385, + "grad_norm": 1.3618284463882446, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.879615306854248, + "num_tokens": 173202746.0, + "step": 4929 + }, + { + "epoch": 0.9155060352831941, + "grad_norm": 1.5032987594604492, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8691582679748535, + "num_tokens": 173235026.0, + "step": 4930 + }, + { + "epoch": 0.9156917363045497, + "grad_norm": 1.430853247642517, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.870192289352417, + "num_tokens": 173269354.0, + "step": 4931 + }, + { + "epoch": 0.9158774373259053, + "grad_norm": 1.3047659397125244, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8694593906402588, + "num_tokens": 173312156.0, + "step": 4932 + }, + { + "epoch": 0.9160631383472609, + "grad_norm": 1.4631844758987427, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8676141500473022, + "num_tokens": 173348269.0, + "step": 4933 + }, + { + "epoch": 0.9162488393686166, + "grad_norm": 1.5664291381835938, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8737000226974487, + "num_tokens": 173376106.0, + "step": 4934 + }, + { + "epoch": 0.9164345403899722, + "grad_norm": 1.372469425201416, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8783729076385498, + "num_tokens": 173410654.0, + "step": 4935 + }, + { + "epoch": 0.9166202414113278, + "grad_norm": 1.534890055656433, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8758678436279297, + "num_tokens": 173445365.0, + "step": 4936 + }, + { + "epoch": 0.9168059424326834, + "grad_norm": 1.4153783321380615, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8751932382583618, + "num_tokens": 173482950.0, + "step": 4937 + }, + { + "epoch": 0.916991643454039, + "grad_norm": 1.5690480470657349, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8850613832473755, + "num_tokens": 173513121.0, + "step": 4938 + }, + { + "epoch": 0.9171773444753946, + "grad_norm": 1.351493000984192, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.867531418800354, + "num_tokens": 173552069.0, + "step": 4939 + }, + { + "epoch": 0.9173630454967502, + "grad_norm": 1.4387259483337402, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8703540563583374, + "num_tokens": 173589708.0, + "step": 4940 + }, + { + "epoch": 0.9175487465181058, + "grad_norm": 1.6287236213684082, + "learning_rate": 1e-06, + "loss": 0.4398, + "mean_token_accuracy": 0.8545659184455872, + "num_tokens": 173618224.0, + "step": 4941 + }, + { + "epoch": 0.9177344475394614, + "grad_norm": 1.4995454549789429, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8752073049545288, + "num_tokens": 173652486.0, + "step": 4942 + }, + { + "epoch": 0.917920148560817, + "grad_norm": 1.410377860069275, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8687725067138672, + "num_tokens": 173689072.0, + "step": 4943 + }, + { + "epoch": 0.9181058495821727, + "grad_norm": 1.5703269243240356, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8696544766426086, + "num_tokens": 173719699.0, + "step": 4944 + }, + { + "epoch": 0.9182915506035283, + "grad_norm": 1.3257098197937012, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8635479211807251, + "num_tokens": 173763208.0, + "step": 4945 + }, + { + "epoch": 0.9184772516248839, + "grad_norm": 1.3972809314727783, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8852683305740356, + "num_tokens": 173799051.0, + "step": 4946 + }, + { + "epoch": 0.9186629526462395, + "grad_norm": 1.4463825225830078, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8692339062690735, + "num_tokens": 173834522.0, + "step": 4947 + }, + { + "epoch": 0.9188486536675952, + "grad_norm": 1.356892466545105, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8803384304046631, + "num_tokens": 173868151.0, + "step": 4948 + }, + { + "epoch": 0.9190343546889508, + "grad_norm": 1.307273030281067, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8765271902084351, + "num_tokens": 173909236.0, + "step": 4949 + }, + { + "epoch": 0.9192200557103064, + "grad_norm": 1.5339782238006592, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8845071196556091, + "num_tokens": 173938870.0, + "step": 4950 + }, + { + "epoch": 0.919405756731662, + "grad_norm": 1.5691672563552856, + "learning_rate": 1e-06, + "loss": 0.4657, + "mean_token_accuracy": 0.8450016975402832, + "num_tokens": 173973144.0, + "step": 4951 + }, + { + "epoch": 0.9195914577530176, + "grad_norm": 1.5422042608261108, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.858843207359314, + "num_tokens": 174004701.0, + "step": 4952 + }, + { + "epoch": 0.9197771587743733, + "grad_norm": 1.6325172185897827, + "learning_rate": 1e-06, + "loss": 0.4564, + "mean_token_accuracy": 0.8482384085655212, + "num_tokens": 174036078.0, + "step": 4953 + }, + { + "epoch": 0.9199628597957289, + "grad_norm": 1.4789468050003052, + "learning_rate": 1e-06, + "loss": 0.4458, + "mean_token_accuracy": 0.8530023097991943, + "num_tokens": 174072706.0, + "step": 4954 + }, + { + "epoch": 0.9201485608170845, + "grad_norm": 1.4700846672058105, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.867114782333374, + "num_tokens": 174106410.0, + "step": 4955 + }, + { + "epoch": 0.9203342618384401, + "grad_norm": 1.3439648151397705, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8779826760292053, + "num_tokens": 174142277.0, + "step": 4956 + }, + { + "epoch": 0.9205199628597958, + "grad_norm": 1.5049488544464111, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.857547402381897, + "num_tokens": 174173246.0, + "step": 4957 + }, + { + "epoch": 0.9207056638811514, + "grad_norm": 1.6246687173843384, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8730027675628662, + "num_tokens": 174198109.0, + "step": 4958 + }, + { + "epoch": 0.920891364902507, + "grad_norm": 1.3800349235534668, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8846307396888733, + "num_tokens": 174232694.0, + "step": 4959 + }, + { + "epoch": 0.9210770659238626, + "grad_norm": 1.412406325340271, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8730732202529907, + "num_tokens": 174268333.0, + "step": 4960 + }, + { + "epoch": 0.9212627669452182, + "grad_norm": 1.3812021017074585, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8615652918815613, + "num_tokens": 174307617.0, + "step": 4961 + }, + { + "epoch": 0.9214484679665739, + "grad_norm": 1.4309953451156616, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.860256016254425, + "num_tokens": 174347527.0, + "step": 4962 + }, + { + "epoch": 0.9216341689879294, + "grad_norm": 1.5525932312011719, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8715665340423584, + "num_tokens": 174377256.0, + "step": 4963 + }, + { + "epoch": 0.921819870009285, + "grad_norm": 1.528438687324524, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.869483232498169, + "num_tokens": 174406178.0, + "step": 4964 + }, + { + "epoch": 0.9220055710306406, + "grad_norm": 1.2326549291610718, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8743150234222412, + "num_tokens": 174452908.0, + "step": 4965 + }, + { + "epoch": 0.9221912720519962, + "grad_norm": 1.5639928579330444, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8711459636688232, + "num_tokens": 174481797.0, + "step": 4966 + }, + { + "epoch": 0.9223769730733519, + "grad_norm": 1.4625450372695923, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8878655433654785, + "num_tokens": 174514205.0, + "step": 4967 + }, + { + "epoch": 0.9225626740947075, + "grad_norm": 1.4617551565170288, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8583163619041443, + "num_tokens": 174551029.0, + "step": 4968 + }, + { + "epoch": 0.9227483751160631, + "grad_norm": 1.4393305778503418, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8535213470458984, + "num_tokens": 174587109.0, + "step": 4969 + }, + { + "epoch": 0.9229340761374187, + "grad_norm": 1.45901620388031, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8584653735160828, + "num_tokens": 174623967.0, + "step": 4970 + }, + { + "epoch": 0.9231197771587744, + "grad_norm": 1.4291337728500366, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8630480170249939, + "num_tokens": 174664290.0, + "step": 4971 + }, + { + "epoch": 0.92330547818013, + "grad_norm": 1.240376591682434, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8872973918914795, + "num_tokens": 174703754.0, + "step": 4972 + }, + { + "epoch": 0.9234911792014856, + "grad_norm": 1.6871724128723145, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8684481382369995, + "num_tokens": 174732957.0, + "step": 4973 + }, + { + "epoch": 0.9236768802228412, + "grad_norm": 1.5075323581695557, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8711148500442505, + "num_tokens": 174762401.0, + "step": 4974 + }, + { + "epoch": 0.9238625812441968, + "grad_norm": 1.411726951599121, + "learning_rate": 1e-06, + "loss": 0.4185, + "mean_token_accuracy": 0.8567491769790649, + "num_tokens": 174800287.0, + "step": 4975 + }, + { + "epoch": 0.9240482822655525, + "grad_norm": 1.561189889907837, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8710503578186035, + "num_tokens": 174832148.0, + "step": 4976 + }, + { + "epoch": 0.9242339832869081, + "grad_norm": 1.5077184438705444, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8678378462791443, + "num_tokens": 174862095.0, + "step": 4977 + }, + { + "epoch": 0.9244196843082637, + "grad_norm": 1.4792884588241577, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8761406540870667, + "num_tokens": 174899044.0, + "step": 4978 + }, + { + "epoch": 0.9246053853296193, + "grad_norm": 1.4470802545547485, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.8736120462417603, + "num_tokens": 174932964.0, + "step": 4979 + }, + { + "epoch": 0.924791086350975, + "grad_norm": 1.414332389831543, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8781906366348267, + "num_tokens": 174969161.0, + "step": 4980 + }, + { + "epoch": 0.9249767873723306, + "grad_norm": 1.3969001770019531, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8695565462112427, + "num_tokens": 175007368.0, + "step": 4981 + }, + { + "epoch": 0.9251624883936862, + "grad_norm": 1.4224934577941895, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8799677491188049, + "num_tokens": 175043299.0, + "step": 4982 + }, + { + "epoch": 0.9253481894150418, + "grad_norm": 1.6183122396469116, + "learning_rate": 1e-06, + "loss": 0.4585, + "mean_token_accuracy": 0.8532540202140808, + "num_tokens": 175073553.0, + "step": 4983 + }, + { + "epoch": 0.9255338904363974, + "grad_norm": 1.648074746131897, + "learning_rate": 1e-06, + "loss": 0.5057, + "mean_token_accuracy": 0.8382142186164856, + "num_tokens": 175106736.0, + "step": 4984 + }, + { + "epoch": 0.9257195914577531, + "grad_norm": 1.4858474731445312, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8737185001373291, + "num_tokens": 175139113.0, + "step": 4985 + }, + { + "epoch": 0.9259052924791087, + "grad_norm": 1.466038703918457, + "learning_rate": 1e-06, + "loss": 0.4366, + "mean_token_accuracy": 0.858888566493988, + "num_tokens": 175177941.0, + "step": 4986 + }, + { + "epoch": 0.9260909935004642, + "grad_norm": 1.6412838697433472, + "learning_rate": 1e-06, + "loss": 0.4229, + "mean_token_accuracy": 0.8602118492126465, + "num_tokens": 175206479.0, + "step": 4987 + }, + { + "epoch": 0.9262766945218198, + "grad_norm": 1.6234440803527832, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8672899007797241, + "num_tokens": 175237696.0, + "step": 4988 + }, + { + "epoch": 0.9264623955431754, + "grad_norm": 1.4853419065475464, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8693872690200806, + "num_tokens": 175274910.0, + "step": 4989 + }, + { + "epoch": 0.9266480965645311, + "grad_norm": 1.4028120040893555, + "learning_rate": 1e-06, + "loss": 0.4139, + "mean_token_accuracy": 0.8621498346328735, + "num_tokens": 175316019.0, + "step": 4990 + }, + { + "epoch": 0.9268337975858867, + "grad_norm": 1.3866920471191406, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8764128684997559, + "num_tokens": 175350828.0, + "step": 4991 + }, + { + "epoch": 0.9270194986072423, + "grad_norm": 1.370148777961731, + "learning_rate": 1e-06, + "loss": 0.4255, + "mean_token_accuracy": 0.8570061922073364, + "num_tokens": 175391382.0, + "step": 4992 + }, + { + "epoch": 0.9272051996285979, + "grad_norm": 1.4207653999328613, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8593552112579346, + "num_tokens": 175427802.0, + "step": 4993 + }, + { + "epoch": 0.9273909006499536, + "grad_norm": 1.3630688190460205, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8805211782455444, + "num_tokens": 175463720.0, + "step": 4994 + }, + { + "epoch": 0.9275766016713092, + "grad_norm": 1.409995675086975, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8696231842041016, + "num_tokens": 175499921.0, + "step": 4995 + }, + { + "epoch": 0.9277623026926648, + "grad_norm": 1.60280442237854, + "learning_rate": 1e-06, + "loss": 0.4365, + "mean_token_accuracy": 0.8537460565567017, + "num_tokens": 175531110.0, + "step": 4996 + }, + { + "epoch": 0.9279480037140204, + "grad_norm": 1.3883142471313477, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.8619826436042786, + "num_tokens": 175569027.0, + "step": 4997 + }, + { + "epoch": 0.928133704735376, + "grad_norm": 1.3809220790863037, + "learning_rate": 1e-06, + "loss": 0.4428, + "mean_token_accuracy": 0.8561722636222839, + "num_tokens": 175606267.0, + "step": 4998 + }, + { + "epoch": 0.9283194057567317, + "grad_norm": 1.5619728565216064, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8643726110458374, + "num_tokens": 175640249.0, + "step": 4999 + }, + { + "epoch": 0.9285051067780873, + "grad_norm": 1.4100369215011597, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8843997716903687, + "num_tokens": 175672346.0, + "step": 5000 + }, + { + "epoch": 0.9286908077994429, + "grad_norm": 1.407963752746582, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.871039092540741, + "num_tokens": 175709368.0, + "step": 5001 + }, + { + "epoch": 0.9288765088207985, + "grad_norm": 1.3475942611694336, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8658833503723145, + "num_tokens": 175751341.0, + "step": 5002 + }, + { + "epoch": 0.9290622098421542, + "grad_norm": 1.511647343635559, + "learning_rate": 1e-06, + "loss": 0.4431, + "mean_token_accuracy": 0.858526349067688, + "num_tokens": 175787009.0, + "step": 5003 + }, + { + "epoch": 0.9292479108635098, + "grad_norm": 1.4469656944274902, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8682669401168823, + "num_tokens": 175825339.0, + "step": 5004 + }, + { + "epoch": 0.9294336118848654, + "grad_norm": 1.4301517009735107, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.8670169115066528, + "num_tokens": 175863072.0, + "step": 5005 + }, + { + "epoch": 0.929619312906221, + "grad_norm": 1.466586947441101, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8745894432067871, + "num_tokens": 175897361.0, + "step": 5006 + }, + { + "epoch": 0.9298050139275766, + "grad_norm": 1.374436616897583, + "learning_rate": 1e-06, + "loss": 0.4136, + "mean_token_accuracy": 0.8581793308258057, + "num_tokens": 175938425.0, + "step": 5007 + }, + { + "epoch": 0.9299907149489323, + "grad_norm": 1.3146746158599854, + "learning_rate": 1e-06, + "loss": 0.4096, + "mean_token_accuracy": 0.8671867847442627, + "num_tokens": 175981095.0, + "step": 5008 + }, + { + "epoch": 0.9301764159702879, + "grad_norm": 1.3302379846572876, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8732496500015259, + "num_tokens": 176022315.0, + "step": 5009 + }, + { + "epoch": 0.9303621169916435, + "grad_norm": 1.4365321397781372, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8746404647827148, + "num_tokens": 176057256.0, + "step": 5010 + }, + { + "epoch": 0.930547818012999, + "grad_norm": 1.3431042432785034, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8708387017250061, + "num_tokens": 176098304.0, + "step": 5011 + }, + { + "epoch": 0.9307335190343546, + "grad_norm": 1.5335981845855713, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8829606175422668, + "num_tokens": 176128759.0, + "step": 5012 + }, + { + "epoch": 0.9309192200557103, + "grad_norm": 1.4581165313720703, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8765500783920288, + "num_tokens": 176165785.0, + "step": 5013 + }, + { + "epoch": 0.9311049210770659, + "grad_norm": 1.7909051179885864, + "learning_rate": 1e-06, + "loss": 0.3983, + "mean_token_accuracy": 0.8589972257614136, + "num_tokens": 176190465.0, + "step": 5014 + }, + { + "epoch": 0.9312906220984215, + "grad_norm": 1.4652903079986572, + "learning_rate": 1e-06, + "loss": 0.4331, + "mean_token_accuracy": 0.8544781804084778, + "num_tokens": 176227215.0, + "step": 5015 + }, + { + "epoch": 0.9314763231197771, + "grad_norm": 1.5487860441207886, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8713907599449158, + "num_tokens": 176258239.0, + "step": 5016 + }, + { + "epoch": 0.9316620241411327, + "grad_norm": 1.4760571718215942, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8605858087539673, + "num_tokens": 176298509.0, + "step": 5017 + }, + { + "epoch": 0.9318477251624884, + "grad_norm": 1.658028244972229, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8680711984634399, + "num_tokens": 176326892.0, + "step": 5018 + }, + { + "epoch": 0.932033426183844, + "grad_norm": 1.9132883548736572, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8652383685112, + "num_tokens": 176358363.0, + "step": 5019 + }, + { + "epoch": 0.9322191272051996, + "grad_norm": 1.4471434354782104, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8687251806259155, + "num_tokens": 176391574.0, + "step": 5020 + }, + { + "epoch": 0.9324048282265552, + "grad_norm": 1.601973533630371, + "learning_rate": 1e-06, + "loss": 0.4227, + "mean_token_accuracy": 0.8571823835372925, + "num_tokens": 176422031.0, + "step": 5021 + }, + { + "epoch": 0.9325905292479109, + "grad_norm": 1.728652000427246, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8693041801452637, + "num_tokens": 176450748.0, + "step": 5022 + }, + { + "epoch": 0.9327762302692665, + "grad_norm": 1.3870643377304077, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8741489052772522, + "num_tokens": 176487404.0, + "step": 5023 + }, + { + "epoch": 0.9329619312906221, + "grad_norm": 1.364319086074829, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8748753070831299, + "num_tokens": 176523418.0, + "step": 5024 + }, + { + "epoch": 0.9331476323119777, + "grad_norm": 1.375122308731079, + "learning_rate": 1e-06, + "loss": 0.4393, + "mean_token_accuracy": 0.8549298048019409, + "num_tokens": 176563716.0, + "step": 5025 + }, + { + "epoch": 0.9333333333333333, + "grad_norm": 1.5783467292785645, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8594962358474731, + "num_tokens": 176599431.0, + "step": 5026 + }, + { + "epoch": 0.933519034354689, + "grad_norm": 1.5317879915237427, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8758688569068909, + "num_tokens": 176630014.0, + "step": 5027 + }, + { + "epoch": 0.9337047353760446, + "grad_norm": 1.5238242149353027, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8697353601455688, + "num_tokens": 176660786.0, + "step": 5028 + }, + { + "epoch": 0.9338904363974002, + "grad_norm": 1.2837231159210205, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8764584064483643, + "num_tokens": 176700315.0, + "step": 5029 + }, + { + "epoch": 0.9340761374187558, + "grad_norm": 1.3959203958511353, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.8685944676399231, + "num_tokens": 176737122.0, + "step": 5030 + }, + { + "epoch": 0.9342618384401115, + "grad_norm": 1.5763534307479858, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.8681836128234863, + "num_tokens": 176764834.0, + "step": 5031 + }, + { + "epoch": 0.9344475394614671, + "grad_norm": 1.5975595712661743, + "learning_rate": 1e-06, + "loss": 0.4161, + "mean_token_accuracy": 0.86021888256073, + "num_tokens": 176796212.0, + "step": 5032 + }, + { + "epoch": 0.9346332404828227, + "grad_norm": 1.3978066444396973, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8775561451911926, + "num_tokens": 176832902.0, + "step": 5033 + }, + { + "epoch": 0.9348189415041783, + "grad_norm": 1.429618000984192, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8704568147659302, + "num_tokens": 176870356.0, + "step": 5034 + }, + { + "epoch": 0.9350046425255338, + "grad_norm": 1.4266667366027832, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.875784158706665, + "num_tokens": 176906749.0, + "step": 5035 + }, + { + "epoch": 0.9351903435468895, + "grad_norm": 1.4997389316558838, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8692097663879395, + "num_tokens": 176939684.0, + "step": 5036 + }, + { + "epoch": 0.9353760445682451, + "grad_norm": 1.385286808013916, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8714891672134399, + "num_tokens": 176976338.0, + "step": 5037 + }, + { + "epoch": 0.9355617455896007, + "grad_norm": 1.3147722482681274, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8825748562812805, + "num_tokens": 177015952.0, + "step": 5038 + }, + { + "epoch": 0.9357474466109563, + "grad_norm": 1.4269975423812866, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8684746026992798, + "num_tokens": 177049844.0, + "step": 5039 + }, + { + "epoch": 0.935933147632312, + "grad_norm": 1.50432550907135, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8628018498420715, + "num_tokens": 177085907.0, + "step": 5040 + }, + { + "epoch": 0.9361188486536676, + "grad_norm": 1.4005802869796753, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8663648366928101, + "num_tokens": 177123100.0, + "step": 5041 + }, + { + "epoch": 0.9363045496750232, + "grad_norm": 1.6320109367370605, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8841714859008789, + "num_tokens": 177150065.0, + "step": 5042 + }, + { + "epoch": 0.9364902506963788, + "grad_norm": 1.3514138460159302, + "learning_rate": 1e-06, + "loss": 0.4239, + "mean_token_accuracy": 0.8602744340896606, + "num_tokens": 177190689.0, + "step": 5043 + }, + { + "epoch": 0.9366759517177344, + "grad_norm": 1.4525244235992432, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8886697292327881, + "num_tokens": 177221000.0, + "step": 5044 + }, + { + "epoch": 0.9368616527390901, + "grad_norm": 1.6551557779312134, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.855148434638977, + "num_tokens": 177250714.0, + "step": 5045 + }, + { + "epoch": 0.9370473537604457, + "grad_norm": 1.5357075929641724, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8663307428359985, + "num_tokens": 177282945.0, + "step": 5046 + }, + { + "epoch": 0.9372330547818013, + "grad_norm": 1.4467871189117432, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8701858520507812, + "num_tokens": 177317774.0, + "step": 5047 + }, + { + "epoch": 0.9374187558031569, + "grad_norm": 1.4292620420455933, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8657475709915161, + "num_tokens": 177357657.0, + "step": 5048 + }, + { + "epoch": 0.9376044568245125, + "grad_norm": 1.4327971935272217, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.8678846955299377, + "num_tokens": 177393952.0, + "step": 5049 + }, + { + "epoch": 0.9377901578458682, + "grad_norm": 1.4463971853256226, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8642196655273438, + "num_tokens": 177430681.0, + "step": 5050 + }, + { + "epoch": 0.9379758588672238, + "grad_norm": 1.4342055320739746, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.8636785745620728, + "num_tokens": 177468793.0, + "step": 5051 + }, + { + "epoch": 0.9381615598885794, + "grad_norm": 1.5621256828308105, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8609489798545837, + "num_tokens": 177503204.0, + "step": 5052 + }, + { + "epoch": 0.938347260909935, + "grad_norm": 1.4246997833251953, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8784788846969604, + "num_tokens": 177544587.0, + "step": 5053 + }, + { + "epoch": 0.9385329619312907, + "grad_norm": 1.4664860963821411, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.879531979560852, + "num_tokens": 177577191.0, + "step": 5054 + }, + { + "epoch": 0.9387186629526463, + "grad_norm": 1.4415130615234375, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8731204271316528, + "num_tokens": 177613197.0, + "step": 5055 + }, + { + "epoch": 0.9389043639740019, + "grad_norm": 1.353251576423645, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8740212917327881, + "num_tokens": 177649484.0, + "step": 5056 + }, + { + "epoch": 0.9390900649953575, + "grad_norm": 1.526384711265564, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8817595839500427, + "num_tokens": 177686099.0, + "step": 5057 + }, + { + "epoch": 0.9392757660167131, + "grad_norm": 1.4773850440979004, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8698797225952148, + "num_tokens": 177720223.0, + "step": 5058 + }, + { + "epoch": 0.9394614670380688, + "grad_norm": 1.5508275032043457, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8739452362060547, + "num_tokens": 177752852.0, + "step": 5059 + }, + { + "epoch": 0.9396471680594243, + "grad_norm": 1.4453117847442627, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8700028657913208, + "num_tokens": 177786041.0, + "step": 5060 + }, + { + "epoch": 0.9398328690807799, + "grad_norm": 1.4861292839050293, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8775972127914429, + "num_tokens": 177817678.0, + "step": 5061 + }, + { + "epoch": 0.9400185701021355, + "grad_norm": 1.5105359554290771, + "learning_rate": 1e-06, + "loss": 0.4022, + "mean_token_accuracy": 0.8700745105743408, + "num_tokens": 177850114.0, + "step": 5062 + }, + { + "epoch": 0.9402042711234911, + "grad_norm": 1.418580174446106, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8742178082466125, + "num_tokens": 177889835.0, + "step": 5063 + }, + { + "epoch": 0.9403899721448468, + "grad_norm": 1.456478238105774, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8567616939544678, + "num_tokens": 177927459.0, + "step": 5064 + }, + { + "epoch": 0.9405756731662024, + "grad_norm": 1.370377779006958, + "learning_rate": 1e-06, + "loss": 0.4121, + "mean_token_accuracy": 0.8626214265823364, + "num_tokens": 177971120.0, + "step": 5065 + }, + { + "epoch": 0.940761374187558, + "grad_norm": 1.5695444345474243, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8571277856826782, + "num_tokens": 178006797.0, + "step": 5066 + }, + { + "epoch": 0.9409470752089136, + "grad_norm": 1.5587009191513062, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8593634366989136, + "num_tokens": 178041291.0, + "step": 5067 + }, + { + "epoch": 0.9411327762302693, + "grad_norm": 1.4967515468597412, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8673674464225769, + "num_tokens": 178073634.0, + "step": 5068 + }, + { + "epoch": 0.9413184772516249, + "grad_norm": 1.5417157411575317, + "learning_rate": 1e-06, + "loss": 0.4219, + "mean_token_accuracy": 0.8619913458824158, + "num_tokens": 178109146.0, + "step": 5069 + }, + { + "epoch": 0.9415041782729805, + "grad_norm": 1.361025094985962, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8866482377052307, + "num_tokens": 178142849.0, + "step": 5070 + }, + { + "epoch": 0.9416898792943361, + "grad_norm": 1.4266057014465332, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8764233589172363, + "num_tokens": 178176146.0, + "step": 5071 + }, + { + "epoch": 0.9418755803156917, + "grad_norm": 1.476240873336792, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8613570928573608, + "num_tokens": 178213327.0, + "step": 5072 + }, + { + "epoch": 0.9420612813370474, + "grad_norm": 1.5568501949310303, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8581820130348206, + "num_tokens": 178246213.0, + "step": 5073 + }, + { + "epoch": 0.942246982358403, + "grad_norm": 1.5341675281524658, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8822206258773804, + "num_tokens": 178279474.0, + "step": 5074 + }, + { + "epoch": 0.9424326833797586, + "grad_norm": 1.4327954053878784, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8765943646430969, + "num_tokens": 178313781.0, + "step": 5075 + }, + { + "epoch": 0.9426183844011142, + "grad_norm": 1.308393120765686, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8625706434249878, + "num_tokens": 178357641.0, + "step": 5076 + }, + { + "epoch": 0.9428040854224699, + "grad_norm": 1.383129596710205, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8689983487129211, + "num_tokens": 178400698.0, + "step": 5077 + }, + { + "epoch": 0.9429897864438255, + "grad_norm": 1.4008510112762451, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8797133564949036, + "num_tokens": 178434394.0, + "step": 5078 + }, + { + "epoch": 0.9431754874651811, + "grad_norm": 1.4049115180969238, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8788732290267944, + "num_tokens": 178473434.0, + "step": 5079 + }, + { + "epoch": 0.9433611884865367, + "grad_norm": 1.6876025199890137, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8724003434181213, + "num_tokens": 178502873.0, + "step": 5080 + }, + { + "epoch": 0.9435468895078923, + "grad_norm": 1.4590336084365845, + "learning_rate": 1e-06, + "loss": 0.4428, + "mean_token_accuracy": 0.8526991605758667, + "num_tokens": 178539612.0, + "step": 5081 + }, + { + "epoch": 0.943732590529248, + "grad_norm": 1.5352758169174194, + "learning_rate": 1e-06, + "loss": 0.4862, + "mean_token_accuracy": 0.837799072265625, + "num_tokens": 178577105.0, + "step": 5082 + }, + { + "epoch": 0.9439182915506036, + "grad_norm": 1.4779314994812012, + "learning_rate": 1e-06, + "loss": 0.4367, + "mean_token_accuracy": 0.8510938882827759, + "num_tokens": 178615498.0, + "step": 5083 + }, + { + "epoch": 0.9441039925719591, + "grad_norm": 1.418889045715332, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8792877197265625, + "num_tokens": 178647177.0, + "step": 5084 + }, + { + "epoch": 0.9442896935933147, + "grad_norm": 1.4034831523895264, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8830685615539551, + "num_tokens": 178684599.0, + "step": 5085 + }, + { + "epoch": 0.9444753946146703, + "grad_norm": 1.4511926174163818, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8668234348297119, + "num_tokens": 178719736.0, + "step": 5086 + }, + { + "epoch": 0.944661095636026, + "grad_norm": 1.4151650667190552, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8733828067779541, + "num_tokens": 178757251.0, + "step": 5087 + }, + { + "epoch": 0.9448467966573816, + "grad_norm": 1.5079792737960815, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8752143979072571, + "num_tokens": 178785037.0, + "step": 5088 + }, + { + "epoch": 0.9450324976787372, + "grad_norm": 1.3644697666168213, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8735030889511108, + "num_tokens": 178823149.0, + "step": 5089 + }, + { + "epoch": 0.9452181987000928, + "grad_norm": 1.4677925109863281, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8754176497459412, + "num_tokens": 178852819.0, + "step": 5090 + }, + { + "epoch": 0.9454038997214484, + "grad_norm": 1.613322138786316, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.8548524379730225, + "num_tokens": 178882733.0, + "step": 5091 + }, + { + "epoch": 0.9455896007428041, + "grad_norm": 1.486252784729004, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8615891337394714, + "num_tokens": 178917310.0, + "step": 5092 + }, + { + "epoch": 0.9457753017641597, + "grad_norm": 1.6579309701919556, + "learning_rate": 1e-06, + "loss": 0.4251, + "mean_token_accuracy": 0.8582958579063416, + "num_tokens": 178944682.0, + "step": 5093 + }, + { + "epoch": 0.9459610027855153, + "grad_norm": 1.4161323308944702, + "learning_rate": 1e-06, + "loss": 0.4405, + "mean_token_accuracy": 0.8533908724784851, + "num_tokens": 178985000.0, + "step": 5094 + }, + { + "epoch": 0.9461467038068709, + "grad_norm": 1.477448582649231, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8799047470092773, + "num_tokens": 179020332.0, + "step": 5095 + }, + { + "epoch": 0.9463324048282266, + "grad_norm": 1.4285783767700195, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8693763017654419, + "num_tokens": 179056202.0, + "step": 5096 + }, + { + "epoch": 0.9465181058495822, + "grad_norm": 1.471045732498169, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8708124160766602, + "num_tokens": 179088173.0, + "step": 5097 + }, + { + "epoch": 0.9467038068709378, + "grad_norm": 1.3356237411499023, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8815932273864746, + "num_tokens": 179126015.0, + "step": 5098 + }, + { + "epoch": 0.9468895078922934, + "grad_norm": 1.5531753301620483, + "learning_rate": 1e-06, + "loss": 0.4514, + "mean_token_accuracy": 0.8506553173065186, + "num_tokens": 179164917.0, + "step": 5099 + }, + { + "epoch": 0.947075208913649, + "grad_norm": 1.5137543678283691, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8772310018539429, + "num_tokens": 179193168.0, + "step": 5100 + }, + { + "epoch": 0.9472609099350047, + "grad_norm": 1.4937968254089355, + "learning_rate": 1e-06, + "loss": 0.4123, + "mean_token_accuracy": 0.8591462969779968, + "num_tokens": 179224846.0, + "step": 5101 + }, + { + "epoch": 0.9474466109563603, + "grad_norm": 1.5596786737442017, + "learning_rate": 1e-06, + "loss": 0.4365, + "mean_token_accuracy": 0.8522921204566956, + "num_tokens": 179256365.0, + "step": 5102 + }, + { + "epoch": 0.9476323119777159, + "grad_norm": 1.3285222053527832, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8678321838378906, + "num_tokens": 179299460.0, + "step": 5103 + }, + { + "epoch": 0.9478180129990715, + "grad_norm": 1.4088207483291626, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8780889511108398, + "num_tokens": 179332056.0, + "step": 5104 + }, + { + "epoch": 0.9480037140204272, + "grad_norm": 1.5099488496780396, + "learning_rate": 1e-06, + "loss": 0.4074, + "mean_token_accuracy": 0.8619740009307861, + "num_tokens": 179364331.0, + "step": 5105 + }, + { + "epoch": 0.9481894150417828, + "grad_norm": 1.434309720993042, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8819582462310791, + "num_tokens": 179399663.0, + "step": 5106 + }, + { + "epoch": 0.9483751160631384, + "grad_norm": 1.4390894174575806, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8743462562561035, + "num_tokens": 179430736.0, + "step": 5107 + }, + { + "epoch": 0.9485608170844939, + "grad_norm": 1.4123916625976562, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8753450512886047, + "num_tokens": 179464761.0, + "step": 5108 + }, + { + "epoch": 0.9487465181058495, + "grad_norm": 1.6492964029312134, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8724576234817505, + "num_tokens": 179494272.0, + "step": 5109 + }, + { + "epoch": 0.9489322191272052, + "grad_norm": 1.4808413982391357, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8734464049339294, + "num_tokens": 179529068.0, + "step": 5110 + }, + { + "epoch": 0.9491179201485608, + "grad_norm": 1.2810800075531006, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8799657225608826, + "num_tokens": 179572085.0, + "step": 5111 + }, + { + "epoch": 0.9493036211699164, + "grad_norm": 1.390533208847046, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8689458966255188, + "num_tokens": 179609930.0, + "step": 5112 + }, + { + "epoch": 0.949489322191272, + "grad_norm": 1.5045156478881836, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8592968583106995, + "num_tokens": 179646078.0, + "step": 5113 + }, + { + "epoch": 0.9496750232126276, + "grad_norm": 1.5176029205322266, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8765955567359924, + "num_tokens": 179681858.0, + "step": 5114 + }, + { + "epoch": 0.9498607242339833, + "grad_norm": 1.6469892263412476, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.8656705617904663, + "num_tokens": 179710511.0, + "step": 5115 + }, + { + "epoch": 0.9500464252553389, + "grad_norm": 1.424102544784546, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8818991184234619, + "num_tokens": 179748448.0, + "step": 5116 + }, + { + "epoch": 0.9502321262766945, + "grad_norm": 1.8401272296905518, + "learning_rate": 1e-06, + "loss": 0.426, + "mean_token_accuracy": 0.8569121360778809, + "num_tokens": 179773630.0, + "step": 5117 + }, + { + "epoch": 0.9504178272980501, + "grad_norm": 1.4510138034820557, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8736351132392883, + "num_tokens": 179805755.0, + "step": 5118 + }, + { + "epoch": 0.9506035283194058, + "grad_norm": 1.638214111328125, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8721376061439514, + "num_tokens": 179839059.0, + "step": 5119 + }, + { + "epoch": 0.9507892293407614, + "grad_norm": 1.4233802556991577, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8563164472579956, + "num_tokens": 179876124.0, + "step": 5120 + }, + { + "epoch": 0.950974930362117, + "grad_norm": 1.6471670866012573, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8712888956069946, + "num_tokens": 179906492.0, + "step": 5121 + }, + { + "epoch": 0.9511606313834726, + "grad_norm": 1.5885820388793945, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8671219348907471, + "num_tokens": 179935239.0, + "step": 5122 + }, + { + "epoch": 0.9513463324048282, + "grad_norm": 1.3796839714050293, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8864338397979736, + "num_tokens": 179970539.0, + "step": 5123 + }, + { + "epoch": 0.9515320334261839, + "grad_norm": 1.2304072380065918, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8838292360305786, + "num_tokens": 180009155.0, + "step": 5124 + }, + { + "epoch": 0.9517177344475395, + "grad_norm": 1.3627674579620361, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8715221285820007, + "num_tokens": 180048862.0, + "step": 5125 + }, + { + "epoch": 0.9519034354688951, + "grad_norm": 1.477474331855774, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8763229250907898, + "num_tokens": 180080011.0, + "step": 5126 + }, + { + "epoch": 0.9520891364902507, + "grad_norm": 1.478013515472412, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8661443591117859, + "num_tokens": 180113121.0, + "step": 5127 + }, + { + "epoch": 0.9522748375116064, + "grad_norm": 1.5164636373519897, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8697446584701538, + "num_tokens": 180145673.0, + "step": 5128 + }, + { + "epoch": 0.952460538532962, + "grad_norm": 1.5098479986190796, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8586585521697998, + "num_tokens": 180180379.0, + "step": 5129 + }, + { + "epoch": 0.9526462395543176, + "grad_norm": 1.5183616876602173, + "learning_rate": 1e-06, + "loss": 0.4339, + "mean_token_accuracy": 0.852772057056427, + "num_tokens": 180216594.0, + "step": 5130 + }, + { + "epoch": 0.9528319405756732, + "grad_norm": 1.4377185106277466, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8843138217926025, + "num_tokens": 180251419.0, + "step": 5131 + }, + { + "epoch": 0.9530176415970287, + "grad_norm": 1.5909767150878906, + "learning_rate": 1e-06, + "loss": 0.4599, + "mean_token_accuracy": 0.8483703136444092, + "num_tokens": 180282689.0, + "step": 5132 + }, + { + "epoch": 0.9532033426183844, + "grad_norm": 1.5117329359054565, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8715271949768066, + "num_tokens": 180318325.0, + "step": 5133 + }, + { + "epoch": 0.95338904363974, + "grad_norm": 1.444234848022461, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8672879934310913, + "num_tokens": 180353892.0, + "step": 5134 + }, + { + "epoch": 0.9535747446610956, + "grad_norm": 1.3655614852905273, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8912755250930786, + "num_tokens": 180389658.0, + "step": 5135 + }, + { + "epoch": 0.9537604456824512, + "grad_norm": 1.4932422637939453, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8793154954910278, + "num_tokens": 180422908.0, + "step": 5136 + }, + { + "epoch": 0.9539461467038068, + "grad_norm": 1.532041311264038, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8699250221252441, + "num_tokens": 180452836.0, + "step": 5137 + }, + { + "epoch": 0.9541318477251625, + "grad_norm": 1.402854323387146, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8793582916259766, + "num_tokens": 180485372.0, + "step": 5138 + }, + { + "epoch": 0.9543175487465181, + "grad_norm": 1.4964405298233032, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8642435669898987, + "num_tokens": 180517988.0, + "step": 5139 + }, + { + "epoch": 0.9545032497678737, + "grad_norm": 1.5101591348648071, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8635145425796509, + "num_tokens": 180551778.0, + "step": 5140 + }, + { + "epoch": 0.9546889507892293, + "grad_norm": 1.3742448091506958, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8715783357620239, + "num_tokens": 180592534.0, + "step": 5141 + }, + { + "epoch": 0.954874651810585, + "grad_norm": 1.3892364501953125, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.8628052473068237, + "num_tokens": 180631715.0, + "step": 5142 + }, + { + "epoch": 0.9550603528319406, + "grad_norm": 1.5726226568222046, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8665986061096191, + "num_tokens": 180662858.0, + "step": 5143 + }, + { + "epoch": 0.9552460538532962, + "grad_norm": 1.4259251356124878, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8750301003456116, + "num_tokens": 180697378.0, + "step": 5144 + }, + { + "epoch": 0.9554317548746518, + "grad_norm": 1.5099188089370728, + "learning_rate": 1e-06, + "loss": 0.4478, + "mean_token_accuracy": 0.8513258695602417, + "num_tokens": 180730990.0, + "step": 5145 + }, + { + "epoch": 0.9556174558960074, + "grad_norm": 1.6003392934799194, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8773341178894043, + "num_tokens": 180762864.0, + "step": 5146 + }, + { + "epoch": 0.9558031569173631, + "grad_norm": 1.3400434255599976, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8820880055427551, + "num_tokens": 180800996.0, + "step": 5147 + }, + { + "epoch": 0.9559888579387187, + "grad_norm": 1.5137102603912354, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.859889030456543, + "num_tokens": 180837462.0, + "step": 5148 + }, + { + "epoch": 0.9561745589600743, + "grad_norm": 1.4313108921051025, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8658289909362793, + "num_tokens": 180873729.0, + "step": 5149 + }, + { + "epoch": 0.9563602599814299, + "grad_norm": 1.4447051286697388, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8751591444015503, + "num_tokens": 180907981.0, + "step": 5150 + }, + { + "epoch": 0.9565459610027855, + "grad_norm": 1.5256023406982422, + "learning_rate": 1e-06, + "loss": 0.4255, + "mean_token_accuracy": 0.8600846529006958, + "num_tokens": 180949323.0, + "step": 5151 + }, + { + "epoch": 0.9567316620241412, + "grad_norm": 1.3959203958511353, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8883994221687317, + "num_tokens": 180980286.0, + "step": 5152 + }, + { + "epoch": 0.9569173630454968, + "grad_norm": 1.3876042366027832, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.882358193397522, + "num_tokens": 181015428.0, + "step": 5153 + }, + { + "epoch": 0.9571030640668524, + "grad_norm": 1.3594090938568115, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8783210515975952, + "num_tokens": 181052272.0, + "step": 5154 + }, + { + "epoch": 0.957288765088208, + "grad_norm": 1.509698748588562, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8785080909729004, + "num_tokens": 181079826.0, + "step": 5155 + }, + { + "epoch": 0.9574744661095635, + "grad_norm": 1.3460745811462402, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8844555616378784, + "num_tokens": 181116817.0, + "step": 5156 + }, + { + "epoch": 0.9576601671309192, + "grad_norm": 1.5338953733444214, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8745561838150024, + "num_tokens": 181145440.0, + "step": 5157 + }, + { + "epoch": 0.9578458681522748, + "grad_norm": 1.4498741626739502, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8730481863021851, + "num_tokens": 181177576.0, + "step": 5158 + }, + { + "epoch": 0.9580315691736304, + "grad_norm": 1.5266990661621094, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8775583505630493, + "num_tokens": 181210183.0, + "step": 5159 + }, + { + "epoch": 0.958217270194986, + "grad_norm": 1.416068434715271, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8749439120292664, + "num_tokens": 181243140.0, + "step": 5160 + }, + { + "epoch": 0.9584029712163417, + "grad_norm": 1.2485778331756592, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8839852213859558, + "num_tokens": 181283845.0, + "step": 5161 + }, + { + "epoch": 0.9585886722376973, + "grad_norm": 1.3796966075897217, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8745607733726501, + "num_tokens": 181322534.0, + "step": 5162 + }, + { + "epoch": 0.9587743732590529, + "grad_norm": 1.3983128070831299, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8726146221160889, + "num_tokens": 181357437.0, + "step": 5163 + }, + { + "epoch": 0.9589600742804085, + "grad_norm": 1.4936861991882324, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8731881976127625, + "num_tokens": 181388103.0, + "step": 5164 + }, + { + "epoch": 0.9591457753017641, + "grad_norm": 1.428161859512329, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8665416240692139, + "num_tokens": 181422069.0, + "step": 5165 + }, + { + "epoch": 0.9593314763231198, + "grad_norm": 1.351631999015808, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8761305212974548, + "num_tokens": 181459344.0, + "step": 5166 + }, + { + "epoch": 0.9595171773444754, + "grad_norm": 1.3656262159347534, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8635439872741699, + "num_tokens": 181502021.0, + "step": 5167 + }, + { + "epoch": 0.959702878365831, + "grad_norm": 1.3785001039505005, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8783414959907532, + "num_tokens": 181537314.0, + "step": 5168 + }, + { + "epoch": 0.9598885793871866, + "grad_norm": 1.4620404243469238, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8816800713539124, + "num_tokens": 181570511.0, + "step": 5169 + }, + { + "epoch": 0.9600742804085423, + "grad_norm": 1.4801665544509888, + "learning_rate": 1e-06, + "loss": 0.4479, + "mean_token_accuracy": 0.8474879264831543, + "num_tokens": 181605721.0, + "step": 5170 + }, + { + "epoch": 0.9602599814298979, + "grad_norm": 1.4772354364395142, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8844376802444458, + "num_tokens": 181637048.0, + "step": 5171 + }, + { + "epoch": 0.9604456824512535, + "grad_norm": 1.4375407695770264, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.862209677696228, + "num_tokens": 181674412.0, + "step": 5172 + }, + { + "epoch": 0.9606313834726091, + "grad_norm": 1.430795669555664, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8837194442749023, + "num_tokens": 181708963.0, + "step": 5173 + }, + { + "epoch": 0.9608170844939647, + "grad_norm": 1.4594019651412964, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8566949367523193, + "num_tokens": 181742650.0, + "step": 5174 + }, + { + "epoch": 0.9610027855153204, + "grad_norm": 1.3643029928207397, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8786970376968384, + "num_tokens": 181777345.0, + "step": 5175 + }, + { + "epoch": 0.961188486536676, + "grad_norm": 1.2739711999893188, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8830767273902893, + "num_tokens": 181818671.0, + "step": 5176 + }, + { + "epoch": 0.9613741875580316, + "grad_norm": 1.4248592853546143, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8719673156738281, + "num_tokens": 181851839.0, + "step": 5177 + }, + { + "epoch": 0.9615598885793872, + "grad_norm": 1.4244494438171387, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8843088150024414, + "num_tokens": 181882110.0, + "step": 5178 + }, + { + "epoch": 0.9617455896007429, + "grad_norm": 1.4655007123947144, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8697361946105957, + "num_tokens": 181917591.0, + "step": 5179 + }, + { + "epoch": 0.9619312906220984, + "grad_norm": 1.5445239543914795, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8741282224655151, + "num_tokens": 181946196.0, + "step": 5180 + }, + { + "epoch": 0.962116991643454, + "grad_norm": 1.3408008813858032, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8607895970344543, + "num_tokens": 181989328.0, + "step": 5181 + }, + { + "epoch": 0.9623026926648096, + "grad_norm": 1.3621807098388672, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8804617524147034, + "num_tokens": 182026358.0, + "step": 5182 + }, + { + "epoch": 0.9624883936861652, + "grad_norm": 1.4821152687072754, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.878227174282074, + "num_tokens": 182061072.0, + "step": 5183 + }, + { + "epoch": 0.9626740947075209, + "grad_norm": 1.6174744367599487, + "learning_rate": 1e-06, + "loss": 0.454, + "mean_token_accuracy": 0.8536865711212158, + "num_tokens": 182095461.0, + "step": 5184 + }, + { + "epoch": 0.9628597957288765, + "grad_norm": 1.3499516248703003, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8754017949104309, + "num_tokens": 182134565.0, + "step": 5185 + }, + { + "epoch": 0.9630454967502321, + "grad_norm": 1.4086178541183472, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8750206828117371, + "num_tokens": 182175419.0, + "step": 5186 + }, + { + "epoch": 0.9632311977715877, + "grad_norm": 1.4784998893737793, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8671372532844543, + "num_tokens": 182206064.0, + "step": 5187 + }, + { + "epoch": 0.9634168987929433, + "grad_norm": 1.591097354888916, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8573296070098877, + "num_tokens": 182238660.0, + "step": 5188 + }, + { + "epoch": 0.963602599814299, + "grad_norm": 1.3789905309677124, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.877350389957428, + "num_tokens": 182273114.0, + "step": 5189 + }, + { + "epoch": 0.9637883008356546, + "grad_norm": 1.3980306386947632, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8867393732070923, + "num_tokens": 182310241.0, + "step": 5190 + }, + { + "epoch": 0.9639740018570102, + "grad_norm": 1.5022077560424805, + "learning_rate": 1e-06, + "loss": 0.4368, + "mean_token_accuracy": 0.8542553186416626, + "num_tokens": 182347157.0, + "step": 5191 + }, + { + "epoch": 0.9641597028783658, + "grad_norm": 1.6028915643692017, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8777045011520386, + "num_tokens": 182376259.0, + "step": 5192 + }, + { + "epoch": 0.9643454038997215, + "grad_norm": 1.4004206657409668, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8673064708709717, + "num_tokens": 182416169.0, + "step": 5193 + }, + { + "epoch": 0.9645311049210771, + "grad_norm": 1.477899432182312, + "learning_rate": 1e-06, + "loss": 0.4033, + "mean_token_accuracy": 0.8674569725990295, + "num_tokens": 182449117.0, + "step": 5194 + }, + { + "epoch": 0.9647168059424327, + "grad_norm": 1.490601897239685, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8764505386352539, + "num_tokens": 182478450.0, + "step": 5195 + }, + { + "epoch": 0.9649025069637883, + "grad_norm": 1.4102689027786255, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8755413293838501, + "num_tokens": 182515097.0, + "step": 5196 + }, + { + "epoch": 0.9650882079851439, + "grad_norm": 1.4460110664367676, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.880478024482727, + "num_tokens": 182552053.0, + "step": 5197 + }, + { + "epoch": 0.9652739090064996, + "grad_norm": 1.5640181303024292, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8775483965873718, + "num_tokens": 182585206.0, + "step": 5198 + }, + { + "epoch": 0.9654596100278552, + "grad_norm": 1.5245592594146729, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8619323968887329, + "num_tokens": 182618680.0, + "step": 5199 + }, + { + "epoch": 0.9656453110492108, + "grad_norm": 1.4546384811401367, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8718738555908203, + "num_tokens": 182654848.0, + "step": 5200 + }, + { + "epoch": 0.9658310120705664, + "grad_norm": 1.3560689687728882, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8750371336936951, + "num_tokens": 182691170.0, + "step": 5201 + }, + { + "epoch": 0.966016713091922, + "grad_norm": 1.372901201248169, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8741986751556396, + "num_tokens": 182729673.0, + "step": 5202 + }, + { + "epoch": 0.9662024141132777, + "grad_norm": 1.4800480604171753, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8654886484146118, + "num_tokens": 182771484.0, + "step": 5203 + }, + { + "epoch": 0.9663881151346332, + "grad_norm": 1.5301710367202759, + "learning_rate": 1e-06, + "loss": 0.3958, + "mean_token_accuracy": 0.8688786029815674, + "num_tokens": 182804673.0, + "step": 5204 + }, + { + "epoch": 0.9665738161559888, + "grad_norm": 1.590329647064209, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8683009147644043, + "num_tokens": 182834805.0, + "step": 5205 + }, + { + "epoch": 0.9667595171773444, + "grad_norm": 1.3709737062454224, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8594579100608826, + "num_tokens": 182874311.0, + "step": 5206 + }, + { + "epoch": 0.9669452181987, + "grad_norm": 1.4606635570526123, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.872127890586853, + "num_tokens": 182909918.0, + "step": 5207 + }, + { + "epoch": 0.9671309192200557, + "grad_norm": 1.4518725872039795, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.8605650663375854, + "num_tokens": 182945790.0, + "step": 5208 + }, + { + "epoch": 0.9673166202414113, + "grad_norm": 1.43079674243927, + "learning_rate": 1e-06, + "loss": 0.4134, + "mean_token_accuracy": 0.8600772619247437, + "num_tokens": 182980748.0, + "step": 5209 + }, + { + "epoch": 0.9675023212627669, + "grad_norm": 1.4700196981430054, + "learning_rate": 1e-06, + "loss": 0.4197, + "mean_token_accuracy": 0.8534225225448608, + "num_tokens": 183013943.0, + "step": 5210 + }, + { + "epoch": 0.9676880222841225, + "grad_norm": 1.5548709630966187, + "learning_rate": 1e-06, + "loss": 0.424, + "mean_token_accuracy": 0.8576052784919739, + "num_tokens": 183047193.0, + "step": 5211 + }, + { + "epoch": 0.9678737233054782, + "grad_norm": 1.3860700130462646, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8843250274658203, + "num_tokens": 183084994.0, + "step": 5212 + }, + { + "epoch": 0.9680594243268338, + "grad_norm": 1.4435642957687378, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8696509599685669, + "num_tokens": 183118681.0, + "step": 5213 + }, + { + "epoch": 0.9682451253481894, + "grad_norm": 1.5194380283355713, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.8627910614013672, + "num_tokens": 183156163.0, + "step": 5214 + }, + { + "epoch": 0.968430826369545, + "grad_norm": 1.3342092037200928, + "learning_rate": 1e-06, + "loss": 0.4022, + "mean_token_accuracy": 0.8648965358734131, + "num_tokens": 183201049.0, + "step": 5215 + }, + { + "epoch": 0.9686165273909007, + "grad_norm": 1.3580302000045776, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8702391982078552, + "num_tokens": 183242492.0, + "step": 5216 + }, + { + "epoch": 0.9688022284122563, + "grad_norm": 1.4859449863433838, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8635455965995789, + "num_tokens": 183275722.0, + "step": 5217 + }, + { + "epoch": 0.9689879294336119, + "grad_norm": 1.495589256286621, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8812134861946106, + "num_tokens": 183306135.0, + "step": 5218 + }, + { + "epoch": 0.9691736304549675, + "grad_norm": 1.4143339395523071, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8789347410202026, + "num_tokens": 183339777.0, + "step": 5219 + }, + { + "epoch": 0.9693593314763231, + "grad_norm": 1.605307936668396, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8725187182426453, + "num_tokens": 183368586.0, + "step": 5220 + }, + { + "epoch": 0.9695450324976788, + "grad_norm": 1.6454417705535889, + "learning_rate": 1e-06, + "loss": 0.4275, + "mean_token_accuracy": 0.8575664758682251, + "num_tokens": 183396735.0, + "step": 5221 + }, + { + "epoch": 0.9697307335190344, + "grad_norm": 1.6732257604599, + "learning_rate": 1e-06, + "loss": 0.4413, + "mean_token_accuracy": 0.8528527021408081, + "num_tokens": 183425878.0, + "step": 5222 + }, + { + "epoch": 0.96991643454039, + "grad_norm": 1.5400521755218506, + "learning_rate": 1e-06, + "loss": 0.4116, + "mean_token_accuracy": 0.8624688386917114, + "num_tokens": 183458115.0, + "step": 5223 + }, + { + "epoch": 0.9701021355617456, + "grad_norm": 1.3508388996124268, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8626964092254639, + "num_tokens": 183502462.0, + "step": 5224 + }, + { + "epoch": 0.9702878365831012, + "grad_norm": 1.4846079349517822, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8677332997322083, + "num_tokens": 183535248.0, + "step": 5225 + }, + { + "epoch": 0.9704735376044569, + "grad_norm": 1.3717793226242065, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8864395022392273, + "num_tokens": 183569748.0, + "step": 5226 + }, + { + "epoch": 0.9706592386258125, + "grad_norm": 1.6110358238220215, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8624457120895386, + "num_tokens": 183598725.0, + "step": 5227 + }, + { + "epoch": 0.9708449396471681, + "grad_norm": 1.3599752187728882, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8758878707885742, + "num_tokens": 183643070.0, + "step": 5228 + }, + { + "epoch": 0.9710306406685236, + "grad_norm": 1.4306672811508179, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.8854066133499146, + "num_tokens": 183672549.0, + "step": 5229 + }, + { + "epoch": 0.9712163416898792, + "grad_norm": 1.5153993368148804, + "learning_rate": 1e-06, + "loss": 0.4207, + "mean_token_accuracy": 0.858213484287262, + "num_tokens": 183703826.0, + "step": 5230 + }, + { + "epoch": 0.9714020427112349, + "grad_norm": 1.4755486249923706, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.8635424971580505, + "num_tokens": 183738403.0, + "step": 5231 + }, + { + "epoch": 0.9715877437325905, + "grad_norm": 1.5022683143615723, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8775860071182251, + "num_tokens": 183771300.0, + "step": 5232 + }, + { + "epoch": 0.9717734447539461, + "grad_norm": 1.4721624851226807, + "learning_rate": 1e-06, + "loss": 0.4365, + "mean_token_accuracy": 0.8546047806739807, + "num_tokens": 183808398.0, + "step": 5233 + }, + { + "epoch": 0.9719591457753017, + "grad_norm": 1.474733591079712, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8755311965942383, + "num_tokens": 183842545.0, + "step": 5234 + }, + { + "epoch": 0.9721448467966574, + "grad_norm": 1.5088284015655518, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8612220287322998, + "num_tokens": 183875425.0, + "step": 5235 + }, + { + "epoch": 0.972330547818013, + "grad_norm": 1.4180604219436646, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8649975657463074, + "num_tokens": 183914036.0, + "step": 5236 + }, + { + "epoch": 0.9725162488393686, + "grad_norm": 1.644181728363037, + "learning_rate": 1e-06, + "loss": 0.4354, + "mean_token_accuracy": 0.854767918586731, + "num_tokens": 183942393.0, + "step": 5237 + }, + { + "epoch": 0.9727019498607242, + "grad_norm": 1.3586703538894653, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.875281572341919, + "num_tokens": 183975815.0, + "step": 5238 + }, + { + "epoch": 0.9728876508820798, + "grad_norm": 1.4905176162719727, + "learning_rate": 1e-06, + "loss": 0.4304, + "mean_token_accuracy": 0.8590363264083862, + "num_tokens": 184009930.0, + "step": 5239 + }, + { + "epoch": 0.9730733519034355, + "grad_norm": 1.4856128692626953, + "learning_rate": 1e-06, + "loss": 0.4664, + "mean_token_accuracy": 0.8448481559753418, + "num_tokens": 184047570.0, + "step": 5240 + }, + { + "epoch": 0.9732590529247911, + "grad_norm": 1.3893723487854004, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8788614273071289, + "num_tokens": 184083368.0, + "step": 5241 + }, + { + "epoch": 0.9734447539461467, + "grad_norm": 1.6387948989868164, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8697987198829651, + "num_tokens": 184111772.0, + "step": 5242 + }, + { + "epoch": 0.9736304549675023, + "grad_norm": 1.6269807815551758, + "learning_rate": 1e-06, + "loss": 0.4031, + "mean_token_accuracy": 0.8648592233657837, + "num_tokens": 184143054.0, + "step": 5243 + }, + { + "epoch": 0.973816155988858, + "grad_norm": 1.6119457483291626, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.876069962978363, + "num_tokens": 184178152.0, + "step": 5244 + }, + { + "epoch": 0.9740018570102136, + "grad_norm": 1.4768102169036865, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8648257255554199, + "num_tokens": 184211369.0, + "step": 5245 + }, + { + "epoch": 0.9741875580315692, + "grad_norm": 1.50498366355896, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.8615497946739197, + "num_tokens": 184247222.0, + "step": 5246 + }, + { + "epoch": 0.9743732590529248, + "grad_norm": 1.4417849779129028, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8787328004837036, + "num_tokens": 184284554.0, + "step": 5247 + }, + { + "epoch": 0.9745589600742804, + "grad_norm": 1.5178232192993164, + "learning_rate": 1e-06, + "loss": 0.4271, + "mean_token_accuracy": 0.8572224378585815, + "num_tokens": 184320575.0, + "step": 5248 + }, + { + "epoch": 0.9747446610956361, + "grad_norm": 1.3646259307861328, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.882856011390686, + "num_tokens": 184356661.0, + "step": 5249 + }, + { + "epoch": 0.9749303621169917, + "grad_norm": 1.43464195728302, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8747769594192505, + "num_tokens": 184396446.0, + "step": 5250 + }, + { + "epoch": 0.9751160631383473, + "grad_norm": 1.5257354974746704, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8748300075531006, + "num_tokens": 184426657.0, + "step": 5251 + }, + { + "epoch": 0.9753017641597029, + "grad_norm": 1.5186913013458252, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8698779940605164, + "num_tokens": 184460164.0, + "step": 5252 + }, + { + "epoch": 0.9754874651810584, + "grad_norm": 1.3911422491073608, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8768417835235596, + "num_tokens": 184497934.0, + "step": 5253 + }, + { + "epoch": 0.9756731662024141, + "grad_norm": 1.4357129335403442, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8745490908622742, + "num_tokens": 184534122.0, + "step": 5254 + }, + { + "epoch": 0.9758588672237697, + "grad_norm": 1.5812894105911255, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8802964687347412, + "num_tokens": 184565984.0, + "step": 5255 + }, + { + "epoch": 0.9760445682451253, + "grad_norm": 1.5190128087997437, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8616237044334412, + "num_tokens": 184602511.0, + "step": 5256 + }, + { + "epoch": 0.9762302692664809, + "grad_norm": 1.5202181339263916, + "learning_rate": 1e-06, + "loss": 0.409, + "mean_token_accuracy": 0.8703714609146118, + "num_tokens": 184639991.0, + "step": 5257 + }, + { + "epoch": 0.9764159702878366, + "grad_norm": 1.2833540439605713, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8810915350914001, + "num_tokens": 184681115.0, + "step": 5258 + }, + { + "epoch": 0.9766016713091922, + "grad_norm": 1.40339195728302, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8688122034072876, + "num_tokens": 184717826.0, + "step": 5259 + }, + { + "epoch": 0.9767873723305478, + "grad_norm": 1.4263296127319336, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.879948079586029, + "num_tokens": 184754622.0, + "step": 5260 + }, + { + "epoch": 0.9769730733519034, + "grad_norm": 1.2830508947372437, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8796960711479187, + "num_tokens": 184794448.0, + "step": 5261 + }, + { + "epoch": 0.977158774373259, + "grad_norm": 1.5103105306625366, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8780869245529175, + "num_tokens": 184827792.0, + "step": 5262 + }, + { + "epoch": 0.9773444753946147, + "grad_norm": 1.5920766592025757, + "learning_rate": 1e-06, + "loss": 0.4268, + "mean_token_accuracy": 0.8600342273712158, + "num_tokens": 184858880.0, + "step": 5263 + }, + { + "epoch": 0.9775301764159703, + "grad_norm": 1.3985248804092407, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8635797500610352, + "num_tokens": 184897936.0, + "step": 5264 + }, + { + "epoch": 0.9777158774373259, + "grad_norm": 1.4730373620986938, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.8624048829078674, + "num_tokens": 184933312.0, + "step": 5265 + }, + { + "epoch": 0.9779015784586815, + "grad_norm": 1.3002103567123413, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8801999092102051, + "num_tokens": 184975281.0, + "step": 5266 + }, + { + "epoch": 0.9780872794800372, + "grad_norm": 1.3137930631637573, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8700951337814331, + "num_tokens": 185017273.0, + "step": 5267 + }, + { + "epoch": 0.9782729805013928, + "grad_norm": 1.4394729137420654, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8769375085830688, + "num_tokens": 185050762.0, + "step": 5268 + }, + { + "epoch": 0.9784586815227484, + "grad_norm": 1.43952476978302, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8727227449417114, + "num_tokens": 185085724.0, + "step": 5269 + }, + { + "epoch": 0.978644382544104, + "grad_norm": 1.5447746515274048, + "learning_rate": 1e-06, + "loss": 0.4055, + "mean_token_accuracy": 0.8652925491333008, + "num_tokens": 185118908.0, + "step": 5270 + }, + { + "epoch": 0.9788300835654596, + "grad_norm": 1.3820436000823975, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8803034424781799, + "num_tokens": 185154533.0, + "step": 5271 + }, + { + "epoch": 0.9790157845868153, + "grad_norm": 1.4460656642913818, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8741625547409058, + "num_tokens": 185186582.0, + "step": 5272 + }, + { + "epoch": 0.9792014856081709, + "grad_norm": 1.3886774778366089, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8770797252655029, + "num_tokens": 185221867.0, + "step": 5273 + }, + { + "epoch": 0.9793871866295265, + "grad_norm": 1.4564813375473022, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.866910994052887, + "num_tokens": 185252222.0, + "step": 5274 + }, + { + "epoch": 0.9795728876508821, + "grad_norm": 1.6065385341644287, + "learning_rate": 1e-06, + "loss": 0.4834, + "mean_token_accuracy": 0.8402773141860962, + "num_tokens": 185284343.0, + "step": 5275 + }, + { + "epoch": 0.9797585886722378, + "grad_norm": 1.3954262733459473, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8791297674179077, + "num_tokens": 185319410.0, + "step": 5276 + }, + { + "epoch": 0.9799442896935933, + "grad_norm": 1.4529763460159302, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8647692203521729, + "num_tokens": 185354740.0, + "step": 5277 + }, + { + "epoch": 0.9801299907149489, + "grad_norm": 1.5051630735397339, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8671990036964417, + "num_tokens": 185387758.0, + "step": 5278 + }, + { + "epoch": 0.9803156917363045, + "grad_norm": 1.479979395866394, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8640668392181396, + "num_tokens": 185422306.0, + "step": 5279 + }, + { + "epoch": 0.9805013927576601, + "grad_norm": 1.4950815439224243, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8704575300216675, + "num_tokens": 185453795.0, + "step": 5280 + }, + { + "epoch": 0.9806870937790158, + "grad_norm": 1.5102813243865967, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8628913164138794, + "num_tokens": 185485084.0, + "step": 5281 + }, + { + "epoch": 0.9808727948003714, + "grad_norm": 1.4954581260681152, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8616681694984436, + "num_tokens": 185519763.0, + "step": 5282 + }, + { + "epoch": 0.981058495821727, + "grad_norm": 1.4323065280914307, + "learning_rate": 1e-06, + "loss": 0.4303, + "mean_token_accuracy": 0.8556600213050842, + "num_tokens": 185560364.0, + "step": 5283 + }, + { + "epoch": 0.9812441968430826, + "grad_norm": 1.4729102849960327, + "learning_rate": 1e-06, + "loss": 0.4403, + "mean_token_accuracy": 0.8575341701507568, + "num_tokens": 185595465.0, + "step": 5284 + }, + { + "epoch": 0.9814298978644382, + "grad_norm": 1.5734490156173706, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.871677041053772, + "num_tokens": 185624893.0, + "step": 5285 + }, + { + "epoch": 0.9816155988857939, + "grad_norm": 1.4479652643203735, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8663639426231384, + "num_tokens": 185660648.0, + "step": 5286 + }, + { + "epoch": 0.9818012999071495, + "grad_norm": 1.37771737575531, + "learning_rate": 1e-06, + "loss": 0.4362, + "mean_token_accuracy": 0.8518566489219666, + "num_tokens": 185702070.0, + "step": 5287 + }, + { + "epoch": 0.9819870009285051, + "grad_norm": 1.5054552555084229, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8692610859870911, + "num_tokens": 185735062.0, + "step": 5288 + }, + { + "epoch": 0.9821727019498607, + "grad_norm": 1.447628378868103, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8684496283531189, + "num_tokens": 185772831.0, + "step": 5289 + }, + { + "epoch": 0.9823584029712163, + "grad_norm": 1.5094151496887207, + "learning_rate": 1e-06, + "loss": 0.4059, + "mean_token_accuracy": 0.8622677326202393, + "num_tokens": 185810105.0, + "step": 5290 + }, + { + "epoch": 0.982544103992572, + "grad_norm": 1.3684170246124268, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8804234266281128, + "num_tokens": 185845776.0, + "step": 5291 + }, + { + "epoch": 0.9827298050139276, + "grad_norm": 1.5192428827285767, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.861443281173706, + "num_tokens": 185879695.0, + "step": 5292 + }, + { + "epoch": 0.9829155060352832, + "grad_norm": 1.4116714000701904, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8766698241233826, + "num_tokens": 185916139.0, + "step": 5293 + }, + { + "epoch": 0.9831012070566388, + "grad_norm": 1.5000942945480347, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8716303706169128, + "num_tokens": 185949589.0, + "step": 5294 + }, + { + "epoch": 0.9832869080779945, + "grad_norm": 1.463810920715332, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8692319393157959, + "num_tokens": 185987822.0, + "step": 5295 + }, + { + "epoch": 0.9834726090993501, + "grad_norm": 1.402625560760498, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8682270050048828, + "num_tokens": 186026855.0, + "step": 5296 + }, + { + "epoch": 0.9836583101207057, + "grad_norm": 1.5281033515930176, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8768954873085022, + "num_tokens": 186055561.0, + "step": 5297 + }, + { + "epoch": 0.9838440111420613, + "grad_norm": 1.4019336700439453, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8558738231658936, + "num_tokens": 186096076.0, + "step": 5298 + }, + { + "epoch": 0.984029712163417, + "grad_norm": 1.4227644205093384, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8750999569892883, + "num_tokens": 186132453.0, + "step": 5299 + }, + { + "epoch": 0.9842154131847726, + "grad_norm": 1.4656981229782104, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8734064102172852, + "num_tokens": 186164084.0, + "step": 5300 + }, + { + "epoch": 0.9844011142061281, + "grad_norm": 1.4264713525772095, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8730159997940063, + "num_tokens": 186198602.0, + "step": 5301 + }, + { + "epoch": 0.9845868152274837, + "grad_norm": 1.3339893817901611, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8771908283233643, + "num_tokens": 186239502.0, + "step": 5302 + }, + { + "epoch": 0.9847725162488393, + "grad_norm": 1.5145223140716553, + "learning_rate": 1e-06, + "loss": 0.4336, + "mean_token_accuracy": 0.8584300875663757, + "num_tokens": 186273722.0, + "step": 5303 + }, + { + "epoch": 0.984958217270195, + "grad_norm": 1.3534057140350342, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8845525979995728, + "num_tokens": 186312478.0, + "step": 5304 + }, + { + "epoch": 0.9851439182915506, + "grad_norm": 1.508715033531189, + "learning_rate": 1e-06, + "loss": 0.4214, + "mean_token_accuracy": 0.8599225878715515, + "num_tokens": 186349707.0, + "step": 5305 + }, + { + "epoch": 0.9853296193129062, + "grad_norm": 1.5536576509475708, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8587791919708252, + "num_tokens": 186384520.0, + "step": 5306 + }, + { + "epoch": 0.9855153203342618, + "grad_norm": 1.421593427658081, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8829923868179321, + "num_tokens": 186418565.0, + "step": 5307 + }, + { + "epoch": 0.9857010213556174, + "grad_norm": 1.538393497467041, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8762496709823608, + "num_tokens": 186448702.0, + "step": 5308 + }, + { + "epoch": 0.9858867223769731, + "grad_norm": 1.4665321111679077, + "learning_rate": 1e-06, + "loss": 0.43, + "mean_token_accuracy": 0.8550829291343689, + "num_tokens": 186483993.0, + "step": 5309 + }, + { + "epoch": 0.9860724233983287, + "grad_norm": 1.5703567266464233, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8665387034416199, + "num_tokens": 186520042.0, + "step": 5310 + }, + { + "epoch": 0.9862581244196843, + "grad_norm": 1.41335129737854, + "learning_rate": 1e-06, + "loss": 0.401, + "mean_token_accuracy": 0.8622291684150696, + "num_tokens": 186558484.0, + "step": 5311 + }, + { + "epoch": 0.9864438254410399, + "grad_norm": 1.3574230670928955, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8739572763442993, + "num_tokens": 186596766.0, + "step": 5312 + }, + { + "epoch": 0.9866295264623955, + "grad_norm": 1.448453664779663, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8586654663085938, + "num_tokens": 186633328.0, + "step": 5313 + }, + { + "epoch": 0.9868152274837512, + "grad_norm": 1.3987300395965576, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8690453767776489, + "num_tokens": 186669567.0, + "step": 5314 + }, + { + "epoch": 0.9870009285051068, + "grad_norm": 1.3723323345184326, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8815815448760986, + "num_tokens": 186705293.0, + "step": 5315 + }, + { + "epoch": 0.9871866295264624, + "grad_norm": 1.4897876977920532, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8689229488372803, + "num_tokens": 186741383.0, + "step": 5316 + }, + { + "epoch": 0.987372330547818, + "grad_norm": 1.3941986560821533, + "learning_rate": 1e-06, + "loss": 0.386, + "mean_token_accuracy": 0.8690705895423889, + "num_tokens": 186781737.0, + "step": 5317 + }, + { + "epoch": 0.9875580315691737, + "grad_norm": 1.633399486541748, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8759326934814453, + "num_tokens": 186809185.0, + "step": 5318 + }, + { + "epoch": 0.9877437325905293, + "grad_norm": 1.5412514209747314, + "learning_rate": 1e-06, + "loss": 0.3963, + "mean_token_accuracy": 0.8698092699050903, + "num_tokens": 186838962.0, + "step": 5319 + }, + { + "epoch": 0.9879294336118849, + "grad_norm": 1.3764487504959106, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8743670582771301, + "num_tokens": 186874616.0, + "step": 5320 + }, + { + "epoch": 0.9881151346332405, + "grad_norm": 1.67879056930542, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8636561632156372, + "num_tokens": 186906632.0, + "step": 5321 + }, + { + "epoch": 0.9883008356545961, + "grad_norm": 1.4653117656707764, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8715826272964478, + "num_tokens": 186941952.0, + "step": 5322 + }, + { + "epoch": 0.9884865366759518, + "grad_norm": 1.6846582889556885, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.8647761344909668, + "num_tokens": 186971328.0, + "step": 5323 + }, + { + "epoch": 0.9886722376973074, + "grad_norm": 1.6349385976791382, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8604552149772644, + "num_tokens": 186999667.0, + "step": 5324 + }, + { + "epoch": 0.9888579387186629, + "grad_norm": 1.4956748485565186, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8626207709312439, + "num_tokens": 187033022.0, + "step": 5325 + }, + { + "epoch": 0.9890436397400185, + "grad_norm": 1.3424760103225708, + "learning_rate": 1e-06, + "loss": 0.3991, + "mean_token_accuracy": 0.8631126880645752, + "num_tokens": 187076557.0, + "step": 5326 + }, + { + "epoch": 0.9892293407613741, + "grad_norm": 1.3573150634765625, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8828563690185547, + "num_tokens": 187116289.0, + "step": 5327 + }, + { + "epoch": 0.9894150417827298, + "grad_norm": 1.4696071147918701, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8762606382369995, + "num_tokens": 187149049.0, + "step": 5328 + }, + { + "epoch": 0.9896007428040854, + "grad_norm": 1.6098296642303467, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8671563863754272, + "num_tokens": 187180853.0, + "step": 5329 + }, + { + "epoch": 0.989786443825441, + "grad_norm": 1.3558982610702515, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.871870756149292, + "num_tokens": 187218353.0, + "step": 5330 + }, + { + "epoch": 0.9899721448467966, + "grad_norm": 1.515208125114441, + "learning_rate": 1e-06, + "loss": 0.4419, + "mean_token_accuracy": 0.8579310178756714, + "num_tokens": 187249318.0, + "step": 5331 + }, + { + "epoch": 0.9901578458681523, + "grad_norm": 1.4687390327453613, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.875011682510376, + "num_tokens": 187288630.0, + "step": 5332 + }, + { + "epoch": 0.9903435468895079, + "grad_norm": 1.416159749031067, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8801817893981934, + "num_tokens": 187321367.0, + "step": 5333 + }, + { + "epoch": 0.9905292479108635, + "grad_norm": 1.345489740371704, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.877868115901947, + "num_tokens": 187359917.0, + "step": 5334 + }, + { + "epoch": 0.9907149489322191, + "grad_norm": 1.788629412651062, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.866598904132843, + "num_tokens": 187384586.0, + "step": 5335 + }, + { + "epoch": 0.9909006499535747, + "grad_norm": 1.461147665977478, + "learning_rate": 1e-06, + "loss": 0.4244, + "mean_token_accuracy": 0.8570559024810791, + "num_tokens": 187421922.0, + "step": 5336 + }, + { + "epoch": 0.9910863509749304, + "grad_norm": 1.3708165884017944, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.880487859249115, + "num_tokens": 187457608.0, + "step": 5337 + }, + { + "epoch": 0.991272051996286, + "grad_norm": 1.334999442100525, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8783519864082336, + "num_tokens": 187494208.0, + "step": 5338 + }, + { + "epoch": 0.9914577530176416, + "grad_norm": 1.4320768117904663, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8715946078300476, + "num_tokens": 187532243.0, + "step": 5339 + }, + { + "epoch": 0.9916434540389972, + "grad_norm": 1.4344407320022583, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.875766396522522, + "num_tokens": 187567691.0, + "step": 5340 + }, + { + "epoch": 0.9918291550603529, + "grad_norm": 1.3597403764724731, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8688135743141174, + "num_tokens": 187604953.0, + "step": 5341 + }, + { + "epoch": 0.9920148560817085, + "grad_norm": 1.4384353160858154, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8696197271347046, + "num_tokens": 187640412.0, + "step": 5342 + }, + { + "epoch": 0.9922005571030641, + "grad_norm": 1.5126457214355469, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8782507181167603, + "num_tokens": 187674653.0, + "step": 5343 + }, + { + "epoch": 0.9923862581244197, + "grad_norm": 1.3244341611862183, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8803319931030273, + "num_tokens": 187715067.0, + "step": 5344 + }, + { + "epoch": 0.9925719591457753, + "grad_norm": 1.382576823234558, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8828356266021729, + "num_tokens": 187749059.0, + "step": 5345 + }, + { + "epoch": 0.992757660167131, + "grad_norm": 1.438704252243042, + "learning_rate": 1e-06, + "loss": 0.4243, + "mean_token_accuracy": 0.8581927418708801, + "num_tokens": 187792164.0, + "step": 5346 + }, + { + "epoch": 0.9929433611884866, + "grad_norm": 1.562269926071167, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.869911789894104, + "num_tokens": 187823533.0, + "step": 5347 + }, + { + "epoch": 0.9931290622098422, + "grad_norm": 1.470500111579895, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8575619459152222, + "num_tokens": 187862020.0, + "step": 5348 + }, + { + "epoch": 0.9933147632311977, + "grad_norm": 1.403733491897583, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8694601058959961, + "num_tokens": 187899927.0, + "step": 5349 + }, + { + "epoch": 0.9935004642525533, + "grad_norm": 1.6082158088684082, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8676449060440063, + "num_tokens": 187932483.0, + "step": 5350 + }, + { + "epoch": 0.993686165273909, + "grad_norm": 1.691640019416809, + "learning_rate": 1e-06, + "loss": 0.4873, + "mean_token_accuracy": 0.8451138138771057, + "num_tokens": 187965372.0, + "step": 5351 + }, + { + "epoch": 0.9938718662952646, + "grad_norm": 2.202547550201416, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8705910444259644, + "num_tokens": 188000301.0, + "step": 5352 + }, + { + "epoch": 0.9940575673166202, + "grad_norm": 1.296202301979065, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8732933402061462, + "num_tokens": 188040564.0, + "step": 5353 + }, + { + "epoch": 0.9942432683379758, + "grad_norm": 1.4476642608642578, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.883365273475647, + "num_tokens": 188071655.0, + "step": 5354 + }, + { + "epoch": 0.9944289693593314, + "grad_norm": 1.3758927583694458, + "learning_rate": 1e-06, + "loss": 0.3641, + "mean_token_accuracy": 0.8742330074310303, + "num_tokens": 188111107.0, + "step": 5355 + }, + { + "epoch": 0.9946146703806871, + "grad_norm": 1.4794764518737793, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8821918368339539, + "num_tokens": 188140780.0, + "step": 5356 + }, + { + "epoch": 0.9948003714020427, + "grad_norm": 1.5714116096496582, + "learning_rate": 1e-06, + "loss": 0.4067, + "mean_token_accuracy": 0.8668542504310608, + "num_tokens": 188172190.0, + "step": 5357 + }, + { + "epoch": 0.9949860724233983, + "grad_norm": 1.4589834213256836, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.872414231300354, + "num_tokens": 188204929.0, + "step": 5358 + }, + { + "epoch": 0.9951717734447539, + "grad_norm": 1.4969501495361328, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8580783605575562, + "num_tokens": 188239789.0, + "step": 5359 + }, + { + "epoch": 0.9953574744661096, + "grad_norm": 1.519660234451294, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8729651570320129, + "num_tokens": 188273402.0, + "step": 5360 + }, + { + "epoch": 0.9955431754874652, + "grad_norm": 1.5625369548797607, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.877361536026001, + "num_tokens": 188303660.0, + "step": 5361 + }, + { + "epoch": 0.9957288765088208, + "grad_norm": 1.379440426826477, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8904434442520142, + "num_tokens": 188335372.0, + "step": 5362 + }, + { + "epoch": 0.9959145775301764, + "grad_norm": 1.414282202720642, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8632676601409912, + "num_tokens": 188373394.0, + "step": 5363 + }, + { + "epoch": 0.996100278551532, + "grad_norm": 1.4736629724502563, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8670932054519653, + "num_tokens": 188407301.0, + "step": 5364 + }, + { + "epoch": 0.9962859795728877, + "grad_norm": 1.4988746643066406, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8705700039863586, + "num_tokens": 188438073.0, + "step": 5365 + }, + { + "epoch": 0.9964716805942433, + "grad_norm": 1.6411421298980713, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.869399905204773, + "num_tokens": 188465438.0, + "step": 5366 + }, + { + "epoch": 0.9966573816155989, + "grad_norm": 1.4253019094467163, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8725957870483398, + "num_tokens": 188503247.0, + "step": 5367 + }, + { + "epoch": 0.9968430826369545, + "grad_norm": 1.5132884979248047, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.875910758972168, + "num_tokens": 188536050.0, + "step": 5368 + }, + { + "epoch": 0.9970287836583102, + "grad_norm": 1.5115886926651, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.8695688247680664, + "num_tokens": 188570281.0, + "step": 5369 + }, + { + "epoch": 0.9972144846796658, + "grad_norm": 1.5883336067199707, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8682093024253845, + "num_tokens": 188602368.0, + "step": 5370 + }, + { + "epoch": 0.9974001857010214, + "grad_norm": 1.4184367656707764, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8768448829650879, + "num_tokens": 188637459.0, + "step": 5371 + }, + { + "epoch": 0.997585886722377, + "grad_norm": 1.4964321851730347, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.867814302444458, + "num_tokens": 188674409.0, + "step": 5372 + }, + { + "epoch": 0.9977715877437325, + "grad_norm": 1.4879506826400757, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.853550374507904, + "num_tokens": 188709943.0, + "step": 5373 + }, + { + "epoch": 0.9979572887650882, + "grad_norm": 1.4357364177703857, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8728936910629272, + "num_tokens": 188746639.0, + "step": 5374 + }, + { + "epoch": 0.9981429897864438, + "grad_norm": 1.4187275171279907, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8796674013137817, + "num_tokens": 188782102.0, + "step": 5375 + }, + { + "epoch": 0.9983286908077994, + "grad_norm": 1.4546865224838257, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8730252981185913, + "num_tokens": 188816425.0, + "step": 5376 + }, + { + "epoch": 0.998514391829155, + "grad_norm": 1.4791351556777954, + "learning_rate": 1e-06, + "loss": 0.4358, + "mean_token_accuracy": 0.8557997941970825, + "num_tokens": 188852963.0, + "step": 5377 + }, + { + "epoch": 0.9987000928505106, + "grad_norm": 1.4419963359832764, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8662954568862915, + "num_tokens": 188888759.0, + "step": 5378 + }, + { + "epoch": 0.9988857938718663, + "grad_norm": 1.4987488985061646, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8580164909362793, + "num_tokens": 188920676.0, + "step": 5379 + }, + { + "epoch": 0.9990714948932219, + "grad_norm": 1.4755582809448242, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8736426830291748, + "num_tokens": 188954944.0, + "step": 5380 + }, + { + "epoch": 0.9992571959145775, + "grad_norm": 1.4112598896026611, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8905578851699829, + "num_tokens": 188986729.0, + "step": 5381 + }, + { + "epoch": 0.9994428969359331, + "grad_norm": 1.7989100217819214, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8662295937538147, + "num_tokens": 189010020.0, + "step": 5382 + }, + { + "epoch": 0.9996285979572888, + "grad_norm": 1.3978811502456665, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.880031943321228, + "num_tokens": 189044580.0, + "step": 5383 + }, + { + "epoch": 0.9998142989786444, + "grad_norm": 1.3672611713409424, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.864138126373291, + "num_tokens": 189085789.0, + "step": 5384 + }, + { + "epoch": 1.0, + "grad_norm": 1.4040892124176025, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8799138069152832, + "num_tokens": 189120066.0, + "step": 5385 + }, + { + "epoch": 1.0001857010213555, + "grad_norm": 1.3863563537597656, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8833967447280884, + "num_tokens": 189156216.0, + "step": 5386 + }, + { + "epoch": 1.0003714020427112, + "grad_norm": 1.3346136808395386, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8803529739379883, + "num_tokens": 189195012.0, + "step": 5387 + }, + { + "epoch": 1.0005571030640668, + "grad_norm": 1.346903920173645, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8810161352157593, + "num_tokens": 189233088.0, + "step": 5388 + }, + { + "epoch": 1.0007428040854225, + "grad_norm": 1.462659478187561, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8729438781738281, + "num_tokens": 189265638.0, + "step": 5389 + }, + { + "epoch": 1.000928505106778, + "grad_norm": 1.3988744020462036, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8759135007858276, + "num_tokens": 189303383.0, + "step": 5390 + }, + { + "epoch": 1.0011142061281337, + "grad_norm": 1.301560401916504, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8769925236701965, + "num_tokens": 189344108.0, + "step": 5391 + }, + { + "epoch": 1.0012999071494892, + "grad_norm": 1.4792841672897339, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8712151646614075, + "num_tokens": 189377702.0, + "step": 5392 + }, + { + "epoch": 1.001485608170845, + "grad_norm": 1.6252968311309814, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8638826608657837, + "num_tokens": 189408136.0, + "step": 5393 + }, + { + "epoch": 1.0016713091922005, + "grad_norm": 1.4748634099960327, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.8628091812133789, + "num_tokens": 189441627.0, + "step": 5394 + }, + { + "epoch": 1.0018570102135562, + "grad_norm": 1.5048614740371704, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8734723329544067, + "num_tokens": 189475908.0, + "step": 5395 + }, + { + "epoch": 1.0020427112349117, + "grad_norm": 1.6269534826278687, + "learning_rate": 1e-06, + "loss": 0.4317, + "mean_token_accuracy": 0.8536231517791748, + "num_tokens": 189514754.0, + "step": 5396 + }, + { + "epoch": 1.0022284122562675, + "grad_norm": 1.4508410692214966, + "learning_rate": 1e-06, + "loss": 0.3016, + "mean_token_accuracy": 0.8934864401817322, + "num_tokens": 189545586.0, + "step": 5397 + }, + { + "epoch": 1.002414113277623, + "grad_norm": 1.3819564580917358, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8818497061729431, + "num_tokens": 189583998.0, + "step": 5398 + }, + { + "epoch": 1.0025998142989787, + "grad_norm": 1.5417511463165283, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8732666373252869, + "num_tokens": 189619484.0, + "step": 5399 + }, + { + "epoch": 1.0027855153203342, + "grad_norm": 1.5399583578109741, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.87944495677948, + "num_tokens": 189652295.0, + "step": 5400 + }, + { + "epoch": 1.00297121634169, + "grad_norm": 1.4731745719909668, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8847065567970276, + "num_tokens": 189687331.0, + "step": 5401 + }, + { + "epoch": 1.0031569173630455, + "grad_norm": 1.4903210401535034, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8722947835922241, + "num_tokens": 189725780.0, + "step": 5402 + }, + { + "epoch": 1.0033426183844012, + "grad_norm": 1.6119499206542969, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8918524384498596, + "num_tokens": 189758331.0, + "step": 5403 + }, + { + "epoch": 1.0035283194057567, + "grad_norm": 1.5720272064208984, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8797087073326111, + "num_tokens": 189793886.0, + "step": 5404 + }, + { + "epoch": 1.0037140204271124, + "grad_norm": 1.4559872150421143, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.8757525682449341, + "num_tokens": 189831119.0, + "step": 5405 + }, + { + "epoch": 1.003899721448468, + "grad_norm": 1.4051685333251953, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8785285353660583, + "num_tokens": 189867538.0, + "step": 5406 + }, + { + "epoch": 1.0040854224698237, + "grad_norm": 1.5303020477294922, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8821573257446289, + "num_tokens": 189905936.0, + "step": 5407 + }, + { + "epoch": 1.0042711234911792, + "grad_norm": 1.4327678680419922, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.861773669719696, + "num_tokens": 189942062.0, + "step": 5408 + }, + { + "epoch": 1.004456824512535, + "grad_norm": 1.363932490348816, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8918092250823975, + "num_tokens": 189978754.0, + "step": 5409 + }, + { + "epoch": 1.0046425255338904, + "grad_norm": 1.4255828857421875, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.876562237739563, + "num_tokens": 190016221.0, + "step": 5410 + }, + { + "epoch": 1.004828226555246, + "grad_norm": 1.379294753074646, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8702208399772644, + "num_tokens": 190055705.0, + "step": 5411 + }, + { + "epoch": 1.0050139275766017, + "grad_norm": 1.3794723749160767, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8732714653015137, + "num_tokens": 190097069.0, + "step": 5412 + }, + { + "epoch": 1.0051996285979572, + "grad_norm": 1.6207841634750366, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8762618899345398, + "num_tokens": 190126902.0, + "step": 5413 + }, + { + "epoch": 1.005385329619313, + "grad_norm": 1.3899319171905518, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8757907152175903, + "num_tokens": 190165017.0, + "step": 5414 + }, + { + "epoch": 1.0055710306406684, + "grad_norm": 1.5606892108917236, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8660931587219238, + "num_tokens": 190201403.0, + "step": 5415 + }, + { + "epoch": 1.0057567316620242, + "grad_norm": 1.5701972246170044, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8687175512313843, + "num_tokens": 190238926.0, + "step": 5416 + }, + { + "epoch": 1.0059424326833797, + "grad_norm": 1.461909294128418, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8890343904495239, + "num_tokens": 190275433.0, + "step": 5417 + }, + { + "epoch": 1.0061281337047354, + "grad_norm": 1.5124584436416626, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8749874830245972, + "num_tokens": 190308673.0, + "step": 5418 + }, + { + "epoch": 1.006313834726091, + "grad_norm": 1.4430643320083618, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8746151924133301, + "num_tokens": 190350870.0, + "step": 5419 + }, + { + "epoch": 1.0064995357474467, + "grad_norm": 1.4140225648880005, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8576833605766296, + "num_tokens": 190390472.0, + "step": 5420 + }, + { + "epoch": 1.0066852367688022, + "grad_norm": 1.5269616842269897, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8770757913589478, + "num_tokens": 190425276.0, + "step": 5421 + }, + { + "epoch": 1.006870937790158, + "grad_norm": 1.530984878540039, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8897672891616821, + "num_tokens": 190456966.0, + "step": 5422 + }, + { + "epoch": 1.0070566388115134, + "grad_norm": 1.5186779499053955, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.876110315322876, + "num_tokens": 190489735.0, + "step": 5423 + }, + { + "epoch": 1.0072423398328691, + "grad_norm": 1.5233893394470215, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.872472882270813, + "num_tokens": 190521412.0, + "step": 5424 + }, + { + "epoch": 1.0074280408542247, + "grad_norm": 1.515679955482483, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8704313635826111, + "num_tokens": 190557455.0, + "step": 5425 + }, + { + "epoch": 1.0076137418755804, + "grad_norm": 1.501060962677002, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.874047040939331, + "num_tokens": 190593541.0, + "step": 5426 + }, + { + "epoch": 1.007799442896936, + "grad_norm": 1.5770436525344849, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8695198893547058, + "num_tokens": 190625269.0, + "step": 5427 + }, + { + "epoch": 1.0079851439182916, + "grad_norm": 1.5107803344726562, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8745372295379639, + "num_tokens": 190660787.0, + "step": 5428 + }, + { + "epoch": 1.0081708449396471, + "grad_norm": 1.3310773372650146, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8888475894927979, + "num_tokens": 190698031.0, + "step": 5429 + }, + { + "epoch": 1.0083565459610029, + "grad_norm": 1.6567744016647339, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.864435613155365, + "num_tokens": 190730311.0, + "step": 5430 + }, + { + "epoch": 1.0085422469823584, + "grad_norm": 1.604754090309143, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8667821884155273, + "num_tokens": 190765664.0, + "step": 5431 + }, + { + "epoch": 1.0087279480037141, + "grad_norm": 1.5032529830932617, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8654604554176331, + "num_tokens": 190799579.0, + "step": 5432 + }, + { + "epoch": 1.0089136490250696, + "grad_norm": 1.4963213205337524, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.886363685131073, + "num_tokens": 190834989.0, + "step": 5433 + }, + { + "epoch": 1.0090993500464251, + "grad_norm": 1.2918819189071655, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.886908769607544, + "num_tokens": 190875770.0, + "step": 5434 + }, + { + "epoch": 1.0092850510677809, + "grad_norm": 1.597578763961792, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8650544285774231, + "num_tokens": 190904596.0, + "step": 5435 + }, + { + "epoch": 1.0094707520891364, + "grad_norm": 1.588492751121521, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8773235082626343, + "num_tokens": 190932920.0, + "step": 5436 + }, + { + "epoch": 1.0096564531104921, + "grad_norm": 1.4707528352737427, + "learning_rate": 1e-06, + "loss": 0.4058, + "mean_token_accuracy": 0.859771728515625, + "num_tokens": 190971768.0, + "step": 5437 + }, + { + "epoch": 1.0098421541318476, + "grad_norm": 1.3454416990280151, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8828823566436768, + "num_tokens": 191015841.0, + "step": 5438 + }, + { + "epoch": 1.0100278551532034, + "grad_norm": 1.6291842460632324, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.88334059715271, + "num_tokens": 191043475.0, + "step": 5439 + }, + { + "epoch": 1.0102135561745589, + "grad_norm": 1.5841991901397705, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8684626817703247, + "num_tokens": 191078168.0, + "step": 5440 + }, + { + "epoch": 1.0103992571959146, + "grad_norm": 1.4029743671417236, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8642297983169556, + "num_tokens": 191119943.0, + "step": 5441 + }, + { + "epoch": 1.0105849582172701, + "grad_norm": 1.5104793310165405, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8734504580497742, + "num_tokens": 191152428.0, + "step": 5442 + }, + { + "epoch": 1.0107706592386259, + "grad_norm": 1.6846745014190674, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8655255436897278, + "num_tokens": 191182566.0, + "step": 5443 + }, + { + "epoch": 1.0109563602599814, + "grad_norm": 1.5764983892440796, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8708589673042297, + "num_tokens": 191212974.0, + "step": 5444 + }, + { + "epoch": 1.011142061281337, + "grad_norm": 1.48721182346344, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8614786863327026, + "num_tokens": 191248685.0, + "step": 5445 + }, + { + "epoch": 1.0113277623026926, + "grad_norm": 1.52055025100708, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8741292953491211, + "num_tokens": 191282765.0, + "step": 5446 + }, + { + "epoch": 1.0115134633240483, + "grad_norm": 1.572190523147583, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.867891788482666, + "num_tokens": 191313311.0, + "step": 5447 + }, + { + "epoch": 1.0116991643454039, + "grad_norm": 1.5395265817642212, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8827743530273438, + "num_tokens": 191343195.0, + "step": 5448 + }, + { + "epoch": 1.0118848653667596, + "grad_norm": 1.6374859809875488, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8886654376983643, + "num_tokens": 191372013.0, + "step": 5449 + }, + { + "epoch": 1.012070566388115, + "grad_norm": 1.570701241493225, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8656599521636963, + "num_tokens": 191403958.0, + "step": 5450 + }, + { + "epoch": 1.0122562674094708, + "grad_norm": 1.4693282842636108, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8760923147201538, + "num_tokens": 191437425.0, + "step": 5451 + }, + { + "epoch": 1.0124419684308263, + "grad_norm": 1.3819143772125244, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8820725083351135, + "num_tokens": 191472671.0, + "step": 5452 + }, + { + "epoch": 1.012627669452182, + "grad_norm": 1.409324049949646, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8774281144142151, + "num_tokens": 191510972.0, + "step": 5453 + }, + { + "epoch": 1.0128133704735376, + "grad_norm": 1.4007748365402222, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8838028907775879, + "num_tokens": 191551795.0, + "step": 5454 + }, + { + "epoch": 1.0129990714948933, + "grad_norm": 1.4351377487182617, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8765045404434204, + "num_tokens": 191590218.0, + "step": 5455 + }, + { + "epoch": 1.0131847725162488, + "grad_norm": 1.420915126800537, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8841153383255005, + "num_tokens": 191626122.0, + "step": 5456 + }, + { + "epoch": 1.0133704735376046, + "grad_norm": 1.4463671445846558, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8718088269233704, + "num_tokens": 191663477.0, + "step": 5457 + }, + { + "epoch": 1.01355617455896, + "grad_norm": 1.4407846927642822, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8750844597816467, + "num_tokens": 191699359.0, + "step": 5458 + }, + { + "epoch": 1.0137418755803156, + "grad_norm": 1.3469786643981934, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8812816739082336, + "num_tokens": 191739561.0, + "step": 5459 + }, + { + "epoch": 1.0139275766016713, + "grad_norm": 1.4650027751922607, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8752926588058472, + "num_tokens": 191773489.0, + "step": 5460 + }, + { + "epoch": 1.0141132776230268, + "grad_norm": 1.4509015083312988, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8824528455734253, + "num_tokens": 191808270.0, + "step": 5461 + }, + { + "epoch": 1.0142989786443826, + "grad_norm": 1.688681960105896, + "learning_rate": 1e-06, + "loss": 0.437, + "mean_token_accuracy": 0.8532242178916931, + "num_tokens": 191841342.0, + "step": 5462 + }, + { + "epoch": 1.014484679665738, + "grad_norm": 1.4257309436798096, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8793834447860718, + "num_tokens": 191878534.0, + "step": 5463 + }, + { + "epoch": 1.0146703806870938, + "grad_norm": 1.6702682971954346, + "learning_rate": 1e-06, + "loss": 0.3623, + "mean_token_accuracy": 0.8744570016860962, + "num_tokens": 191912023.0, + "step": 5464 + }, + { + "epoch": 1.0148560817084493, + "grad_norm": 1.6454356908798218, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8816533088684082, + "num_tokens": 191945589.0, + "step": 5465 + }, + { + "epoch": 1.015041782729805, + "grad_norm": 1.566811442375183, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8745496273040771, + "num_tokens": 191979545.0, + "step": 5466 + }, + { + "epoch": 1.0152274837511606, + "grad_norm": 1.6117240190505981, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8731569051742554, + "num_tokens": 192011560.0, + "step": 5467 + }, + { + "epoch": 1.0154131847725163, + "grad_norm": 1.5054608583450317, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8787681460380554, + "num_tokens": 192049879.0, + "step": 5468 + }, + { + "epoch": 1.0155988857938718, + "grad_norm": 1.5961533784866333, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8857662677764893, + "num_tokens": 192080958.0, + "step": 5469 + }, + { + "epoch": 1.0157845868152275, + "grad_norm": 1.5948233604431152, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.875533938407898, + "num_tokens": 192112883.0, + "step": 5470 + }, + { + "epoch": 1.015970287836583, + "grad_norm": 1.450539469718933, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8799257278442383, + "num_tokens": 192148507.0, + "step": 5471 + }, + { + "epoch": 1.0161559888579388, + "grad_norm": 1.4711108207702637, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8774193525314331, + "num_tokens": 192183561.0, + "step": 5472 + }, + { + "epoch": 1.0163416898792943, + "grad_norm": 1.5114721059799194, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.878132700920105, + "num_tokens": 192215563.0, + "step": 5473 + }, + { + "epoch": 1.01652739090065, + "grad_norm": 1.7497894763946533, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8792484402656555, + "num_tokens": 192239749.0, + "step": 5474 + }, + { + "epoch": 1.0167130919220055, + "grad_norm": 1.578348159790039, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8717942237854004, + "num_tokens": 192276832.0, + "step": 5475 + }, + { + "epoch": 1.0168987929433613, + "grad_norm": 1.511757493019104, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.8515457510948181, + "num_tokens": 192316936.0, + "step": 5476 + }, + { + "epoch": 1.0170844939647168, + "grad_norm": 1.482738733291626, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8684337735176086, + "num_tokens": 192352925.0, + "step": 5477 + }, + { + "epoch": 1.0172701949860725, + "grad_norm": 1.488189458847046, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8695486783981323, + "num_tokens": 192391402.0, + "step": 5478 + }, + { + "epoch": 1.017455896007428, + "grad_norm": 1.4105169773101807, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8835881948471069, + "num_tokens": 192429335.0, + "step": 5479 + }, + { + "epoch": 1.0176415970287838, + "grad_norm": 1.7245508432388306, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.8694912791252136, + "num_tokens": 192461170.0, + "step": 5480 + }, + { + "epoch": 1.0178272980501393, + "grad_norm": 1.4206064939498901, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8813905119895935, + "num_tokens": 192503393.0, + "step": 5481 + }, + { + "epoch": 1.0180129990714948, + "grad_norm": 1.5284591913223267, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.887869656085968, + "num_tokens": 192537671.0, + "step": 5482 + }, + { + "epoch": 1.0181987000928505, + "grad_norm": 1.5429513454437256, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8859288096427917, + "num_tokens": 192570482.0, + "step": 5483 + }, + { + "epoch": 1.018384401114206, + "grad_norm": 1.5745093822479248, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8742251396179199, + "num_tokens": 192606197.0, + "step": 5484 + }, + { + "epoch": 1.0185701021355618, + "grad_norm": 1.4925826787948608, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8708276748657227, + "num_tokens": 192644639.0, + "step": 5485 + }, + { + "epoch": 1.0187558031569173, + "grad_norm": 1.483158826828003, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8751817345619202, + "num_tokens": 192681461.0, + "step": 5486 + }, + { + "epoch": 1.018941504178273, + "grad_norm": 1.4848071336746216, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8890878558158875, + "num_tokens": 192713365.0, + "step": 5487 + }, + { + "epoch": 1.0191272051996285, + "grad_norm": 1.5259900093078613, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8783403635025024, + "num_tokens": 192752423.0, + "step": 5488 + }, + { + "epoch": 1.0193129062209842, + "grad_norm": 1.4101669788360596, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8854272365570068, + "num_tokens": 192787962.0, + "step": 5489 + }, + { + "epoch": 1.0194986072423398, + "grad_norm": 1.443893551826477, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8838468790054321, + "num_tokens": 192823023.0, + "step": 5490 + }, + { + "epoch": 1.0196843082636955, + "grad_norm": 1.5090878009796143, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8837398886680603, + "num_tokens": 192859222.0, + "step": 5491 + }, + { + "epoch": 1.019870009285051, + "grad_norm": 1.488301157951355, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8925476670265198, + "num_tokens": 192888753.0, + "step": 5492 + }, + { + "epoch": 1.0200557103064067, + "grad_norm": 1.621415615081787, + "learning_rate": 1e-06, + "loss": 0.401, + "mean_token_accuracy": 0.8635736107826233, + "num_tokens": 192927516.0, + "step": 5493 + }, + { + "epoch": 1.0202414113277622, + "grad_norm": 1.506105899810791, + "learning_rate": 1e-06, + "loss": 0.4467, + "mean_token_accuracy": 0.8469381332397461, + "num_tokens": 192967616.0, + "step": 5494 + }, + { + "epoch": 1.020427112349118, + "grad_norm": 1.480953574180603, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8751264810562134, + "num_tokens": 193003729.0, + "step": 5495 + }, + { + "epoch": 1.0206128133704735, + "grad_norm": 2.0112085342407227, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8747255206108093, + "num_tokens": 193038827.0, + "step": 5496 + }, + { + "epoch": 1.0207985143918292, + "grad_norm": 1.4213443994522095, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8780534863471985, + "num_tokens": 193076576.0, + "step": 5497 + }, + { + "epoch": 1.0209842154131847, + "grad_norm": 1.293293833732605, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8841137290000916, + "num_tokens": 193119159.0, + "step": 5498 + }, + { + "epoch": 1.0211699164345405, + "grad_norm": 1.5617458820343018, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8893775939941406, + "num_tokens": 193148693.0, + "step": 5499 + }, + { + "epoch": 1.021355617455896, + "grad_norm": 1.4761884212493896, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8715344667434692, + "num_tokens": 193184701.0, + "step": 5500 + }, + { + "epoch": 1.0215413184772517, + "grad_norm": 1.4763915538787842, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8718695640563965, + "num_tokens": 193220411.0, + "step": 5501 + }, + { + "epoch": 1.0217270194986072, + "grad_norm": 1.4493016004562378, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8712942600250244, + "num_tokens": 193257549.0, + "step": 5502 + }, + { + "epoch": 1.021912720519963, + "grad_norm": 1.5227854251861572, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8714430332183838, + "num_tokens": 193290728.0, + "step": 5503 + }, + { + "epoch": 1.0220984215413185, + "grad_norm": 1.4247405529022217, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8845659494400024, + "num_tokens": 193327662.0, + "step": 5504 + }, + { + "epoch": 1.0222841225626742, + "grad_norm": 1.475941777229309, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8708567023277283, + "num_tokens": 193362723.0, + "step": 5505 + }, + { + "epoch": 1.0224698235840297, + "grad_norm": 1.478058099746704, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8862667679786682, + "num_tokens": 193396666.0, + "step": 5506 + }, + { + "epoch": 1.0226555246053852, + "grad_norm": 1.3980149030685425, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8655173778533936, + "num_tokens": 193440183.0, + "step": 5507 + }, + { + "epoch": 1.022841225626741, + "grad_norm": 1.425475001335144, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8762259483337402, + "num_tokens": 193477723.0, + "step": 5508 + }, + { + "epoch": 1.0230269266480965, + "grad_norm": 1.6728252172470093, + "learning_rate": 1e-06, + "loss": 0.4203, + "mean_token_accuracy": 0.8581196665763855, + "num_tokens": 193509783.0, + "step": 5509 + }, + { + "epoch": 1.0232126276694522, + "grad_norm": 1.6209051609039307, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.860675036907196, + "num_tokens": 193544482.0, + "step": 5510 + }, + { + "epoch": 1.0233983286908077, + "grad_norm": 1.6198005676269531, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8745417594909668, + "num_tokens": 193571418.0, + "step": 5511 + }, + { + "epoch": 1.0235840297121634, + "grad_norm": 1.4877912998199463, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8870561122894287, + "num_tokens": 193609310.0, + "step": 5512 + }, + { + "epoch": 1.023769730733519, + "grad_norm": 1.46190345287323, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8734143972396851, + "num_tokens": 193645315.0, + "step": 5513 + }, + { + "epoch": 1.0239554317548747, + "grad_norm": 1.5167371034622192, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8800960183143616, + "num_tokens": 193680455.0, + "step": 5514 + }, + { + "epoch": 1.0241411327762302, + "grad_norm": 1.4684542417526245, + "learning_rate": 1e-06, + "loss": 0.4178, + "mean_token_accuracy": 0.8595702052116394, + "num_tokens": 193718343.0, + "step": 5515 + }, + { + "epoch": 1.024326833797586, + "grad_norm": 1.4481867551803589, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8729504942893982, + "num_tokens": 193756429.0, + "step": 5516 + }, + { + "epoch": 1.0245125348189414, + "grad_norm": 1.521158218383789, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8777762651443481, + "num_tokens": 193791250.0, + "step": 5517 + }, + { + "epoch": 1.0246982358402972, + "grad_norm": 1.4449797868728638, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8721290826797485, + "num_tokens": 193832636.0, + "step": 5518 + }, + { + "epoch": 1.0248839368616527, + "grad_norm": 1.5258996486663818, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8784767985343933, + "num_tokens": 193864250.0, + "step": 5519 + }, + { + "epoch": 1.0250696378830084, + "grad_norm": 1.5353227853775024, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8676165342330933, + "num_tokens": 193900446.0, + "step": 5520 + }, + { + "epoch": 1.025255338904364, + "grad_norm": 1.4862172603607178, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8882905840873718, + "num_tokens": 193931012.0, + "step": 5521 + }, + { + "epoch": 1.0254410399257197, + "grad_norm": 1.4800682067871094, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8795384168624878, + "num_tokens": 193964150.0, + "step": 5522 + }, + { + "epoch": 1.0256267409470752, + "grad_norm": 1.4128706455230713, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8662278056144714, + "num_tokens": 194000770.0, + "step": 5523 + }, + { + "epoch": 1.025812441968431, + "grad_norm": 1.4369398355484009, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8858211040496826, + "num_tokens": 194032622.0, + "step": 5524 + }, + { + "epoch": 1.0259981429897864, + "grad_norm": 1.5185966491699219, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8728557825088501, + "num_tokens": 194069019.0, + "step": 5525 + }, + { + "epoch": 1.0261838440111422, + "grad_norm": 1.5030821561813354, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8930507898330688, + "num_tokens": 194096176.0, + "step": 5526 + }, + { + "epoch": 1.0263695450324977, + "grad_norm": 1.4925788640975952, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8777061700820923, + "num_tokens": 194133064.0, + "step": 5527 + }, + { + "epoch": 1.0265552460538534, + "grad_norm": 1.4182720184326172, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8718979358673096, + "num_tokens": 194175734.0, + "step": 5528 + }, + { + "epoch": 1.026740947075209, + "grad_norm": 1.4826359748840332, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8815480470657349, + "num_tokens": 194208280.0, + "step": 5529 + }, + { + "epoch": 1.0269266480965644, + "grad_norm": 1.4994258880615234, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8648228645324707, + "num_tokens": 194243506.0, + "step": 5530 + }, + { + "epoch": 1.0271123491179202, + "grad_norm": 1.6166545152664185, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8687307834625244, + "num_tokens": 194278158.0, + "step": 5531 + }, + { + "epoch": 1.0272980501392757, + "grad_norm": 1.4776464700698853, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8814356923103333, + "num_tokens": 194314587.0, + "step": 5532 + }, + { + "epoch": 1.0274837511606314, + "grad_norm": 1.9181703329086304, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8730190992355347, + "num_tokens": 194344981.0, + "step": 5533 + }, + { + "epoch": 1.027669452181987, + "grad_norm": 1.4582526683807373, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8867192268371582, + "num_tokens": 194383781.0, + "step": 5534 + }, + { + "epoch": 1.0278551532033426, + "grad_norm": 1.7069382667541504, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8632599115371704, + "num_tokens": 194412502.0, + "step": 5535 + }, + { + "epoch": 1.0280408542246982, + "grad_norm": 1.5158056020736694, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8698378205299377, + "num_tokens": 194445685.0, + "step": 5536 + }, + { + "epoch": 1.0282265552460539, + "grad_norm": 1.4496127367019653, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8812952041625977, + "num_tokens": 194481649.0, + "step": 5537 + }, + { + "epoch": 1.0284122562674094, + "grad_norm": 1.4143401384353638, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8806989192962646, + "num_tokens": 194522687.0, + "step": 5538 + }, + { + "epoch": 1.0285979572887651, + "grad_norm": 1.4691174030303955, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8800452947616577, + "num_tokens": 194558635.0, + "step": 5539 + }, + { + "epoch": 1.0287836583101206, + "grad_norm": 1.6083828210830688, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8799508810043335, + "num_tokens": 194588395.0, + "step": 5540 + }, + { + "epoch": 1.0289693593314764, + "grad_norm": 1.535585641860962, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8702098727226257, + "num_tokens": 194620730.0, + "step": 5541 + }, + { + "epoch": 1.0291550603528319, + "grad_norm": 1.4013971090316772, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.872079610824585, + "num_tokens": 194661836.0, + "step": 5542 + }, + { + "epoch": 1.0293407613741876, + "grad_norm": 1.5430412292480469, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8682072758674622, + "num_tokens": 194697737.0, + "step": 5543 + }, + { + "epoch": 1.0295264623955431, + "grad_norm": 1.377529263496399, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8898992538452148, + "num_tokens": 194741624.0, + "step": 5544 + }, + { + "epoch": 1.0297121634168989, + "grad_norm": 1.4982147216796875, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8747230768203735, + "num_tokens": 194773908.0, + "step": 5545 + }, + { + "epoch": 1.0298978644382544, + "grad_norm": 1.4459757804870605, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.886204719543457, + "num_tokens": 194812787.0, + "step": 5546 + }, + { + "epoch": 1.03008356545961, + "grad_norm": 1.5807548761367798, + "learning_rate": 1e-06, + "loss": 0.4039, + "mean_token_accuracy": 0.8665438294410706, + "num_tokens": 194850827.0, + "step": 5547 + }, + { + "epoch": 1.0302692664809656, + "grad_norm": 1.5869911909103394, + "learning_rate": 1e-06, + "loss": 0.4144, + "mean_token_accuracy": 0.8602121472358704, + "num_tokens": 194887925.0, + "step": 5548 + }, + { + "epoch": 1.0304549675023214, + "grad_norm": 1.323337197303772, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8900523781776428, + "num_tokens": 194927460.0, + "step": 5549 + }, + { + "epoch": 1.0306406685236769, + "grad_norm": 1.5940021276474, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8890354037284851, + "num_tokens": 194958442.0, + "step": 5550 + }, + { + "epoch": 1.0308263695450326, + "grad_norm": 1.3956677913665771, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8831552863121033, + "num_tokens": 194995266.0, + "step": 5551 + }, + { + "epoch": 1.031012070566388, + "grad_norm": 1.5482378005981445, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8913301825523376, + "num_tokens": 195027618.0, + "step": 5552 + }, + { + "epoch": 1.0311977715877438, + "grad_norm": 1.4925684928894043, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8848024606704712, + "num_tokens": 195060826.0, + "step": 5553 + }, + { + "epoch": 1.0313834726090993, + "grad_norm": 1.547526478767395, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.879204511642456, + "num_tokens": 195093897.0, + "step": 5554 + }, + { + "epoch": 1.0315691736304549, + "grad_norm": 1.4741774797439575, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8758847713470459, + "num_tokens": 195129624.0, + "step": 5555 + }, + { + "epoch": 1.0317548746518106, + "grad_norm": 1.385788083076477, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.889068603515625, + "num_tokens": 195163548.0, + "step": 5556 + }, + { + "epoch": 1.031940575673166, + "grad_norm": 1.687965750694275, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8764619827270508, + "num_tokens": 195191597.0, + "step": 5557 + }, + { + "epoch": 1.0321262766945218, + "grad_norm": 1.4979400634765625, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8729010820388794, + "num_tokens": 195226471.0, + "step": 5558 + }, + { + "epoch": 1.0323119777158773, + "grad_norm": 1.4283194541931152, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8796550035476685, + "num_tokens": 195265196.0, + "step": 5559 + }, + { + "epoch": 1.032497678737233, + "grad_norm": 1.385852575302124, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8822374939918518, + "num_tokens": 195304853.0, + "step": 5560 + }, + { + "epoch": 1.0326833797585886, + "grad_norm": 1.4532904624938965, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8867987394332886, + "num_tokens": 195340632.0, + "step": 5561 + }, + { + "epoch": 1.0328690807799443, + "grad_norm": 1.5238065719604492, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.874683678150177, + "num_tokens": 195374239.0, + "step": 5562 + }, + { + "epoch": 1.0330547818012998, + "grad_norm": 1.4441670179367065, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8783420920372009, + "num_tokens": 195411061.0, + "step": 5563 + }, + { + "epoch": 1.0332404828226556, + "grad_norm": 1.4858099222183228, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8896259069442749, + "num_tokens": 195444505.0, + "step": 5564 + }, + { + "epoch": 1.033426183844011, + "grad_norm": 1.4537975788116455, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8670949935913086, + "num_tokens": 195481975.0, + "step": 5565 + }, + { + "epoch": 1.0336118848653668, + "grad_norm": 1.4963446855545044, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8830817341804504, + "num_tokens": 195518201.0, + "step": 5566 + }, + { + "epoch": 1.0337975858867223, + "grad_norm": 1.611393690109253, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8850167989730835, + "num_tokens": 195548131.0, + "step": 5567 + }, + { + "epoch": 1.033983286908078, + "grad_norm": 1.4491803646087646, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8907052278518677, + "num_tokens": 195581590.0, + "step": 5568 + }, + { + "epoch": 1.0341689879294336, + "grad_norm": 1.365389347076416, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8769632577896118, + "num_tokens": 195623684.0, + "step": 5569 + }, + { + "epoch": 1.0343546889507893, + "grad_norm": 1.411921501159668, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8873007297515869, + "num_tokens": 195660195.0, + "step": 5570 + }, + { + "epoch": 1.0345403899721448, + "grad_norm": 1.4548399448394775, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8763676285743713, + "num_tokens": 195695091.0, + "step": 5571 + }, + { + "epoch": 1.0347260909935005, + "grad_norm": 1.5612173080444336, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8647428750991821, + "num_tokens": 195726616.0, + "step": 5572 + }, + { + "epoch": 1.034911792014856, + "grad_norm": 1.4711296558380127, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8798799514770508, + "num_tokens": 195760237.0, + "step": 5573 + }, + { + "epoch": 1.0350974930362118, + "grad_norm": 1.49591863155365, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8671504855155945, + "num_tokens": 195797675.0, + "step": 5574 + }, + { + "epoch": 1.0352831940575673, + "grad_norm": 1.4694244861602783, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.878594160079956, + "num_tokens": 195830712.0, + "step": 5575 + }, + { + "epoch": 1.035468895078923, + "grad_norm": 1.4636448621749878, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8796948194503784, + "num_tokens": 195867711.0, + "step": 5576 + }, + { + "epoch": 1.0356545961002785, + "grad_norm": 1.408408284187317, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8681275248527527, + "num_tokens": 195910983.0, + "step": 5577 + }, + { + "epoch": 1.0358402971216343, + "grad_norm": 1.5248759984970093, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8832369446754456, + "num_tokens": 195944631.0, + "step": 5578 + }, + { + "epoch": 1.0360259981429898, + "grad_norm": 1.4222228527069092, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8760267496109009, + "num_tokens": 195985140.0, + "step": 5579 + }, + { + "epoch": 1.0362116991643453, + "grad_norm": 1.5004827976226807, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8683564066886902, + "num_tokens": 196020842.0, + "step": 5580 + }, + { + "epoch": 1.036397400185701, + "grad_norm": 1.6932451725006104, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8646585941314697, + "num_tokens": 196049917.0, + "step": 5581 + }, + { + "epoch": 1.0365831012070565, + "grad_norm": 1.4719077348709106, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8914735317230225, + "num_tokens": 196084271.0, + "step": 5582 + }, + { + "epoch": 1.0367688022284123, + "grad_norm": 1.4629191160202026, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8809248208999634, + "num_tokens": 196123911.0, + "step": 5583 + }, + { + "epoch": 1.0369545032497678, + "grad_norm": 1.3976686000823975, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8815298080444336, + "num_tokens": 196162810.0, + "step": 5584 + }, + { + "epoch": 1.0371402042711235, + "grad_norm": 1.4374464750289917, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8800193071365356, + "num_tokens": 196203139.0, + "step": 5585 + }, + { + "epoch": 1.037325905292479, + "grad_norm": 1.482297658920288, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8759222030639648, + "num_tokens": 196239674.0, + "step": 5586 + }, + { + "epoch": 1.0375116063138348, + "grad_norm": 1.470171570777893, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8735896348953247, + "num_tokens": 196275378.0, + "step": 5587 + }, + { + "epoch": 1.0376973073351903, + "grad_norm": 1.3758941888809204, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8892917633056641, + "num_tokens": 196313227.0, + "step": 5588 + }, + { + "epoch": 1.037883008356546, + "grad_norm": 1.4581172466278076, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8677759170532227, + "num_tokens": 196353623.0, + "step": 5589 + }, + { + "epoch": 1.0380687093779015, + "grad_norm": 1.5054268836975098, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8843649625778198, + "num_tokens": 196388402.0, + "step": 5590 + }, + { + "epoch": 1.0382544103992573, + "grad_norm": 1.4356964826583862, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8875855207443237, + "num_tokens": 196423561.0, + "step": 5591 + }, + { + "epoch": 1.0384401114206128, + "grad_norm": 1.4546560049057007, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8833717107772827, + "num_tokens": 196461025.0, + "step": 5592 + }, + { + "epoch": 1.0386258124419685, + "grad_norm": 1.478480339050293, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8807888031005859, + "num_tokens": 196494962.0, + "step": 5593 + }, + { + "epoch": 1.038811513463324, + "grad_norm": 1.4784938097000122, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8729681372642517, + "num_tokens": 196532966.0, + "step": 5594 + }, + { + "epoch": 1.0389972144846797, + "grad_norm": 1.5723520517349243, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.886188805103302, + "num_tokens": 196564228.0, + "step": 5595 + }, + { + "epoch": 1.0391829155060353, + "grad_norm": 1.4396394491195679, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8629696369171143, + "num_tokens": 196602670.0, + "step": 5596 + }, + { + "epoch": 1.039368616527391, + "grad_norm": 1.5052878856658936, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8746228218078613, + "num_tokens": 196644320.0, + "step": 5597 + }, + { + "epoch": 1.0395543175487465, + "grad_norm": 1.5972776412963867, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8793976306915283, + "num_tokens": 196673838.0, + "step": 5598 + }, + { + "epoch": 1.0397400185701022, + "grad_norm": 1.5293081998825073, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.897844672203064, + "num_tokens": 196704467.0, + "step": 5599 + }, + { + "epoch": 1.0399257195914577, + "grad_norm": 1.4294655323028564, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8802437782287598, + "num_tokens": 196743613.0, + "step": 5600 + }, + { + "epoch": 1.0401114206128135, + "grad_norm": 1.7673368453979492, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8882476091384888, + "num_tokens": 196773601.0, + "step": 5601 + }, + { + "epoch": 1.040297121634169, + "grad_norm": 1.4806554317474365, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8745733499526978, + "num_tokens": 196809248.0, + "step": 5602 + }, + { + "epoch": 1.0404828226555245, + "grad_norm": 1.4554295539855957, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8899103999137878, + "num_tokens": 196841155.0, + "step": 5603 + }, + { + "epoch": 1.0406685236768802, + "grad_norm": 1.5894993543624878, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8721758127212524, + "num_tokens": 196874309.0, + "step": 5604 + }, + { + "epoch": 1.0408542246982357, + "grad_norm": 1.6510717868804932, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8823561668395996, + "num_tokens": 196906899.0, + "step": 5605 + }, + { + "epoch": 1.0410399257195915, + "grad_norm": 1.5924924612045288, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.87516188621521, + "num_tokens": 196940098.0, + "step": 5606 + }, + { + "epoch": 1.041225626740947, + "grad_norm": 1.4288074970245361, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.878895103931427, + "num_tokens": 196977194.0, + "step": 5607 + }, + { + "epoch": 1.0414113277623027, + "grad_norm": 1.5599093437194824, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.876404345035553, + "num_tokens": 197006190.0, + "step": 5608 + }, + { + "epoch": 1.0415970287836582, + "grad_norm": 1.5908339023590088, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8790169954299927, + "num_tokens": 197042536.0, + "step": 5609 + }, + { + "epoch": 1.041782729805014, + "grad_norm": 1.4886916875839233, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.872525155544281, + "num_tokens": 197082428.0, + "step": 5610 + }, + { + "epoch": 1.0419684308263695, + "grad_norm": 1.5724207162857056, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8742086887359619, + "num_tokens": 197115643.0, + "step": 5611 + }, + { + "epoch": 1.0421541318477252, + "grad_norm": 1.4580039978027344, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8764957189559937, + "num_tokens": 197154277.0, + "step": 5612 + }, + { + "epoch": 1.0423398328690807, + "grad_norm": 1.6289628744125366, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8663231134414673, + "num_tokens": 197187549.0, + "step": 5613 + }, + { + "epoch": 1.0425255338904365, + "grad_norm": 1.5334296226501465, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8661797642707825, + "num_tokens": 197227390.0, + "step": 5614 + }, + { + "epoch": 1.042711234911792, + "grad_norm": 1.5663925409317017, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8591490983963013, + "num_tokens": 197262299.0, + "step": 5615 + }, + { + "epoch": 1.0428969359331477, + "grad_norm": 1.6051456928253174, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8743520975112915, + "num_tokens": 197292040.0, + "step": 5616 + }, + { + "epoch": 1.0430826369545032, + "grad_norm": 1.472816824913025, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8757544755935669, + "num_tokens": 197328400.0, + "step": 5617 + }, + { + "epoch": 1.043268337975859, + "grad_norm": 1.5385538339614868, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8675781488418579, + "num_tokens": 197365677.0, + "step": 5618 + }, + { + "epoch": 1.0434540389972145, + "grad_norm": 1.573015570640564, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8737132549285889, + "num_tokens": 197399218.0, + "step": 5619 + }, + { + "epoch": 1.0436397400185702, + "grad_norm": 1.5129623413085938, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8668134212493896, + "num_tokens": 197433516.0, + "step": 5620 + }, + { + "epoch": 1.0438254410399257, + "grad_norm": 1.4078272581100464, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8827353715896606, + "num_tokens": 197472706.0, + "step": 5621 + }, + { + "epoch": 1.0440111420612814, + "grad_norm": 1.4561635255813599, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8801224231719971, + "num_tokens": 197508612.0, + "step": 5622 + }, + { + "epoch": 1.044196843082637, + "grad_norm": 1.4497753381729126, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.884566068649292, + "num_tokens": 197545725.0, + "step": 5623 + }, + { + "epoch": 1.0443825441039927, + "grad_norm": 1.4508506059646606, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8766950368881226, + "num_tokens": 197584329.0, + "step": 5624 + }, + { + "epoch": 1.0445682451253482, + "grad_norm": 1.5264537334442139, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8663829565048218, + "num_tokens": 197620764.0, + "step": 5625 + }, + { + "epoch": 1.0447539461467037, + "grad_norm": 1.5459415912628174, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8732455372810364, + "num_tokens": 197655996.0, + "step": 5626 + }, + { + "epoch": 1.0449396471680594, + "grad_norm": 1.5428128242492676, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8743218183517456, + "num_tokens": 197688692.0, + "step": 5627 + }, + { + "epoch": 1.045125348189415, + "grad_norm": 1.6517179012298584, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8843929767608643, + "num_tokens": 197717068.0, + "step": 5628 + }, + { + "epoch": 1.0453110492107707, + "grad_norm": 1.3358501195907593, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8745279908180237, + "num_tokens": 197760194.0, + "step": 5629 + }, + { + "epoch": 1.0454967502321262, + "grad_norm": 1.4984123706817627, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8830893039703369, + "num_tokens": 197794010.0, + "step": 5630 + }, + { + "epoch": 1.045682451253482, + "grad_norm": 1.5171823501586914, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8810229301452637, + "num_tokens": 197833165.0, + "step": 5631 + }, + { + "epoch": 1.0458681522748374, + "grad_norm": 1.520777702331543, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8688719868659973, + "num_tokens": 197866406.0, + "step": 5632 + }, + { + "epoch": 1.0460538532961932, + "grad_norm": 1.761427879333496, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8805229067802429, + "num_tokens": 197891514.0, + "step": 5633 + }, + { + "epoch": 1.0462395543175487, + "grad_norm": 1.4456298351287842, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8929115533828735, + "num_tokens": 197924451.0, + "step": 5634 + }, + { + "epoch": 1.0464252553389044, + "grad_norm": 1.5726271867752075, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8898442983627319, + "num_tokens": 197953848.0, + "step": 5635 + }, + { + "epoch": 1.04661095636026, + "grad_norm": 1.569968819618225, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8705313205718994, + "num_tokens": 197988562.0, + "step": 5636 + }, + { + "epoch": 1.0467966573816156, + "grad_norm": 1.3321175575256348, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.8982961177825928, + "num_tokens": 198027609.0, + "step": 5637 + }, + { + "epoch": 1.0469823584029712, + "grad_norm": 1.472320556640625, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.871870756149292, + "num_tokens": 198063851.0, + "step": 5638 + }, + { + "epoch": 1.047168059424327, + "grad_norm": 1.5063345432281494, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8820735216140747, + "num_tokens": 198096459.0, + "step": 5639 + }, + { + "epoch": 1.0473537604456824, + "grad_norm": 1.4967321157455444, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8806202411651611, + "num_tokens": 198134341.0, + "step": 5640 + }, + { + "epoch": 1.0475394614670381, + "grad_norm": 1.4919589757919312, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.8984838724136353, + "num_tokens": 198169643.0, + "step": 5641 + }, + { + "epoch": 1.0477251624883936, + "grad_norm": 1.5264086723327637, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8827312588691711, + "num_tokens": 198205207.0, + "step": 5642 + }, + { + "epoch": 1.0479108635097494, + "grad_norm": 1.4670300483703613, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8791845440864563, + "num_tokens": 198238829.0, + "step": 5643 + }, + { + "epoch": 1.048096564531105, + "grad_norm": 1.5221573114395142, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8746345639228821, + "num_tokens": 198273316.0, + "step": 5644 + }, + { + "epoch": 1.0482822655524606, + "grad_norm": 1.770961880683899, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8745867013931274, + "num_tokens": 198299437.0, + "step": 5645 + }, + { + "epoch": 1.0484679665738161, + "grad_norm": 1.4685444831848145, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8701110482215881, + "num_tokens": 198338394.0, + "step": 5646 + }, + { + "epoch": 1.0486536675951719, + "grad_norm": 1.5621118545532227, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8773671388626099, + "num_tokens": 198375849.0, + "step": 5647 + }, + { + "epoch": 1.0488393686165274, + "grad_norm": 1.658837914466858, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8722388744354248, + "num_tokens": 198407752.0, + "step": 5648 + }, + { + "epoch": 1.0490250696378831, + "grad_norm": 1.5917844772338867, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8805934190750122, + "num_tokens": 198445164.0, + "step": 5649 + }, + { + "epoch": 1.0492107706592386, + "grad_norm": 1.546252727508545, + "learning_rate": 1e-06, + "loss": 0.4483, + "mean_token_accuracy": 0.8511605858802795, + "num_tokens": 198484114.0, + "step": 5650 + }, + { + "epoch": 1.0493964716805944, + "grad_norm": 1.4285904169082642, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8776204586029053, + "num_tokens": 198521733.0, + "step": 5651 + }, + { + "epoch": 1.0495821727019499, + "grad_norm": 1.723678708076477, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8626127243041992, + "num_tokens": 198553277.0, + "step": 5652 + }, + { + "epoch": 1.0497678737233054, + "grad_norm": 1.50156831741333, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8752999305725098, + "num_tokens": 198592485.0, + "step": 5653 + }, + { + "epoch": 1.0499535747446611, + "grad_norm": 1.4698622226715088, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8854637742042542, + "num_tokens": 198627345.0, + "step": 5654 + }, + { + "epoch": 1.0501392757660166, + "grad_norm": 1.6220314502716064, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8764277696609497, + "num_tokens": 198663065.0, + "step": 5655 + }, + { + "epoch": 1.0503249767873724, + "grad_norm": 1.5041961669921875, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8690682649612427, + "num_tokens": 198695103.0, + "step": 5656 + }, + { + "epoch": 1.0505106778087279, + "grad_norm": 1.491912603378296, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8639935851097107, + "num_tokens": 198731445.0, + "step": 5657 + }, + { + "epoch": 1.0506963788300836, + "grad_norm": 1.4077666997909546, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8741999864578247, + "num_tokens": 198766945.0, + "step": 5658 + }, + { + "epoch": 1.050882079851439, + "grad_norm": 1.5139353275299072, + "learning_rate": 1e-06, + "loss": 0.4159, + "mean_token_accuracy": 0.8585041165351868, + "num_tokens": 198807349.0, + "step": 5659 + }, + { + "epoch": 1.0510677808727948, + "grad_norm": 1.5659526586532593, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8744990825653076, + "num_tokens": 198839737.0, + "step": 5660 + }, + { + "epoch": 1.0512534818941504, + "grad_norm": 1.4181456565856934, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8893529772758484, + "num_tokens": 198877261.0, + "step": 5661 + }, + { + "epoch": 1.051439182915506, + "grad_norm": 1.5722742080688477, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8682111501693726, + "num_tokens": 198911351.0, + "step": 5662 + }, + { + "epoch": 1.0516248839368616, + "grad_norm": 1.5291597843170166, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8815038204193115, + "num_tokens": 198945100.0, + "step": 5663 + }, + { + "epoch": 1.0518105849582173, + "grad_norm": 1.493167757987976, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8828000426292419, + "num_tokens": 198978216.0, + "step": 5664 + }, + { + "epoch": 1.0519962859795728, + "grad_norm": 1.4082320928573608, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8768181204795837, + "num_tokens": 199014889.0, + "step": 5665 + }, + { + "epoch": 1.0521819870009286, + "grad_norm": 1.4216551780700684, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8786654472351074, + "num_tokens": 199054813.0, + "step": 5666 + }, + { + "epoch": 1.052367688022284, + "grad_norm": 1.4403858184814453, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8665581345558167, + "num_tokens": 199091208.0, + "step": 5667 + }, + { + "epoch": 1.0525533890436398, + "grad_norm": 1.5599361658096313, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8730387687683105, + "num_tokens": 199125626.0, + "step": 5668 + }, + { + "epoch": 1.0527390900649953, + "grad_norm": 1.42124342918396, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8623110055923462, + "num_tokens": 199169311.0, + "step": 5669 + }, + { + "epoch": 1.052924791086351, + "grad_norm": 1.5054771900177002, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8784421682357788, + "num_tokens": 199202168.0, + "step": 5670 + }, + { + "epoch": 1.0531104921077066, + "grad_norm": 1.5230882167816162, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8850154876708984, + "num_tokens": 199232828.0, + "step": 5671 + }, + { + "epoch": 1.0532961931290623, + "grad_norm": 1.434062123298645, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8810179233551025, + "num_tokens": 199270477.0, + "step": 5672 + }, + { + "epoch": 1.0534818941504178, + "grad_norm": 1.3034424781799316, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8702379465103149, + "num_tokens": 199314450.0, + "step": 5673 + }, + { + "epoch": 1.0536675951717736, + "grad_norm": 1.6596797704696655, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8617254495620728, + "num_tokens": 199344703.0, + "step": 5674 + }, + { + "epoch": 1.053853296193129, + "grad_norm": 1.3686820268630981, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8814112544059753, + "num_tokens": 199383163.0, + "step": 5675 + }, + { + "epoch": 1.0540389972144846, + "grad_norm": 1.4861791133880615, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8700334429740906, + "num_tokens": 199422783.0, + "step": 5676 + }, + { + "epoch": 1.0542246982358403, + "grad_norm": 1.3692922592163086, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8763157725334167, + "num_tokens": 199464525.0, + "step": 5677 + }, + { + "epoch": 1.0544103992571958, + "grad_norm": 1.5386791229248047, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8655040264129639, + "num_tokens": 199500460.0, + "step": 5678 + }, + { + "epoch": 1.0545961002785516, + "grad_norm": 1.3215237855911255, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8843536376953125, + "num_tokens": 199540547.0, + "step": 5679 + }, + { + "epoch": 1.054781801299907, + "grad_norm": 1.5408682823181152, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8728480339050293, + "num_tokens": 199575416.0, + "step": 5680 + }, + { + "epoch": 1.0549675023212628, + "grad_norm": 1.400537371635437, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.874926745891571, + "num_tokens": 199613609.0, + "step": 5681 + }, + { + "epoch": 1.0551532033426183, + "grad_norm": 1.4039729833602905, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8846880197525024, + "num_tokens": 199649606.0, + "step": 5682 + }, + { + "epoch": 1.055338904363974, + "grad_norm": 1.4519140720367432, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8853996396064758, + "num_tokens": 199688575.0, + "step": 5683 + }, + { + "epoch": 1.0555246053853296, + "grad_norm": 1.4751474857330322, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8886539340019226, + "num_tokens": 199723004.0, + "step": 5684 + }, + { + "epoch": 1.0557103064066853, + "grad_norm": 1.5092397928237915, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8891017436981201, + "num_tokens": 199757592.0, + "step": 5685 + }, + { + "epoch": 1.0558960074280408, + "grad_norm": 1.6034883260726929, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8771584033966064, + "num_tokens": 199786717.0, + "step": 5686 + }, + { + "epoch": 1.0560817084493965, + "grad_norm": 1.5004010200500488, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8870585560798645, + "num_tokens": 199819815.0, + "step": 5687 + }, + { + "epoch": 1.056267409470752, + "grad_norm": 1.485079288482666, + "learning_rate": 1e-06, + "loss": 0.4072, + "mean_token_accuracy": 0.862770676612854, + "num_tokens": 199857129.0, + "step": 5688 + }, + { + "epoch": 1.0564531104921078, + "grad_norm": 1.529738426208496, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8794072270393372, + "num_tokens": 199889256.0, + "step": 5689 + }, + { + "epoch": 1.0566388115134633, + "grad_norm": 1.5203815698623657, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.8616401553153992, + "num_tokens": 199931356.0, + "step": 5690 + }, + { + "epoch": 1.056824512534819, + "grad_norm": 1.5261974334716797, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.873664379119873, + "num_tokens": 199965426.0, + "step": 5691 + }, + { + "epoch": 1.0570102135561745, + "grad_norm": 1.4069726467132568, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.88807213306427, + "num_tokens": 200001626.0, + "step": 5692 + }, + { + "epoch": 1.0571959145775303, + "grad_norm": 1.6053880453109741, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8720927238464355, + "num_tokens": 200034304.0, + "step": 5693 + }, + { + "epoch": 1.0573816155988858, + "grad_norm": 1.5935384035110474, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8705431818962097, + "num_tokens": 200064318.0, + "step": 5694 + }, + { + "epoch": 1.0575673166202415, + "grad_norm": 1.4554401636123657, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8925708532333374, + "num_tokens": 200096684.0, + "step": 5695 + }, + { + "epoch": 1.057753017641597, + "grad_norm": 1.680983304977417, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8902332782745361, + "num_tokens": 200122795.0, + "step": 5696 + }, + { + "epoch": 1.0579387186629527, + "grad_norm": 1.559607744216919, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8723336458206177, + "num_tokens": 200154830.0, + "step": 5697 + }, + { + "epoch": 1.0581244196843083, + "grad_norm": 1.4770092964172363, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8651756644248962, + "num_tokens": 200192920.0, + "step": 5698 + }, + { + "epoch": 1.0583101207056638, + "grad_norm": 1.424254059791565, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8894791603088379, + "num_tokens": 200227715.0, + "step": 5699 + }, + { + "epoch": 1.0584958217270195, + "grad_norm": 1.436787486076355, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8771089911460876, + "num_tokens": 200265268.0, + "step": 5700 + }, + { + "epoch": 1.058681522748375, + "grad_norm": 1.5095109939575195, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8831254839897156, + "num_tokens": 200297999.0, + "step": 5701 + }, + { + "epoch": 1.0588672237697307, + "grad_norm": 1.7382665872573853, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8807876110076904, + "num_tokens": 200324076.0, + "step": 5702 + }, + { + "epoch": 1.0590529247910863, + "grad_norm": 1.6116642951965332, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8874298334121704, + "num_tokens": 200354465.0, + "step": 5703 + }, + { + "epoch": 1.059238625812442, + "grad_norm": 1.4582864046096802, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.8648241758346558, + "num_tokens": 200393503.0, + "step": 5704 + }, + { + "epoch": 1.0594243268337975, + "grad_norm": 1.5248547792434692, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8763754367828369, + "num_tokens": 200428167.0, + "step": 5705 + }, + { + "epoch": 1.0596100278551532, + "grad_norm": 1.4885772466659546, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8803492784500122, + "num_tokens": 200461187.0, + "step": 5706 + }, + { + "epoch": 1.0597957288765087, + "grad_norm": 1.7015349864959717, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8720709085464478, + "num_tokens": 200493405.0, + "step": 5707 + }, + { + "epoch": 1.0599814298978645, + "grad_norm": 1.577070951461792, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8806190490722656, + "num_tokens": 200525325.0, + "step": 5708 + }, + { + "epoch": 1.06016713091922, + "grad_norm": 1.4061347246170044, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8780776858329773, + "num_tokens": 200564881.0, + "step": 5709 + }, + { + "epoch": 1.0603528319405757, + "grad_norm": 1.5009790658950806, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8742682933807373, + "num_tokens": 200600353.0, + "step": 5710 + }, + { + "epoch": 1.0605385329619312, + "grad_norm": 1.589409351348877, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8802509307861328, + "num_tokens": 200627721.0, + "step": 5711 + }, + { + "epoch": 1.060724233983287, + "grad_norm": 1.3719117641448975, + "learning_rate": 1e-06, + "loss": 0.2807, + "mean_token_accuracy": 0.8987245559692383, + "num_tokens": 200662316.0, + "step": 5712 + }, + { + "epoch": 1.0609099350046425, + "grad_norm": 1.454550862312317, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8767189979553223, + "num_tokens": 200700493.0, + "step": 5713 + }, + { + "epoch": 1.0610956360259982, + "grad_norm": 1.4608080387115479, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8660495281219482, + "num_tokens": 200737165.0, + "step": 5714 + }, + { + "epoch": 1.0612813370473537, + "grad_norm": 1.5949264764785767, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.872768759727478, + "num_tokens": 200768237.0, + "step": 5715 + }, + { + "epoch": 1.0614670380687095, + "grad_norm": 1.5233240127563477, + "learning_rate": 1e-06, + "loss": 0.4308, + "mean_token_accuracy": 0.8598244190216064, + "num_tokens": 200805925.0, + "step": 5716 + }, + { + "epoch": 1.061652739090065, + "grad_norm": 1.556274652481079, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8682030439376831, + "num_tokens": 200839417.0, + "step": 5717 + }, + { + "epoch": 1.0618384401114207, + "grad_norm": 1.602666974067688, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8770252466201782, + "num_tokens": 200871309.0, + "step": 5718 + }, + { + "epoch": 1.0620241411327762, + "grad_norm": 1.5455886125564575, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8691843748092651, + "num_tokens": 200903994.0, + "step": 5719 + }, + { + "epoch": 1.062209842154132, + "grad_norm": 1.5513031482696533, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8680331707000732, + "num_tokens": 200934485.0, + "step": 5720 + }, + { + "epoch": 1.0623955431754875, + "grad_norm": 1.507189154624939, + "learning_rate": 1e-06, + "loss": 0.3623, + "mean_token_accuracy": 0.8741603493690491, + "num_tokens": 200971792.0, + "step": 5721 + }, + { + "epoch": 1.062581244196843, + "grad_norm": 1.5837174654006958, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8700718879699707, + "num_tokens": 201004019.0, + "step": 5722 + }, + { + "epoch": 1.0627669452181987, + "grad_norm": 1.4098132848739624, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8697106838226318, + "num_tokens": 201043981.0, + "step": 5723 + }, + { + "epoch": 1.0629526462395544, + "grad_norm": 1.4911556243896484, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8739422559738159, + "num_tokens": 201078071.0, + "step": 5724 + }, + { + "epoch": 1.06313834726091, + "grad_norm": 1.5588934421539307, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8792544007301331, + "num_tokens": 201109176.0, + "step": 5725 + }, + { + "epoch": 1.0633240482822655, + "grad_norm": 1.5297788381576538, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8733485341072083, + "num_tokens": 201141649.0, + "step": 5726 + }, + { + "epoch": 1.0635097493036212, + "grad_norm": 1.5801254510879517, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8634750247001648, + "num_tokens": 201175197.0, + "step": 5727 + }, + { + "epoch": 1.0636954503249767, + "grad_norm": 1.5063880681991577, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8729127645492554, + "num_tokens": 201209104.0, + "step": 5728 + }, + { + "epoch": 1.0638811513463324, + "grad_norm": 1.5125586986541748, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8766183853149414, + "num_tokens": 201244770.0, + "step": 5729 + }, + { + "epoch": 1.064066852367688, + "grad_norm": 1.4001752138137817, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8693193197250366, + "num_tokens": 201288316.0, + "step": 5730 + }, + { + "epoch": 1.0642525533890437, + "grad_norm": 1.3677231073379517, + "learning_rate": 1e-06, + "loss": 0.301, + "mean_token_accuracy": 0.8952454328536987, + "num_tokens": 201328678.0, + "step": 5731 + }, + { + "epoch": 1.0644382544103992, + "grad_norm": 1.3669832944869995, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8821093440055847, + "num_tokens": 201366843.0, + "step": 5732 + }, + { + "epoch": 1.064623955431755, + "grad_norm": 1.5114349126815796, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8750883340835571, + "num_tokens": 201403454.0, + "step": 5733 + }, + { + "epoch": 1.0648096564531104, + "grad_norm": 1.6093858480453491, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8782164454460144, + "num_tokens": 201433511.0, + "step": 5734 + }, + { + "epoch": 1.0649953574744662, + "grad_norm": 1.4507017135620117, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8775900602340698, + "num_tokens": 201470097.0, + "step": 5735 + }, + { + "epoch": 1.0651810584958217, + "grad_norm": 1.5709831714630127, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8897596597671509, + "num_tokens": 201502772.0, + "step": 5736 + }, + { + "epoch": 1.0653667595171774, + "grad_norm": 1.4189879894256592, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8891018033027649, + "num_tokens": 201539624.0, + "step": 5737 + }, + { + "epoch": 1.065552460538533, + "grad_norm": 1.4673367738723755, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8741600513458252, + "num_tokens": 201576558.0, + "step": 5738 + }, + { + "epoch": 1.0657381615598887, + "grad_norm": 1.6258174180984497, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8661477565765381, + "num_tokens": 201609471.0, + "step": 5739 + }, + { + "epoch": 1.0659238625812442, + "grad_norm": 1.5439330339431763, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8758879899978638, + "num_tokens": 201645550.0, + "step": 5740 + }, + { + "epoch": 1.0661095636026, + "grad_norm": 1.4811487197875977, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8748922348022461, + "num_tokens": 201681731.0, + "step": 5741 + }, + { + "epoch": 1.0662952646239554, + "grad_norm": 1.4435505867004395, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8813597559928894, + "num_tokens": 201725688.0, + "step": 5742 + }, + { + "epoch": 1.0664809656453111, + "grad_norm": 1.5071239471435547, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8801180124282837, + "num_tokens": 201760184.0, + "step": 5743 + }, + { + "epoch": 1.0666666666666667, + "grad_norm": 1.4030816555023193, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8745659589767456, + "num_tokens": 201804064.0, + "step": 5744 + }, + { + "epoch": 1.0668523676880224, + "grad_norm": 1.3899158239364624, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8842043876647949, + "num_tokens": 201842542.0, + "step": 5745 + }, + { + "epoch": 1.067038068709378, + "grad_norm": 1.5670809745788574, + "learning_rate": 1e-06, + "loss": 0.4324, + "mean_token_accuracy": 0.8579078316688538, + "num_tokens": 201878761.0, + "step": 5746 + }, + { + "epoch": 1.0672237697307336, + "grad_norm": 1.501583456993103, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.869811475276947, + "num_tokens": 201913591.0, + "step": 5747 + }, + { + "epoch": 1.0674094707520891, + "grad_norm": 1.6950687170028687, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8687372207641602, + "num_tokens": 201942563.0, + "step": 5748 + }, + { + "epoch": 1.0675951717734447, + "grad_norm": 1.3839571475982666, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.881934404373169, + "num_tokens": 201977376.0, + "step": 5749 + }, + { + "epoch": 1.0677808727948004, + "grad_norm": 1.4737061262130737, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8814302682876587, + "num_tokens": 202011569.0, + "step": 5750 + }, + { + "epoch": 1.067966573816156, + "grad_norm": 1.3698759078979492, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8741399049758911, + "num_tokens": 202053692.0, + "step": 5751 + }, + { + "epoch": 1.0681522748375116, + "grad_norm": 1.4497478008270264, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8869370222091675, + "num_tokens": 202087883.0, + "step": 5752 + }, + { + "epoch": 1.0683379758588671, + "grad_norm": 1.47286057472229, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8854112029075623, + "num_tokens": 202121503.0, + "step": 5753 + }, + { + "epoch": 1.0685236768802229, + "grad_norm": 1.3800925016403198, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.881104588508606, + "num_tokens": 202159579.0, + "step": 5754 + }, + { + "epoch": 1.0687093779015784, + "grad_norm": 1.5859299898147583, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8683645129203796, + "num_tokens": 202195245.0, + "step": 5755 + }, + { + "epoch": 1.0688950789229341, + "grad_norm": 1.5164830684661865, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8706415891647339, + "num_tokens": 202232045.0, + "step": 5756 + }, + { + "epoch": 1.0690807799442896, + "grad_norm": 1.427904486656189, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8928797245025635, + "num_tokens": 202268465.0, + "step": 5757 + }, + { + "epoch": 1.0692664809656454, + "grad_norm": 1.3926297426223755, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8871442079544067, + "num_tokens": 202307013.0, + "step": 5758 + }, + { + "epoch": 1.0694521819870009, + "grad_norm": 1.391404390335083, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8874807357788086, + "num_tokens": 202343618.0, + "step": 5759 + }, + { + "epoch": 1.0696378830083566, + "grad_norm": 1.5005539655685425, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8696593642234802, + "num_tokens": 202382943.0, + "step": 5760 + }, + { + "epoch": 1.0698235840297121, + "grad_norm": 1.4836620092391968, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8789830207824707, + "num_tokens": 202421731.0, + "step": 5761 + }, + { + "epoch": 1.0700092850510678, + "grad_norm": 1.5592180490493774, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8796066045761108, + "num_tokens": 202453478.0, + "step": 5762 + }, + { + "epoch": 1.0701949860724234, + "grad_norm": 1.3862931728363037, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8816350698471069, + "num_tokens": 202490622.0, + "step": 5763 + }, + { + "epoch": 1.070380687093779, + "grad_norm": 1.4805073738098145, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8893190622329712, + "num_tokens": 202523673.0, + "step": 5764 + }, + { + "epoch": 1.0705663881151346, + "grad_norm": 1.467509388923645, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8848171234130859, + "num_tokens": 202557334.0, + "step": 5765 + }, + { + "epoch": 1.0707520891364903, + "grad_norm": 1.5019350051879883, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.875347375869751, + "num_tokens": 202590861.0, + "step": 5766 + }, + { + "epoch": 1.0709377901578458, + "grad_norm": 1.389778733253479, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.888339638710022, + "num_tokens": 202627164.0, + "step": 5767 + }, + { + "epoch": 1.0711234911792016, + "grad_norm": 1.4857064485549927, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8742763996124268, + "num_tokens": 202659881.0, + "step": 5768 + }, + { + "epoch": 1.071309192200557, + "grad_norm": 1.7965508699417114, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8837147951126099, + "num_tokens": 202685575.0, + "step": 5769 + }, + { + "epoch": 1.0714948932219128, + "grad_norm": 1.5731570720672607, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8734208345413208, + "num_tokens": 202718271.0, + "step": 5770 + }, + { + "epoch": 1.0716805942432683, + "grad_norm": 1.5158895254135132, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8842948079109192, + "num_tokens": 202750445.0, + "step": 5771 + }, + { + "epoch": 1.0718662952646238, + "grad_norm": 1.5822968482971191, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8782837986946106, + "num_tokens": 202779539.0, + "step": 5772 + }, + { + "epoch": 1.0720519962859796, + "grad_norm": 1.6558467149734497, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8720150589942932, + "num_tokens": 202810380.0, + "step": 5773 + }, + { + "epoch": 1.072237697307335, + "grad_norm": 1.5096020698547363, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.872329831123352, + "num_tokens": 202849116.0, + "step": 5774 + }, + { + "epoch": 1.0724233983286908, + "grad_norm": 1.377768874168396, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8843870162963867, + "num_tokens": 202889881.0, + "step": 5775 + }, + { + "epoch": 1.0726090993500463, + "grad_norm": 1.6454925537109375, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.869443953037262, + "num_tokens": 202922317.0, + "step": 5776 + }, + { + "epoch": 1.072794800371402, + "grad_norm": 1.5177401304244995, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8885250687599182, + "num_tokens": 202953889.0, + "step": 5777 + }, + { + "epoch": 1.0729805013927576, + "grad_norm": 1.6037683486938477, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.87874436378479, + "num_tokens": 202983830.0, + "step": 5778 + }, + { + "epoch": 1.0731662024141133, + "grad_norm": 1.8173903226852417, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8664676547050476, + "num_tokens": 203008844.0, + "step": 5779 + }, + { + "epoch": 1.0733519034354688, + "grad_norm": 1.5559625625610352, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8645095229148865, + "num_tokens": 203042945.0, + "step": 5780 + }, + { + "epoch": 1.0735376044568246, + "grad_norm": 1.64024019241333, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8769923448562622, + "num_tokens": 203071758.0, + "step": 5781 + }, + { + "epoch": 1.07372330547818, + "grad_norm": 1.604136347770691, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8770052194595337, + "num_tokens": 203100742.0, + "step": 5782 + }, + { + "epoch": 1.0739090064995358, + "grad_norm": 1.4680410623550415, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8824127912521362, + "num_tokens": 203139282.0, + "step": 5783 + }, + { + "epoch": 1.0740947075208913, + "grad_norm": 1.5062326192855835, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8805999755859375, + "num_tokens": 203171147.0, + "step": 5784 + }, + { + "epoch": 1.074280408542247, + "grad_norm": 1.4396132230758667, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8758242130279541, + "num_tokens": 203207858.0, + "step": 5785 + }, + { + "epoch": 1.0744661095636026, + "grad_norm": 1.5870890617370605, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8853176832199097, + "num_tokens": 203240030.0, + "step": 5786 + }, + { + "epoch": 1.0746518105849583, + "grad_norm": 1.6907798051834106, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8681432604789734, + "num_tokens": 203269149.0, + "step": 5787 + }, + { + "epoch": 1.0748375116063138, + "grad_norm": 1.575829267501831, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8604574203491211, + "num_tokens": 203299631.0, + "step": 5788 + }, + { + "epoch": 1.0750232126276695, + "grad_norm": 1.5649417638778687, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8878703117370605, + "num_tokens": 203330703.0, + "step": 5789 + }, + { + "epoch": 1.075208913649025, + "grad_norm": 1.630924940109253, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8635275363922119, + "num_tokens": 203364182.0, + "step": 5790 + }, + { + "epoch": 1.0753946146703808, + "grad_norm": 1.5052796602249146, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.880089282989502, + "num_tokens": 203394877.0, + "step": 5791 + }, + { + "epoch": 1.0755803156917363, + "grad_norm": 1.6569141149520874, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.889581561088562, + "num_tokens": 203426641.0, + "step": 5792 + }, + { + "epoch": 1.075766016713092, + "grad_norm": 1.5632282495498657, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.865480899810791, + "num_tokens": 203467951.0, + "step": 5793 + }, + { + "epoch": 1.0759517177344475, + "grad_norm": 1.390842080116272, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8910492062568665, + "num_tokens": 203506101.0, + "step": 5794 + }, + { + "epoch": 1.076137418755803, + "grad_norm": 1.429764986038208, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8813360929489136, + "num_tokens": 203544321.0, + "step": 5795 + }, + { + "epoch": 1.0763231197771588, + "grad_norm": 1.5945125818252563, + "learning_rate": 1e-06, + "loss": 0.4251, + "mean_token_accuracy": 0.8583328723907471, + "num_tokens": 203578245.0, + "step": 5796 + }, + { + "epoch": 1.0765088207985145, + "grad_norm": 1.3607503175735474, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8789380192756653, + "num_tokens": 203619078.0, + "step": 5797 + }, + { + "epoch": 1.07669452181987, + "grad_norm": 1.3704110383987427, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8794480562210083, + "num_tokens": 203659544.0, + "step": 5798 + }, + { + "epoch": 1.0768802228412255, + "grad_norm": 1.6875025033950806, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8937122225761414, + "num_tokens": 203687750.0, + "step": 5799 + }, + { + "epoch": 1.0770659238625813, + "grad_norm": 1.7487562894821167, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8847599625587463, + "num_tokens": 203713692.0, + "step": 5800 + }, + { + "epoch": 1.0772516248839368, + "grad_norm": 1.5690208673477173, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8683157563209534, + "num_tokens": 203748511.0, + "step": 5801 + }, + { + "epoch": 1.0774373259052925, + "grad_norm": 1.633874773979187, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8686860799789429, + "num_tokens": 203780292.0, + "step": 5802 + }, + { + "epoch": 1.077623026926648, + "grad_norm": 1.441157341003418, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8736087083816528, + "num_tokens": 203818903.0, + "step": 5803 + }, + { + "epoch": 1.0778087279480038, + "grad_norm": 1.5674175024032593, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8702856302261353, + "num_tokens": 203853073.0, + "step": 5804 + }, + { + "epoch": 1.0779944289693593, + "grad_norm": 1.4440128803253174, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8944253921508789, + "num_tokens": 203889059.0, + "step": 5805 + }, + { + "epoch": 1.078180129990715, + "grad_norm": 1.4130144119262695, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8835144639015198, + "num_tokens": 203923860.0, + "step": 5806 + }, + { + "epoch": 1.0783658310120705, + "grad_norm": 1.4737107753753662, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8754976987838745, + "num_tokens": 203965507.0, + "step": 5807 + }, + { + "epoch": 1.0785515320334262, + "grad_norm": 1.562637448310852, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8864045739173889, + "num_tokens": 203999323.0, + "step": 5808 + }, + { + "epoch": 1.0787372330547818, + "grad_norm": 1.5428845882415771, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8642986416816711, + "num_tokens": 204036528.0, + "step": 5809 + }, + { + "epoch": 1.0789229340761375, + "grad_norm": 1.4743220806121826, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8791994452476501, + "num_tokens": 204070163.0, + "step": 5810 + }, + { + "epoch": 1.079108635097493, + "grad_norm": 1.4763160943984985, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8745156526565552, + "num_tokens": 204109443.0, + "step": 5811 + }, + { + "epoch": 1.0792943361188487, + "grad_norm": 1.6567661762237549, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8750790953636169, + "num_tokens": 204139592.0, + "step": 5812 + }, + { + "epoch": 1.0794800371402042, + "grad_norm": 1.6138919591903687, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8804352283477783, + "num_tokens": 204176999.0, + "step": 5813 + }, + { + "epoch": 1.07966573816156, + "grad_norm": 1.6265779733657837, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8817774057388306, + "num_tokens": 204205723.0, + "step": 5814 + }, + { + "epoch": 1.0798514391829155, + "grad_norm": 1.7250897884368896, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8735501766204834, + "num_tokens": 204234486.0, + "step": 5815 + }, + { + "epoch": 1.0800371402042712, + "grad_norm": 1.5320700407028198, + "learning_rate": 1e-06, + "loss": 0.3911, + "mean_token_accuracy": 0.868559718132019, + "num_tokens": 204270889.0, + "step": 5816 + }, + { + "epoch": 1.0802228412256267, + "grad_norm": 1.4750062227249146, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.875954806804657, + "num_tokens": 204309233.0, + "step": 5817 + }, + { + "epoch": 1.0804085422469825, + "grad_norm": 1.4223055839538574, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8903635144233704, + "num_tokens": 204345321.0, + "step": 5818 + }, + { + "epoch": 1.080594243268338, + "grad_norm": 1.4894371032714844, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8746677041053772, + "num_tokens": 204383742.0, + "step": 5819 + }, + { + "epoch": 1.0807799442896937, + "grad_norm": 1.4512193202972412, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8753927946090698, + "num_tokens": 204421400.0, + "step": 5820 + }, + { + "epoch": 1.0809656453110492, + "grad_norm": 1.5514230728149414, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8768953084945679, + "num_tokens": 204454914.0, + "step": 5821 + }, + { + "epoch": 1.0811513463324047, + "grad_norm": 1.5236858129501343, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8852607011795044, + "num_tokens": 204485138.0, + "step": 5822 + }, + { + "epoch": 1.0813370473537605, + "grad_norm": 1.3526617288589478, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8752167224884033, + "num_tokens": 204530953.0, + "step": 5823 + }, + { + "epoch": 1.081522748375116, + "grad_norm": 1.5273534059524536, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8820770978927612, + "num_tokens": 204563740.0, + "step": 5824 + }, + { + "epoch": 1.0817084493964717, + "grad_norm": 1.5064786672592163, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.871866762638092, + "num_tokens": 204595524.0, + "step": 5825 + }, + { + "epoch": 1.0818941504178272, + "grad_norm": 1.5231810808181763, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8887312412261963, + "num_tokens": 204630208.0, + "step": 5826 + }, + { + "epoch": 1.082079851439183, + "grad_norm": 1.5480842590332031, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8702581524848938, + "num_tokens": 204665480.0, + "step": 5827 + }, + { + "epoch": 1.0822655524605385, + "grad_norm": 1.374861717224121, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8868345618247986, + "num_tokens": 204705892.0, + "step": 5828 + }, + { + "epoch": 1.0824512534818942, + "grad_norm": 1.5932402610778809, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8798328042030334, + "num_tokens": 204737713.0, + "step": 5829 + }, + { + "epoch": 1.0826369545032497, + "grad_norm": 1.4970908164978027, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8658607006072998, + "num_tokens": 204773540.0, + "step": 5830 + }, + { + "epoch": 1.0828226555246054, + "grad_norm": 1.564389944076538, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8859442472457886, + "num_tokens": 204807394.0, + "step": 5831 + }, + { + "epoch": 1.083008356545961, + "grad_norm": 1.3685067892074585, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8901964426040649, + "num_tokens": 204845108.0, + "step": 5832 + }, + { + "epoch": 1.0831940575673167, + "grad_norm": 1.5753307342529297, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8756607174873352, + "num_tokens": 204877124.0, + "step": 5833 + }, + { + "epoch": 1.0833797585886722, + "grad_norm": 1.4806972742080688, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8802632093429565, + "num_tokens": 204911089.0, + "step": 5834 + }, + { + "epoch": 1.083565459610028, + "grad_norm": 1.489268183708191, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8650748133659363, + "num_tokens": 204951800.0, + "step": 5835 + }, + { + "epoch": 1.0837511606313834, + "grad_norm": 1.3913648128509521, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8880492448806763, + "num_tokens": 204988915.0, + "step": 5836 + }, + { + "epoch": 1.0839368616527392, + "grad_norm": 1.45298171043396, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8764589428901672, + "num_tokens": 205024732.0, + "step": 5837 + }, + { + "epoch": 1.0841225626740947, + "grad_norm": 1.4552960395812988, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8994033336639404, + "num_tokens": 205057663.0, + "step": 5838 + }, + { + "epoch": 1.0843082636954504, + "grad_norm": 1.4472118616104126, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8839871883392334, + "num_tokens": 205091814.0, + "step": 5839 + }, + { + "epoch": 1.084493964716806, + "grad_norm": 1.4214776754379272, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8820945024490356, + "num_tokens": 205127765.0, + "step": 5840 + }, + { + "epoch": 1.0846796657381617, + "grad_norm": 1.405024766921997, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8837783336639404, + "num_tokens": 205164101.0, + "step": 5841 + }, + { + "epoch": 1.0848653667595172, + "grad_norm": 1.4781649112701416, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8733445405960083, + "num_tokens": 205199538.0, + "step": 5842 + }, + { + "epoch": 1.085051067780873, + "grad_norm": 1.4820303916931152, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8931499719619751, + "num_tokens": 205233111.0, + "step": 5843 + }, + { + "epoch": 1.0852367688022284, + "grad_norm": 1.4242922067642212, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8850956559181213, + "num_tokens": 205267131.0, + "step": 5844 + }, + { + "epoch": 1.085422469823584, + "grad_norm": 1.5489850044250488, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8791077136993408, + "num_tokens": 205298441.0, + "step": 5845 + }, + { + "epoch": 1.0856081708449397, + "grad_norm": 1.5601038932800293, + "learning_rate": 1e-06, + "loss": 0.41, + "mean_token_accuracy": 0.8642684817314148, + "num_tokens": 205331650.0, + "step": 5846 + }, + { + "epoch": 1.0857938718662952, + "grad_norm": 1.5498452186584473, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8765606880187988, + "num_tokens": 205364768.0, + "step": 5847 + }, + { + "epoch": 1.085979572887651, + "grad_norm": 1.7563337087631226, + "learning_rate": 1e-06, + "loss": 0.4058, + "mean_token_accuracy": 0.8636216521263123, + "num_tokens": 205391032.0, + "step": 5848 + }, + { + "epoch": 1.0861652739090064, + "grad_norm": 1.6002609729766846, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8844919204711914, + "num_tokens": 205419799.0, + "step": 5849 + }, + { + "epoch": 1.0863509749303621, + "grad_norm": 1.4415675401687622, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8733523488044739, + "num_tokens": 205457399.0, + "step": 5850 + }, + { + "epoch": 1.0865366759517177, + "grad_norm": 1.4274709224700928, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8838642835617065, + "num_tokens": 205496513.0, + "step": 5851 + }, + { + "epoch": 1.0867223769730734, + "grad_norm": 1.4670284986495972, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8709672689437866, + "num_tokens": 205534460.0, + "step": 5852 + }, + { + "epoch": 1.086908077994429, + "grad_norm": 1.362634301185608, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8878629803657532, + "num_tokens": 205571780.0, + "step": 5853 + }, + { + "epoch": 1.0870937790157846, + "grad_norm": 1.5205860137939453, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8777634501457214, + "num_tokens": 205603941.0, + "step": 5854 + }, + { + "epoch": 1.0872794800371401, + "grad_norm": 1.5882084369659424, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8626888990402222, + "num_tokens": 205638107.0, + "step": 5855 + }, + { + "epoch": 1.0874651810584959, + "grad_norm": 1.4405137300491333, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8737365007400513, + "num_tokens": 205677812.0, + "step": 5856 + }, + { + "epoch": 1.0876508820798514, + "grad_norm": 1.533617377281189, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8746106624603271, + "num_tokens": 205709974.0, + "step": 5857 + }, + { + "epoch": 1.0878365831012071, + "grad_norm": 1.465806245803833, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8696222305297852, + "num_tokens": 205746294.0, + "step": 5858 + }, + { + "epoch": 1.0880222841225626, + "grad_norm": 1.6040208339691162, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8672613501548767, + "num_tokens": 205780005.0, + "step": 5859 + }, + { + "epoch": 1.0882079851439184, + "grad_norm": 1.7196235656738281, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8774375915527344, + "num_tokens": 205810598.0, + "step": 5860 + }, + { + "epoch": 1.0883936861652739, + "grad_norm": 1.4143882989883423, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.891581654548645, + "num_tokens": 205845764.0, + "step": 5861 + }, + { + "epoch": 1.0885793871866296, + "grad_norm": 1.5576322078704834, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8755103349685669, + "num_tokens": 205880143.0, + "step": 5862 + }, + { + "epoch": 1.0887650882079851, + "grad_norm": 1.4323714971542358, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.88312166929245, + "num_tokens": 205918710.0, + "step": 5863 + }, + { + "epoch": 1.0889507892293409, + "grad_norm": 1.493335247039795, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8667210340499878, + "num_tokens": 205958365.0, + "step": 5864 + }, + { + "epoch": 1.0891364902506964, + "grad_norm": 1.5594992637634277, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8864963054656982, + "num_tokens": 205989197.0, + "step": 5865 + }, + { + "epoch": 1.089322191272052, + "grad_norm": 1.4319454431533813, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8826519250869751, + "num_tokens": 206028871.0, + "step": 5866 + }, + { + "epoch": 1.0895078922934076, + "grad_norm": 1.499596118927002, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8810806274414062, + "num_tokens": 206063378.0, + "step": 5867 + }, + { + "epoch": 1.0896935933147631, + "grad_norm": 1.558270812034607, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.855941653251648, + "num_tokens": 206100598.0, + "step": 5868 + }, + { + "epoch": 1.0898792943361189, + "grad_norm": 1.5847910642623901, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8627533316612244, + "num_tokens": 206134847.0, + "step": 5869 + }, + { + "epoch": 1.0900649953574744, + "grad_norm": 1.5353435277938843, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8712945580482483, + "num_tokens": 206171469.0, + "step": 5870 + }, + { + "epoch": 1.09025069637883, + "grad_norm": 1.446256399154663, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8771607279777527, + "num_tokens": 206212422.0, + "step": 5871 + }, + { + "epoch": 1.0904363974001856, + "grad_norm": 1.5269842147827148, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8786184787750244, + "num_tokens": 206247457.0, + "step": 5872 + }, + { + "epoch": 1.0906220984215413, + "grad_norm": 1.363268494606018, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8778844475746155, + "num_tokens": 206291818.0, + "step": 5873 + }, + { + "epoch": 1.0908077994428969, + "grad_norm": 1.6031827926635742, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8653308749198914, + "num_tokens": 206326477.0, + "step": 5874 + }, + { + "epoch": 1.0909935004642526, + "grad_norm": 1.844075083732605, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.866972804069519, + "num_tokens": 206352653.0, + "step": 5875 + }, + { + "epoch": 1.091179201485608, + "grad_norm": 1.4653085470199585, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8750882148742676, + "num_tokens": 206385505.0, + "step": 5876 + }, + { + "epoch": 1.0913649025069638, + "grad_norm": 1.4364659786224365, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8771434426307678, + "num_tokens": 206422430.0, + "step": 5877 + }, + { + "epoch": 1.0915506035283193, + "grad_norm": 1.48157799243927, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8827872276306152, + "num_tokens": 206456151.0, + "step": 5878 + }, + { + "epoch": 1.091736304549675, + "grad_norm": 1.3968188762664795, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8873822689056396, + "num_tokens": 206491166.0, + "step": 5879 + }, + { + "epoch": 1.0919220055710306, + "grad_norm": 1.5939558744430542, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8805267214775085, + "num_tokens": 206524372.0, + "step": 5880 + }, + { + "epoch": 1.0921077065923863, + "grad_norm": 1.5722492933273315, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8766603469848633, + "num_tokens": 206558678.0, + "step": 5881 + }, + { + "epoch": 1.0922934076137418, + "grad_norm": 1.5049265623092651, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8957614898681641, + "num_tokens": 206590512.0, + "step": 5882 + }, + { + "epoch": 1.0924791086350976, + "grad_norm": 1.6327084302902222, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8891262412071228, + "num_tokens": 206616253.0, + "step": 5883 + }, + { + "epoch": 1.092664809656453, + "grad_norm": 1.3495267629623413, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8722455501556396, + "num_tokens": 206657646.0, + "step": 5884 + }, + { + "epoch": 1.0928505106778088, + "grad_norm": 1.497987985610962, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.86916583776474, + "num_tokens": 206691756.0, + "step": 5885 + }, + { + "epoch": 1.0930362116991643, + "grad_norm": 1.4960330724716187, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8741499781608582, + "num_tokens": 206727756.0, + "step": 5886 + }, + { + "epoch": 1.09322191272052, + "grad_norm": 1.3546181917190552, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8902511596679688, + "num_tokens": 206767879.0, + "step": 5887 + }, + { + "epoch": 1.0934076137418756, + "grad_norm": 1.4663046598434448, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8729866743087769, + "num_tokens": 206806919.0, + "step": 5888 + }, + { + "epoch": 1.0935933147632313, + "grad_norm": 1.6108670234680176, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8658026456832886, + "num_tokens": 206840809.0, + "step": 5889 + }, + { + "epoch": 1.0937790157845868, + "grad_norm": 1.6612255573272705, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8907336592674255, + "num_tokens": 206868108.0, + "step": 5890 + }, + { + "epoch": 1.0939647168059423, + "grad_norm": 1.5072147846221924, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.883764386177063, + "num_tokens": 206903338.0, + "step": 5891 + }, + { + "epoch": 1.094150417827298, + "grad_norm": 1.4216797351837158, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8753172159194946, + "num_tokens": 206942855.0, + "step": 5892 + }, + { + "epoch": 1.0943361188486538, + "grad_norm": 1.4552359580993652, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8759984970092773, + "num_tokens": 206976939.0, + "step": 5893 + }, + { + "epoch": 1.0945218198700093, + "grad_norm": 1.4477359056472778, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8719367980957031, + "num_tokens": 207012540.0, + "step": 5894 + }, + { + "epoch": 1.0947075208913648, + "grad_norm": 1.5859534740447998, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8653920292854309, + "num_tokens": 207047540.0, + "step": 5895 + }, + { + "epoch": 1.0948932219127205, + "grad_norm": 1.4935321807861328, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8721250891685486, + "num_tokens": 207082381.0, + "step": 5896 + }, + { + "epoch": 1.095078922934076, + "grad_norm": 1.5656955242156982, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8844887018203735, + "num_tokens": 207112642.0, + "step": 5897 + }, + { + "epoch": 1.0952646239554318, + "grad_norm": 1.514583706855774, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8713065385818481, + "num_tokens": 207152285.0, + "step": 5898 + }, + { + "epoch": 1.0954503249767873, + "grad_norm": 1.5925549268722534, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8731263279914856, + "num_tokens": 207186567.0, + "step": 5899 + }, + { + "epoch": 1.095636025998143, + "grad_norm": 1.62965989112854, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8682918548583984, + "num_tokens": 207218993.0, + "step": 5900 + }, + { + "epoch": 1.0958217270194985, + "grad_norm": 1.5159093141555786, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8701358437538147, + "num_tokens": 207254793.0, + "step": 5901 + }, + { + "epoch": 1.0960074280408543, + "grad_norm": 1.483203649520874, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.871780276298523, + "num_tokens": 207292474.0, + "step": 5902 + }, + { + "epoch": 1.0961931290622098, + "grad_norm": 1.5072176456451416, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.878839373588562, + "num_tokens": 207325320.0, + "step": 5903 + }, + { + "epoch": 1.0963788300835655, + "grad_norm": 1.4967424869537354, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8856728076934814, + "num_tokens": 207360835.0, + "step": 5904 + }, + { + "epoch": 1.096564531104921, + "grad_norm": 1.5946581363677979, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8788789510726929, + "num_tokens": 207392213.0, + "step": 5905 + }, + { + "epoch": 1.0967502321262768, + "grad_norm": 1.5364373922348022, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8625112771987915, + "num_tokens": 207427472.0, + "step": 5906 + }, + { + "epoch": 1.0969359331476323, + "grad_norm": 1.6900813579559326, + "learning_rate": 1e-06, + "loss": 0.4207, + "mean_token_accuracy": 0.8596439361572266, + "num_tokens": 207456237.0, + "step": 5907 + }, + { + "epoch": 1.097121634168988, + "grad_norm": 1.4657237529754639, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8851544857025146, + "num_tokens": 207488654.0, + "step": 5908 + }, + { + "epoch": 1.0973073351903435, + "grad_norm": 1.489354133605957, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8824414610862732, + "num_tokens": 207523422.0, + "step": 5909 + }, + { + "epoch": 1.0974930362116992, + "grad_norm": 1.7113523483276367, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8688569068908691, + "num_tokens": 207551223.0, + "step": 5910 + }, + { + "epoch": 1.0976787372330548, + "grad_norm": 1.5846492052078247, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8699711561203003, + "num_tokens": 207586662.0, + "step": 5911 + }, + { + "epoch": 1.0978644382544105, + "grad_norm": 1.4176913499832153, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8758682608604431, + "num_tokens": 207623884.0, + "step": 5912 + }, + { + "epoch": 1.098050139275766, + "grad_norm": 1.4324973821640015, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8854936361312866, + "num_tokens": 207658896.0, + "step": 5913 + }, + { + "epoch": 1.0982358402971217, + "grad_norm": 1.464416742324829, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8796156644821167, + "num_tokens": 207693086.0, + "step": 5914 + }, + { + "epoch": 1.0984215413184772, + "grad_norm": 1.560755968093872, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8711956739425659, + "num_tokens": 207728102.0, + "step": 5915 + }, + { + "epoch": 1.098607242339833, + "grad_norm": 1.5545461177825928, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8780747056007385, + "num_tokens": 207760323.0, + "step": 5916 + }, + { + "epoch": 1.0987929433611885, + "grad_norm": 1.4587494134902954, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8846384882926941, + "num_tokens": 207796387.0, + "step": 5917 + }, + { + "epoch": 1.098978644382544, + "grad_norm": 1.4850472211837769, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8878718614578247, + "num_tokens": 207830738.0, + "step": 5918 + }, + { + "epoch": 1.0991643454038997, + "grad_norm": 1.496140480041504, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8910176753997803, + "num_tokens": 207862362.0, + "step": 5919 + }, + { + "epoch": 1.0993500464252552, + "grad_norm": 1.4820924997329712, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8828219771385193, + "num_tokens": 207896361.0, + "step": 5920 + }, + { + "epoch": 1.099535747446611, + "grad_norm": 1.4274083375930786, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8745412826538086, + "num_tokens": 207939365.0, + "step": 5921 + }, + { + "epoch": 1.0997214484679665, + "grad_norm": 1.4879150390625, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8580487370491028, + "num_tokens": 207976318.0, + "step": 5922 + }, + { + "epoch": 1.0999071494893222, + "grad_norm": 1.5391689538955688, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.866278350353241, + "num_tokens": 208009077.0, + "step": 5923 + }, + { + "epoch": 1.1000928505106777, + "grad_norm": 1.3586344718933105, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8816988468170166, + "num_tokens": 208051976.0, + "step": 5924 + }, + { + "epoch": 1.1002785515320335, + "grad_norm": 1.458830714225769, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8881670236587524, + "num_tokens": 208090210.0, + "step": 5925 + }, + { + "epoch": 1.100464252553389, + "grad_norm": 1.5178943872451782, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8766627311706543, + "num_tokens": 208128786.0, + "step": 5926 + }, + { + "epoch": 1.1006499535747447, + "grad_norm": 1.53665030002594, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8716760873794556, + "num_tokens": 208164684.0, + "step": 5927 + }, + { + "epoch": 1.1008356545961002, + "grad_norm": 1.5347713232040405, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8794881701469421, + "num_tokens": 208196960.0, + "step": 5928 + }, + { + "epoch": 1.101021355617456, + "grad_norm": 1.5954807996749878, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8902831077575684, + "num_tokens": 208226766.0, + "step": 5929 + }, + { + "epoch": 1.1012070566388115, + "grad_norm": 1.4066016674041748, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.875995397567749, + "num_tokens": 208265924.0, + "step": 5930 + }, + { + "epoch": 1.1013927576601672, + "grad_norm": 1.4439748525619507, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8772649765014648, + "num_tokens": 208300454.0, + "step": 5931 + }, + { + "epoch": 1.1015784586815227, + "grad_norm": 1.4793341159820557, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8638505935668945, + "num_tokens": 208338067.0, + "step": 5932 + }, + { + "epoch": 1.1017641597028784, + "grad_norm": 1.4978265762329102, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8757349252700806, + "num_tokens": 208372564.0, + "step": 5933 + }, + { + "epoch": 1.101949860724234, + "grad_norm": 1.4744352102279663, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8731659650802612, + "num_tokens": 208407822.0, + "step": 5934 + }, + { + "epoch": 1.1021355617455897, + "grad_norm": 1.601772665977478, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.87314373254776, + "num_tokens": 208439279.0, + "step": 5935 + }, + { + "epoch": 1.1023212627669452, + "grad_norm": 1.546479344367981, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8774033784866333, + "num_tokens": 208470903.0, + "step": 5936 + }, + { + "epoch": 1.102506963788301, + "grad_norm": 1.41701078414917, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8884207010269165, + "num_tokens": 208508180.0, + "step": 5937 + }, + { + "epoch": 1.1026926648096564, + "grad_norm": 1.4850480556488037, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8701978325843811, + "num_tokens": 208544045.0, + "step": 5938 + }, + { + "epoch": 1.1028783658310122, + "grad_norm": 1.4056133031845093, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8859118819236755, + "num_tokens": 208581281.0, + "step": 5939 + }, + { + "epoch": 1.1030640668523677, + "grad_norm": 1.5252587795257568, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8697940111160278, + "num_tokens": 208613550.0, + "step": 5940 + }, + { + "epoch": 1.1032497678737232, + "grad_norm": 1.5032861232757568, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8849235773086548, + "num_tokens": 208646075.0, + "step": 5941 + }, + { + "epoch": 1.103435468895079, + "grad_norm": 1.3872735500335693, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8749877214431763, + "num_tokens": 208685853.0, + "step": 5942 + }, + { + "epoch": 1.1036211699164344, + "grad_norm": 1.5874459743499756, + "learning_rate": 1e-06, + "loss": 0.403, + "mean_token_accuracy": 0.8618367910385132, + "num_tokens": 208720927.0, + "step": 5943 + }, + { + "epoch": 1.1038068709377902, + "grad_norm": 1.3849619626998901, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8937473893165588, + "num_tokens": 208754818.0, + "step": 5944 + }, + { + "epoch": 1.1039925719591457, + "grad_norm": 1.6342644691467285, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.879485011100769, + "num_tokens": 208783178.0, + "step": 5945 + }, + { + "epoch": 1.1041782729805014, + "grad_norm": 1.5264588594436646, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8898119926452637, + "num_tokens": 208817510.0, + "step": 5946 + }, + { + "epoch": 1.104363974001857, + "grad_norm": 1.4331705570220947, + "learning_rate": 1e-06, + "loss": 0.3883, + "mean_token_accuracy": 0.8678218722343445, + "num_tokens": 208859943.0, + "step": 5947 + }, + { + "epoch": 1.1045496750232127, + "grad_norm": 1.4677468538284302, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.870463490486145, + "num_tokens": 208896825.0, + "step": 5948 + }, + { + "epoch": 1.1047353760445682, + "grad_norm": 1.571723461151123, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8808366060256958, + "num_tokens": 208928039.0, + "step": 5949 + }, + { + "epoch": 1.104921077065924, + "grad_norm": 1.5529989004135132, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8719471096992493, + "num_tokens": 208960004.0, + "step": 5950 + }, + { + "epoch": 1.1051067780872794, + "grad_norm": 1.8615097999572754, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8660904765129089, + "num_tokens": 208994724.0, + "step": 5951 + }, + { + "epoch": 1.1052924791086352, + "grad_norm": 1.410343050956726, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8610209226608276, + "num_tokens": 209034628.0, + "step": 5952 + }, + { + "epoch": 1.1054781801299907, + "grad_norm": 1.5724111795425415, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.889198362827301, + "num_tokens": 209064686.0, + "step": 5953 + }, + { + "epoch": 1.1056638811513464, + "grad_norm": 1.4029840230941772, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8775300979614258, + "num_tokens": 209101986.0, + "step": 5954 + }, + { + "epoch": 1.105849582172702, + "grad_norm": 1.4704240560531616, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8670010566711426, + "num_tokens": 209136540.0, + "step": 5955 + }, + { + "epoch": 1.1060352831940576, + "grad_norm": 1.5191069841384888, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8774352073669434, + "num_tokens": 209169320.0, + "step": 5956 + }, + { + "epoch": 1.1062209842154132, + "grad_norm": 1.71133553981781, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.873208224773407, + "num_tokens": 209195972.0, + "step": 5957 + }, + { + "epoch": 1.1064066852367689, + "grad_norm": 1.5404607057571411, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8671351671218872, + "num_tokens": 209230633.0, + "step": 5958 + }, + { + "epoch": 1.1065923862581244, + "grad_norm": 1.4613666534423828, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8777303099632263, + "num_tokens": 209268532.0, + "step": 5959 + }, + { + "epoch": 1.1067780872794801, + "grad_norm": 1.48037850856781, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8705013990402222, + "num_tokens": 209309811.0, + "step": 5960 + }, + { + "epoch": 1.1069637883008356, + "grad_norm": 1.6408355236053467, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8678353428840637, + "num_tokens": 209339436.0, + "step": 5961 + }, + { + "epoch": 1.1071494893221914, + "grad_norm": 1.5428898334503174, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8827564120292664, + "num_tokens": 209370946.0, + "step": 5962 + }, + { + "epoch": 1.1073351903435469, + "grad_norm": 1.4113733768463135, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.874495267868042, + "num_tokens": 209408797.0, + "step": 5963 + }, + { + "epoch": 1.1075208913649024, + "grad_norm": 1.474048137664795, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8734828233718872, + "num_tokens": 209441023.0, + "step": 5964 + }, + { + "epoch": 1.1077065923862581, + "grad_norm": 1.531802773475647, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8860209584236145, + "num_tokens": 209478767.0, + "step": 5965 + }, + { + "epoch": 1.1078922934076139, + "grad_norm": 1.6379809379577637, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8684118390083313, + "num_tokens": 209507876.0, + "step": 5966 + }, + { + "epoch": 1.1080779944289694, + "grad_norm": 1.6753822565078735, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8746433258056641, + "num_tokens": 209538690.0, + "step": 5967 + }, + { + "epoch": 1.1082636954503249, + "grad_norm": 1.708606481552124, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.875511646270752, + "num_tokens": 209568095.0, + "step": 5968 + }, + { + "epoch": 1.1084493964716806, + "grad_norm": 1.6304088830947876, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8578807711601257, + "num_tokens": 209598338.0, + "step": 5969 + }, + { + "epoch": 1.1086350974930361, + "grad_norm": 1.4595650434494019, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8897778987884521, + "num_tokens": 209635244.0, + "step": 5970 + }, + { + "epoch": 1.1088207985143919, + "grad_norm": 1.534048318862915, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8754469752311707, + "num_tokens": 209671601.0, + "step": 5971 + }, + { + "epoch": 1.1090064995357474, + "grad_norm": 1.6099317073822021, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8571189641952515, + "num_tokens": 209704176.0, + "step": 5972 + }, + { + "epoch": 1.109192200557103, + "grad_norm": 1.6933012008666992, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8835959434509277, + "num_tokens": 209735857.0, + "step": 5973 + }, + { + "epoch": 1.1093779015784586, + "grad_norm": 1.580355167388916, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8737224340438843, + "num_tokens": 209766883.0, + "step": 5974 + }, + { + "epoch": 1.1095636025998143, + "grad_norm": 1.535504937171936, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8876283764839172, + "num_tokens": 209806653.0, + "step": 5975 + }, + { + "epoch": 1.1097493036211699, + "grad_norm": 1.4677809476852417, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8762632012367249, + "num_tokens": 209842731.0, + "step": 5976 + }, + { + "epoch": 1.1099350046425256, + "grad_norm": 1.5829437971115112, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8640259504318237, + "num_tokens": 209877257.0, + "step": 5977 + }, + { + "epoch": 1.110120705663881, + "grad_norm": 1.6160213947296143, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8680046796798706, + "num_tokens": 209911619.0, + "step": 5978 + }, + { + "epoch": 1.1103064066852368, + "grad_norm": 1.6564199924468994, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8883885741233826, + "num_tokens": 209939401.0, + "step": 5979 + }, + { + "epoch": 1.1104921077065923, + "grad_norm": 1.4997518062591553, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.879384458065033, + "num_tokens": 209979213.0, + "step": 5980 + }, + { + "epoch": 1.110677808727948, + "grad_norm": 1.5775254964828491, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8774901628494263, + "num_tokens": 210014043.0, + "step": 5981 + }, + { + "epoch": 1.1108635097493036, + "grad_norm": 1.520025372505188, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8931883573532104, + "num_tokens": 210043721.0, + "step": 5982 + }, + { + "epoch": 1.1110492107706593, + "grad_norm": 1.5315641164779663, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8719804286956787, + "num_tokens": 210077751.0, + "step": 5983 + }, + { + "epoch": 1.1112349117920148, + "grad_norm": 1.4987269639968872, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8920881152153015, + "num_tokens": 210111322.0, + "step": 5984 + }, + { + "epoch": 1.1114206128133706, + "grad_norm": 1.4710673093795776, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8658488392829895, + "num_tokens": 210153547.0, + "step": 5985 + }, + { + "epoch": 1.111606313834726, + "grad_norm": 1.707053780555725, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8640748262405396, + "num_tokens": 210182224.0, + "step": 5986 + }, + { + "epoch": 1.1117920148560818, + "grad_norm": 1.5400551557540894, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8871995210647583, + "num_tokens": 210213768.0, + "step": 5987 + }, + { + "epoch": 1.1119777158774373, + "grad_norm": 1.5619558095932007, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8689512014389038, + "num_tokens": 210248553.0, + "step": 5988 + }, + { + "epoch": 1.112163416898793, + "grad_norm": 1.4854134321212769, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.884339451789856, + "num_tokens": 210281978.0, + "step": 5989 + }, + { + "epoch": 1.1123491179201486, + "grad_norm": 1.4084956645965576, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8802001476287842, + "num_tokens": 210322607.0, + "step": 5990 + }, + { + "epoch": 1.112534818941504, + "grad_norm": 1.5917357206344604, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8744910955429077, + "num_tokens": 210353873.0, + "step": 5991 + }, + { + "epoch": 1.1127205199628598, + "grad_norm": 1.4051904678344727, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.881650984287262, + "num_tokens": 210389979.0, + "step": 5992 + }, + { + "epoch": 1.1129062209842153, + "grad_norm": 1.4377015829086304, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8811500668525696, + "num_tokens": 210429899.0, + "step": 5993 + }, + { + "epoch": 1.113091922005571, + "grad_norm": 1.3851228952407837, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8837122917175293, + "num_tokens": 210468527.0, + "step": 5994 + }, + { + "epoch": 1.1132776230269266, + "grad_norm": 1.5961065292358398, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8776682615280151, + "num_tokens": 210499966.0, + "step": 5995 + }, + { + "epoch": 1.1134633240482823, + "grad_norm": 1.386384129524231, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.879899263381958, + "num_tokens": 210544746.0, + "step": 5996 + }, + { + "epoch": 1.1136490250696378, + "grad_norm": 1.56294846534729, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8836647272109985, + "num_tokens": 210574734.0, + "step": 5997 + }, + { + "epoch": 1.1138347260909935, + "grad_norm": 1.4620901346206665, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8828290700912476, + "num_tokens": 210609269.0, + "step": 5998 + }, + { + "epoch": 1.114020427112349, + "grad_norm": 1.530638337135315, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8817393183708191, + "num_tokens": 210642854.0, + "step": 5999 + }, + { + "epoch": 1.1142061281337048, + "grad_norm": 1.554016351699829, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.877947211265564, + "num_tokens": 210672413.0, + "step": 6000 + }, + { + "epoch": 1.1143918291550603, + "grad_norm": 1.5205880403518677, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8786494135856628, + "num_tokens": 210707188.0, + "step": 6001 + }, + { + "epoch": 1.114577530176416, + "grad_norm": 1.5888113975524902, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.863142192363739, + "num_tokens": 210743047.0, + "step": 6002 + }, + { + "epoch": 1.1147632311977715, + "grad_norm": 1.47112238407135, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8723721504211426, + "num_tokens": 210779986.0, + "step": 6003 + }, + { + "epoch": 1.1149489322191273, + "grad_norm": 1.4448586702346802, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8733972907066345, + "num_tokens": 210818156.0, + "step": 6004 + }, + { + "epoch": 1.1151346332404828, + "grad_norm": 1.4881956577301025, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8768337368965149, + "num_tokens": 210853532.0, + "step": 6005 + }, + { + "epoch": 1.1153203342618385, + "grad_norm": 1.4774078130722046, + "learning_rate": 1e-06, + "loss": 0.4327, + "mean_token_accuracy": 0.8561031818389893, + "num_tokens": 210894815.0, + "step": 6006 + }, + { + "epoch": 1.115506035283194, + "grad_norm": 1.730303168296814, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.88067626953125, + "num_tokens": 210922501.0, + "step": 6007 + }, + { + "epoch": 1.1156917363045498, + "grad_norm": 1.3570153713226318, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8834456205368042, + "num_tokens": 210959413.0, + "step": 6008 + }, + { + "epoch": 1.1158774373259053, + "grad_norm": 1.5279438495635986, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8669490814208984, + "num_tokens": 210994222.0, + "step": 6009 + }, + { + "epoch": 1.116063138347261, + "grad_norm": 1.5724396705627441, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8773424625396729, + "num_tokens": 211028328.0, + "step": 6010 + }, + { + "epoch": 1.1162488393686165, + "grad_norm": 1.5782015323638916, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8845283389091492, + "num_tokens": 211060215.0, + "step": 6011 + }, + { + "epoch": 1.1164345403899723, + "grad_norm": 1.4697505235671997, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8685519695281982, + "num_tokens": 211096953.0, + "step": 6012 + }, + { + "epoch": 1.1166202414113278, + "grad_norm": 1.5403575897216797, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8660225868225098, + "num_tokens": 211132962.0, + "step": 6013 + }, + { + "epoch": 1.1168059424326833, + "grad_norm": 1.502651333808899, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.878108024597168, + "num_tokens": 211166261.0, + "step": 6014 + }, + { + "epoch": 1.116991643454039, + "grad_norm": 1.4900102615356445, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8643400073051453, + "num_tokens": 211202205.0, + "step": 6015 + }, + { + "epoch": 1.1171773444753945, + "grad_norm": 1.396584153175354, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8834806680679321, + "num_tokens": 211239668.0, + "step": 6016 + }, + { + "epoch": 1.1173630454967503, + "grad_norm": 1.3836135864257812, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8858360052108765, + "num_tokens": 211276800.0, + "step": 6017 + }, + { + "epoch": 1.1175487465181058, + "grad_norm": 1.5063762664794922, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8753236532211304, + "num_tokens": 211311370.0, + "step": 6018 + }, + { + "epoch": 1.1177344475394615, + "grad_norm": 1.5772862434387207, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8770757913589478, + "num_tokens": 211341995.0, + "step": 6019 + }, + { + "epoch": 1.117920148560817, + "grad_norm": 1.4308186769485474, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8763313293457031, + "num_tokens": 211377680.0, + "step": 6020 + }, + { + "epoch": 1.1181058495821727, + "grad_norm": 1.5550726652145386, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.859277606010437, + "num_tokens": 211413102.0, + "step": 6021 + }, + { + "epoch": 1.1182915506035283, + "grad_norm": 1.4166542291641235, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8847514390945435, + "num_tokens": 211450095.0, + "step": 6022 + }, + { + "epoch": 1.118477251624884, + "grad_norm": 1.4364895820617676, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.874836266040802, + "num_tokens": 211487526.0, + "step": 6023 + }, + { + "epoch": 1.1186629526462395, + "grad_norm": 1.4246701002120972, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.873310923576355, + "num_tokens": 211527088.0, + "step": 6024 + }, + { + "epoch": 1.1188486536675952, + "grad_norm": 1.702454686164856, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.876726508140564, + "num_tokens": 211554515.0, + "step": 6025 + }, + { + "epoch": 1.1190343546889507, + "grad_norm": 1.544278621673584, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8741586208343506, + "num_tokens": 211585028.0, + "step": 6026 + }, + { + "epoch": 1.1192200557103065, + "grad_norm": 1.5370266437530518, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8822989463806152, + "num_tokens": 211617162.0, + "step": 6027 + }, + { + "epoch": 1.119405756731662, + "grad_norm": 1.554390788078308, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8807154893875122, + "num_tokens": 211649422.0, + "step": 6028 + }, + { + "epoch": 1.1195914577530177, + "grad_norm": 1.4237089157104492, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8786948919296265, + "num_tokens": 211688764.0, + "step": 6029 + }, + { + "epoch": 1.1197771587743732, + "grad_norm": 1.3612366914749146, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8864455223083496, + "num_tokens": 211727175.0, + "step": 6030 + }, + { + "epoch": 1.119962859795729, + "grad_norm": 1.5547072887420654, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8759759068489075, + "num_tokens": 211760094.0, + "step": 6031 + }, + { + "epoch": 1.1201485608170845, + "grad_norm": 1.6802526712417603, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8704097270965576, + "num_tokens": 211789203.0, + "step": 6032 + }, + { + "epoch": 1.1203342618384402, + "grad_norm": 1.5129261016845703, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.874208927154541, + "num_tokens": 211822896.0, + "step": 6033 + }, + { + "epoch": 1.1205199628597957, + "grad_norm": 1.4955167770385742, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8744387626647949, + "num_tokens": 211861574.0, + "step": 6034 + }, + { + "epoch": 1.1207056638811514, + "grad_norm": 1.4275319576263428, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8875288367271423, + "num_tokens": 211898126.0, + "step": 6035 + }, + { + "epoch": 1.120891364902507, + "grad_norm": 1.5478124618530273, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8786735534667969, + "num_tokens": 211927963.0, + "step": 6036 + }, + { + "epoch": 1.1210770659238625, + "grad_norm": 1.565898060798645, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8870452642440796, + "num_tokens": 211956810.0, + "step": 6037 + }, + { + "epoch": 1.1212627669452182, + "grad_norm": 1.48691987991333, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8867776393890381, + "num_tokens": 211991036.0, + "step": 6038 + }, + { + "epoch": 1.1214484679665737, + "grad_norm": 1.5327047109603882, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8838269710540771, + "num_tokens": 212028065.0, + "step": 6039 + }, + { + "epoch": 1.1216341689879294, + "grad_norm": 1.5770450830459595, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8750706315040588, + "num_tokens": 212062831.0, + "step": 6040 + }, + { + "epoch": 1.121819870009285, + "grad_norm": 1.677138328552246, + "learning_rate": 1e-06, + "loss": 0.427, + "mean_token_accuracy": 0.8564176559448242, + "num_tokens": 212095258.0, + "step": 6041 + }, + { + "epoch": 1.1220055710306407, + "grad_norm": 1.5086309909820557, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8770054578781128, + "num_tokens": 212132547.0, + "step": 6042 + }, + { + "epoch": 1.1221912720519962, + "grad_norm": 1.5147408246994019, + "learning_rate": 1e-06, + "loss": 0.4287, + "mean_token_accuracy": 0.8534449338912964, + "num_tokens": 212173254.0, + "step": 6043 + }, + { + "epoch": 1.122376973073352, + "grad_norm": 1.5370237827301025, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8689702749252319, + "num_tokens": 212210069.0, + "step": 6044 + }, + { + "epoch": 1.1225626740947074, + "grad_norm": 1.489016056060791, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8805556297302246, + "num_tokens": 212249593.0, + "step": 6045 + }, + { + "epoch": 1.1227483751160632, + "grad_norm": 1.517145037651062, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.879854679107666, + "num_tokens": 212287160.0, + "step": 6046 + }, + { + "epoch": 1.1229340761374187, + "grad_norm": 1.5785359144210815, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.878897488117218, + "num_tokens": 212318439.0, + "step": 6047 + }, + { + "epoch": 1.1231197771587744, + "grad_norm": 1.731072187423706, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8680053949356079, + "num_tokens": 212348022.0, + "step": 6048 + }, + { + "epoch": 1.12330547818013, + "grad_norm": 1.5202946662902832, + "learning_rate": 1e-06, + "loss": 0.4158, + "mean_token_accuracy": 0.85826176404953, + "num_tokens": 212386574.0, + "step": 6049 + }, + { + "epoch": 1.1234911792014857, + "grad_norm": 1.551810383796692, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8750346899032593, + "num_tokens": 212418634.0, + "step": 6050 + }, + { + "epoch": 1.1236768802228412, + "grad_norm": 1.7481507062911987, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8808435797691345, + "num_tokens": 212450363.0, + "step": 6051 + }, + { + "epoch": 1.123862581244197, + "grad_norm": 1.6114825010299683, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8692522048950195, + "num_tokens": 212480055.0, + "step": 6052 + }, + { + "epoch": 1.1240482822655524, + "grad_norm": 1.6204168796539307, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8659973740577698, + "num_tokens": 212513672.0, + "step": 6053 + }, + { + "epoch": 1.1242339832869082, + "grad_norm": 1.456518292427063, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8775743246078491, + "num_tokens": 212551368.0, + "step": 6054 + }, + { + "epoch": 1.1244196843082637, + "grad_norm": 1.5281612873077393, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8936620950698853, + "num_tokens": 212582709.0, + "step": 6055 + }, + { + "epoch": 1.1246053853296194, + "grad_norm": 1.5503596067428589, + "learning_rate": 1e-06, + "loss": 0.444, + "mean_token_accuracy": 0.8484480381011963, + "num_tokens": 212621488.0, + "step": 6056 + }, + { + "epoch": 1.124791086350975, + "grad_norm": 1.5434969663619995, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8806055784225464, + "num_tokens": 212653069.0, + "step": 6057 + }, + { + "epoch": 1.1249767873723306, + "grad_norm": 1.6235203742980957, + "learning_rate": 1e-06, + "loss": 0.4182, + "mean_token_accuracy": 0.858731746673584, + "num_tokens": 212686305.0, + "step": 6058 + }, + { + "epoch": 1.1251624883936862, + "grad_norm": 1.507554531097412, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8702607154846191, + "num_tokens": 212727040.0, + "step": 6059 + }, + { + "epoch": 1.1253481894150417, + "grad_norm": 1.6332414150238037, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.886794924736023, + "num_tokens": 212759630.0, + "step": 6060 + }, + { + "epoch": 1.1255338904363974, + "grad_norm": 1.5548862218856812, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8694965839385986, + "num_tokens": 212794541.0, + "step": 6061 + }, + { + "epoch": 1.1257195914577531, + "grad_norm": 1.5517261028289795, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8769853115081787, + "num_tokens": 212829650.0, + "step": 6062 + }, + { + "epoch": 1.1259052924791086, + "grad_norm": 1.6279079914093018, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8677119016647339, + "num_tokens": 212863170.0, + "step": 6063 + }, + { + "epoch": 1.1260909935004642, + "grad_norm": 1.5451079607009888, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8847682476043701, + "num_tokens": 212894116.0, + "step": 6064 + }, + { + "epoch": 1.1262766945218199, + "grad_norm": 1.448315143585205, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8640875816345215, + "num_tokens": 212932804.0, + "step": 6065 + }, + { + "epoch": 1.1264623955431754, + "grad_norm": 1.5457602739334106, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8650327920913696, + "num_tokens": 212968697.0, + "step": 6066 + }, + { + "epoch": 1.1266480965645311, + "grad_norm": 1.5512993335723877, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.871270477771759, + "num_tokens": 213004467.0, + "step": 6067 + }, + { + "epoch": 1.1268337975858866, + "grad_norm": 1.4346952438354492, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8705122470855713, + "num_tokens": 213042453.0, + "step": 6068 + }, + { + "epoch": 1.1270194986072424, + "grad_norm": 1.4931776523590088, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8842929601669312, + "num_tokens": 213074952.0, + "step": 6069 + }, + { + "epoch": 1.1272051996285979, + "grad_norm": 1.3956637382507324, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.880584716796875, + "num_tokens": 213112628.0, + "step": 6070 + }, + { + "epoch": 1.1273909006499536, + "grad_norm": 1.5207710266113281, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8699508905410767, + "num_tokens": 213150083.0, + "step": 6071 + }, + { + "epoch": 1.1275766016713091, + "grad_norm": 1.5066663026809692, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8767004013061523, + "num_tokens": 213183741.0, + "step": 6072 + }, + { + "epoch": 1.1277623026926649, + "grad_norm": 1.4861506223678589, + "learning_rate": 1e-06, + "loss": 0.4174, + "mean_token_accuracy": 0.8584840893745422, + "num_tokens": 213224522.0, + "step": 6073 + }, + { + "epoch": 1.1279480037140204, + "grad_norm": 1.3619667291641235, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8825677037239075, + "num_tokens": 213264836.0, + "step": 6074 + }, + { + "epoch": 1.128133704735376, + "grad_norm": 1.3725699186325073, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8867040872573853, + "num_tokens": 213302452.0, + "step": 6075 + }, + { + "epoch": 1.1283194057567316, + "grad_norm": 1.5644232034683228, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8726688623428345, + "num_tokens": 213337767.0, + "step": 6076 + }, + { + "epoch": 1.1285051067780874, + "grad_norm": 1.4733705520629883, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8706676959991455, + "num_tokens": 213377237.0, + "step": 6077 + }, + { + "epoch": 1.1286908077994429, + "grad_norm": 1.6688990592956543, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8767415881156921, + "num_tokens": 213402843.0, + "step": 6078 + }, + { + "epoch": 1.1288765088207986, + "grad_norm": 1.412515640258789, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8652923107147217, + "num_tokens": 213445954.0, + "step": 6079 + }, + { + "epoch": 1.129062209842154, + "grad_norm": 1.5776968002319336, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8746333122253418, + "num_tokens": 213479366.0, + "step": 6080 + }, + { + "epoch": 1.1292479108635098, + "grad_norm": 1.6634833812713623, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8618612885475159, + "num_tokens": 213510552.0, + "step": 6081 + }, + { + "epoch": 1.1294336118848654, + "grad_norm": 1.556816577911377, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8831098079681396, + "num_tokens": 213542442.0, + "step": 6082 + }, + { + "epoch": 1.1296193129062209, + "grad_norm": 1.3714922666549683, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8947877883911133, + "num_tokens": 213577102.0, + "step": 6083 + }, + { + "epoch": 1.1298050139275766, + "grad_norm": 1.5318832397460938, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.869171142578125, + "num_tokens": 213614168.0, + "step": 6084 + }, + { + "epoch": 1.1299907149489323, + "grad_norm": 1.4148937463760376, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8736178874969482, + "num_tokens": 213655213.0, + "step": 6085 + }, + { + "epoch": 1.1301764159702878, + "grad_norm": 1.454590082168579, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8685210943222046, + "num_tokens": 213696230.0, + "step": 6086 + }, + { + "epoch": 1.1303621169916434, + "grad_norm": 1.5649802684783936, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8751164674758911, + "num_tokens": 213734320.0, + "step": 6087 + }, + { + "epoch": 1.130547818012999, + "grad_norm": 1.5412706136703491, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.870266854763031, + "num_tokens": 213766463.0, + "step": 6088 + }, + { + "epoch": 1.1307335190343546, + "grad_norm": 1.635835886001587, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8848782181739807, + "num_tokens": 213797425.0, + "step": 6089 + }, + { + "epoch": 1.1309192200557103, + "grad_norm": 1.5324583053588867, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.879015326499939, + "num_tokens": 213831768.0, + "step": 6090 + }, + { + "epoch": 1.1311049210770658, + "grad_norm": 1.5120607614517212, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8651676177978516, + "num_tokens": 213865361.0, + "step": 6091 + }, + { + "epoch": 1.1312906220984216, + "grad_norm": 1.5368409156799316, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8799005150794983, + "num_tokens": 213898694.0, + "step": 6092 + }, + { + "epoch": 1.131476323119777, + "grad_norm": 1.5563791990280151, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8689438700675964, + "num_tokens": 213934551.0, + "step": 6093 + }, + { + "epoch": 1.1316620241411328, + "grad_norm": 1.59931480884552, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8854504823684692, + "num_tokens": 213967392.0, + "step": 6094 + }, + { + "epoch": 1.1318477251624883, + "grad_norm": 1.5736545324325562, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8748464584350586, + "num_tokens": 213999530.0, + "step": 6095 + }, + { + "epoch": 1.132033426183844, + "grad_norm": 1.5354435443878174, + "learning_rate": 1e-06, + "loss": 0.3991, + "mean_token_accuracy": 0.8618227243423462, + "num_tokens": 214036510.0, + "step": 6096 + }, + { + "epoch": 1.1322191272051996, + "grad_norm": 1.5640771389007568, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8677549362182617, + "num_tokens": 214072898.0, + "step": 6097 + }, + { + "epoch": 1.1324048282265553, + "grad_norm": 1.5775904655456543, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8890747427940369, + "num_tokens": 214102523.0, + "step": 6098 + }, + { + "epoch": 1.1325905292479108, + "grad_norm": 1.5195590257644653, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8833721280097961, + "num_tokens": 214136697.0, + "step": 6099 + }, + { + "epoch": 1.1327762302692665, + "grad_norm": 1.5904136896133423, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8705743551254272, + "num_tokens": 214170831.0, + "step": 6100 + }, + { + "epoch": 1.132961931290622, + "grad_norm": 1.4727672338485718, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8851091861724854, + "num_tokens": 214207390.0, + "step": 6101 + }, + { + "epoch": 1.1331476323119778, + "grad_norm": 1.3900679349899292, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8768482804298401, + "num_tokens": 214247014.0, + "step": 6102 + }, + { + "epoch": 1.1333333333333333, + "grad_norm": 1.5912197828292847, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.881554901599884, + "num_tokens": 214278966.0, + "step": 6103 + }, + { + "epoch": 1.133519034354689, + "grad_norm": 1.5641462802886963, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8763387203216553, + "num_tokens": 214312289.0, + "step": 6104 + }, + { + "epoch": 1.1337047353760445, + "grad_norm": 1.5837372541427612, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8837912678718567, + "num_tokens": 214348404.0, + "step": 6105 + }, + { + "epoch": 1.1338904363974003, + "grad_norm": 1.589844822883606, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8693153858184814, + "num_tokens": 214382042.0, + "step": 6106 + }, + { + "epoch": 1.1340761374187558, + "grad_norm": 1.493747353553772, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.887027382850647, + "num_tokens": 214415761.0, + "step": 6107 + }, + { + "epoch": 1.1342618384401115, + "grad_norm": 1.4623510837554932, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8763042092323303, + "num_tokens": 214450321.0, + "step": 6108 + }, + { + "epoch": 1.134447539461467, + "grad_norm": 1.5544394254684448, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8697885870933533, + "num_tokens": 214483213.0, + "step": 6109 + }, + { + "epoch": 1.1346332404828225, + "grad_norm": 1.662015438079834, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.873248815536499, + "num_tokens": 214512270.0, + "step": 6110 + }, + { + "epoch": 1.1348189415041783, + "grad_norm": 1.650424599647522, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8663721680641174, + "num_tokens": 214551647.0, + "step": 6111 + }, + { + "epoch": 1.135004642525534, + "grad_norm": 1.6081486940383911, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8769475817680359, + "num_tokens": 214587177.0, + "step": 6112 + }, + { + "epoch": 1.1351903435468895, + "grad_norm": 1.4744700193405151, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8823279142379761, + "num_tokens": 214622144.0, + "step": 6113 + }, + { + "epoch": 1.135376044568245, + "grad_norm": 1.4080474376678467, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8876625299453735, + "num_tokens": 214659481.0, + "step": 6114 + }, + { + "epoch": 1.1355617455896008, + "grad_norm": 1.3712977170944214, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8873659372329712, + "num_tokens": 214700846.0, + "step": 6115 + }, + { + "epoch": 1.1357474466109563, + "grad_norm": 1.5829018354415894, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.8741015195846558, + "num_tokens": 214731344.0, + "step": 6116 + }, + { + "epoch": 1.135933147632312, + "grad_norm": 1.6057637929916382, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8807497620582581, + "num_tokens": 214759329.0, + "step": 6117 + }, + { + "epoch": 1.1361188486536675, + "grad_norm": 1.3896392583847046, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8707215785980225, + "num_tokens": 214801497.0, + "step": 6118 + }, + { + "epoch": 1.1363045496750233, + "grad_norm": 1.4825217723846436, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8827866315841675, + "num_tokens": 214835535.0, + "step": 6119 + }, + { + "epoch": 1.1364902506963788, + "grad_norm": 1.5344420671463013, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8884412050247192, + "num_tokens": 214863567.0, + "step": 6120 + }, + { + "epoch": 1.1366759517177345, + "grad_norm": 1.4421306848526, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8814422488212585, + "num_tokens": 214902527.0, + "step": 6121 + }, + { + "epoch": 1.13686165273909, + "grad_norm": 1.5763912200927734, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8711251020431519, + "num_tokens": 214935328.0, + "step": 6122 + }, + { + "epoch": 1.1370473537604457, + "grad_norm": 1.2923029661178589, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.886648952960968, + "num_tokens": 214976472.0, + "step": 6123 + }, + { + "epoch": 1.1372330547818013, + "grad_norm": 1.3998273611068726, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8785523772239685, + "num_tokens": 215014448.0, + "step": 6124 + }, + { + "epoch": 1.137418755803157, + "grad_norm": 1.6712229251861572, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8751003742218018, + "num_tokens": 215045976.0, + "step": 6125 + }, + { + "epoch": 1.1376044568245125, + "grad_norm": 1.520359754562378, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8790343999862671, + "num_tokens": 215079431.0, + "step": 6126 + }, + { + "epoch": 1.1377901578458682, + "grad_norm": 1.522878885269165, + "learning_rate": 1e-06, + "loss": 0.4134, + "mean_token_accuracy": 0.8611141443252563, + "num_tokens": 215115000.0, + "step": 6127 + }, + { + "epoch": 1.1379758588672237, + "grad_norm": 1.3580520153045654, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8853393793106079, + "num_tokens": 215154728.0, + "step": 6128 + }, + { + "epoch": 1.1381615598885795, + "grad_norm": 1.482231616973877, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8763800263404846, + "num_tokens": 215192896.0, + "step": 6129 + }, + { + "epoch": 1.138347260909935, + "grad_norm": 1.4103999137878418, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8856724500656128, + "num_tokens": 215232334.0, + "step": 6130 + }, + { + "epoch": 1.1385329619312907, + "grad_norm": 1.3735038042068481, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8832426071166992, + "num_tokens": 215271482.0, + "step": 6131 + }, + { + "epoch": 1.1387186629526462, + "grad_norm": 1.4557582139968872, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8720953464508057, + "num_tokens": 215308890.0, + "step": 6132 + }, + { + "epoch": 1.1389043639740017, + "grad_norm": 1.389542818069458, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8817252516746521, + "num_tokens": 215347322.0, + "step": 6133 + }, + { + "epoch": 1.1390900649953575, + "grad_norm": 1.6689432859420776, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8711686134338379, + "num_tokens": 215379738.0, + "step": 6134 + }, + { + "epoch": 1.1392757660167132, + "grad_norm": 1.4694865942001343, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8766266107559204, + "num_tokens": 215417957.0, + "step": 6135 + }, + { + "epoch": 1.1394614670380687, + "grad_norm": 1.6209158897399902, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8740971088409424, + "num_tokens": 215451478.0, + "step": 6136 + }, + { + "epoch": 1.1396471680594242, + "grad_norm": 1.5773285627365112, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8703871369361877, + "num_tokens": 215483630.0, + "step": 6137 + }, + { + "epoch": 1.13983286908078, + "grad_norm": 1.5960015058517456, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8769839406013489, + "num_tokens": 215519298.0, + "step": 6138 + }, + { + "epoch": 1.1400185701021355, + "grad_norm": 1.5121756792068481, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8707330226898193, + "num_tokens": 215557064.0, + "step": 6139 + }, + { + "epoch": 1.1402042711234912, + "grad_norm": 1.4130353927612305, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.88303142786026, + "num_tokens": 215595556.0, + "step": 6140 + }, + { + "epoch": 1.1403899721448467, + "grad_norm": 1.5600311756134033, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.871495246887207, + "num_tokens": 215627053.0, + "step": 6141 + }, + { + "epoch": 1.1405756731662025, + "grad_norm": 1.4807072877883911, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.875552237033844, + "num_tokens": 215661611.0, + "step": 6142 + }, + { + "epoch": 1.140761374187558, + "grad_norm": 1.4790844917297363, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8763114213943481, + "num_tokens": 215700848.0, + "step": 6143 + }, + { + "epoch": 1.1409470752089137, + "grad_norm": 1.319753646850586, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8832882046699524, + "num_tokens": 215745152.0, + "step": 6144 + }, + { + "epoch": 1.1411327762302692, + "grad_norm": 1.592490792274475, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8901035785675049, + "num_tokens": 215775683.0, + "step": 6145 + }, + { + "epoch": 1.141318477251625, + "grad_norm": 1.443196177482605, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8804271221160889, + "num_tokens": 215813271.0, + "step": 6146 + }, + { + "epoch": 1.1415041782729805, + "grad_norm": 1.4249017238616943, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8829779028892517, + "num_tokens": 215850891.0, + "step": 6147 + }, + { + "epoch": 1.1416898792943362, + "grad_norm": 1.5624021291732788, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8788393139839172, + "num_tokens": 215886040.0, + "step": 6148 + }, + { + "epoch": 1.1418755803156917, + "grad_norm": 1.3920648097991943, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8890132308006287, + "num_tokens": 215922197.0, + "step": 6149 + }, + { + "epoch": 1.1420612813370474, + "grad_norm": 1.5019994974136353, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8827100992202759, + "num_tokens": 215953527.0, + "step": 6150 + }, + { + "epoch": 1.142246982358403, + "grad_norm": 1.5148355960845947, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8644007444381714, + "num_tokens": 215989065.0, + "step": 6151 + }, + { + "epoch": 1.1424326833797587, + "grad_norm": 1.5248897075653076, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8704535365104675, + "num_tokens": 216024885.0, + "step": 6152 + }, + { + "epoch": 1.1426183844011142, + "grad_norm": 1.4930545091629028, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8765910863876343, + "num_tokens": 216058602.0, + "step": 6153 + }, + { + "epoch": 1.14280408542247, + "grad_norm": 1.607391357421875, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.883071780204773, + "num_tokens": 216089835.0, + "step": 6154 + }, + { + "epoch": 1.1429897864438254, + "grad_norm": 1.6329954862594604, + "learning_rate": 1e-06, + "loss": 0.4156, + "mean_token_accuracy": 0.8619791269302368, + "num_tokens": 216126018.0, + "step": 6155 + }, + { + "epoch": 1.143175487465181, + "grad_norm": 1.3222743272781372, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8863401412963867, + "num_tokens": 216165114.0, + "step": 6156 + }, + { + "epoch": 1.1433611884865367, + "grad_norm": 1.5899038314819336, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8780737519264221, + "num_tokens": 216193946.0, + "step": 6157 + }, + { + "epoch": 1.1435468895078924, + "grad_norm": 1.5793343782424927, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8693245649337769, + "num_tokens": 216227987.0, + "step": 6158 + }, + { + "epoch": 1.143732590529248, + "grad_norm": 1.582663655281067, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8829878568649292, + "num_tokens": 216258987.0, + "step": 6159 + }, + { + "epoch": 1.1439182915506034, + "grad_norm": 1.4427708387374878, + "learning_rate": 1e-06, + "loss": 0.4225, + "mean_token_accuracy": 0.855154275894165, + "num_tokens": 216300379.0, + "step": 6160 + }, + { + "epoch": 1.1441039925719592, + "grad_norm": 1.4815541505813599, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.8596699237823486, + "num_tokens": 216338301.0, + "step": 6161 + }, + { + "epoch": 1.1442896935933147, + "grad_norm": 1.6346523761749268, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8703151941299438, + "num_tokens": 216366851.0, + "step": 6162 + }, + { + "epoch": 1.1444753946146704, + "grad_norm": 1.6412949562072754, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.869233250617981, + "num_tokens": 216396708.0, + "step": 6163 + }, + { + "epoch": 1.144661095636026, + "grad_norm": 1.412384271621704, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8834498524665833, + "num_tokens": 216433501.0, + "step": 6164 + }, + { + "epoch": 1.1448467966573816, + "grad_norm": 1.4698406457901, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8770188093185425, + "num_tokens": 216470646.0, + "step": 6165 + }, + { + "epoch": 1.1450324976787372, + "grad_norm": 1.4378890991210938, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8749678134918213, + "num_tokens": 216506136.0, + "step": 6166 + }, + { + "epoch": 1.145218198700093, + "grad_norm": 1.4465082883834839, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8767690658569336, + "num_tokens": 216544731.0, + "step": 6167 + }, + { + "epoch": 1.1454038997214484, + "grad_norm": 1.4923871755599976, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8824774026870728, + "num_tokens": 216579880.0, + "step": 6168 + }, + { + "epoch": 1.1455896007428041, + "grad_norm": 1.5815331935882568, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.871519148349762, + "num_tokens": 216611920.0, + "step": 6169 + }, + { + "epoch": 1.1457753017641596, + "grad_norm": 1.462241530418396, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8882760405540466, + "num_tokens": 216642650.0, + "step": 6170 + }, + { + "epoch": 1.1459610027855154, + "grad_norm": 1.436468482017517, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8750758171081543, + "num_tokens": 216681893.0, + "step": 6171 + }, + { + "epoch": 1.146146703806871, + "grad_norm": 1.3320895433425903, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8879037499427795, + "num_tokens": 216723712.0, + "step": 6172 + }, + { + "epoch": 1.1463324048282266, + "grad_norm": 1.6965512037277222, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8760755658149719, + "num_tokens": 216754287.0, + "step": 6173 + }, + { + "epoch": 1.1465181058495821, + "grad_norm": 1.7266486883163452, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8647743463516235, + "num_tokens": 216781824.0, + "step": 6174 + }, + { + "epoch": 1.1467038068709379, + "grad_norm": 1.5301190614700317, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8821535110473633, + "num_tokens": 216816415.0, + "step": 6175 + }, + { + "epoch": 1.1468895078922934, + "grad_norm": 1.66214120388031, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8607467412948608, + "num_tokens": 216850810.0, + "step": 6176 + }, + { + "epoch": 1.1470752089136491, + "grad_norm": 1.5125751495361328, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8842369318008423, + "num_tokens": 216884317.0, + "step": 6177 + }, + { + "epoch": 1.1472609099350046, + "grad_norm": 1.560027003288269, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.8624322414398193, + "num_tokens": 216921416.0, + "step": 6178 + }, + { + "epoch": 1.1474466109563601, + "grad_norm": 1.5423239469528198, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8801643252372742, + "num_tokens": 216957455.0, + "step": 6179 + }, + { + "epoch": 1.1476323119777159, + "grad_norm": 1.6348522901535034, + "learning_rate": 1e-06, + "loss": 0.3878, + "mean_token_accuracy": 0.8647985458374023, + "num_tokens": 216989272.0, + "step": 6180 + }, + { + "epoch": 1.1478180129990716, + "grad_norm": 1.4113258123397827, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.878776490688324, + "num_tokens": 217026600.0, + "step": 6181 + }, + { + "epoch": 1.1480037140204271, + "grad_norm": 1.470896601676941, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.880480170249939, + "num_tokens": 217064787.0, + "step": 6182 + }, + { + "epoch": 1.1481894150417826, + "grad_norm": 1.3653326034545898, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8761277198791504, + "num_tokens": 217104030.0, + "step": 6183 + }, + { + "epoch": 1.1483751160631384, + "grad_norm": 1.467655062675476, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8880069851875305, + "num_tokens": 217139664.0, + "step": 6184 + }, + { + "epoch": 1.1485608170844939, + "grad_norm": 1.426463007926941, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8858376741409302, + "num_tokens": 217177478.0, + "step": 6185 + }, + { + "epoch": 1.1487465181058496, + "grad_norm": 1.663012146949768, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8708683252334595, + "num_tokens": 217207502.0, + "step": 6186 + }, + { + "epoch": 1.1489322191272051, + "grad_norm": 1.86063814163208, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.865353524684906, + "num_tokens": 217243416.0, + "step": 6187 + }, + { + "epoch": 1.1491179201485608, + "grad_norm": 1.3333231210708618, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8817175030708313, + "num_tokens": 217282373.0, + "step": 6188 + }, + { + "epoch": 1.1493036211699164, + "grad_norm": 1.3634718656539917, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8868893980979919, + "num_tokens": 217321050.0, + "step": 6189 + }, + { + "epoch": 1.149489322191272, + "grad_norm": 1.4226192235946655, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8883299231529236, + "num_tokens": 217358672.0, + "step": 6190 + }, + { + "epoch": 1.1496750232126276, + "grad_norm": 1.4986158609390259, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8809268474578857, + "num_tokens": 217392084.0, + "step": 6191 + }, + { + "epoch": 1.1498607242339833, + "grad_norm": 1.687914252281189, + "learning_rate": 1e-06, + "loss": 0.4058, + "mean_token_accuracy": 0.8641064167022705, + "num_tokens": 217423445.0, + "step": 6192 + }, + { + "epoch": 1.1500464252553388, + "grad_norm": 1.451730489730835, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8724900484085083, + "num_tokens": 217460205.0, + "step": 6193 + }, + { + "epoch": 1.1502321262766946, + "grad_norm": 1.2986253499984741, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8882613778114319, + "num_tokens": 217499838.0, + "step": 6194 + }, + { + "epoch": 1.15041782729805, + "grad_norm": 1.4428081512451172, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8890136480331421, + "num_tokens": 217534490.0, + "step": 6195 + }, + { + "epoch": 1.1506035283194058, + "grad_norm": 1.6166579723358154, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8788166642189026, + "num_tokens": 217565504.0, + "step": 6196 + }, + { + "epoch": 1.1507892293407613, + "grad_norm": 1.4568029642105103, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8714113235473633, + "num_tokens": 217604118.0, + "step": 6197 + }, + { + "epoch": 1.150974930362117, + "grad_norm": 1.4496171474456787, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8799222111701965, + "num_tokens": 217639688.0, + "step": 6198 + }, + { + "epoch": 1.1511606313834726, + "grad_norm": 1.487195611000061, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8809952735900879, + "num_tokens": 217676568.0, + "step": 6199 + }, + { + "epoch": 1.1513463324048283, + "grad_norm": 1.4841896295547485, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8738889694213867, + "num_tokens": 217711255.0, + "step": 6200 + }, + { + "epoch": 1.1515320334261838, + "grad_norm": 1.5143970251083374, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8901908993721008, + "num_tokens": 217742078.0, + "step": 6201 + }, + { + "epoch": 1.1517177344475396, + "grad_norm": 1.4392627477645874, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8680745363235474, + "num_tokens": 217783360.0, + "step": 6202 + }, + { + "epoch": 1.151903435468895, + "grad_norm": 1.4750914573669434, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8714423179626465, + "num_tokens": 217819993.0, + "step": 6203 + }, + { + "epoch": 1.1520891364902508, + "grad_norm": 1.6856194734573364, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8747060298919678, + "num_tokens": 217847223.0, + "step": 6204 + }, + { + "epoch": 1.1522748375116063, + "grad_norm": 1.5728740692138672, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8717032670974731, + "num_tokens": 217881931.0, + "step": 6205 + }, + { + "epoch": 1.1524605385329618, + "grad_norm": 1.5989420413970947, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8821985125541687, + "num_tokens": 217911004.0, + "step": 6206 + }, + { + "epoch": 1.1526462395543176, + "grad_norm": 1.4475620985031128, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8782821297645569, + "num_tokens": 217946630.0, + "step": 6207 + }, + { + "epoch": 1.1528319405756733, + "grad_norm": 1.4475548267364502, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8748005032539368, + "num_tokens": 217981161.0, + "step": 6208 + }, + { + "epoch": 1.1530176415970288, + "grad_norm": 1.429524302482605, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8544663786888123, + "num_tokens": 218023196.0, + "step": 6209 + }, + { + "epoch": 1.1532033426183843, + "grad_norm": 1.3934271335601807, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8903676867485046, + "num_tokens": 218060054.0, + "step": 6210 + }, + { + "epoch": 1.15338904363974, + "grad_norm": 1.4507865905761719, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8728130459785461, + "num_tokens": 218099482.0, + "step": 6211 + }, + { + "epoch": 1.1535747446610956, + "grad_norm": 1.5057321786880493, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8681129217147827, + "num_tokens": 218135777.0, + "step": 6212 + }, + { + "epoch": 1.1537604456824513, + "grad_norm": 1.411873698234558, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8894491791725159, + "num_tokens": 218170167.0, + "step": 6213 + }, + { + "epoch": 1.1539461467038068, + "grad_norm": 1.354366421699524, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8891136646270752, + "num_tokens": 218214404.0, + "step": 6214 + }, + { + "epoch": 1.1541318477251625, + "grad_norm": 1.399402379989624, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8792462348937988, + "num_tokens": 218250407.0, + "step": 6215 + }, + { + "epoch": 1.154317548746518, + "grad_norm": 1.3980449438095093, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8664335012435913, + "num_tokens": 218290779.0, + "step": 6216 + }, + { + "epoch": 1.1545032497678738, + "grad_norm": 1.452061414718628, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8839794993400574, + "num_tokens": 218329592.0, + "step": 6217 + }, + { + "epoch": 1.1546889507892293, + "grad_norm": 1.3867748975753784, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8877581357955933, + "num_tokens": 218370372.0, + "step": 6218 + }, + { + "epoch": 1.154874651810585, + "grad_norm": 1.5014795064926147, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8656352162361145, + "num_tokens": 218408173.0, + "step": 6219 + }, + { + "epoch": 1.1550603528319405, + "grad_norm": 1.3959509134292603, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8705479502677917, + "num_tokens": 218447950.0, + "step": 6220 + }, + { + "epoch": 1.1552460538532963, + "grad_norm": 1.5536555051803589, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8747013807296753, + "num_tokens": 218482061.0, + "step": 6221 + }, + { + "epoch": 1.1554317548746518, + "grad_norm": 1.462694764137268, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8825687170028687, + "num_tokens": 218516095.0, + "step": 6222 + }, + { + "epoch": 1.1556174558960075, + "grad_norm": 1.5101367235183716, + "learning_rate": 1e-06, + "loss": 0.3958, + "mean_token_accuracy": 0.8651835918426514, + "num_tokens": 218553641.0, + "step": 6223 + }, + { + "epoch": 1.155803156917363, + "grad_norm": 1.4267497062683105, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8744922280311584, + "num_tokens": 218594211.0, + "step": 6224 + }, + { + "epoch": 1.1559888579387188, + "grad_norm": 1.4498118162155151, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8765469789505005, + "num_tokens": 218630123.0, + "step": 6225 + }, + { + "epoch": 1.1561745589600743, + "grad_norm": 1.4814141988754272, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8745987415313721, + "num_tokens": 218667092.0, + "step": 6226 + }, + { + "epoch": 1.15636025998143, + "grad_norm": 1.394643783569336, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8943390846252441, + "num_tokens": 218705313.0, + "step": 6227 + }, + { + "epoch": 1.1565459610027855, + "grad_norm": 1.497518539428711, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8892053961753845, + "num_tokens": 218739311.0, + "step": 6228 + }, + { + "epoch": 1.156731662024141, + "grad_norm": 1.416345477104187, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.890180766582489, + "num_tokens": 218775670.0, + "step": 6229 + }, + { + "epoch": 1.1569173630454967, + "grad_norm": 1.5598417520523071, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8681361675262451, + "num_tokens": 218811359.0, + "step": 6230 + }, + { + "epoch": 1.1571030640668525, + "grad_norm": 1.526903748512268, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8827031850814819, + "num_tokens": 218843933.0, + "step": 6231 + }, + { + "epoch": 1.157288765088208, + "grad_norm": 1.382218599319458, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8785791993141174, + "num_tokens": 218882018.0, + "step": 6232 + }, + { + "epoch": 1.1574744661095635, + "grad_norm": 1.5199193954467773, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8755693435668945, + "num_tokens": 218915165.0, + "step": 6233 + }, + { + "epoch": 1.1576601671309192, + "grad_norm": 1.5346152782440186, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8602724075317383, + "num_tokens": 218952235.0, + "step": 6234 + }, + { + "epoch": 1.1578458681522747, + "grad_norm": 1.4655966758728027, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.884859561920166, + "num_tokens": 218986301.0, + "step": 6235 + }, + { + "epoch": 1.1580315691736305, + "grad_norm": 1.376261591911316, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8949496746063232, + "num_tokens": 219022234.0, + "step": 6236 + }, + { + "epoch": 1.158217270194986, + "grad_norm": 1.5581070184707642, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8867447376251221, + "num_tokens": 219051851.0, + "step": 6237 + }, + { + "epoch": 1.1584029712163417, + "grad_norm": 1.5475335121154785, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8814276456832886, + "num_tokens": 219087686.0, + "step": 6238 + }, + { + "epoch": 1.1585886722376972, + "grad_norm": 1.5520106554031372, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8790680170059204, + "num_tokens": 219119554.0, + "step": 6239 + }, + { + "epoch": 1.158774373259053, + "grad_norm": 1.4747220277786255, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8707781434059143, + "num_tokens": 219156313.0, + "step": 6240 + }, + { + "epoch": 1.1589600742804085, + "grad_norm": 1.40326988697052, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.881770133972168, + "num_tokens": 219191515.0, + "step": 6241 + }, + { + "epoch": 1.1591457753017642, + "grad_norm": 1.5728328227996826, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8665858507156372, + "num_tokens": 219226078.0, + "step": 6242 + }, + { + "epoch": 1.1593314763231197, + "grad_norm": 1.3901067972183228, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8785086870193481, + "num_tokens": 219266382.0, + "step": 6243 + }, + { + "epoch": 1.1595171773444755, + "grad_norm": 1.5501065254211426, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8781595230102539, + "num_tokens": 219298606.0, + "step": 6244 + }, + { + "epoch": 1.159702878365831, + "grad_norm": 1.4922378063201904, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8840903043746948, + "num_tokens": 219334956.0, + "step": 6245 + }, + { + "epoch": 1.1598885793871867, + "grad_norm": 1.5190364122390747, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8681128025054932, + "num_tokens": 219372351.0, + "step": 6246 + }, + { + "epoch": 1.1600742804085422, + "grad_norm": 1.5196236371994019, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8768991231918335, + "num_tokens": 219406124.0, + "step": 6247 + }, + { + "epoch": 1.160259981429898, + "grad_norm": 1.4945036172866821, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.867709755897522, + "num_tokens": 219441882.0, + "step": 6248 + }, + { + "epoch": 1.1604456824512535, + "grad_norm": 1.6430734395980835, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8721396327018738, + "num_tokens": 219479071.0, + "step": 6249 + }, + { + "epoch": 1.1606313834726092, + "grad_norm": 1.4024665355682373, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8817918300628662, + "num_tokens": 219520036.0, + "step": 6250 + }, + { + "epoch": 1.1608170844939647, + "grad_norm": 1.4555174112319946, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8659167289733887, + "num_tokens": 219559204.0, + "step": 6251 + }, + { + "epoch": 1.1610027855153202, + "grad_norm": 1.4423595666885376, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8867771029472351, + "num_tokens": 219596664.0, + "step": 6252 + }, + { + "epoch": 1.161188486536676, + "grad_norm": 1.5765453577041626, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8597609400749207, + "num_tokens": 219632588.0, + "step": 6253 + }, + { + "epoch": 1.1613741875580317, + "grad_norm": 1.5075535774230957, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8721115589141846, + "num_tokens": 219667614.0, + "step": 6254 + }, + { + "epoch": 1.1615598885793872, + "grad_norm": 1.3908231258392334, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8783081769943237, + "num_tokens": 219706074.0, + "step": 6255 + }, + { + "epoch": 1.1617455896007427, + "grad_norm": 1.3979753255844116, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8859146237373352, + "num_tokens": 219742575.0, + "step": 6256 + }, + { + "epoch": 1.1619312906220984, + "grad_norm": 1.5087249279022217, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8681216239929199, + "num_tokens": 219779553.0, + "step": 6257 + }, + { + "epoch": 1.162116991643454, + "grad_norm": 1.4446314573287964, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8774872422218323, + "num_tokens": 219815010.0, + "step": 6258 + }, + { + "epoch": 1.1623026926648097, + "grad_norm": 1.4396463632583618, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8637271523475647, + "num_tokens": 219854461.0, + "step": 6259 + }, + { + "epoch": 1.1624883936861652, + "grad_norm": 1.4558814764022827, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8823508620262146, + "num_tokens": 219890451.0, + "step": 6260 + }, + { + "epoch": 1.162674094707521, + "grad_norm": 1.5303337574005127, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8884174823760986, + "num_tokens": 219919647.0, + "step": 6261 + }, + { + "epoch": 1.1628597957288764, + "grad_norm": 1.333335041999817, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8673401474952698, + "num_tokens": 219960526.0, + "step": 6262 + }, + { + "epoch": 1.1630454967502322, + "grad_norm": 1.6204769611358643, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8712857961654663, + "num_tokens": 219990309.0, + "step": 6263 + }, + { + "epoch": 1.1632311977715877, + "grad_norm": 1.5181238651275635, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8754385709762573, + "num_tokens": 220026239.0, + "step": 6264 + }, + { + "epoch": 1.1634168987929434, + "grad_norm": 1.4352188110351562, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8665781021118164, + "num_tokens": 220065285.0, + "step": 6265 + }, + { + "epoch": 1.163602599814299, + "grad_norm": 1.3593746423721313, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8904349207878113, + "num_tokens": 220101730.0, + "step": 6266 + }, + { + "epoch": 1.1637883008356547, + "grad_norm": 1.5369569063186646, + "learning_rate": 1e-06, + "loss": 0.4108, + "mean_token_accuracy": 0.8630825281143188, + "num_tokens": 220136930.0, + "step": 6267 + }, + { + "epoch": 1.1639740018570102, + "grad_norm": 1.5899055004119873, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8757901191711426, + "num_tokens": 220170209.0, + "step": 6268 + }, + { + "epoch": 1.164159702878366, + "grad_norm": 1.6261403560638428, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8770089149475098, + "num_tokens": 220203607.0, + "step": 6269 + }, + { + "epoch": 1.1643454038997214, + "grad_norm": 1.4769283533096313, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8661551475524902, + "num_tokens": 220240065.0, + "step": 6270 + }, + { + "epoch": 1.1645311049210771, + "grad_norm": 1.647639274597168, + "learning_rate": 1e-06, + "loss": 0.4506, + "mean_token_accuracy": 0.8484672904014587, + "num_tokens": 220273354.0, + "step": 6271 + }, + { + "epoch": 1.1647168059424327, + "grad_norm": 1.5433138608932495, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8762996196746826, + "num_tokens": 220311214.0, + "step": 6272 + }, + { + "epoch": 1.1649025069637884, + "grad_norm": 1.5184639692306519, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8888130187988281, + "num_tokens": 220341820.0, + "step": 6273 + }, + { + "epoch": 1.165088207985144, + "grad_norm": 1.572522759437561, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8725218772888184, + "num_tokens": 220373026.0, + "step": 6274 + }, + { + "epoch": 1.1652739090064996, + "grad_norm": 1.5196787118911743, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8844422101974487, + "num_tokens": 220404647.0, + "step": 6275 + }, + { + "epoch": 1.1654596100278551, + "grad_norm": 1.6189225912094116, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.8590965867042542, + "num_tokens": 220436241.0, + "step": 6276 + }, + { + "epoch": 1.1656453110492109, + "grad_norm": 1.450356364250183, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8849431276321411, + "num_tokens": 220471994.0, + "step": 6277 + }, + { + "epoch": 1.1658310120705664, + "grad_norm": 1.474738359451294, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8704642057418823, + "num_tokens": 220507540.0, + "step": 6278 + }, + { + "epoch": 1.166016713091922, + "grad_norm": 1.4512147903442383, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8836306929588318, + "num_tokens": 220543223.0, + "step": 6279 + }, + { + "epoch": 1.1662024141132776, + "grad_norm": 1.4197498559951782, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8752849102020264, + "num_tokens": 220579357.0, + "step": 6280 + }, + { + "epoch": 1.1663881151346334, + "grad_norm": 1.5447888374328613, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.864433765411377, + "num_tokens": 220614354.0, + "step": 6281 + }, + { + "epoch": 1.1665738161559889, + "grad_norm": 1.5754245519638062, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8662956953048706, + "num_tokens": 220649619.0, + "step": 6282 + }, + { + "epoch": 1.1667595171773444, + "grad_norm": 1.4966635704040527, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8781183958053589, + "num_tokens": 220683866.0, + "step": 6283 + }, + { + "epoch": 1.1669452181987001, + "grad_norm": 1.7611781358718872, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8695722222328186, + "num_tokens": 220710811.0, + "step": 6284 + }, + { + "epoch": 1.1671309192200556, + "grad_norm": 1.4817546606063843, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8746155500411987, + "num_tokens": 220749017.0, + "step": 6285 + }, + { + "epoch": 1.1673166202414114, + "grad_norm": 1.7577381134033203, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.863587498664856, + "num_tokens": 220778865.0, + "step": 6286 + }, + { + "epoch": 1.1675023212627669, + "grad_norm": 1.5086418390274048, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8793960809707642, + "num_tokens": 220812395.0, + "step": 6287 + }, + { + "epoch": 1.1676880222841226, + "grad_norm": 1.6210018396377563, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8613334894180298, + "num_tokens": 220845901.0, + "step": 6288 + }, + { + "epoch": 1.1678737233054781, + "grad_norm": 1.3032965660095215, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8790034055709839, + "num_tokens": 220888619.0, + "step": 6289 + }, + { + "epoch": 1.1680594243268339, + "grad_norm": 1.3920081853866577, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8949296474456787, + "num_tokens": 220928812.0, + "step": 6290 + }, + { + "epoch": 1.1682451253481894, + "grad_norm": 1.4684391021728516, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8702143430709839, + "num_tokens": 220972434.0, + "step": 6291 + }, + { + "epoch": 1.168430826369545, + "grad_norm": 1.6135737895965576, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8738341927528381, + "num_tokens": 221004801.0, + "step": 6292 + }, + { + "epoch": 1.1686165273909006, + "grad_norm": 1.4578114748001099, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8779604434967041, + "num_tokens": 221041541.0, + "step": 6293 + }, + { + "epoch": 1.1688022284122563, + "grad_norm": 1.4916826486587524, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8815218806266785, + "num_tokens": 221077187.0, + "step": 6294 + }, + { + "epoch": 1.1689879294336118, + "grad_norm": 1.448593258857727, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8827378153800964, + "num_tokens": 221114902.0, + "step": 6295 + }, + { + "epoch": 1.1691736304549676, + "grad_norm": 1.4406931400299072, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8727679252624512, + "num_tokens": 221152634.0, + "step": 6296 + }, + { + "epoch": 1.169359331476323, + "grad_norm": 1.408556580543518, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8739693760871887, + "num_tokens": 221188694.0, + "step": 6297 + }, + { + "epoch": 1.1695450324976788, + "grad_norm": 1.5343914031982422, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8796182870864868, + "num_tokens": 221222349.0, + "step": 6298 + }, + { + "epoch": 1.1697307335190343, + "grad_norm": 1.6512868404388428, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8856805562973022, + "num_tokens": 221252186.0, + "step": 6299 + }, + { + "epoch": 1.16991643454039, + "grad_norm": 1.6807897090911865, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8833942413330078, + "num_tokens": 221279651.0, + "step": 6300 + }, + { + "epoch": 1.1701021355617456, + "grad_norm": 1.555859088897705, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8847699165344238, + "num_tokens": 221311743.0, + "step": 6301 + }, + { + "epoch": 1.170287836583101, + "grad_norm": 1.4259865283966064, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8787375688552856, + "num_tokens": 221353755.0, + "step": 6302 + }, + { + "epoch": 1.1704735376044568, + "grad_norm": 1.5074838399887085, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8683480620384216, + "num_tokens": 221388880.0, + "step": 6303 + }, + { + "epoch": 1.1706592386258126, + "grad_norm": 1.5555977821350098, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8785414099693298, + "num_tokens": 221423089.0, + "step": 6304 + }, + { + "epoch": 1.170844939647168, + "grad_norm": 1.5609450340270996, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8784264326095581, + "num_tokens": 221457050.0, + "step": 6305 + }, + { + "epoch": 1.1710306406685236, + "grad_norm": 1.467693567276001, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8718209266662598, + "num_tokens": 221496735.0, + "step": 6306 + }, + { + "epoch": 1.1712163416898793, + "grad_norm": 1.5947903394699097, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8745768070220947, + "num_tokens": 221529136.0, + "step": 6307 + }, + { + "epoch": 1.1714020427112348, + "grad_norm": 1.3775452375411987, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8879200220108032, + "num_tokens": 221565178.0, + "step": 6308 + }, + { + "epoch": 1.1715877437325906, + "grad_norm": 1.5985087156295776, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8744336366653442, + "num_tokens": 221597302.0, + "step": 6309 + }, + { + "epoch": 1.171773444753946, + "grad_norm": 1.56290602684021, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8751503825187683, + "num_tokens": 221632453.0, + "step": 6310 + }, + { + "epoch": 1.1719591457753018, + "grad_norm": 1.5304577350616455, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8738906383514404, + "num_tokens": 221666970.0, + "step": 6311 + }, + { + "epoch": 1.1721448467966573, + "grad_norm": 1.4975485801696777, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8627939224243164, + "num_tokens": 221704479.0, + "step": 6312 + }, + { + "epoch": 1.172330547818013, + "grad_norm": 1.5626789331436157, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8800148367881775, + "num_tokens": 221738119.0, + "step": 6313 + }, + { + "epoch": 1.1725162488393686, + "grad_norm": 1.6567068099975586, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8814041614532471, + "num_tokens": 221767920.0, + "step": 6314 + }, + { + "epoch": 1.1727019498607243, + "grad_norm": 1.4547356367111206, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8778152465820312, + "num_tokens": 221804309.0, + "step": 6315 + }, + { + "epoch": 1.1728876508820798, + "grad_norm": 1.3350481986999512, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8755130767822266, + "num_tokens": 221847740.0, + "step": 6316 + }, + { + "epoch": 1.1730733519034355, + "grad_norm": 1.483486533164978, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8755548000335693, + "num_tokens": 221883063.0, + "step": 6317 + }, + { + "epoch": 1.173259052924791, + "grad_norm": 1.4247097969055176, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8748070597648621, + "num_tokens": 221921079.0, + "step": 6318 + }, + { + "epoch": 1.1734447539461468, + "grad_norm": 1.405509352684021, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8692406415939331, + "num_tokens": 221962186.0, + "step": 6319 + }, + { + "epoch": 1.1736304549675023, + "grad_norm": 1.396898627281189, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.886708676815033, + "num_tokens": 222000238.0, + "step": 6320 + }, + { + "epoch": 1.173816155988858, + "grad_norm": 1.614418387413025, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.865761399269104, + "num_tokens": 222031310.0, + "step": 6321 + }, + { + "epoch": 1.1740018570102135, + "grad_norm": 1.5327818393707275, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8953733444213867, + "num_tokens": 222064573.0, + "step": 6322 + }, + { + "epoch": 1.1741875580315693, + "grad_norm": 1.534902572631836, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8610938787460327, + "num_tokens": 222099359.0, + "step": 6323 + }, + { + "epoch": 1.1743732590529248, + "grad_norm": 1.3920561075210571, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8814626336097717, + "num_tokens": 222135905.0, + "step": 6324 + }, + { + "epoch": 1.1745589600742803, + "grad_norm": 1.4054443836212158, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8796831965446472, + "num_tokens": 222171504.0, + "step": 6325 + }, + { + "epoch": 1.174744661095636, + "grad_norm": 1.4469935894012451, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.867723822593689, + "num_tokens": 222209298.0, + "step": 6326 + }, + { + "epoch": 1.1749303621169918, + "grad_norm": 1.54110586643219, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8815181255340576, + "num_tokens": 222238191.0, + "step": 6327 + }, + { + "epoch": 1.1751160631383473, + "grad_norm": 1.3715038299560547, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8838610649108887, + "num_tokens": 222281679.0, + "step": 6328 + }, + { + "epoch": 1.1753017641597028, + "grad_norm": 1.5237982273101807, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8826946020126343, + "num_tokens": 222312039.0, + "step": 6329 + }, + { + "epoch": 1.1754874651810585, + "grad_norm": 1.5480973720550537, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8594247102737427, + "num_tokens": 222345263.0, + "step": 6330 + }, + { + "epoch": 1.175673166202414, + "grad_norm": 1.5870039463043213, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8709717988967896, + "num_tokens": 222377683.0, + "step": 6331 + }, + { + "epoch": 1.1758588672237698, + "grad_norm": 1.5040173530578613, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.873106062412262, + "num_tokens": 222415269.0, + "step": 6332 + }, + { + "epoch": 1.1760445682451253, + "grad_norm": 1.558713674545288, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8809466361999512, + "num_tokens": 222448653.0, + "step": 6333 + }, + { + "epoch": 1.176230269266481, + "grad_norm": 1.432388186454773, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.891940176486969, + "num_tokens": 222482921.0, + "step": 6334 + }, + { + "epoch": 1.1764159702878365, + "grad_norm": 1.401356816291809, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8767719864845276, + "num_tokens": 222519708.0, + "step": 6335 + }, + { + "epoch": 1.1766016713091922, + "grad_norm": 1.6111403703689575, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8854072093963623, + "num_tokens": 222549683.0, + "step": 6336 + }, + { + "epoch": 1.1767873723305478, + "grad_norm": 1.5358537435531616, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8787120580673218, + "num_tokens": 222582195.0, + "step": 6337 + }, + { + "epoch": 1.1769730733519035, + "grad_norm": 1.4239081144332886, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8900546431541443, + "num_tokens": 222617698.0, + "step": 6338 + }, + { + "epoch": 1.177158774373259, + "grad_norm": 1.648752212524414, + "learning_rate": 1e-06, + "loss": 0.4191, + "mean_token_accuracy": 0.8578639030456543, + "num_tokens": 222651485.0, + "step": 6339 + }, + { + "epoch": 1.1773444753946147, + "grad_norm": 1.4574708938598633, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8889127969741821, + "num_tokens": 222685223.0, + "step": 6340 + }, + { + "epoch": 1.1775301764159702, + "grad_norm": 1.3923208713531494, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8793569207191467, + "num_tokens": 222724237.0, + "step": 6341 + }, + { + "epoch": 1.177715877437326, + "grad_norm": 1.472133994102478, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.886864423751831, + "num_tokens": 222762071.0, + "step": 6342 + }, + { + "epoch": 1.1779015784586815, + "grad_norm": 1.4469150304794312, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8777528405189514, + "num_tokens": 222795723.0, + "step": 6343 + }, + { + "epoch": 1.1780872794800372, + "grad_norm": 1.4440386295318604, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8781976699829102, + "num_tokens": 222834140.0, + "step": 6344 + }, + { + "epoch": 1.1782729805013927, + "grad_norm": 1.491861343383789, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8691340088844299, + "num_tokens": 222868863.0, + "step": 6345 + }, + { + "epoch": 1.1784586815227485, + "grad_norm": 1.528208613395691, + "learning_rate": 1e-06, + "loss": 0.4235, + "mean_token_accuracy": 0.8559727072715759, + "num_tokens": 222906232.0, + "step": 6346 + }, + { + "epoch": 1.178644382544104, + "grad_norm": 1.5542950630187988, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8887286186218262, + "num_tokens": 222936832.0, + "step": 6347 + }, + { + "epoch": 1.1788300835654595, + "grad_norm": 1.5856105089187622, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8651307225227356, + "num_tokens": 222969888.0, + "step": 6348 + }, + { + "epoch": 1.1790157845868152, + "grad_norm": 1.515216588973999, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8757225871086121, + "num_tokens": 223004363.0, + "step": 6349 + }, + { + "epoch": 1.179201485608171, + "grad_norm": 1.4742504358291626, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8770827054977417, + "num_tokens": 223043385.0, + "step": 6350 + }, + { + "epoch": 1.1793871866295265, + "grad_norm": 1.544074535369873, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8693616390228271, + "num_tokens": 223078096.0, + "step": 6351 + }, + { + "epoch": 1.179572887650882, + "grad_norm": 1.5266716480255127, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8775529861450195, + "num_tokens": 223113111.0, + "step": 6352 + }, + { + "epoch": 1.1797585886722377, + "grad_norm": 1.5352628231048584, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8805716633796692, + "num_tokens": 223145052.0, + "step": 6353 + }, + { + "epoch": 1.1799442896935932, + "grad_norm": 1.4167743921279907, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8846964836120605, + "num_tokens": 223180941.0, + "step": 6354 + }, + { + "epoch": 1.180129990714949, + "grad_norm": 1.452090859413147, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8802001476287842, + "num_tokens": 223218619.0, + "step": 6355 + }, + { + "epoch": 1.1803156917363045, + "grad_norm": 1.4612659215927124, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8714566230773926, + "num_tokens": 223259643.0, + "step": 6356 + }, + { + "epoch": 1.1805013927576602, + "grad_norm": 1.4269077777862549, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8728490471839905, + "num_tokens": 223298141.0, + "step": 6357 + }, + { + "epoch": 1.1806870937790157, + "grad_norm": 1.5819487571716309, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8736956119537354, + "num_tokens": 223332759.0, + "step": 6358 + }, + { + "epoch": 1.1808727948003714, + "grad_norm": 1.4447929859161377, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8733123540878296, + "num_tokens": 223371335.0, + "step": 6359 + }, + { + "epoch": 1.181058495821727, + "grad_norm": 1.554586410522461, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8721742630004883, + "num_tokens": 223404774.0, + "step": 6360 + }, + { + "epoch": 1.1812441968430827, + "grad_norm": 1.4892833232879639, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8694353103637695, + "num_tokens": 223443019.0, + "step": 6361 + }, + { + "epoch": 1.1814298978644382, + "grad_norm": 1.328840732574463, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8892735838890076, + "num_tokens": 223482002.0, + "step": 6362 + }, + { + "epoch": 1.181615598885794, + "grad_norm": 1.3776073455810547, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8808799386024475, + "num_tokens": 223519467.0, + "step": 6363 + }, + { + "epoch": 1.1818012999071494, + "grad_norm": 1.627897024154663, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8827009201049805, + "num_tokens": 223548359.0, + "step": 6364 + }, + { + "epoch": 1.1819870009285052, + "grad_norm": 1.7028368711471558, + "learning_rate": 1e-06, + "loss": 0.4133, + "mean_token_accuracy": 0.8645527362823486, + "num_tokens": 223581624.0, + "step": 6365 + }, + { + "epoch": 1.1821727019498607, + "grad_norm": 1.6598619222640991, + "learning_rate": 1e-06, + "loss": 0.413, + "mean_token_accuracy": 0.8618621230125427, + "num_tokens": 223614233.0, + "step": 6366 + }, + { + "epoch": 1.1823584029712164, + "grad_norm": 1.5052181482315063, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8854327201843262, + "num_tokens": 223644197.0, + "step": 6367 + }, + { + "epoch": 1.182544103992572, + "grad_norm": 1.573215126991272, + "learning_rate": 1e-06, + "loss": 0.3983, + "mean_token_accuracy": 0.8628208637237549, + "num_tokens": 223679190.0, + "step": 6368 + }, + { + "epoch": 1.1827298050139277, + "grad_norm": 1.5247076749801636, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8822646141052246, + "num_tokens": 223712855.0, + "step": 6369 + }, + { + "epoch": 1.1829155060352832, + "grad_norm": 1.7884877920150757, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8662678003311157, + "num_tokens": 223745013.0, + "step": 6370 + }, + { + "epoch": 1.183101207056639, + "grad_norm": 1.5341229438781738, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8652927875518799, + "num_tokens": 223776863.0, + "step": 6371 + }, + { + "epoch": 1.1832869080779944, + "grad_norm": 1.5626695156097412, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8787763714790344, + "num_tokens": 223807182.0, + "step": 6372 + }, + { + "epoch": 1.1834726090993501, + "grad_norm": 1.4890908002853394, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.874417781829834, + "num_tokens": 223842053.0, + "step": 6373 + }, + { + "epoch": 1.1836583101207057, + "grad_norm": 1.4579812288284302, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8629485368728638, + "num_tokens": 223884828.0, + "step": 6374 + }, + { + "epoch": 1.1838440111420612, + "grad_norm": 1.4901596307754517, + "learning_rate": 1e-06, + "loss": 0.4228, + "mean_token_accuracy": 0.8572065830230713, + "num_tokens": 223927166.0, + "step": 6375 + }, + { + "epoch": 1.184029712163417, + "grad_norm": 1.5128130912780762, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8838449120521545, + "num_tokens": 223963611.0, + "step": 6376 + }, + { + "epoch": 1.1842154131847726, + "grad_norm": 1.572974443435669, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8767719864845276, + "num_tokens": 224001438.0, + "step": 6377 + }, + { + "epoch": 1.1844011142061281, + "grad_norm": 1.4741088151931763, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8841715455055237, + "num_tokens": 224036897.0, + "step": 6378 + }, + { + "epoch": 1.1845868152274837, + "grad_norm": 1.5064972639083862, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8934416770935059, + "num_tokens": 224070265.0, + "step": 6379 + }, + { + "epoch": 1.1847725162488394, + "grad_norm": 1.552314043045044, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8851587772369385, + "num_tokens": 224101599.0, + "step": 6380 + }, + { + "epoch": 1.184958217270195, + "grad_norm": 1.5410339832305908, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8843676447868347, + "num_tokens": 224133058.0, + "step": 6381 + }, + { + "epoch": 1.1851439182915506, + "grad_norm": 1.721381664276123, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8717254400253296, + "num_tokens": 224161059.0, + "step": 6382 + }, + { + "epoch": 1.1853296193129061, + "grad_norm": 1.6573898792266846, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8715013265609741, + "num_tokens": 224189564.0, + "step": 6383 + }, + { + "epoch": 1.1855153203342619, + "grad_norm": 1.4933021068572998, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8754056692123413, + "num_tokens": 224224666.0, + "step": 6384 + }, + { + "epoch": 1.1857010213556174, + "grad_norm": 1.4576550722122192, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8802236318588257, + "num_tokens": 224259841.0, + "step": 6385 + }, + { + "epoch": 1.1858867223769731, + "grad_norm": 1.59862220287323, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8693022727966309, + "num_tokens": 224294618.0, + "step": 6386 + }, + { + "epoch": 1.1860724233983286, + "grad_norm": 1.5001764297485352, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8770262598991394, + "num_tokens": 224327803.0, + "step": 6387 + }, + { + "epoch": 1.1862581244196844, + "grad_norm": 1.5757981538772583, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8715585470199585, + "num_tokens": 224363371.0, + "step": 6388 + }, + { + "epoch": 1.1864438254410399, + "grad_norm": 1.5248713493347168, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8721683621406555, + "num_tokens": 224399763.0, + "step": 6389 + }, + { + "epoch": 1.1866295264623956, + "grad_norm": 1.6248501539230347, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8684179186820984, + "num_tokens": 224429870.0, + "step": 6390 + }, + { + "epoch": 1.1868152274837511, + "grad_norm": 1.4025386571884155, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8850958347320557, + "num_tokens": 224468888.0, + "step": 6391 + }, + { + "epoch": 1.1870009285051069, + "grad_norm": 1.5093854665756226, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8675479888916016, + "num_tokens": 224507752.0, + "step": 6392 + }, + { + "epoch": 1.1871866295264624, + "grad_norm": 1.5927964448928833, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8731045722961426, + "num_tokens": 224538321.0, + "step": 6393 + }, + { + "epoch": 1.187372330547818, + "grad_norm": 1.4027554988861084, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8821700811386108, + "num_tokens": 224574409.0, + "step": 6394 + }, + { + "epoch": 1.1875580315691736, + "grad_norm": 1.5480653047561646, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8623346090316772, + "num_tokens": 224610925.0, + "step": 6395 + }, + { + "epoch": 1.1877437325905293, + "grad_norm": 1.6660563945770264, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8908720016479492, + "num_tokens": 224637614.0, + "step": 6396 + }, + { + "epoch": 1.1879294336118849, + "grad_norm": 1.4107484817504883, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8757323622703552, + "num_tokens": 224678711.0, + "step": 6397 + }, + { + "epoch": 1.1881151346332404, + "grad_norm": 1.505039930343628, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8753648400306702, + "num_tokens": 224712722.0, + "step": 6398 + }, + { + "epoch": 1.188300835654596, + "grad_norm": 1.492969036102295, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8691287040710449, + "num_tokens": 224752420.0, + "step": 6399 + }, + { + "epoch": 1.1884865366759518, + "grad_norm": 1.7487776279449463, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8648653030395508, + "num_tokens": 224779828.0, + "step": 6400 + }, + { + "epoch": 1.1886722376973073, + "grad_norm": 1.5467782020568848, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8819578886032104, + "num_tokens": 224815197.0, + "step": 6401 + }, + { + "epoch": 1.1888579387186629, + "grad_norm": 1.4927548170089722, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8786972761154175, + "num_tokens": 224848661.0, + "step": 6402 + }, + { + "epoch": 1.1890436397400186, + "grad_norm": 1.5545439720153809, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8616464734077454, + "num_tokens": 224885245.0, + "step": 6403 + }, + { + "epoch": 1.189229340761374, + "grad_norm": 1.5035793781280518, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8769975304603577, + "num_tokens": 224921666.0, + "step": 6404 + }, + { + "epoch": 1.1894150417827298, + "grad_norm": 1.5152517557144165, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8875097632408142, + "num_tokens": 224952780.0, + "step": 6405 + }, + { + "epoch": 1.1896007428040853, + "grad_norm": 1.546742558479309, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8699144721031189, + "num_tokens": 224986636.0, + "step": 6406 + }, + { + "epoch": 1.189786443825441, + "grad_norm": 1.4116065502166748, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8807635307312012, + "num_tokens": 225021027.0, + "step": 6407 + }, + { + "epoch": 1.1899721448467966, + "grad_norm": 1.379302740097046, + "learning_rate": 1e-06, + "loss": 0.2863, + "mean_token_accuracy": 0.8991166949272156, + "num_tokens": 225057096.0, + "step": 6408 + }, + { + "epoch": 1.1901578458681523, + "grad_norm": 1.4468839168548584, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8820707201957703, + "num_tokens": 225094168.0, + "step": 6409 + }, + { + "epoch": 1.1903435468895078, + "grad_norm": 1.4505265951156616, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8719335794448853, + "num_tokens": 225131836.0, + "step": 6410 + }, + { + "epoch": 1.1905292479108636, + "grad_norm": 1.4369432926177979, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8846871852874756, + "num_tokens": 225166751.0, + "step": 6411 + }, + { + "epoch": 1.190714948932219, + "grad_norm": 1.3849581480026245, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8785265684127808, + "num_tokens": 225207417.0, + "step": 6412 + }, + { + "epoch": 1.1909006499535748, + "grad_norm": 1.559944987297058, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8859394192695618, + "num_tokens": 225242103.0, + "step": 6413 + }, + { + "epoch": 1.1910863509749303, + "grad_norm": 1.500199556350708, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8831197023391724, + "num_tokens": 225276545.0, + "step": 6414 + }, + { + "epoch": 1.191272051996286, + "grad_norm": 1.5579121112823486, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8894567489624023, + "num_tokens": 225310194.0, + "step": 6415 + }, + { + "epoch": 1.1914577530176416, + "grad_norm": 1.4272128343582153, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8734503984451294, + "num_tokens": 225349912.0, + "step": 6416 + }, + { + "epoch": 1.1916434540389973, + "grad_norm": 1.473086953163147, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.879936933517456, + "num_tokens": 225389638.0, + "step": 6417 + }, + { + "epoch": 1.1918291550603528, + "grad_norm": 1.6817694902420044, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8592310547828674, + "num_tokens": 225419174.0, + "step": 6418 + }, + { + "epoch": 1.1920148560817085, + "grad_norm": 1.3575448989868164, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8806588649749756, + "num_tokens": 225463321.0, + "step": 6419 + }, + { + "epoch": 1.192200557103064, + "grad_norm": 1.476590633392334, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8732863068580627, + "num_tokens": 225497540.0, + "step": 6420 + }, + { + "epoch": 1.1923862581244196, + "grad_norm": 1.6074957847595215, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.880020260810852, + "num_tokens": 225527623.0, + "step": 6421 + }, + { + "epoch": 1.1925719591457753, + "grad_norm": 1.572472333908081, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8763510584831238, + "num_tokens": 225562092.0, + "step": 6422 + }, + { + "epoch": 1.192757660167131, + "grad_norm": 1.6899489164352417, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8623006939888, + "num_tokens": 225592853.0, + "step": 6423 + }, + { + "epoch": 1.1929433611884865, + "grad_norm": 1.4276775121688843, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8792803287506104, + "num_tokens": 225630979.0, + "step": 6424 + }, + { + "epoch": 1.193129062209842, + "grad_norm": 1.5249791145324707, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8908103108406067, + "num_tokens": 225662616.0, + "step": 6425 + }, + { + "epoch": 1.1933147632311978, + "grad_norm": 1.5038371086120605, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8660721182823181, + "num_tokens": 225698285.0, + "step": 6426 + }, + { + "epoch": 1.1935004642525533, + "grad_norm": 1.5967954397201538, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8809349536895752, + "num_tokens": 225729459.0, + "step": 6427 + }, + { + "epoch": 1.193686165273909, + "grad_norm": 1.5525875091552734, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8593864440917969, + "num_tokens": 225763675.0, + "step": 6428 + }, + { + "epoch": 1.1938718662952645, + "grad_norm": 1.6143609285354614, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8683378100395203, + "num_tokens": 225795240.0, + "step": 6429 + }, + { + "epoch": 1.1940575673166203, + "grad_norm": 1.6273361444473267, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8701972961425781, + "num_tokens": 225825951.0, + "step": 6430 + }, + { + "epoch": 1.1942432683379758, + "grad_norm": 1.4523217678070068, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8717162609100342, + "num_tokens": 225863509.0, + "step": 6431 + }, + { + "epoch": 1.1944289693593315, + "grad_norm": 1.5749740600585938, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8765414357185364, + "num_tokens": 225895713.0, + "step": 6432 + }, + { + "epoch": 1.194614670380687, + "grad_norm": 1.3345509767532349, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8862799406051636, + "num_tokens": 225936977.0, + "step": 6433 + }, + { + "epoch": 1.1948003714020428, + "grad_norm": 1.521795392036438, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8742591142654419, + "num_tokens": 225971061.0, + "step": 6434 + }, + { + "epoch": 1.1949860724233983, + "grad_norm": 1.4233261346817017, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8771030902862549, + "num_tokens": 226007520.0, + "step": 6435 + }, + { + "epoch": 1.195171773444754, + "grad_norm": 1.457726001739502, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8881233334541321, + "num_tokens": 226040806.0, + "step": 6436 + }, + { + "epoch": 1.1953574744661095, + "grad_norm": 1.512131929397583, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8753958344459534, + "num_tokens": 226073890.0, + "step": 6437 + }, + { + "epoch": 1.1955431754874652, + "grad_norm": 1.6003731489181519, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8710149526596069, + "num_tokens": 226105200.0, + "step": 6438 + }, + { + "epoch": 1.1957288765088208, + "grad_norm": 1.5502899885177612, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8780333995819092, + "num_tokens": 226139428.0, + "step": 6439 + }, + { + "epoch": 1.1959145775301765, + "grad_norm": 1.469132661819458, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.888116180896759, + "num_tokens": 226175414.0, + "step": 6440 + }, + { + "epoch": 1.196100278551532, + "grad_norm": 1.5241620540618896, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8886101245880127, + "num_tokens": 226206668.0, + "step": 6441 + }, + { + "epoch": 1.1962859795728877, + "grad_norm": 1.6941497325897217, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8565781712532043, + "num_tokens": 226237088.0, + "step": 6442 + }, + { + "epoch": 1.1964716805942432, + "grad_norm": 1.4456292390823364, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8816207647323608, + "num_tokens": 226275707.0, + "step": 6443 + }, + { + "epoch": 1.196657381615599, + "grad_norm": 1.4462010860443115, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8600640296936035, + "num_tokens": 226318600.0, + "step": 6444 + }, + { + "epoch": 1.1968430826369545, + "grad_norm": 1.6162275075912476, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8705703020095825, + "num_tokens": 226355495.0, + "step": 6445 + }, + { + "epoch": 1.1970287836583102, + "grad_norm": 1.5309031009674072, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8845989108085632, + "num_tokens": 226390615.0, + "step": 6446 + }, + { + "epoch": 1.1972144846796657, + "grad_norm": 1.435922622680664, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8681526184082031, + "num_tokens": 226431180.0, + "step": 6447 + }, + { + "epoch": 1.1974001857010212, + "grad_norm": 1.6493054628372192, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8723140954971313, + "num_tokens": 226460794.0, + "step": 6448 + }, + { + "epoch": 1.197585886722377, + "grad_norm": 1.2537997961044312, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8858804702758789, + "num_tokens": 226506819.0, + "step": 6449 + }, + { + "epoch": 1.1977715877437327, + "grad_norm": 1.3936392068862915, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.8704046010971069, + "num_tokens": 226545053.0, + "step": 6450 + }, + { + "epoch": 1.1979572887650882, + "grad_norm": 1.6582951545715332, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8841449022293091, + "num_tokens": 226573579.0, + "step": 6451 + }, + { + "epoch": 1.1981429897864437, + "grad_norm": 1.4620426893234253, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8733673691749573, + "num_tokens": 226608590.0, + "step": 6452 + }, + { + "epoch": 1.1983286908077995, + "grad_norm": 1.3641455173492432, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8760015964508057, + "num_tokens": 226646517.0, + "step": 6453 + }, + { + "epoch": 1.198514391829155, + "grad_norm": 1.4662350416183472, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8823168277740479, + "num_tokens": 226682118.0, + "step": 6454 + }, + { + "epoch": 1.1987000928505107, + "grad_norm": 1.4829082489013672, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8768724203109741, + "num_tokens": 226717647.0, + "step": 6455 + }, + { + "epoch": 1.1988857938718662, + "grad_norm": 1.4374419450759888, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8780947327613831, + "num_tokens": 226750969.0, + "step": 6456 + }, + { + "epoch": 1.199071494893222, + "grad_norm": 1.5540635585784912, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8638125658035278, + "num_tokens": 226785777.0, + "step": 6457 + }, + { + "epoch": 1.1992571959145775, + "grad_norm": 1.4219872951507568, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8734918236732483, + "num_tokens": 226823685.0, + "step": 6458 + }, + { + "epoch": 1.1994428969359332, + "grad_norm": 1.5252681970596313, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8740440607070923, + "num_tokens": 226857139.0, + "step": 6459 + }, + { + "epoch": 1.1996285979572887, + "grad_norm": 1.5937185287475586, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8846412301063538, + "num_tokens": 226889956.0, + "step": 6460 + }, + { + "epoch": 1.1998142989786444, + "grad_norm": 1.5360833406448364, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8828866481781006, + "num_tokens": 226921458.0, + "step": 6461 + }, + { + "epoch": 1.2, + "grad_norm": 1.6292810440063477, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8693084716796875, + "num_tokens": 226952459.0, + "step": 6462 + }, + { + "epoch": 1.2001857010213557, + "grad_norm": 1.6942589282989502, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8608269691467285, + "num_tokens": 226986851.0, + "step": 6463 + }, + { + "epoch": 1.2003714020427112, + "grad_norm": 1.5926469564437866, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8667272329330444, + "num_tokens": 227021286.0, + "step": 6464 + }, + { + "epoch": 1.200557103064067, + "grad_norm": 1.3558963537216187, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8719145655632019, + "num_tokens": 227062908.0, + "step": 6465 + }, + { + "epoch": 1.2007428040854224, + "grad_norm": 1.5889137983322144, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8753089904785156, + "num_tokens": 227096308.0, + "step": 6466 + }, + { + "epoch": 1.2009285051067782, + "grad_norm": 1.5795419216156006, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8598716855049133, + "num_tokens": 227128904.0, + "step": 6467 + }, + { + "epoch": 1.2011142061281337, + "grad_norm": 1.48218834400177, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8676974773406982, + "num_tokens": 227166725.0, + "step": 6468 + }, + { + "epoch": 1.2012999071494894, + "grad_norm": 1.4115289449691772, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.880271315574646, + "num_tokens": 227203726.0, + "step": 6469 + }, + { + "epoch": 1.201485608170845, + "grad_norm": 1.3628084659576416, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8630992770195007, + "num_tokens": 227245527.0, + "step": 6470 + }, + { + "epoch": 1.2016713091922004, + "grad_norm": 1.4540818929672241, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8953490257263184, + "num_tokens": 227279134.0, + "step": 6471 + }, + { + "epoch": 1.2018570102135562, + "grad_norm": 1.5083675384521484, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8806301355361938, + "num_tokens": 227314108.0, + "step": 6472 + }, + { + "epoch": 1.202042711234912, + "grad_norm": 1.4151655435562134, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8769001364707947, + "num_tokens": 227351261.0, + "step": 6473 + }, + { + "epoch": 1.2022284122562674, + "grad_norm": 1.5045734643936157, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.884536862373352, + "num_tokens": 227381729.0, + "step": 6474 + }, + { + "epoch": 1.202414113277623, + "grad_norm": 1.6685595512390137, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8753148317337036, + "num_tokens": 227413456.0, + "step": 6475 + }, + { + "epoch": 1.2025998142989787, + "grad_norm": 1.5060969591140747, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.8693645596504211, + "num_tokens": 227449930.0, + "step": 6476 + }, + { + "epoch": 1.2027855153203342, + "grad_norm": 1.4936782121658325, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.880238950252533, + "num_tokens": 227484137.0, + "step": 6477 + }, + { + "epoch": 1.20297121634169, + "grad_norm": 1.4419503211975098, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8755263686180115, + "num_tokens": 227522391.0, + "step": 6478 + }, + { + "epoch": 1.2031569173630454, + "grad_norm": 1.4683477878570557, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.8633691668510437, + "num_tokens": 227561368.0, + "step": 6479 + }, + { + "epoch": 1.2033426183844012, + "grad_norm": 1.4890905618667603, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8644449710845947, + "num_tokens": 227598951.0, + "step": 6480 + }, + { + "epoch": 1.2035283194057567, + "grad_norm": 1.4542081356048584, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.889880895614624, + "num_tokens": 227633975.0, + "step": 6481 + }, + { + "epoch": 1.2037140204271124, + "grad_norm": 1.7526326179504395, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8747351169586182, + "num_tokens": 227665693.0, + "step": 6482 + }, + { + "epoch": 1.203899721448468, + "grad_norm": 1.565083622932434, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8758765459060669, + "num_tokens": 227694922.0, + "step": 6483 + }, + { + "epoch": 1.2040854224698236, + "grad_norm": 1.4635263681411743, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8677135705947876, + "num_tokens": 227731709.0, + "step": 6484 + }, + { + "epoch": 1.2042711234911792, + "grad_norm": 1.5236140489578247, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8706642985343933, + "num_tokens": 227767257.0, + "step": 6485 + }, + { + "epoch": 1.2044568245125349, + "grad_norm": 1.5114367008209229, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8786436319351196, + "num_tokens": 227801102.0, + "step": 6486 + }, + { + "epoch": 1.2046425255338904, + "grad_norm": 1.4648784399032593, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8829663991928101, + "num_tokens": 227834936.0, + "step": 6487 + }, + { + "epoch": 1.2048282265552461, + "grad_norm": 1.548467755317688, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8622526526451111, + "num_tokens": 227870812.0, + "step": 6488 + }, + { + "epoch": 1.2050139275766016, + "grad_norm": 1.5190354585647583, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8757518529891968, + "num_tokens": 227905823.0, + "step": 6489 + }, + { + "epoch": 1.2051996285979574, + "grad_norm": 1.4398273229599, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8792085647583008, + "num_tokens": 227942299.0, + "step": 6490 + }, + { + "epoch": 1.2053853296193129, + "grad_norm": 1.518548846244812, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.88957679271698, + "num_tokens": 227974411.0, + "step": 6491 + }, + { + "epoch": 1.2055710306406686, + "grad_norm": 1.500622272491455, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8785486221313477, + "num_tokens": 228010940.0, + "step": 6492 + }, + { + "epoch": 1.2057567316620241, + "grad_norm": 1.5925768613815308, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8812566995620728, + "num_tokens": 228039944.0, + "step": 6493 + }, + { + "epoch": 1.2059424326833796, + "grad_norm": 1.6791802644729614, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8694393634796143, + "num_tokens": 228069458.0, + "step": 6494 + }, + { + "epoch": 1.2061281337047354, + "grad_norm": 1.5488166809082031, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8787714838981628, + "num_tokens": 228101858.0, + "step": 6495 + }, + { + "epoch": 1.206313834726091, + "grad_norm": 1.677707552909851, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8717541098594666, + "num_tokens": 228132434.0, + "step": 6496 + }, + { + "epoch": 1.2064995357474466, + "grad_norm": 1.4554215669631958, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8787984251976013, + "num_tokens": 228167746.0, + "step": 6497 + }, + { + "epoch": 1.2066852367688021, + "grad_norm": 1.5955866575241089, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.8839735984802246, + "num_tokens": 228198553.0, + "step": 6498 + }, + { + "epoch": 1.2068709377901579, + "grad_norm": 1.5220869779586792, + "learning_rate": 1e-06, + "loss": 0.3018, + "mean_token_accuracy": 0.8922593593597412, + "num_tokens": 228230987.0, + "step": 6499 + }, + { + "epoch": 1.2070566388115134, + "grad_norm": 1.6228561401367188, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8889184594154358, + "num_tokens": 228258157.0, + "step": 6500 + }, + { + "epoch": 1.207242339832869, + "grad_norm": 1.4169433116912842, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8850250244140625, + "num_tokens": 228294064.0, + "step": 6501 + }, + { + "epoch": 1.2074280408542246, + "grad_norm": 1.4038593769073486, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8750203251838684, + "num_tokens": 228332583.0, + "step": 6502 + }, + { + "epoch": 1.2076137418755803, + "grad_norm": 1.4128777980804443, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8671438097953796, + "num_tokens": 228372572.0, + "step": 6503 + }, + { + "epoch": 1.2077994428969359, + "grad_norm": 1.5004220008850098, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8735768795013428, + "num_tokens": 228406397.0, + "step": 6504 + }, + { + "epoch": 1.2079851439182916, + "grad_norm": 1.5904632806777954, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8759695291519165, + "num_tokens": 228438711.0, + "step": 6505 + }, + { + "epoch": 1.208170844939647, + "grad_norm": 1.6924028396606445, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8784927129745483, + "num_tokens": 228465262.0, + "step": 6506 + }, + { + "epoch": 1.2083565459610028, + "grad_norm": 1.6289854049682617, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8660291433334351, + "num_tokens": 228494478.0, + "step": 6507 + }, + { + "epoch": 1.2085422469823583, + "grad_norm": 1.5250741243362427, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8713098764419556, + "num_tokens": 228529580.0, + "step": 6508 + }, + { + "epoch": 1.208727948003714, + "grad_norm": 1.438248634338379, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8739551305770874, + "num_tokens": 228569312.0, + "step": 6509 + }, + { + "epoch": 1.2089136490250696, + "grad_norm": 1.4427611827850342, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8647695779800415, + "num_tokens": 228612737.0, + "step": 6510 + }, + { + "epoch": 1.2090993500464253, + "grad_norm": 1.3877867460250854, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8750956654548645, + "num_tokens": 228651166.0, + "step": 6511 + }, + { + "epoch": 1.2092850510677808, + "grad_norm": 1.476984977722168, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8911615014076233, + "num_tokens": 228682313.0, + "step": 6512 + }, + { + "epoch": 1.2094707520891366, + "grad_norm": 1.5749800205230713, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8754968643188477, + "num_tokens": 228716697.0, + "step": 6513 + }, + { + "epoch": 1.209656453110492, + "grad_norm": 1.6892050504684448, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8663175106048584, + "num_tokens": 228751692.0, + "step": 6514 + }, + { + "epoch": 1.2098421541318478, + "grad_norm": 1.4235635995864868, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.875029981136322, + "num_tokens": 228789433.0, + "step": 6515 + }, + { + "epoch": 1.2100278551532033, + "grad_norm": 1.5361294746398926, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8667075634002686, + "num_tokens": 228826260.0, + "step": 6516 + }, + { + "epoch": 1.2102135561745588, + "grad_norm": 1.4361356496810913, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8739960193634033, + "num_tokens": 228864990.0, + "step": 6517 + }, + { + "epoch": 1.2103992571959146, + "grad_norm": 1.5986474752426147, + "learning_rate": 1e-06, + "loss": 0.4139, + "mean_token_accuracy": 0.8594827651977539, + "num_tokens": 228898363.0, + "step": 6518 + }, + { + "epoch": 1.2105849582172703, + "grad_norm": 1.6835609674453735, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.87386155128479, + "num_tokens": 228930673.0, + "step": 6519 + }, + { + "epoch": 1.2107706592386258, + "grad_norm": 1.5229445695877075, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.870928943157196, + "num_tokens": 228968169.0, + "step": 6520 + }, + { + "epoch": 1.2109563602599813, + "grad_norm": 1.4921059608459473, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8813332319259644, + "num_tokens": 229006276.0, + "step": 6521 + }, + { + "epoch": 1.211142061281337, + "grad_norm": 1.6044782400131226, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.880630373954773, + "num_tokens": 229038989.0, + "step": 6522 + }, + { + "epoch": 1.2113277623026926, + "grad_norm": 1.56429123878479, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8900827765464783, + "num_tokens": 229066286.0, + "step": 6523 + }, + { + "epoch": 1.2115134633240483, + "grad_norm": 1.7597419023513794, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8742961287498474, + "num_tokens": 229091929.0, + "step": 6524 + }, + { + "epoch": 1.2116991643454038, + "grad_norm": 1.477942943572998, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8765933513641357, + "num_tokens": 229130573.0, + "step": 6525 + }, + { + "epoch": 1.2118848653667595, + "grad_norm": 1.4769041538238525, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.894229531288147, + "num_tokens": 229164099.0, + "step": 6526 + }, + { + "epoch": 1.212070566388115, + "grad_norm": 1.438501000404358, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8687026500701904, + "num_tokens": 229202673.0, + "step": 6527 + }, + { + "epoch": 1.2122562674094708, + "grad_norm": 1.4742282629013062, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.8711429834365845, + "num_tokens": 229239048.0, + "step": 6528 + }, + { + "epoch": 1.2124419684308263, + "grad_norm": 1.4015733003616333, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8824080228805542, + "num_tokens": 229278870.0, + "step": 6529 + }, + { + "epoch": 1.212627669452182, + "grad_norm": 1.442690372467041, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8711738586425781, + "num_tokens": 229316315.0, + "step": 6530 + }, + { + "epoch": 1.2128133704735375, + "grad_norm": 1.5090606212615967, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8773264288902283, + "num_tokens": 229350922.0, + "step": 6531 + }, + { + "epoch": 1.2129990714948933, + "grad_norm": 1.5068804025650024, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.866159975528717, + "num_tokens": 229386070.0, + "step": 6532 + }, + { + "epoch": 1.2131847725162488, + "grad_norm": 1.5313524007797241, + "learning_rate": 1e-06, + "loss": 0.4159, + "mean_token_accuracy": 0.859721302986145, + "num_tokens": 229421295.0, + "step": 6533 + }, + { + "epoch": 1.2133704735376045, + "grad_norm": 1.6955991983413696, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8772059082984924, + "num_tokens": 229449426.0, + "step": 6534 + }, + { + "epoch": 1.21355617455896, + "grad_norm": 1.6084656715393066, + "learning_rate": 1e-06, + "loss": 0.4137, + "mean_token_accuracy": 0.858279824256897, + "num_tokens": 229485025.0, + "step": 6535 + }, + { + "epoch": 1.2137418755803158, + "grad_norm": 1.4375789165496826, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8750817775726318, + "num_tokens": 229523367.0, + "step": 6536 + }, + { + "epoch": 1.2139275766016713, + "grad_norm": 1.4787520170211792, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8893049955368042, + "num_tokens": 229553746.0, + "step": 6537 + }, + { + "epoch": 1.214113277623027, + "grad_norm": 1.6046230792999268, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8650680780410767, + "num_tokens": 229588664.0, + "step": 6538 + }, + { + "epoch": 1.2142989786443825, + "grad_norm": 1.5780892372131348, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8756585121154785, + "num_tokens": 229623102.0, + "step": 6539 + }, + { + "epoch": 1.2144846796657383, + "grad_norm": 1.5531734228134155, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.864471435546875, + "num_tokens": 229657277.0, + "step": 6540 + }, + { + "epoch": 1.2146703806870938, + "grad_norm": 1.474776029586792, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8790621757507324, + "num_tokens": 229692473.0, + "step": 6541 + }, + { + "epoch": 1.2148560817084495, + "grad_norm": 1.562596321105957, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8786062002182007, + "num_tokens": 229723332.0, + "step": 6542 + }, + { + "epoch": 1.215041782729805, + "grad_norm": 1.5225976705551147, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8886985182762146, + "num_tokens": 229757710.0, + "step": 6543 + }, + { + "epoch": 1.2152274837511605, + "grad_norm": 1.5795094966888428, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8819776773452759, + "num_tokens": 229789857.0, + "step": 6544 + }, + { + "epoch": 1.2154131847725163, + "grad_norm": 1.5878745317459106, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8801445364952087, + "num_tokens": 229822333.0, + "step": 6545 + }, + { + "epoch": 1.215598885793872, + "grad_norm": 1.4458725452423096, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8786206245422363, + "num_tokens": 229861419.0, + "step": 6546 + }, + { + "epoch": 1.2157845868152275, + "grad_norm": 1.536063551902771, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8762221932411194, + "num_tokens": 229894941.0, + "step": 6547 + }, + { + "epoch": 1.215970287836583, + "grad_norm": 1.7514842748641968, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8716443777084351, + "num_tokens": 229921588.0, + "step": 6548 + }, + { + "epoch": 1.2161559888579387, + "grad_norm": 1.516872763633728, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.884117603302002, + "num_tokens": 229952482.0, + "step": 6549 + }, + { + "epoch": 1.2163416898792943, + "grad_norm": 1.542147159576416, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8853278160095215, + "num_tokens": 229984222.0, + "step": 6550 + }, + { + "epoch": 1.21652739090065, + "grad_norm": 1.580280065536499, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8828791379928589, + "num_tokens": 230013384.0, + "step": 6551 + }, + { + "epoch": 1.2167130919220055, + "grad_norm": 1.4853575229644775, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8768869042396545, + "num_tokens": 230049034.0, + "step": 6552 + }, + { + "epoch": 1.2168987929433612, + "grad_norm": 1.611167311668396, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8861667513847351, + "num_tokens": 230078024.0, + "step": 6553 + }, + { + "epoch": 1.2170844939647167, + "grad_norm": 1.3526252508163452, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8967671394348145, + "num_tokens": 230115664.0, + "step": 6554 + }, + { + "epoch": 1.2172701949860725, + "grad_norm": 1.3773757219314575, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.880391001701355, + "num_tokens": 230153429.0, + "step": 6555 + }, + { + "epoch": 1.217455896007428, + "grad_norm": 1.4499868154525757, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8818522691726685, + "num_tokens": 230188608.0, + "step": 6556 + }, + { + "epoch": 1.2176415970287837, + "grad_norm": 1.4012106657028198, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8860844969749451, + "num_tokens": 230225878.0, + "step": 6557 + }, + { + "epoch": 1.2178272980501392, + "grad_norm": 1.4792108535766602, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8788131475448608, + "num_tokens": 230258961.0, + "step": 6558 + }, + { + "epoch": 1.218012999071495, + "grad_norm": 1.4600586891174316, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8795216679573059, + "num_tokens": 230295640.0, + "step": 6559 + }, + { + "epoch": 1.2181987000928505, + "grad_norm": 1.495332956314087, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8768356442451477, + "num_tokens": 230330555.0, + "step": 6560 + }, + { + "epoch": 1.2183844011142062, + "grad_norm": 1.4578651189804077, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8787903785705566, + "num_tokens": 230365977.0, + "step": 6561 + }, + { + "epoch": 1.2185701021355617, + "grad_norm": 1.6742664575576782, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.891474187374115, + "num_tokens": 230392225.0, + "step": 6562 + }, + { + "epoch": 1.2187558031569174, + "grad_norm": 1.4769506454467773, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.872782289981842, + "num_tokens": 230429793.0, + "step": 6563 + }, + { + "epoch": 1.218941504178273, + "grad_norm": 1.3599193096160889, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8737493753433228, + "num_tokens": 230474236.0, + "step": 6564 + }, + { + "epoch": 1.2191272051996287, + "grad_norm": 1.6291934251785278, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8606582880020142, + "num_tokens": 230508279.0, + "step": 6565 + }, + { + "epoch": 1.2193129062209842, + "grad_norm": 1.5864044427871704, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8769843578338623, + "num_tokens": 230540729.0, + "step": 6566 + }, + { + "epoch": 1.2194986072423397, + "grad_norm": 1.3672889471054077, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8805747032165527, + "num_tokens": 230578500.0, + "step": 6567 + }, + { + "epoch": 1.2196843082636954, + "grad_norm": 1.430881381034851, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8798031806945801, + "num_tokens": 230618145.0, + "step": 6568 + }, + { + "epoch": 1.2198700092850512, + "grad_norm": 1.6417704820632935, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8691706657409668, + "num_tokens": 230650174.0, + "step": 6569 + }, + { + "epoch": 1.2200557103064067, + "grad_norm": 1.4975687265396118, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8774716854095459, + "num_tokens": 230685962.0, + "step": 6570 + }, + { + "epoch": 1.2202414113277622, + "grad_norm": 1.4655060768127441, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8861538767814636, + "num_tokens": 230718648.0, + "step": 6571 + }, + { + "epoch": 1.220427112349118, + "grad_norm": 1.465633749961853, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8731222152709961, + "num_tokens": 230755681.0, + "step": 6572 + }, + { + "epoch": 1.2206128133704734, + "grad_norm": 1.4620048999786377, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8802680373191833, + "num_tokens": 230789604.0, + "step": 6573 + }, + { + "epoch": 1.2207985143918292, + "grad_norm": 1.5201704502105713, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8745073676109314, + "num_tokens": 230825473.0, + "step": 6574 + }, + { + "epoch": 1.2209842154131847, + "grad_norm": 1.4733667373657227, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8706761002540588, + "num_tokens": 230860412.0, + "step": 6575 + }, + { + "epoch": 1.2211699164345404, + "grad_norm": 1.4813138246536255, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8839066028594971, + "num_tokens": 230894599.0, + "step": 6576 + }, + { + "epoch": 1.221355617455896, + "grad_norm": 1.5157582759857178, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.88648521900177, + "num_tokens": 230923751.0, + "step": 6577 + }, + { + "epoch": 1.2215413184772517, + "grad_norm": 1.7024158239364624, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8693616390228271, + "num_tokens": 230952187.0, + "step": 6578 + }, + { + "epoch": 1.2217270194986072, + "grad_norm": 1.5638813972473145, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.885188102722168, + "num_tokens": 230983630.0, + "step": 6579 + }, + { + "epoch": 1.221912720519963, + "grad_norm": 1.3501970767974854, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8729906678199768, + "num_tokens": 231026427.0, + "step": 6580 + }, + { + "epoch": 1.2220984215413184, + "grad_norm": 1.465013861656189, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8710317611694336, + "num_tokens": 231068357.0, + "step": 6581 + }, + { + "epoch": 1.2222841225626742, + "grad_norm": 1.5828380584716797, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8910287618637085, + "num_tokens": 231098068.0, + "step": 6582 + }, + { + "epoch": 1.2224698235840297, + "grad_norm": 1.5988560914993286, + "learning_rate": 1e-06, + "loss": 0.4386, + "mean_token_accuracy": 0.8550739288330078, + "num_tokens": 231136432.0, + "step": 6583 + }, + { + "epoch": 1.2226555246053854, + "grad_norm": 1.521729826927185, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8799704313278198, + "num_tokens": 231173092.0, + "step": 6584 + }, + { + "epoch": 1.222841225626741, + "grad_norm": 1.4688653945922852, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8732303380966187, + "num_tokens": 231210003.0, + "step": 6585 + }, + { + "epoch": 1.2230269266480966, + "grad_norm": 1.6215548515319824, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8693204522132874, + "num_tokens": 231244325.0, + "step": 6586 + }, + { + "epoch": 1.2232126276694522, + "grad_norm": 1.4345934391021729, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8836488723754883, + "num_tokens": 231278139.0, + "step": 6587 + }, + { + "epoch": 1.223398328690808, + "grad_norm": 1.6293792724609375, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8704496622085571, + "num_tokens": 231308867.0, + "step": 6588 + }, + { + "epoch": 1.2235840297121634, + "grad_norm": 1.4860124588012695, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8751081824302673, + "num_tokens": 231342966.0, + "step": 6589 + }, + { + "epoch": 1.223769730733519, + "grad_norm": 1.812464952468872, + "learning_rate": 1e-06, + "loss": 0.3983, + "mean_token_accuracy": 0.8650654554367065, + "num_tokens": 231369507.0, + "step": 6590 + }, + { + "epoch": 1.2239554317548746, + "grad_norm": 1.3829959630966187, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8779876232147217, + "num_tokens": 231407957.0, + "step": 6591 + }, + { + "epoch": 1.2241411327762304, + "grad_norm": 1.588515043258667, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8740712404251099, + "num_tokens": 231438611.0, + "step": 6592 + }, + { + "epoch": 1.224326833797586, + "grad_norm": 1.637856125831604, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8768506050109863, + "num_tokens": 231474267.0, + "step": 6593 + }, + { + "epoch": 1.2245125348189414, + "grad_norm": 1.6317857503890991, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8760910034179688, + "num_tokens": 231504907.0, + "step": 6594 + }, + { + "epoch": 1.2246982358402971, + "grad_norm": 1.4742577075958252, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8779114484786987, + "num_tokens": 231539930.0, + "step": 6595 + }, + { + "epoch": 1.2248839368616526, + "grad_norm": 1.5056945085525513, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8741433024406433, + "num_tokens": 231571677.0, + "step": 6596 + }, + { + "epoch": 1.2250696378830084, + "grad_norm": 1.6170071363449097, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8742701411247253, + "num_tokens": 231602656.0, + "step": 6597 + }, + { + "epoch": 1.2252553389043639, + "grad_norm": 1.3493897914886475, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8958404064178467, + "num_tokens": 231637777.0, + "step": 6598 + }, + { + "epoch": 1.2254410399257196, + "grad_norm": 1.5561084747314453, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8616630434989929, + "num_tokens": 231672814.0, + "step": 6599 + }, + { + "epoch": 1.2256267409470751, + "grad_norm": 1.5695091485977173, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8610329031944275, + "num_tokens": 231706480.0, + "step": 6600 + }, + { + "epoch": 1.2258124419684309, + "grad_norm": 1.5956778526306152, + "learning_rate": 1e-06, + "loss": 0.4285, + "mean_token_accuracy": 0.8580076694488525, + "num_tokens": 231740261.0, + "step": 6601 + }, + { + "epoch": 1.2259981429897864, + "grad_norm": 1.5679115056991577, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8861442804336548, + "num_tokens": 231772024.0, + "step": 6602 + }, + { + "epoch": 1.226183844011142, + "grad_norm": 1.5109422206878662, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8889362812042236, + "num_tokens": 231805425.0, + "step": 6603 + }, + { + "epoch": 1.2263695450324976, + "grad_norm": 1.3639228343963623, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8784659504890442, + "num_tokens": 231845799.0, + "step": 6604 + }, + { + "epoch": 1.2265552460538534, + "grad_norm": 1.4367369413375854, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8810230493545532, + "num_tokens": 231885852.0, + "step": 6605 + }, + { + "epoch": 1.2267409470752089, + "grad_norm": 1.435049057006836, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8812472820281982, + "num_tokens": 231924229.0, + "step": 6606 + }, + { + "epoch": 1.2269266480965646, + "grad_norm": 1.756026029586792, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8783888816833496, + "num_tokens": 231953710.0, + "step": 6607 + }, + { + "epoch": 1.22711234911792, + "grad_norm": 1.5958082675933838, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8734136819839478, + "num_tokens": 231986196.0, + "step": 6608 + }, + { + "epoch": 1.2272980501392758, + "grad_norm": 1.444250464439392, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8719390034675598, + "num_tokens": 232028127.0, + "step": 6609 + }, + { + "epoch": 1.2274837511606314, + "grad_norm": 1.4666715860366821, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8925145864486694, + "num_tokens": 232062126.0, + "step": 6610 + }, + { + "epoch": 1.227669452181987, + "grad_norm": 1.3584375381469727, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8859747052192688, + "num_tokens": 232101187.0, + "step": 6611 + }, + { + "epoch": 1.2278551532033426, + "grad_norm": 1.3081564903259277, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.876119077205658, + "num_tokens": 232145189.0, + "step": 6612 + }, + { + "epoch": 1.2280408542246983, + "grad_norm": 1.6662182807922363, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8723732233047485, + "num_tokens": 232175775.0, + "step": 6613 + }, + { + "epoch": 1.2282265552460538, + "grad_norm": 1.4330642223358154, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8710963726043701, + "num_tokens": 232215759.0, + "step": 6614 + }, + { + "epoch": 1.2284122562674096, + "grad_norm": 1.3991824388504028, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8825434446334839, + "num_tokens": 232253330.0, + "step": 6615 + }, + { + "epoch": 1.228597957288765, + "grad_norm": 1.460585355758667, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8959497213363647, + "num_tokens": 232285960.0, + "step": 6616 + }, + { + "epoch": 1.2287836583101206, + "grad_norm": 1.4227874279022217, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8910447359085083, + "num_tokens": 232318219.0, + "step": 6617 + }, + { + "epoch": 1.2289693593314763, + "grad_norm": 1.4093685150146484, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8760016560554504, + "num_tokens": 232357564.0, + "step": 6618 + }, + { + "epoch": 1.229155060352832, + "grad_norm": 1.4434009790420532, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8866013288497925, + "num_tokens": 232394822.0, + "step": 6619 + }, + { + "epoch": 1.2293407613741876, + "grad_norm": 1.6832200288772583, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8699083924293518, + "num_tokens": 232427152.0, + "step": 6620 + }, + { + "epoch": 1.229526462395543, + "grad_norm": 1.588449478149414, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8734381794929504, + "num_tokens": 232462910.0, + "step": 6621 + }, + { + "epoch": 1.2297121634168988, + "grad_norm": 1.4929500818252563, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8844304084777832, + "num_tokens": 232499816.0, + "step": 6622 + }, + { + "epoch": 1.2298978644382543, + "grad_norm": 1.531235694885254, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8835951089859009, + "num_tokens": 232531139.0, + "step": 6623 + }, + { + "epoch": 1.23008356545961, + "grad_norm": 1.529009222984314, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8855152130126953, + "num_tokens": 232559479.0, + "step": 6624 + }, + { + "epoch": 1.2302692664809656, + "grad_norm": 1.3467522859573364, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8897722959518433, + "num_tokens": 232597617.0, + "step": 6625 + }, + { + "epoch": 1.2304549675023213, + "grad_norm": 1.4454480409622192, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8774042129516602, + "num_tokens": 232634180.0, + "step": 6626 + }, + { + "epoch": 1.2306406685236768, + "grad_norm": 1.4836173057556152, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8781471252441406, + "num_tokens": 232668725.0, + "step": 6627 + }, + { + "epoch": 1.2308263695450326, + "grad_norm": 1.4170897006988525, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8738152980804443, + "num_tokens": 232707037.0, + "step": 6628 + }, + { + "epoch": 1.231012070566388, + "grad_norm": 1.491695761680603, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8739754557609558, + "num_tokens": 232742545.0, + "step": 6629 + }, + { + "epoch": 1.2311977715877438, + "grad_norm": 1.4710111618041992, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.874241292476654, + "num_tokens": 232778992.0, + "step": 6630 + }, + { + "epoch": 1.2313834726090993, + "grad_norm": 1.4078130722045898, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8782546520233154, + "num_tokens": 232817442.0, + "step": 6631 + }, + { + "epoch": 1.231569173630455, + "grad_norm": 1.4603967666625977, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8913575410842896, + "num_tokens": 232852361.0, + "step": 6632 + }, + { + "epoch": 1.2317548746518105, + "grad_norm": 1.6462558507919312, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8567922115325928, + "num_tokens": 232889096.0, + "step": 6633 + }, + { + "epoch": 1.2319405756731663, + "grad_norm": 1.5936100482940674, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8622167706489563, + "num_tokens": 232922602.0, + "step": 6634 + }, + { + "epoch": 1.2321262766945218, + "grad_norm": 1.4745159149169922, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8655106425285339, + "num_tokens": 232961804.0, + "step": 6635 + }, + { + "epoch": 1.2323119777158775, + "grad_norm": 1.526097297668457, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8842858672142029, + "num_tokens": 232994201.0, + "step": 6636 + }, + { + "epoch": 1.232497678737233, + "grad_norm": 1.51850163936615, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8766541481018066, + "num_tokens": 233030109.0, + "step": 6637 + }, + { + "epoch": 1.2326833797585888, + "grad_norm": 1.382828950881958, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8749307990074158, + "num_tokens": 233068413.0, + "step": 6638 + }, + { + "epoch": 1.2328690807799443, + "grad_norm": 1.6421645879745483, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8816843032836914, + "num_tokens": 233096923.0, + "step": 6639 + }, + { + "epoch": 1.2330547818012998, + "grad_norm": 1.5945031642913818, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8848245739936829, + "num_tokens": 233125798.0, + "step": 6640 + }, + { + "epoch": 1.2332404828226555, + "grad_norm": 1.6861438751220703, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8970004320144653, + "num_tokens": 233149441.0, + "step": 6641 + }, + { + "epoch": 1.2334261838440113, + "grad_norm": 1.5253591537475586, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8704805374145508, + "num_tokens": 233183811.0, + "step": 6642 + }, + { + "epoch": 1.2336118848653668, + "grad_norm": 1.3974432945251465, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8852376341819763, + "num_tokens": 233216604.0, + "step": 6643 + }, + { + "epoch": 1.2337975858867223, + "grad_norm": 1.6747537851333618, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8707165122032166, + "num_tokens": 233242260.0, + "step": 6644 + }, + { + "epoch": 1.233983286908078, + "grad_norm": 1.520435094833374, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8929978609085083, + "num_tokens": 233273811.0, + "step": 6645 + }, + { + "epoch": 1.2341689879294335, + "grad_norm": 1.5242117643356323, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.88019198179245, + "num_tokens": 233307693.0, + "step": 6646 + }, + { + "epoch": 1.2343546889507893, + "grad_norm": 1.7091636657714844, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8856700658798218, + "num_tokens": 233333310.0, + "step": 6647 + }, + { + "epoch": 1.2345403899721448, + "grad_norm": 1.849882960319519, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8591421246528625, + "num_tokens": 233360032.0, + "step": 6648 + }, + { + "epoch": 1.2347260909935005, + "grad_norm": 1.4164382219314575, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8897380828857422, + "num_tokens": 233396653.0, + "step": 6649 + }, + { + "epoch": 1.234911792014856, + "grad_norm": 1.5516701936721802, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8593947291374207, + "num_tokens": 233431070.0, + "step": 6650 + }, + { + "epoch": 1.2350974930362117, + "grad_norm": 1.568317174911499, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8789693713188171, + "num_tokens": 233463313.0, + "step": 6651 + }, + { + "epoch": 1.2352831940575673, + "grad_norm": 1.4180524349212646, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8858305215835571, + "num_tokens": 233497188.0, + "step": 6652 + }, + { + "epoch": 1.235468895078923, + "grad_norm": 1.608408808708191, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8695617914199829, + "num_tokens": 233531135.0, + "step": 6653 + }, + { + "epoch": 1.2356545961002785, + "grad_norm": 1.4510059356689453, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8655641078948975, + "num_tokens": 233567841.0, + "step": 6654 + }, + { + "epoch": 1.2358402971216342, + "grad_norm": 1.442339301109314, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8825966119766235, + "num_tokens": 233603610.0, + "step": 6655 + }, + { + "epoch": 1.2360259981429897, + "grad_norm": 1.5121431350708008, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8722270727157593, + "num_tokens": 233638826.0, + "step": 6656 + }, + { + "epoch": 1.2362116991643455, + "grad_norm": 1.5626939535140991, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.87732994556427, + "num_tokens": 233668898.0, + "step": 6657 + }, + { + "epoch": 1.236397400185701, + "grad_norm": 1.4955430030822754, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8787291049957275, + "num_tokens": 233703011.0, + "step": 6658 + }, + { + "epoch": 1.2365831012070567, + "grad_norm": 1.5127995014190674, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8704590797424316, + "num_tokens": 233739573.0, + "step": 6659 + }, + { + "epoch": 1.2367688022284122, + "grad_norm": 1.4504175186157227, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8747586011886597, + "num_tokens": 233776963.0, + "step": 6660 + }, + { + "epoch": 1.236954503249768, + "grad_norm": 1.4504883289337158, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8803873062133789, + "num_tokens": 233811611.0, + "step": 6661 + }, + { + "epoch": 1.2371402042711235, + "grad_norm": 1.5901647806167603, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8685722947120667, + "num_tokens": 233846314.0, + "step": 6662 + }, + { + "epoch": 1.237325905292479, + "grad_norm": 1.5105583667755127, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8775650262832642, + "num_tokens": 233883380.0, + "step": 6663 + }, + { + "epoch": 1.2375116063138347, + "grad_norm": 1.6196471452713013, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.875089168548584, + "num_tokens": 233914117.0, + "step": 6664 + }, + { + "epoch": 1.2376973073351905, + "grad_norm": 1.469067931175232, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8966919183731079, + "num_tokens": 233945133.0, + "step": 6665 + }, + { + "epoch": 1.237883008356546, + "grad_norm": 1.6252602338790894, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8687342405319214, + "num_tokens": 233978300.0, + "step": 6666 + }, + { + "epoch": 1.2380687093779015, + "grad_norm": 1.4515236616134644, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8770722150802612, + "num_tokens": 234016019.0, + "step": 6667 + }, + { + "epoch": 1.2382544103992572, + "grad_norm": 1.5026588439941406, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8866422176361084, + "num_tokens": 234050830.0, + "step": 6668 + }, + { + "epoch": 1.2384401114206127, + "grad_norm": 1.4967375993728638, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8808751702308655, + "num_tokens": 234085653.0, + "step": 6669 + }, + { + "epoch": 1.2386258124419685, + "grad_norm": 1.5136158466339111, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.879784345626831, + "num_tokens": 234120346.0, + "step": 6670 + }, + { + "epoch": 1.238811513463324, + "grad_norm": 1.6965399980545044, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8777217864990234, + "num_tokens": 234152555.0, + "step": 6671 + }, + { + "epoch": 1.2389972144846797, + "grad_norm": 1.401706337928772, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8797824382781982, + "num_tokens": 234190859.0, + "step": 6672 + }, + { + "epoch": 1.2391829155060352, + "grad_norm": 1.5217348337173462, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8784520626068115, + "num_tokens": 234224951.0, + "step": 6673 + }, + { + "epoch": 1.239368616527391, + "grad_norm": 1.5665141344070435, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8856935501098633, + "num_tokens": 234256428.0, + "step": 6674 + }, + { + "epoch": 1.2395543175487465, + "grad_norm": 1.4872580766677856, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8788589835166931, + "num_tokens": 234291314.0, + "step": 6675 + }, + { + "epoch": 1.2397400185701022, + "grad_norm": 1.497972011566162, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8709831237792969, + "num_tokens": 234327029.0, + "step": 6676 + }, + { + "epoch": 1.2399257195914577, + "grad_norm": 1.4400279521942139, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8771340250968933, + "num_tokens": 234362587.0, + "step": 6677 + }, + { + "epoch": 1.2401114206128134, + "grad_norm": 1.5453027486801147, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.8682599663734436, + "num_tokens": 234395524.0, + "step": 6678 + }, + { + "epoch": 1.240297121634169, + "grad_norm": 1.6102176904678345, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8713752031326294, + "num_tokens": 234424787.0, + "step": 6679 + }, + { + "epoch": 1.2404828226555247, + "grad_norm": 1.4204108715057373, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.887338399887085, + "num_tokens": 234459715.0, + "step": 6680 + }, + { + "epoch": 1.2406685236768802, + "grad_norm": 1.5564106702804565, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8645015954971313, + "num_tokens": 234495863.0, + "step": 6681 + }, + { + "epoch": 1.240854224698236, + "grad_norm": 1.5437716245651245, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8701706528663635, + "num_tokens": 234529930.0, + "step": 6682 + }, + { + "epoch": 1.2410399257195914, + "grad_norm": 1.602413535118103, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8637505769729614, + "num_tokens": 234564303.0, + "step": 6683 + }, + { + "epoch": 1.2412256267409472, + "grad_norm": 1.558493733406067, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8708184957504272, + "num_tokens": 234599415.0, + "step": 6684 + }, + { + "epoch": 1.2414113277623027, + "grad_norm": 1.564547061920166, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8763973712921143, + "num_tokens": 234632560.0, + "step": 6685 + }, + { + "epoch": 1.2415970287836582, + "grad_norm": 1.5075660943984985, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.874548077583313, + "num_tokens": 234668903.0, + "step": 6686 + }, + { + "epoch": 1.241782729805014, + "grad_norm": 1.494947075843811, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8802500367164612, + "num_tokens": 234702634.0, + "step": 6687 + }, + { + "epoch": 1.2419684308263697, + "grad_norm": 1.5963867902755737, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8555740714073181, + "num_tokens": 234739058.0, + "step": 6688 + }, + { + "epoch": 1.2421541318477252, + "grad_norm": 1.4743272066116333, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.873735249042511, + "num_tokens": 234777874.0, + "step": 6689 + }, + { + "epoch": 1.2423398328690807, + "grad_norm": 1.5487090349197388, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8831255435943604, + "num_tokens": 234811878.0, + "step": 6690 + }, + { + "epoch": 1.2425255338904364, + "grad_norm": 1.7031999826431274, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8782427310943604, + "num_tokens": 234840438.0, + "step": 6691 + }, + { + "epoch": 1.2427112349117921, + "grad_norm": 1.4352173805236816, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8819445371627808, + "num_tokens": 234875243.0, + "step": 6692 + }, + { + "epoch": 1.2428969359331477, + "grad_norm": 1.4053568840026855, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8726954460144043, + "num_tokens": 234913308.0, + "step": 6693 + }, + { + "epoch": 1.2430826369545032, + "grad_norm": 1.4649851322174072, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.873147189617157, + "num_tokens": 234948840.0, + "step": 6694 + }, + { + "epoch": 1.243268337975859, + "grad_norm": 1.5430594682693481, + "learning_rate": 1e-06, + "loss": 0.4042, + "mean_token_accuracy": 0.8606011867523193, + "num_tokens": 234985139.0, + "step": 6695 + }, + { + "epoch": 1.2434540389972144, + "grad_norm": 1.6007276773452759, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.864721417427063, + "num_tokens": 235018050.0, + "step": 6696 + }, + { + "epoch": 1.2436397400185701, + "grad_norm": 1.38493812084198, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8841519951820374, + "num_tokens": 235057263.0, + "step": 6697 + }, + { + "epoch": 1.2438254410399256, + "grad_norm": 1.4548492431640625, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8691372275352478, + "num_tokens": 235094483.0, + "step": 6698 + }, + { + "epoch": 1.2440111420612814, + "grad_norm": 1.5008865594863892, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8804850578308105, + "num_tokens": 235128766.0, + "step": 6699 + }, + { + "epoch": 1.244196843082637, + "grad_norm": 1.4932708740234375, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8697453737258911, + "num_tokens": 235165571.0, + "step": 6700 + }, + { + "epoch": 1.2443825441039926, + "grad_norm": 1.6102945804595947, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8851862549781799, + "num_tokens": 235195502.0, + "step": 6701 + }, + { + "epoch": 1.2445682451253481, + "grad_norm": 1.5484751462936401, + "learning_rate": 1e-06, + "loss": 0.4074, + "mean_token_accuracy": 0.8657041192054749, + "num_tokens": 235232332.0, + "step": 6702 + }, + { + "epoch": 1.2447539461467039, + "grad_norm": 1.336998701095581, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8820372819900513, + "num_tokens": 235272051.0, + "step": 6703 + }, + { + "epoch": 1.2449396471680594, + "grad_norm": 1.3984479904174805, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8795632123947144, + "num_tokens": 235307045.0, + "step": 6704 + }, + { + "epoch": 1.2451253481894151, + "grad_norm": 1.4941869974136353, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.871466875076294, + "num_tokens": 235346422.0, + "step": 6705 + }, + { + "epoch": 1.2453110492107706, + "grad_norm": 1.7151281833648682, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8671236038208008, + "num_tokens": 235376443.0, + "step": 6706 + }, + { + "epoch": 1.2454967502321264, + "grad_norm": 1.4948692321777344, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8798271417617798, + "num_tokens": 235411218.0, + "step": 6707 + }, + { + "epoch": 1.2456824512534819, + "grad_norm": 1.3977646827697754, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8674104809761047, + "num_tokens": 235454760.0, + "step": 6708 + }, + { + "epoch": 1.2458681522748376, + "grad_norm": 1.5101269483566284, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8879216909408569, + "num_tokens": 235488746.0, + "step": 6709 + }, + { + "epoch": 1.2460538532961931, + "grad_norm": 1.468031644821167, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.868317723274231, + "num_tokens": 235526268.0, + "step": 6710 + }, + { + "epoch": 1.2462395543175488, + "grad_norm": 1.5850698947906494, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8670694828033447, + "num_tokens": 235558703.0, + "step": 6711 + }, + { + "epoch": 1.2464252553389044, + "grad_norm": 1.4707694053649902, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8736496567726135, + "num_tokens": 235595112.0, + "step": 6712 + }, + { + "epoch": 1.2466109563602599, + "grad_norm": 1.5073575973510742, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8737120628356934, + "num_tokens": 235627110.0, + "step": 6713 + }, + { + "epoch": 1.2467966573816156, + "grad_norm": 1.502830147743225, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8755660057067871, + "num_tokens": 235663836.0, + "step": 6714 + }, + { + "epoch": 1.2469823584029713, + "grad_norm": 1.5225119590759277, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8854502439498901, + "num_tokens": 235695865.0, + "step": 6715 + }, + { + "epoch": 1.2471680594243268, + "grad_norm": 1.545855164527893, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8785487413406372, + "num_tokens": 235731849.0, + "step": 6716 + }, + { + "epoch": 1.2473537604456824, + "grad_norm": 1.7194254398345947, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8734993934631348, + "num_tokens": 235764438.0, + "step": 6717 + }, + { + "epoch": 1.247539461467038, + "grad_norm": 1.292885422706604, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8814629316329956, + "num_tokens": 235809485.0, + "step": 6718 + }, + { + "epoch": 1.2477251624883936, + "grad_norm": 1.862919569015503, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8715035915374756, + "num_tokens": 235852583.0, + "step": 6719 + }, + { + "epoch": 1.2479108635097493, + "grad_norm": 1.5524691343307495, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8767932057380676, + "num_tokens": 235884097.0, + "step": 6720 + }, + { + "epoch": 1.2480965645311048, + "grad_norm": 1.4313427209854126, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8867528438568115, + "num_tokens": 235921586.0, + "step": 6721 + }, + { + "epoch": 1.2482822655524606, + "grad_norm": 1.4897348880767822, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8812457323074341, + "num_tokens": 235955598.0, + "step": 6722 + }, + { + "epoch": 1.248467966573816, + "grad_norm": 1.460020661354065, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8753271102905273, + "num_tokens": 235990550.0, + "step": 6723 + }, + { + "epoch": 1.2486536675951718, + "grad_norm": 1.4392622709274292, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8771644830703735, + "num_tokens": 236030568.0, + "step": 6724 + }, + { + "epoch": 1.2488393686165273, + "grad_norm": 1.5803452730178833, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8693121671676636, + "num_tokens": 236065661.0, + "step": 6725 + }, + { + "epoch": 1.249025069637883, + "grad_norm": 1.5127017498016357, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8830766677856445, + "num_tokens": 236105550.0, + "step": 6726 + }, + { + "epoch": 1.2492107706592386, + "grad_norm": 1.598514199256897, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8853886127471924, + "num_tokens": 236133673.0, + "step": 6727 + }, + { + "epoch": 1.2493964716805943, + "grad_norm": 1.4363858699798584, + "learning_rate": 1e-06, + "loss": 0.3016, + "mean_token_accuracy": 0.8962856531143188, + "num_tokens": 236169685.0, + "step": 6728 + }, + { + "epoch": 1.2495821727019498, + "grad_norm": 1.4507215023040771, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8712266683578491, + "num_tokens": 236208516.0, + "step": 6729 + }, + { + "epoch": 1.2497678737233056, + "grad_norm": 1.4099897146224976, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8699606657028198, + "num_tokens": 236247059.0, + "step": 6730 + }, + { + "epoch": 1.249953574744661, + "grad_norm": 1.5847879648208618, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8699512481689453, + "num_tokens": 236283737.0, + "step": 6731 + }, + { + "epoch": 1.2501392757660166, + "grad_norm": 1.5173029899597168, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.883931040763855, + "num_tokens": 236319112.0, + "step": 6732 + }, + { + "epoch": 1.2503249767873723, + "grad_norm": 1.4687309265136719, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8759492039680481, + "num_tokens": 236358683.0, + "step": 6733 + }, + { + "epoch": 1.250510677808728, + "grad_norm": 1.4360483884811401, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8851516246795654, + "num_tokens": 236394321.0, + "step": 6734 + }, + { + "epoch": 1.2506963788300836, + "grad_norm": 1.519976258277893, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8866217136383057, + "num_tokens": 236427146.0, + "step": 6735 + }, + { + "epoch": 1.250882079851439, + "grad_norm": 1.3382494449615479, + "learning_rate": 1e-06, + "loss": 0.2751, + "mean_token_accuracy": 0.9046130180358887, + "num_tokens": 236462615.0, + "step": 6736 + }, + { + "epoch": 1.2510677808727948, + "grad_norm": 1.4475159645080566, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8822386860847473, + "num_tokens": 236498798.0, + "step": 6737 + }, + { + "epoch": 1.2512534818941505, + "grad_norm": 1.401883602142334, + "learning_rate": 1e-06, + "loss": 0.3986, + "mean_token_accuracy": 0.867084264755249, + "num_tokens": 236544487.0, + "step": 6738 + }, + { + "epoch": 1.251439182915506, + "grad_norm": 1.4106519222259521, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8881471157073975, + "num_tokens": 236582279.0, + "step": 6739 + }, + { + "epoch": 1.2516248839368616, + "grad_norm": 1.4321191310882568, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8712282776832581, + "num_tokens": 236622279.0, + "step": 6740 + }, + { + "epoch": 1.2518105849582173, + "grad_norm": 1.4678281545639038, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8655453324317932, + "num_tokens": 236659756.0, + "step": 6741 + }, + { + "epoch": 1.251996285979573, + "grad_norm": 1.485793948173523, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8691275119781494, + "num_tokens": 236697242.0, + "step": 6742 + }, + { + "epoch": 1.2521819870009285, + "grad_norm": 1.5576053857803345, + "learning_rate": 1e-06, + "loss": 0.3873, + "mean_token_accuracy": 0.866163432598114, + "num_tokens": 236732707.0, + "step": 6743 + }, + { + "epoch": 1.252367688022284, + "grad_norm": 1.6410351991653442, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8711494207382202, + "num_tokens": 236764620.0, + "step": 6744 + }, + { + "epoch": 1.2525533890436398, + "grad_norm": 1.6291393041610718, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8689161539077759, + "num_tokens": 236794536.0, + "step": 6745 + }, + { + "epoch": 1.2527390900649953, + "grad_norm": 1.4874722957611084, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.874322235584259, + "num_tokens": 236828931.0, + "step": 6746 + }, + { + "epoch": 1.252924791086351, + "grad_norm": 1.486838936805725, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8637652397155762, + "num_tokens": 236868625.0, + "step": 6747 + }, + { + "epoch": 1.2531104921077065, + "grad_norm": 1.5953236818313599, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8784035444259644, + "num_tokens": 236902354.0, + "step": 6748 + }, + { + "epoch": 1.2532961931290623, + "grad_norm": 1.5765382051467896, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8772387504577637, + "num_tokens": 236940156.0, + "step": 6749 + }, + { + "epoch": 1.2534818941504178, + "grad_norm": 1.6750645637512207, + "learning_rate": 1e-06, + "loss": 0.4277, + "mean_token_accuracy": 0.8627503514289856, + "num_tokens": 236970354.0, + "step": 6750 + }, + { + "epoch": 1.2536675951717735, + "grad_norm": 1.5675907135009766, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8744300007820129, + "num_tokens": 237003387.0, + "step": 6751 + }, + { + "epoch": 1.253853296193129, + "grad_norm": 1.5083794593811035, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8742758631706238, + "num_tokens": 237042457.0, + "step": 6752 + }, + { + "epoch": 1.2540389972144848, + "grad_norm": 1.5593057870864868, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8731629848480225, + "num_tokens": 237076138.0, + "step": 6753 + }, + { + "epoch": 1.2542246982358403, + "grad_norm": 1.496785283088684, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8825737237930298, + "num_tokens": 237110131.0, + "step": 6754 + }, + { + "epoch": 1.254410399257196, + "grad_norm": 1.5690938234329224, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8631783127784729, + "num_tokens": 237144250.0, + "step": 6755 + }, + { + "epoch": 1.2545961002785515, + "grad_norm": 1.5078097581863403, + "learning_rate": 1e-06, + "loss": 0.4077, + "mean_token_accuracy": 0.8632104396820068, + "num_tokens": 237183211.0, + "step": 6756 + }, + { + "epoch": 1.2547818012999072, + "grad_norm": 1.4128907918930054, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8826860785484314, + "num_tokens": 237219192.0, + "step": 6757 + }, + { + "epoch": 1.2549675023212628, + "grad_norm": 1.4611811637878418, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8734926581382751, + "num_tokens": 237253419.0, + "step": 6758 + }, + { + "epoch": 1.2551532033426183, + "grad_norm": 1.595798134803772, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.872382402420044, + "num_tokens": 237286238.0, + "step": 6759 + }, + { + "epoch": 1.255338904363974, + "grad_norm": 1.4280558824539185, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.88636714220047, + "num_tokens": 237320868.0, + "step": 6760 + }, + { + "epoch": 1.2555246053853297, + "grad_norm": 1.7240549325942993, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8857936859130859, + "num_tokens": 237347566.0, + "step": 6761 + }, + { + "epoch": 1.2557103064066852, + "grad_norm": 1.457694172859192, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8779726028442383, + "num_tokens": 237386708.0, + "step": 6762 + }, + { + "epoch": 1.2558960074280408, + "grad_norm": 1.5790975093841553, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.874035656452179, + "num_tokens": 237418298.0, + "step": 6763 + }, + { + "epoch": 1.2560817084493965, + "grad_norm": 1.5510896444320679, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8736977577209473, + "num_tokens": 237454180.0, + "step": 6764 + }, + { + "epoch": 1.2562674094707522, + "grad_norm": 1.387851357460022, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8834006786346436, + "num_tokens": 237494645.0, + "step": 6765 + }, + { + "epoch": 1.2564531104921077, + "grad_norm": 1.3881268501281738, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8707157373428345, + "num_tokens": 237539242.0, + "step": 6766 + }, + { + "epoch": 1.2566388115134632, + "grad_norm": 1.536877155303955, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8735477328300476, + "num_tokens": 237575117.0, + "step": 6767 + }, + { + "epoch": 1.256824512534819, + "grad_norm": 1.5560083389282227, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.872725248336792, + "num_tokens": 237608307.0, + "step": 6768 + }, + { + "epoch": 1.2570102135561745, + "grad_norm": 1.4177871942520142, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8796523809432983, + "num_tokens": 237647557.0, + "step": 6769 + }, + { + "epoch": 1.2571959145775302, + "grad_norm": 1.742913842201233, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8738710880279541, + "num_tokens": 237676478.0, + "step": 6770 + }, + { + "epoch": 1.2573816155988857, + "grad_norm": 1.552789330482483, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8855584859848022, + "num_tokens": 237704520.0, + "step": 6771 + }, + { + "epoch": 1.2575673166202415, + "grad_norm": 1.5420024394989014, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8840990662574768, + "num_tokens": 237739182.0, + "step": 6772 + }, + { + "epoch": 1.257753017641597, + "grad_norm": 1.530544638633728, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8773682117462158, + "num_tokens": 237771087.0, + "step": 6773 + }, + { + "epoch": 1.2579387186629527, + "grad_norm": 1.48085355758667, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.865791916847229, + "num_tokens": 237808736.0, + "step": 6774 + }, + { + "epoch": 1.2581244196843082, + "grad_norm": 1.5057015419006348, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8746337890625, + "num_tokens": 237843300.0, + "step": 6775 + }, + { + "epoch": 1.258310120705664, + "grad_norm": 1.5887943506240845, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8643743991851807, + "num_tokens": 237876859.0, + "step": 6776 + }, + { + "epoch": 1.2584958217270195, + "grad_norm": 1.660355806350708, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.874981701374054, + "num_tokens": 237908178.0, + "step": 6777 + }, + { + "epoch": 1.2586815227483752, + "grad_norm": 1.3360141515731812, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8876209259033203, + "num_tokens": 237945368.0, + "step": 6778 + }, + { + "epoch": 1.2588672237697307, + "grad_norm": 1.5806851387023926, + "learning_rate": 1e-06, + "loss": 0.3954, + "mean_token_accuracy": 0.8698827028274536, + "num_tokens": 237977865.0, + "step": 6779 + }, + { + "epoch": 1.2590529247910864, + "grad_norm": 1.3913224935531616, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8677053451538086, + "num_tokens": 238019947.0, + "step": 6780 + }, + { + "epoch": 1.259238625812442, + "grad_norm": 1.4726245403289795, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.866641640663147, + "num_tokens": 238058413.0, + "step": 6781 + }, + { + "epoch": 1.2594243268337975, + "grad_norm": 1.6735721826553345, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8723058700561523, + "num_tokens": 238088503.0, + "step": 6782 + }, + { + "epoch": 1.2596100278551532, + "grad_norm": 1.5564879179000854, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8795570135116577, + "num_tokens": 238122144.0, + "step": 6783 + }, + { + "epoch": 1.259795728876509, + "grad_norm": 1.35383141040802, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8766801357269287, + "num_tokens": 238161851.0, + "step": 6784 + }, + { + "epoch": 1.2599814298978644, + "grad_norm": 1.429100513458252, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8722813129425049, + "num_tokens": 238201914.0, + "step": 6785 + }, + { + "epoch": 1.26016713091922, + "grad_norm": 1.3892430067062378, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8796021938323975, + "num_tokens": 238240140.0, + "step": 6786 + }, + { + "epoch": 1.2603528319405757, + "grad_norm": 1.483271598815918, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8781293034553528, + "num_tokens": 238281229.0, + "step": 6787 + }, + { + "epoch": 1.2605385329619314, + "grad_norm": 1.5835511684417725, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.8677455186843872, + "num_tokens": 238315984.0, + "step": 6788 + }, + { + "epoch": 1.260724233983287, + "grad_norm": 1.4780768156051636, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8746520280838013, + "num_tokens": 238351137.0, + "step": 6789 + }, + { + "epoch": 1.2609099350046424, + "grad_norm": 1.4495095014572144, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8863250017166138, + "num_tokens": 238388398.0, + "step": 6790 + }, + { + "epoch": 1.2610956360259982, + "grad_norm": 1.4512152671813965, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8809767961502075, + "num_tokens": 238423884.0, + "step": 6791 + }, + { + "epoch": 1.2612813370473537, + "grad_norm": 1.4204360246658325, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8878706693649292, + "num_tokens": 238460244.0, + "step": 6792 + }, + { + "epoch": 1.2614670380687094, + "grad_norm": 1.385815978050232, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8826553821563721, + "num_tokens": 238496696.0, + "step": 6793 + }, + { + "epoch": 1.261652739090065, + "grad_norm": 1.5002237558364868, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8837041854858398, + "num_tokens": 238531672.0, + "step": 6794 + }, + { + "epoch": 1.2618384401114207, + "grad_norm": 1.498830795288086, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8954005837440491, + "num_tokens": 238566231.0, + "step": 6795 + }, + { + "epoch": 1.2620241411327762, + "grad_norm": 1.525546908378601, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8774229288101196, + "num_tokens": 238599139.0, + "step": 6796 + }, + { + "epoch": 1.262209842154132, + "grad_norm": 1.4166715145111084, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8856334686279297, + "num_tokens": 238634782.0, + "step": 6797 + }, + { + "epoch": 1.2623955431754874, + "grad_norm": 1.7512041330337524, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8694113492965698, + "num_tokens": 238663327.0, + "step": 6798 + }, + { + "epoch": 1.2625812441968431, + "grad_norm": 1.5993784666061401, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8749708533287048, + "num_tokens": 238695731.0, + "step": 6799 + }, + { + "epoch": 1.2627669452181987, + "grad_norm": 1.4795385599136353, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.878278374671936, + "num_tokens": 238733990.0, + "step": 6800 + }, + { + "epoch": 1.2629526462395544, + "grad_norm": 1.4365768432617188, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8786939382553101, + "num_tokens": 238771459.0, + "step": 6801 + }, + { + "epoch": 1.26313834726091, + "grad_norm": 1.5485665798187256, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8764588832855225, + "num_tokens": 238803373.0, + "step": 6802 + }, + { + "epoch": 1.2633240482822656, + "grad_norm": 1.3713232278823853, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8767186403274536, + "num_tokens": 238841790.0, + "step": 6803 + }, + { + "epoch": 1.2635097493036211, + "grad_norm": 1.447481632232666, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8830162286758423, + "num_tokens": 238876075.0, + "step": 6804 + }, + { + "epoch": 1.2636954503249767, + "grad_norm": 1.6063268184661865, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8869299292564392, + "num_tokens": 238907132.0, + "step": 6805 + }, + { + "epoch": 1.2638811513463324, + "grad_norm": 1.5376942157745361, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8711087107658386, + "num_tokens": 238946729.0, + "step": 6806 + }, + { + "epoch": 1.2640668523676881, + "grad_norm": 1.4762678146362305, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8637722730636597, + "num_tokens": 238983731.0, + "step": 6807 + }, + { + "epoch": 1.2642525533890436, + "grad_norm": 1.397332787513733, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8884977102279663, + "num_tokens": 239020729.0, + "step": 6808 + }, + { + "epoch": 1.2644382544103991, + "grad_norm": 1.3983029127120972, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8886008262634277, + "num_tokens": 239054972.0, + "step": 6809 + }, + { + "epoch": 1.2646239554317549, + "grad_norm": 1.4418559074401855, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8807740211486816, + "num_tokens": 239088943.0, + "step": 6810 + }, + { + "epoch": 1.2648096564531106, + "grad_norm": 1.4086534976959229, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8845991492271423, + "num_tokens": 239126314.0, + "step": 6811 + }, + { + "epoch": 1.2649953574744661, + "grad_norm": 1.5491728782653809, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8869277834892273, + "num_tokens": 239163722.0, + "step": 6812 + }, + { + "epoch": 1.2651810584958216, + "grad_norm": 1.5458165407180786, + "learning_rate": 1e-06, + "loss": 0.4322, + "mean_token_accuracy": 0.8543100357055664, + "num_tokens": 239199419.0, + "step": 6813 + }, + { + "epoch": 1.2653667595171774, + "grad_norm": 1.5589483976364136, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8840084075927734, + "num_tokens": 239230285.0, + "step": 6814 + }, + { + "epoch": 1.265552460538533, + "grad_norm": 1.4310753345489502, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8895542621612549, + "num_tokens": 239265081.0, + "step": 6815 + }, + { + "epoch": 1.2657381615598886, + "grad_norm": 1.4475808143615723, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8778824210166931, + "num_tokens": 239305799.0, + "step": 6816 + }, + { + "epoch": 1.2659238625812441, + "grad_norm": 1.5325803756713867, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8767212629318237, + "num_tokens": 239341141.0, + "step": 6817 + }, + { + "epoch": 1.2661095636025999, + "grad_norm": 1.6122174263000488, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8788484930992126, + "num_tokens": 239372467.0, + "step": 6818 + }, + { + "epoch": 1.2662952646239554, + "grad_norm": 1.4606972932815552, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8833215832710266, + "num_tokens": 239407137.0, + "step": 6819 + }, + { + "epoch": 1.266480965645311, + "grad_norm": 1.5081270933151245, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.8955586552619934, + "num_tokens": 239438598.0, + "step": 6820 + }, + { + "epoch": 1.2666666666666666, + "grad_norm": 1.4426050186157227, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8794975280761719, + "num_tokens": 239474166.0, + "step": 6821 + }, + { + "epoch": 1.2668523676880223, + "grad_norm": 1.4579613208770752, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8656054139137268, + "num_tokens": 239514314.0, + "step": 6822 + }, + { + "epoch": 1.2670380687093779, + "grad_norm": 1.574568748474121, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8634569644927979, + "num_tokens": 239548957.0, + "step": 6823 + }, + { + "epoch": 1.2672237697307336, + "grad_norm": 1.4730795621871948, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8697789907455444, + "num_tokens": 239584659.0, + "step": 6824 + }, + { + "epoch": 1.267409470752089, + "grad_norm": 1.547875165939331, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8724671006202698, + "num_tokens": 239618530.0, + "step": 6825 + }, + { + "epoch": 1.2675951717734448, + "grad_norm": 1.4952826499938965, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8816603422164917, + "num_tokens": 239651713.0, + "step": 6826 + }, + { + "epoch": 1.2677808727948003, + "grad_norm": 1.5535411834716797, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8730682134628296, + "num_tokens": 239686986.0, + "step": 6827 + }, + { + "epoch": 1.2679665738161559, + "grad_norm": 1.557234287261963, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8699507117271423, + "num_tokens": 239723585.0, + "step": 6828 + }, + { + "epoch": 1.2681522748375116, + "grad_norm": 1.3923757076263428, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8826972246170044, + "num_tokens": 239763591.0, + "step": 6829 + }, + { + "epoch": 1.2683379758588673, + "grad_norm": 1.56520414352417, + "learning_rate": 1e-06, + "loss": 0.4065, + "mean_token_accuracy": 0.8638303279876709, + "num_tokens": 239800853.0, + "step": 6830 + }, + { + "epoch": 1.2685236768802228, + "grad_norm": 1.5979688167572021, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8793784379959106, + "num_tokens": 239833857.0, + "step": 6831 + }, + { + "epoch": 1.2687093779015783, + "grad_norm": 1.4737423658370972, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8820306062698364, + "num_tokens": 239868286.0, + "step": 6832 + }, + { + "epoch": 1.268895078922934, + "grad_norm": 1.5596375465393066, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8651214838027954, + "num_tokens": 239902536.0, + "step": 6833 + }, + { + "epoch": 1.2690807799442898, + "grad_norm": 1.4504609107971191, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8867430686950684, + "num_tokens": 239935437.0, + "step": 6834 + }, + { + "epoch": 1.2692664809656453, + "grad_norm": 1.4716241359710693, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8790432214736938, + "num_tokens": 239969533.0, + "step": 6835 + }, + { + "epoch": 1.2694521819870008, + "grad_norm": 1.451022982597351, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8646635413169861, + "num_tokens": 240007165.0, + "step": 6836 + }, + { + "epoch": 1.2696378830083566, + "grad_norm": 1.574206829071045, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8790087103843689, + "num_tokens": 240040070.0, + "step": 6837 + }, + { + "epoch": 1.2698235840297123, + "grad_norm": 1.4885590076446533, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.876440703868866, + "num_tokens": 240074605.0, + "step": 6838 + }, + { + "epoch": 1.2700092850510678, + "grad_norm": 1.4495817422866821, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8798670172691345, + "num_tokens": 240114434.0, + "step": 6839 + }, + { + "epoch": 1.2701949860724233, + "grad_norm": 1.4038691520690918, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8742591142654419, + "num_tokens": 240153771.0, + "step": 6840 + }, + { + "epoch": 1.270380687093779, + "grad_norm": 1.556999921798706, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8805546760559082, + "num_tokens": 240187503.0, + "step": 6841 + }, + { + "epoch": 1.2705663881151346, + "grad_norm": 1.4227001667022705, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8712782859802246, + "num_tokens": 240228418.0, + "step": 6842 + }, + { + "epoch": 1.2707520891364903, + "grad_norm": 1.6422548294067383, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8675836324691772, + "num_tokens": 240260125.0, + "step": 6843 + }, + { + "epoch": 1.2709377901578458, + "grad_norm": 1.6607283353805542, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.8608206510543823, + "num_tokens": 240291835.0, + "step": 6844 + }, + { + "epoch": 1.2711234911792015, + "grad_norm": 1.5935630798339844, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8829224109649658, + "num_tokens": 240320847.0, + "step": 6845 + }, + { + "epoch": 1.271309192200557, + "grad_norm": 1.4573051929473877, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.878312885761261, + "num_tokens": 240358725.0, + "step": 6846 + }, + { + "epoch": 1.2714948932219128, + "grad_norm": 1.4103769063949585, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8782212138175964, + "num_tokens": 240396580.0, + "step": 6847 + }, + { + "epoch": 1.2716805942432683, + "grad_norm": 1.3990745544433594, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8799886703491211, + "num_tokens": 240441867.0, + "step": 6848 + }, + { + "epoch": 1.271866295264624, + "grad_norm": 1.5583689212799072, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.869208574295044, + "num_tokens": 240471790.0, + "step": 6849 + }, + { + "epoch": 1.2720519962859795, + "grad_norm": 1.4424235820770264, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.8638777136802673, + "num_tokens": 240513215.0, + "step": 6850 + }, + { + "epoch": 1.2722376973073353, + "grad_norm": 1.4077128171920776, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8746403455734253, + "num_tokens": 240552836.0, + "step": 6851 + }, + { + "epoch": 1.2724233983286908, + "grad_norm": 1.3540419340133667, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8758442401885986, + "num_tokens": 240591630.0, + "step": 6852 + }, + { + "epoch": 1.2726090993500465, + "grad_norm": 1.3279579877853394, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8723766803741455, + "num_tokens": 240636842.0, + "step": 6853 + }, + { + "epoch": 1.272794800371402, + "grad_norm": 1.5133140087127686, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8682612180709839, + "num_tokens": 240671896.0, + "step": 6854 + }, + { + "epoch": 1.2729805013927575, + "grad_norm": 1.4991205930709839, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8873127698898315, + "num_tokens": 240707491.0, + "step": 6855 + }, + { + "epoch": 1.2731662024141133, + "grad_norm": 1.4941904544830322, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8730851411819458, + "num_tokens": 240745640.0, + "step": 6856 + }, + { + "epoch": 1.273351903435469, + "grad_norm": 1.6005584001541138, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.868179440498352, + "num_tokens": 240779948.0, + "step": 6857 + }, + { + "epoch": 1.2735376044568245, + "grad_norm": 1.434977650642395, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.871589720249176, + "num_tokens": 240819545.0, + "step": 6858 + }, + { + "epoch": 1.27372330547818, + "grad_norm": 1.3837945461273193, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8984194397926331, + "num_tokens": 240853631.0, + "step": 6859 + }, + { + "epoch": 1.2739090064995358, + "grad_norm": 1.4952633380889893, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8749987483024597, + "num_tokens": 240886533.0, + "step": 6860 + }, + { + "epoch": 1.2740947075208915, + "grad_norm": 1.559412956237793, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8733114004135132, + "num_tokens": 240919482.0, + "step": 6861 + }, + { + "epoch": 1.274280408542247, + "grad_norm": 1.7551630735397339, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8772799372673035, + "num_tokens": 240947950.0, + "step": 6862 + }, + { + "epoch": 1.2744661095636025, + "grad_norm": 1.5910831689834595, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8765542507171631, + "num_tokens": 240981203.0, + "step": 6863 + }, + { + "epoch": 1.2746518105849582, + "grad_norm": 1.4538542032241821, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8733206987380981, + "num_tokens": 241021612.0, + "step": 6864 + }, + { + "epoch": 1.2748375116063138, + "grad_norm": 1.4683340787887573, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8886932730674744, + "num_tokens": 241055465.0, + "step": 6865 + }, + { + "epoch": 1.2750232126276695, + "grad_norm": 1.3884413242340088, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8724846839904785, + "num_tokens": 241098270.0, + "step": 6866 + }, + { + "epoch": 1.275208913649025, + "grad_norm": 1.6302951574325562, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8808900117874146, + "num_tokens": 241126440.0, + "step": 6867 + }, + { + "epoch": 1.2753946146703807, + "grad_norm": 1.660235047340393, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8720197677612305, + "num_tokens": 241159601.0, + "step": 6868 + }, + { + "epoch": 1.2755803156917362, + "grad_norm": 1.4552663564682007, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8895431756973267, + "num_tokens": 241197587.0, + "step": 6869 + }, + { + "epoch": 1.275766016713092, + "grad_norm": 1.4911508560180664, + "learning_rate": 1e-06, + "loss": 0.4006, + "mean_token_accuracy": 0.8643481731414795, + "num_tokens": 241234234.0, + "step": 6870 + }, + { + "epoch": 1.2759517177344475, + "grad_norm": 1.5911544561386108, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.8574592471122742, + "num_tokens": 241269560.0, + "step": 6871 + }, + { + "epoch": 1.2761374187558032, + "grad_norm": 1.4280109405517578, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8856673836708069, + "num_tokens": 241303596.0, + "step": 6872 + }, + { + "epoch": 1.2763231197771587, + "grad_norm": 1.5323821306228638, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8847383260726929, + "num_tokens": 241338643.0, + "step": 6873 + }, + { + "epoch": 1.2765088207985145, + "grad_norm": 1.4873405694961548, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8680382966995239, + "num_tokens": 241376544.0, + "step": 6874 + }, + { + "epoch": 1.27669452181987, + "grad_norm": 1.42031729221344, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8859941959381104, + "num_tokens": 241416087.0, + "step": 6875 + }, + { + "epoch": 1.2768802228412257, + "grad_norm": 1.6146034002304077, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8808579444885254, + "num_tokens": 241447952.0, + "step": 6876 + }, + { + "epoch": 1.2770659238625812, + "grad_norm": 1.6028449535369873, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8756747841835022, + "num_tokens": 241476308.0, + "step": 6877 + }, + { + "epoch": 1.2772516248839367, + "grad_norm": 1.5554287433624268, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8771206140518188, + "num_tokens": 241507632.0, + "step": 6878 + }, + { + "epoch": 1.2774373259052925, + "grad_norm": 1.6230809688568115, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8642903566360474, + "num_tokens": 241541512.0, + "step": 6879 + }, + { + "epoch": 1.2776230269266482, + "grad_norm": 1.4886524677276611, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8772132396697998, + "num_tokens": 241577693.0, + "step": 6880 + }, + { + "epoch": 1.2778087279480037, + "grad_norm": 1.5870455503463745, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8843176960945129, + "num_tokens": 241608194.0, + "step": 6881 + }, + { + "epoch": 1.2779944289693592, + "grad_norm": 1.4701014757156372, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8647928833961487, + "num_tokens": 241645495.0, + "step": 6882 + }, + { + "epoch": 1.278180129990715, + "grad_norm": 1.4875001907348633, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8734062910079956, + "num_tokens": 241681900.0, + "step": 6883 + }, + { + "epoch": 1.2783658310120707, + "grad_norm": 1.4634920358657837, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.868808388710022, + "num_tokens": 241723142.0, + "step": 6884 + }, + { + "epoch": 1.2785515320334262, + "grad_norm": 1.479560375213623, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8650161027908325, + "num_tokens": 241763082.0, + "step": 6885 + }, + { + "epoch": 1.2787372330547817, + "grad_norm": 1.4327149391174316, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.877534806728363, + "num_tokens": 241800353.0, + "step": 6886 + }, + { + "epoch": 1.2789229340761374, + "grad_norm": 1.4478670358657837, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8795599937438965, + "num_tokens": 241837100.0, + "step": 6887 + }, + { + "epoch": 1.2791086350974932, + "grad_norm": 1.594594120979309, + "learning_rate": 1e-06, + "loss": 0.4197, + "mean_token_accuracy": 0.8568519949913025, + "num_tokens": 241872742.0, + "step": 6888 + }, + { + "epoch": 1.2792943361188487, + "grad_norm": 1.4591034650802612, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8720484972000122, + "num_tokens": 241914052.0, + "step": 6889 + }, + { + "epoch": 1.2794800371402042, + "grad_norm": 1.4771085977554321, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8770014643669128, + "num_tokens": 241953247.0, + "step": 6890 + }, + { + "epoch": 1.27966573816156, + "grad_norm": 1.735108733177185, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8796139359474182, + "num_tokens": 241978289.0, + "step": 6891 + }, + { + "epoch": 1.2798514391829154, + "grad_norm": 1.4539958238601685, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.892390251159668, + "num_tokens": 242007172.0, + "step": 6892 + }, + { + "epoch": 1.2800371402042712, + "grad_norm": 1.4185242652893066, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.882944643497467, + "num_tokens": 242042437.0, + "step": 6893 + }, + { + "epoch": 1.2802228412256267, + "grad_norm": 1.5832157135009766, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.887040913105011, + "num_tokens": 242070964.0, + "step": 6894 + }, + { + "epoch": 1.2804085422469824, + "grad_norm": 1.7076166868209839, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8666144609451294, + "num_tokens": 242099659.0, + "step": 6895 + }, + { + "epoch": 1.280594243268338, + "grad_norm": 1.4666907787322998, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8714379072189331, + "num_tokens": 242136020.0, + "step": 6896 + }, + { + "epoch": 1.2807799442896937, + "grad_norm": 1.6132090091705322, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8785766363143921, + "num_tokens": 242166749.0, + "step": 6897 + }, + { + "epoch": 1.2809656453110492, + "grad_norm": 1.5339597463607788, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8740500211715698, + "num_tokens": 242199366.0, + "step": 6898 + }, + { + "epoch": 1.281151346332405, + "grad_norm": 1.4748997688293457, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8660292625427246, + "num_tokens": 242241750.0, + "step": 6899 + }, + { + "epoch": 1.2813370473537604, + "grad_norm": 1.5541212558746338, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.859025776386261, + "num_tokens": 242279036.0, + "step": 6900 + }, + { + "epoch": 1.281522748375116, + "grad_norm": 1.3382624387741089, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.8681607246398926, + "num_tokens": 242326647.0, + "step": 6901 + }, + { + "epoch": 1.2817084493964717, + "grad_norm": 1.568185806274414, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8651937246322632, + "num_tokens": 242365003.0, + "step": 6902 + }, + { + "epoch": 1.2818941504178274, + "grad_norm": 1.4160031080245972, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.887647271156311, + "num_tokens": 242400976.0, + "step": 6903 + }, + { + "epoch": 1.282079851439183, + "grad_norm": 1.429945945739746, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8801434636116028, + "num_tokens": 242438407.0, + "step": 6904 + }, + { + "epoch": 1.2822655524605384, + "grad_norm": 1.5223896503448486, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8843175172805786, + "num_tokens": 242473310.0, + "step": 6905 + }, + { + "epoch": 1.2824512534818941, + "grad_norm": 1.5270395278930664, + "learning_rate": 1e-06, + "loss": 0.4051, + "mean_token_accuracy": 0.8644424676895142, + "num_tokens": 242508567.0, + "step": 6906 + }, + { + "epoch": 1.2826369545032499, + "grad_norm": 1.4088481664657593, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8813680410385132, + "num_tokens": 242547717.0, + "step": 6907 + }, + { + "epoch": 1.2828226555246054, + "grad_norm": 1.7574434280395508, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8800222873687744, + "num_tokens": 242570245.0, + "step": 6908 + }, + { + "epoch": 1.283008356545961, + "grad_norm": 1.5516624450683594, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8780235052108765, + "num_tokens": 242602065.0, + "step": 6909 + }, + { + "epoch": 1.2831940575673166, + "grad_norm": 1.6031179428100586, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.8660619258880615, + "num_tokens": 242636374.0, + "step": 6910 + }, + { + "epoch": 1.2833797585886724, + "grad_norm": 1.3954315185546875, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8647224307060242, + "num_tokens": 242676814.0, + "step": 6911 + }, + { + "epoch": 1.2835654596100279, + "grad_norm": 1.4615232944488525, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8862650394439697, + "num_tokens": 242708625.0, + "step": 6912 + }, + { + "epoch": 1.2837511606313834, + "grad_norm": 1.446783423423767, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8898097276687622, + "num_tokens": 242745309.0, + "step": 6913 + }, + { + "epoch": 1.2839368616527391, + "grad_norm": 1.5497876405715942, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8695653676986694, + "num_tokens": 242781468.0, + "step": 6914 + }, + { + "epoch": 1.2841225626740946, + "grad_norm": 1.6426024436950684, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8684285879135132, + "num_tokens": 242812991.0, + "step": 6915 + }, + { + "epoch": 1.2843082636954504, + "grad_norm": 1.5001896619796753, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8841408491134644, + "num_tokens": 242844911.0, + "step": 6916 + }, + { + "epoch": 1.2844939647168059, + "grad_norm": 1.4015939235687256, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.878767192363739, + "num_tokens": 242880884.0, + "step": 6917 + }, + { + "epoch": 1.2846796657381616, + "grad_norm": 1.4588513374328613, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8717591762542725, + "num_tokens": 242920200.0, + "step": 6918 + }, + { + "epoch": 1.2848653667595171, + "grad_norm": 1.2956770658493042, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.89054274559021, + "num_tokens": 242961434.0, + "step": 6919 + }, + { + "epoch": 1.2850510677808729, + "grad_norm": 1.6772481203079224, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8759716749191284, + "num_tokens": 242987820.0, + "step": 6920 + }, + { + "epoch": 1.2852367688022284, + "grad_norm": 1.4893405437469482, + "learning_rate": 1e-06, + "loss": 0.3878, + "mean_token_accuracy": 0.8692598938941956, + "num_tokens": 243020999.0, + "step": 6921 + }, + { + "epoch": 1.285422469823584, + "grad_norm": 1.4473410844802856, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.886944055557251, + "num_tokens": 243054807.0, + "step": 6922 + }, + { + "epoch": 1.2856081708449396, + "grad_norm": 1.5496790409088135, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8707383275032043, + "num_tokens": 243085807.0, + "step": 6923 + }, + { + "epoch": 1.2857938718662953, + "grad_norm": 1.5233445167541504, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8575493097305298, + "num_tokens": 243122428.0, + "step": 6924 + }, + { + "epoch": 1.2859795728876509, + "grad_norm": 1.5177419185638428, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8742021322250366, + "num_tokens": 243155256.0, + "step": 6925 + }, + { + "epoch": 1.2861652739090066, + "grad_norm": 1.478198766708374, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8849437832832336, + "num_tokens": 243189278.0, + "step": 6926 + }, + { + "epoch": 1.286350974930362, + "grad_norm": 1.4863289594650269, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8692249059677124, + "num_tokens": 243224802.0, + "step": 6927 + }, + { + "epoch": 1.2865366759517176, + "grad_norm": 1.3899728059768677, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8773283958435059, + "num_tokens": 243262106.0, + "step": 6928 + }, + { + "epoch": 1.2867223769730733, + "grad_norm": 1.4931236505508423, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8770269155502319, + "num_tokens": 243296599.0, + "step": 6929 + }, + { + "epoch": 1.286908077994429, + "grad_norm": 1.489561915397644, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8766312599182129, + "num_tokens": 243329836.0, + "step": 6930 + }, + { + "epoch": 1.2870937790157846, + "grad_norm": 1.4380711317062378, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8870102167129517, + "num_tokens": 243364376.0, + "step": 6931 + }, + { + "epoch": 1.28727948003714, + "grad_norm": 1.3551380634307861, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8857531547546387, + "num_tokens": 243406353.0, + "step": 6932 + }, + { + "epoch": 1.2874651810584958, + "grad_norm": 1.49611496925354, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.8763782382011414, + "num_tokens": 243442521.0, + "step": 6933 + }, + { + "epoch": 1.2876508820798516, + "grad_norm": 1.469119668006897, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.870586633682251, + "num_tokens": 243478845.0, + "step": 6934 + }, + { + "epoch": 1.287836583101207, + "grad_norm": 1.4161229133605957, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.889499843120575, + "num_tokens": 243512466.0, + "step": 6935 + }, + { + "epoch": 1.2880222841225626, + "grad_norm": 1.6333799362182617, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8693356513977051, + "num_tokens": 243545084.0, + "step": 6936 + }, + { + "epoch": 1.2882079851439183, + "grad_norm": 1.4404206275939941, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8788204789161682, + "num_tokens": 243582472.0, + "step": 6937 + }, + { + "epoch": 1.2883936861652738, + "grad_norm": 1.819332242012024, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8745135068893433, + "num_tokens": 243613010.0, + "step": 6938 + }, + { + "epoch": 1.2885793871866296, + "grad_norm": 1.5759316682815552, + "learning_rate": 1e-06, + "loss": 0.4059, + "mean_token_accuracy": 0.8638285398483276, + "num_tokens": 243647028.0, + "step": 6939 + }, + { + "epoch": 1.288765088207985, + "grad_norm": 1.4769786596298218, + "learning_rate": 1e-06, + "loss": 0.4263, + "mean_token_accuracy": 0.8562523126602173, + "num_tokens": 243688727.0, + "step": 6940 + }, + { + "epoch": 1.2889507892293408, + "grad_norm": 1.4457296133041382, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8723385334014893, + "num_tokens": 243726054.0, + "step": 6941 + }, + { + "epoch": 1.2891364902506963, + "grad_norm": 1.4873312711715698, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8828151226043701, + "num_tokens": 243762858.0, + "step": 6942 + }, + { + "epoch": 1.289322191272052, + "grad_norm": 1.4349238872528076, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8742227554321289, + "num_tokens": 243800791.0, + "step": 6943 + }, + { + "epoch": 1.2895078922934076, + "grad_norm": 1.6346790790557861, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8778685331344604, + "num_tokens": 243834566.0, + "step": 6944 + }, + { + "epoch": 1.2896935933147633, + "grad_norm": 1.5773698091506958, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8843634128570557, + "num_tokens": 243865886.0, + "step": 6945 + }, + { + "epoch": 1.2898792943361188, + "grad_norm": 1.6529226303100586, + "learning_rate": 1e-06, + "loss": 0.4374, + "mean_token_accuracy": 0.8524757027626038, + "num_tokens": 243897265.0, + "step": 6946 + }, + { + "epoch": 1.2900649953574745, + "grad_norm": 1.43312406539917, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8664696216583252, + "num_tokens": 243939099.0, + "step": 6947 + }, + { + "epoch": 1.29025069637883, + "grad_norm": 1.549261212348938, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8851809501647949, + "num_tokens": 243969480.0, + "step": 6948 + }, + { + "epoch": 1.2904363974001858, + "grad_norm": 1.4414478540420532, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8913542032241821, + "num_tokens": 244004110.0, + "step": 6949 + }, + { + "epoch": 1.2906220984215413, + "grad_norm": 1.5140968561172485, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8817852735519409, + "num_tokens": 244034182.0, + "step": 6950 + }, + { + "epoch": 1.2908077994428968, + "grad_norm": 1.4676132202148438, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8726538419723511, + "num_tokens": 244071129.0, + "step": 6951 + }, + { + "epoch": 1.2909935004642525, + "grad_norm": 1.438625693321228, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8744082450866699, + "num_tokens": 244109682.0, + "step": 6952 + }, + { + "epoch": 1.2911792014856083, + "grad_norm": 1.4091120958328247, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8717231750488281, + "num_tokens": 244149053.0, + "step": 6953 + }, + { + "epoch": 1.2913649025069638, + "grad_norm": 1.6013413667678833, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8817805051803589, + "num_tokens": 244178073.0, + "step": 6954 + }, + { + "epoch": 1.2915506035283193, + "grad_norm": 1.4789745807647705, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8885607123374939, + "num_tokens": 244213693.0, + "step": 6955 + }, + { + "epoch": 1.291736304549675, + "grad_norm": 1.436160683631897, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8773053884506226, + "num_tokens": 244249552.0, + "step": 6956 + }, + { + "epoch": 1.2919220055710308, + "grad_norm": 1.5764098167419434, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8731738924980164, + "num_tokens": 244281246.0, + "step": 6957 + }, + { + "epoch": 1.2921077065923863, + "grad_norm": 1.4432885646820068, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.866326093673706, + "num_tokens": 244321810.0, + "step": 6958 + }, + { + "epoch": 1.2922934076137418, + "grad_norm": 1.5675345659255981, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8737564086914062, + "num_tokens": 244351399.0, + "step": 6959 + }, + { + "epoch": 1.2924791086350975, + "grad_norm": 1.5688575506210327, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8796883821487427, + "num_tokens": 244383037.0, + "step": 6960 + }, + { + "epoch": 1.292664809656453, + "grad_norm": 1.5562502145767212, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8841668367385864, + "num_tokens": 244415308.0, + "step": 6961 + }, + { + "epoch": 1.2928505106778088, + "grad_norm": 1.4417595863342285, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8710833787918091, + "num_tokens": 244456170.0, + "step": 6962 + }, + { + "epoch": 1.2930362116991643, + "grad_norm": 1.4760633707046509, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8895409107208252, + "num_tokens": 244489491.0, + "step": 6963 + }, + { + "epoch": 1.29322191272052, + "grad_norm": 1.4936647415161133, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.878221333026886, + "num_tokens": 244523326.0, + "step": 6964 + }, + { + "epoch": 1.2934076137418755, + "grad_norm": 1.4919179677963257, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8719339370727539, + "num_tokens": 244561142.0, + "step": 6965 + }, + { + "epoch": 1.2935933147632313, + "grad_norm": 1.7208212614059448, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8640881180763245, + "num_tokens": 244589791.0, + "step": 6966 + }, + { + "epoch": 1.2937790157845868, + "grad_norm": 1.441597580909729, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8780016899108887, + "num_tokens": 244625690.0, + "step": 6967 + }, + { + "epoch": 1.2939647168059425, + "grad_norm": 1.4543967247009277, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8645853996276855, + "num_tokens": 244662083.0, + "step": 6968 + }, + { + "epoch": 1.294150417827298, + "grad_norm": 1.328259825706482, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8881398439407349, + "num_tokens": 244696088.0, + "step": 6969 + }, + { + "epoch": 1.2943361188486537, + "grad_norm": 1.377204179763794, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8739277720451355, + "num_tokens": 244735670.0, + "step": 6970 + }, + { + "epoch": 1.2945218198700092, + "grad_norm": 1.469720721244812, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8922306299209595, + "num_tokens": 244769482.0, + "step": 6971 + }, + { + "epoch": 1.294707520891365, + "grad_norm": 1.4837687015533447, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8743407726287842, + "num_tokens": 244804429.0, + "step": 6972 + }, + { + "epoch": 1.2948932219127205, + "grad_norm": 1.5698049068450928, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8851974010467529, + "num_tokens": 244833551.0, + "step": 6973 + }, + { + "epoch": 1.295078922934076, + "grad_norm": 1.7068895101547241, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8739420175552368, + "num_tokens": 244860251.0, + "step": 6974 + }, + { + "epoch": 1.2952646239554317, + "grad_norm": 1.420257568359375, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8823565244674683, + "num_tokens": 244896264.0, + "step": 6975 + }, + { + "epoch": 1.2954503249767875, + "grad_norm": 1.4775933027267456, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8776731491088867, + "num_tokens": 244931516.0, + "step": 6976 + }, + { + "epoch": 1.295636025998143, + "grad_norm": 1.573033094406128, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8791012763977051, + "num_tokens": 244965138.0, + "step": 6977 + }, + { + "epoch": 1.2958217270194985, + "grad_norm": 1.5228561162948608, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8634223341941833, + "num_tokens": 245000911.0, + "step": 6978 + }, + { + "epoch": 1.2960074280408542, + "grad_norm": 1.6900228261947632, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8852312564849854, + "num_tokens": 245027509.0, + "step": 6979 + }, + { + "epoch": 1.29619312906221, + "grad_norm": 1.4991761445999146, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8720226883888245, + "num_tokens": 245060445.0, + "step": 6980 + }, + { + "epoch": 1.2963788300835655, + "grad_norm": 1.6716138124465942, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.873387336730957, + "num_tokens": 245091088.0, + "step": 6981 + }, + { + "epoch": 1.296564531104921, + "grad_norm": 1.3537076711654663, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8966186046600342, + "num_tokens": 245125949.0, + "step": 6982 + }, + { + "epoch": 1.2967502321262767, + "grad_norm": 1.5373293161392212, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.877360463142395, + "num_tokens": 245163355.0, + "step": 6983 + }, + { + "epoch": 1.2969359331476324, + "grad_norm": 1.6400477886199951, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8830859661102295, + "num_tokens": 245190492.0, + "step": 6984 + }, + { + "epoch": 1.297121634168988, + "grad_norm": 1.4884754419326782, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.870293915271759, + "num_tokens": 245229929.0, + "step": 6985 + }, + { + "epoch": 1.2973073351903435, + "grad_norm": 1.558194637298584, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8711562156677246, + "num_tokens": 245262309.0, + "step": 6986 + }, + { + "epoch": 1.2974930362116992, + "grad_norm": 1.5062031745910645, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8661973476409912, + "num_tokens": 245299963.0, + "step": 6987 + }, + { + "epoch": 1.2976787372330547, + "grad_norm": 1.4703460931777954, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8750114440917969, + "num_tokens": 245332798.0, + "step": 6988 + }, + { + "epoch": 1.2978644382544104, + "grad_norm": 1.4998699426651, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8797887563705444, + "num_tokens": 245368374.0, + "step": 6989 + }, + { + "epoch": 1.298050139275766, + "grad_norm": 1.317431092262268, + "learning_rate": 1e-06, + "loss": 0.2869, + "mean_token_accuracy": 0.9000287055969238, + "num_tokens": 245405173.0, + "step": 6990 + }, + { + "epoch": 1.2982358402971217, + "grad_norm": 1.7710450887680054, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8809779286384583, + "num_tokens": 245431810.0, + "step": 6991 + }, + { + "epoch": 1.2984215413184772, + "grad_norm": 1.5443437099456787, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8828758001327515, + "num_tokens": 245463453.0, + "step": 6992 + }, + { + "epoch": 1.298607242339833, + "grad_norm": 1.6441264152526855, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8826616406440735, + "num_tokens": 245493144.0, + "step": 6993 + }, + { + "epoch": 1.2987929433611884, + "grad_norm": 1.4176851511001587, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8886116743087769, + "num_tokens": 245531444.0, + "step": 6994 + }, + { + "epoch": 1.2989786443825442, + "grad_norm": 1.4542515277862549, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8817439079284668, + "num_tokens": 245568607.0, + "step": 6995 + }, + { + "epoch": 1.2991643454038997, + "grad_norm": 1.5113871097564697, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.866478443145752, + "num_tokens": 245605407.0, + "step": 6996 + }, + { + "epoch": 1.2993500464252552, + "grad_norm": 1.4529476165771484, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8700599670410156, + "num_tokens": 245644242.0, + "step": 6997 + }, + { + "epoch": 1.299535747446611, + "grad_norm": 1.6183112859725952, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.869763970375061, + "num_tokens": 245673685.0, + "step": 6998 + }, + { + "epoch": 1.2997214484679667, + "grad_norm": 1.560955286026001, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8837909698486328, + "num_tokens": 245707047.0, + "step": 6999 + }, + { + "epoch": 1.2999071494893222, + "grad_norm": 1.3226557970046997, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8726080656051636, + "num_tokens": 245752152.0, + "step": 7000 + }, + { + "epoch": 1.3000928505106777, + "grad_norm": 1.5808371305465698, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8784840703010559, + "num_tokens": 245787175.0, + "step": 7001 + }, + { + "epoch": 1.3002785515320334, + "grad_norm": 1.3917392492294312, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8890395164489746, + "num_tokens": 245823646.0, + "step": 7002 + }, + { + "epoch": 1.3004642525533892, + "grad_norm": 1.4367964267730713, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8765418529510498, + "num_tokens": 245862632.0, + "step": 7003 + }, + { + "epoch": 1.3006499535747447, + "grad_norm": 1.3873381614685059, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.877792477607727, + "num_tokens": 245902298.0, + "step": 7004 + }, + { + "epoch": 1.3008356545961002, + "grad_norm": 1.400968313217163, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.86758953332901, + "num_tokens": 245943808.0, + "step": 7005 + }, + { + "epoch": 1.301021355617456, + "grad_norm": 1.6430108547210693, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8754657506942749, + "num_tokens": 245974372.0, + "step": 7006 + }, + { + "epoch": 1.3012070566388116, + "grad_norm": 1.4747165441513062, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8825870752334595, + "num_tokens": 246008706.0, + "step": 7007 + }, + { + "epoch": 1.3013927576601672, + "grad_norm": 1.4852166175842285, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8770251870155334, + "num_tokens": 246045546.0, + "step": 7008 + }, + { + "epoch": 1.3015784586815227, + "grad_norm": 1.4515405893325806, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.880867063999176, + "num_tokens": 246080841.0, + "step": 7009 + }, + { + "epoch": 1.3017641597028784, + "grad_norm": 1.578676462173462, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8765304684638977, + "num_tokens": 246111303.0, + "step": 7010 + }, + { + "epoch": 1.301949860724234, + "grad_norm": 1.5171759128570557, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.868968665599823, + "num_tokens": 246146522.0, + "step": 7011 + }, + { + "epoch": 1.3021355617455896, + "grad_norm": 1.3410316705703735, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8795332908630371, + "num_tokens": 246188115.0, + "step": 7012 + }, + { + "epoch": 1.3023212627669452, + "grad_norm": 1.4318737983703613, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8855957984924316, + "num_tokens": 246222958.0, + "step": 7013 + }, + { + "epoch": 1.3025069637883009, + "grad_norm": 1.6422568559646606, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8740350008010864, + "num_tokens": 246253857.0, + "step": 7014 + }, + { + "epoch": 1.3026926648096564, + "grad_norm": 1.5944815874099731, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8747326731681824, + "num_tokens": 246287728.0, + "step": 7015 + }, + { + "epoch": 1.3028783658310121, + "grad_norm": 1.5961823463439941, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8723477125167847, + "num_tokens": 246321112.0, + "step": 7016 + }, + { + "epoch": 1.3030640668523676, + "grad_norm": 1.4621528387069702, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8850520849227905, + "num_tokens": 246355513.0, + "step": 7017 + }, + { + "epoch": 1.3032497678737234, + "grad_norm": 1.6280611753463745, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8752124309539795, + "num_tokens": 246386645.0, + "step": 7018 + }, + { + "epoch": 1.3034354688950789, + "grad_norm": 1.3849672079086304, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.880974292755127, + "num_tokens": 246424204.0, + "step": 7019 + }, + { + "epoch": 1.3036211699164346, + "grad_norm": 1.5763821601867676, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8656254410743713, + "num_tokens": 246460310.0, + "step": 7020 + }, + { + "epoch": 1.3038068709377901, + "grad_norm": 1.4866560697555542, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8778420686721802, + "num_tokens": 246496925.0, + "step": 7021 + }, + { + "epoch": 1.3039925719591459, + "grad_norm": 1.534196138381958, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8881609439849854, + "num_tokens": 246531227.0, + "step": 7022 + }, + { + "epoch": 1.3041782729805014, + "grad_norm": 1.4023360013961792, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8849018812179565, + "num_tokens": 246568963.0, + "step": 7023 + }, + { + "epoch": 1.3043639740018569, + "grad_norm": 1.4604365825653076, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8808297514915466, + "num_tokens": 246603076.0, + "step": 7024 + }, + { + "epoch": 1.3045496750232126, + "grad_norm": 1.4826300144195557, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8771728873252869, + "num_tokens": 246641746.0, + "step": 7025 + }, + { + "epoch": 1.3047353760445684, + "grad_norm": 1.4962939023971558, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8721882700920105, + "num_tokens": 246675657.0, + "step": 7026 + }, + { + "epoch": 1.3049210770659239, + "grad_norm": 1.5340789556503296, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8754439949989319, + "num_tokens": 246706433.0, + "step": 7027 + }, + { + "epoch": 1.3051067780872794, + "grad_norm": 1.6249750852584839, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.899512529373169, + "num_tokens": 246736684.0, + "step": 7028 + }, + { + "epoch": 1.305292479108635, + "grad_norm": 1.5391513109207153, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8611441850662231, + "num_tokens": 246774375.0, + "step": 7029 + }, + { + "epoch": 1.3054781801299908, + "grad_norm": 1.5288397073745728, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8709355592727661, + "num_tokens": 246813499.0, + "step": 7030 + }, + { + "epoch": 1.3056638811513464, + "grad_norm": 1.451535940170288, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8759920001029968, + "num_tokens": 246848835.0, + "step": 7031 + }, + { + "epoch": 1.3058495821727019, + "grad_norm": 1.5509159564971924, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8699368238449097, + "num_tokens": 246881787.0, + "step": 7032 + }, + { + "epoch": 1.3060352831940576, + "grad_norm": 1.5372393131256104, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8710556626319885, + "num_tokens": 246913043.0, + "step": 7033 + }, + { + "epoch": 1.306220984215413, + "grad_norm": 1.3895992040634155, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.902320384979248, + "num_tokens": 246947267.0, + "step": 7034 + }, + { + "epoch": 1.3064066852367688, + "grad_norm": 1.498801350593567, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.881745457649231, + "num_tokens": 246979601.0, + "step": 7035 + }, + { + "epoch": 1.3065923862581243, + "grad_norm": 1.557427167892456, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8758726119995117, + "num_tokens": 247014510.0, + "step": 7036 + }, + { + "epoch": 1.30677808727948, + "grad_norm": 1.6592907905578613, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8780888319015503, + "num_tokens": 247043678.0, + "step": 7037 + }, + { + "epoch": 1.3069637883008356, + "grad_norm": 1.5902659893035889, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8795597553253174, + "num_tokens": 247073509.0, + "step": 7038 + }, + { + "epoch": 1.3071494893221913, + "grad_norm": 1.6000046730041504, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8854620456695557, + "num_tokens": 247103155.0, + "step": 7039 + }, + { + "epoch": 1.3073351903435468, + "grad_norm": 1.6035211086273193, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8714237213134766, + "num_tokens": 247132421.0, + "step": 7040 + }, + { + "epoch": 1.3075208913649026, + "grad_norm": 1.6474932432174683, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8750536441802979, + "num_tokens": 247160341.0, + "step": 7041 + }, + { + "epoch": 1.307706592386258, + "grad_norm": 1.4074244499206543, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8603197336196899, + "num_tokens": 247203365.0, + "step": 7042 + }, + { + "epoch": 1.3078922934076138, + "grad_norm": 1.5867453813552856, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8846228122711182, + "num_tokens": 247231544.0, + "step": 7043 + }, + { + "epoch": 1.3080779944289693, + "grad_norm": 1.516640305519104, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8669363260269165, + "num_tokens": 247267166.0, + "step": 7044 + }, + { + "epoch": 1.308263695450325, + "grad_norm": 1.4211187362670898, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8811995983123779, + "num_tokens": 247303241.0, + "step": 7045 + }, + { + "epoch": 1.3084493964716806, + "grad_norm": 1.4648059606552124, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8797142505645752, + "num_tokens": 247338631.0, + "step": 7046 + }, + { + "epoch": 1.308635097493036, + "grad_norm": 1.405646562576294, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8851789236068726, + "num_tokens": 247373308.0, + "step": 7047 + }, + { + "epoch": 1.3088207985143918, + "grad_norm": 1.550587773323059, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8709529638290405, + "num_tokens": 247410384.0, + "step": 7048 + }, + { + "epoch": 1.3090064995357475, + "grad_norm": 1.4901223182678223, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8823084831237793, + "num_tokens": 247447989.0, + "step": 7049 + }, + { + "epoch": 1.309192200557103, + "grad_norm": 1.4582895040512085, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8707968592643738, + "num_tokens": 247486264.0, + "step": 7050 + }, + { + "epoch": 1.3093779015784586, + "grad_norm": 1.7076042890548706, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8792402148246765, + "num_tokens": 247511388.0, + "step": 7051 + }, + { + "epoch": 1.3095636025998143, + "grad_norm": 1.4966943264007568, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8854838609695435, + "num_tokens": 247546217.0, + "step": 7052 + }, + { + "epoch": 1.30974930362117, + "grad_norm": 1.6082215309143066, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.8603736758232117, + "num_tokens": 247578183.0, + "step": 7053 + }, + { + "epoch": 1.3099350046425255, + "grad_norm": 1.4539493322372437, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8788429498672485, + "num_tokens": 247613885.0, + "step": 7054 + }, + { + "epoch": 1.310120705663881, + "grad_norm": 1.564658284187317, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8691731691360474, + "num_tokens": 247647365.0, + "step": 7055 + }, + { + "epoch": 1.3103064066852368, + "grad_norm": 1.5016268491744995, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8744001388549805, + "num_tokens": 247682528.0, + "step": 7056 + }, + { + "epoch": 1.3104921077065925, + "grad_norm": 1.5367798805236816, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.872527003288269, + "num_tokens": 247716221.0, + "step": 7057 + }, + { + "epoch": 1.310677808727948, + "grad_norm": 1.4999679327011108, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8747579455375671, + "num_tokens": 247750089.0, + "step": 7058 + }, + { + "epoch": 1.3108635097493035, + "grad_norm": 1.4825044870376587, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8758288621902466, + "num_tokens": 247789505.0, + "step": 7059 + }, + { + "epoch": 1.3110492107706593, + "grad_norm": 1.4405670166015625, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8784586191177368, + "num_tokens": 247826195.0, + "step": 7060 + }, + { + "epoch": 1.3112349117920148, + "grad_norm": 1.5273879766464233, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8766521215438843, + "num_tokens": 247858117.0, + "step": 7061 + }, + { + "epoch": 1.3114206128133705, + "grad_norm": 1.4019502401351929, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8735818862915039, + "num_tokens": 247900756.0, + "step": 7062 + }, + { + "epoch": 1.311606313834726, + "grad_norm": 1.4276704788208008, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8801178932189941, + "num_tokens": 247940403.0, + "step": 7063 + }, + { + "epoch": 1.3117920148560818, + "grad_norm": 1.6127506494522095, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.864501953125, + "num_tokens": 247971368.0, + "step": 7064 + }, + { + "epoch": 1.3119777158774373, + "grad_norm": 1.3954558372497559, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8852699995040894, + "num_tokens": 248009245.0, + "step": 7065 + }, + { + "epoch": 1.312163416898793, + "grad_norm": 1.5213762521743774, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8822686672210693, + "num_tokens": 248045336.0, + "step": 7066 + }, + { + "epoch": 1.3123491179201485, + "grad_norm": 1.4630740880966187, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8832085132598877, + "num_tokens": 248079039.0, + "step": 7067 + }, + { + "epoch": 1.3125348189415043, + "grad_norm": 1.548782229423523, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8727381229400635, + "num_tokens": 248113928.0, + "step": 7068 + }, + { + "epoch": 1.3127205199628598, + "grad_norm": 1.5376261472702026, + "learning_rate": 1e-06, + "loss": 0.4146, + "mean_token_accuracy": 0.8574991226196289, + "num_tokens": 248155710.0, + "step": 7069 + }, + { + "epoch": 1.3129062209842153, + "grad_norm": 1.4588062763214111, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8739187717437744, + "num_tokens": 248193631.0, + "step": 7070 + }, + { + "epoch": 1.313091922005571, + "grad_norm": 1.406530499458313, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8841061592102051, + "num_tokens": 248229790.0, + "step": 7071 + }, + { + "epoch": 1.3132776230269267, + "grad_norm": 1.5397679805755615, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8710473775863647, + "num_tokens": 248269978.0, + "step": 7072 + }, + { + "epoch": 1.3134633240482823, + "grad_norm": 1.5392107963562012, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8810352087020874, + "num_tokens": 248303497.0, + "step": 7073 + }, + { + "epoch": 1.3136490250696378, + "grad_norm": 1.337077260017395, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8890560865402222, + "num_tokens": 248346284.0, + "step": 7074 + }, + { + "epoch": 1.3138347260909935, + "grad_norm": 1.514177918434143, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8793034553527832, + "num_tokens": 248378714.0, + "step": 7075 + }, + { + "epoch": 1.3140204271123492, + "grad_norm": 1.4161453247070312, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8739789724349976, + "num_tokens": 248417816.0, + "step": 7076 + }, + { + "epoch": 1.3142061281337047, + "grad_norm": 1.3419806957244873, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8815800547599792, + "num_tokens": 248459259.0, + "step": 7077 + }, + { + "epoch": 1.3143918291550603, + "grad_norm": 1.5268993377685547, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8656117916107178, + "num_tokens": 248494370.0, + "step": 7078 + }, + { + "epoch": 1.314577530176416, + "grad_norm": 1.5380717515945435, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8769060373306274, + "num_tokens": 248529076.0, + "step": 7079 + }, + { + "epoch": 1.3147632311977717, + "grad_norm": 1.5515621900558472, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8667154312133789, + "num_tokens": 248560839.0, + "step": 7080 + }, + { + "epoch": 1.3149489322191272, + "grad_norm": 1.3386220932006836, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8924024105072021, + "num_tokens": 248600448.0, + "step": 7081 + }, + { + "epoch": 1.3151346332404827, + "grad_norm": 1.6366853713989258, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.8683696985244751, + "num_tokens": 248631921.0, + "step": 7082 + }, + { + "epoch": 1.3153203342618385, + "grad_norm": 1.5862812995910645, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8788639307022095, + "num_tokens": 248666648.0, + "step": 7083 + }, + { + "epoch": 1.315506035283194, + "grad_norm": 1.517806887626648, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8776792883872986, + "num_tokens": 248700053.0, + "step": 7084 + }, + { + "epoch": 1.3156917363045497, + "grad_norm": 1.5146610736846924, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8897490501403809, + "num_tokens": 248729552.0, + "step": 7085 + }, + { + "epoch": 1.3158774373259052, + "grad_norm": 1.507302165031433, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8694149255752563, + "num_tokens": 248766245.0, + "step": 7086 + }, + { + "epoch": 1.316063138347261, + "grad_norm": 1.339963436126709, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.897848904132843, + "num_tokens": 248804000.0, + "step": 7087 + }, + { + "epoch": 1.3162488393686165, + "grad_norm": 1.4488757848739624, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.888573944568634, + "num_tokens": 248839920.0, + "step": 7088 + }, + { + "epoch": 1.3164345403899722, + "grad_norm": 1.4143489599227905, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8661089539527893, + "num_tokens": 248885389.0, + "step": 7089 + }, + { + "epoch": 1.3166202414113277, + "grad_norm": 1.4768036603927612, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8938288688659668, + "num_tokens": 248918091.0, + "step": 7090 + }, + { + "epoch": 1.3168059424326835, + "grad_norm": 1.4708656072616577, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8869198560714722, + "num_tokens": 248952079.0, + "step": 7091 + }, + { + "epoch": 1.316991643454039, + "grad_norm": 1.618085265159607, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8753499984741211, + "num_tokens": 248982967.0, + "step": 7092 + }, + { + "epoch": 1.3171773444753947, + "grad_norm": 1.4350719451904297, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8786450624465942, + "num_tokens": 249018995.0, + "step": 7093 + }, + { + "epoch": 1.3173630454967502, + "grad_norm": 1.4379003047943115, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8808987140655518, + "num_tokens": 249053361.0, + "step": 7094 + }, + { + "epoch": 1.317548746518106, + "grad_norm": 1.3531826734542847, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8875032067298889, + "num_tokens": 249091305.0, + "step": 7095 + }, + { + "epoch": 1.3177344475394615, + "grad_norm": 1.603652834892273, + "learning_rate": 1e-06, + "loss": 0.3962, + "mean_token_accuracy": 0.8647458553314209, + "num_tokens": 249123776.0, + "step": 7096 + }, + { + "epoch": 1.317920148560817, + "grad_norm": 1.5492266416549683, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8762804865837097, + "num_tokens": 249155988.0, + "step": 7097 + }, + { + "epoch": 1.3181058495821727, + "grad_norm": 1.5620803833007812, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8741081357002258, + "num_tokens": 249187542.0, + "step": 7098 + }, + { + "epoch": 1.3182915506035284, + "grad_norm": 1.37507963180542, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8816023468971252, + "num_tokens": 249226439.0, + "step": 7099 + }, + { + "epoch": 1.318477251624884, + "grad_norm": 1.4363585710525513, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8768614530563354, + "num_tokens": 249259246.0, + "step": 7100 + }, + { + "epoch": 1.3186629526462395, + "grad_norm": 1.4247760772705078, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8737061023712158, + "num_tokens": 249297081.0, + "step": 7101 + }, + { + "epoch": 1.3188486536675952, + "grad_norm": 1.3991352319717407, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8801360726356506, + "num_tokens": 249336832.0, + "step": 7102 + }, + { + "epoch": 1.319034354688951, + "grad_norm": 1.5685359239578247, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8674957752227783, + "num_tokens": 249372384.0, + "step": 7103 + }, + { + "epoch": 1.3192200557103064, + "grad_norm": 1.4812020063400269, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8764340877532959, + "num_tokens": 249408390.0, + "step": 7104 + }, + { + "epoch": 1.319405756731662, + "grad_norm": 1.5284926891326904, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8800311088562012, + "num_tokens": 249437825.0, + "step": 7105 + }, + { + "epoch": 1.3195914577530177, + "grad_norm": 1.5150628089904785, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8737733364105225, + "num_tokens": 249471159.0, + "step": 7106 + }, + { + "epoch": 1.3197771587743732, + "grad_norm": 1.4690290689468384, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8789321184158325, + "num_tokens": 249508734.0, + "step": 7107 + }, + { + "epoch": 1.319962859795729, + "grad_norm": 1.4460973739624023, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8924057483673096, + "num_tokens": 249542144.0, + "step": 7108 + }, + { + "epoch": 1.3201485608170844, + "grad_norm": 1.50566828250885, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8688868880271912, + "num_tokens": 249580264.0, + "step": 7109 + }, + { + "epoch": 1.3203342618384402, + "grad_norm": 1.4708690643310547, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8699268102645874, + "num_tokens": 249618242.0, + "step": 7110 + }, + { + "epoch": 1.3205199628597957, + "grad_norm": 1.4339542388916016, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8835399746894836, + "num_tokens": 249652909.0, + "step": 7111 + }, + { + "epoch": 1.3207056638811514, + "grad_norm": 1.616236686706543, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8767937421798706, + "num_tokens": 249682608.0, + "step": 7112 + }, + { + "epoch": 1.320891364902507, + "grad_norm": 1.472259759902954, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8831984996795654, + "num_tokens": 249719055.0, + "step": 7113 + }, + { + "epoch": 1.3210770659238626, + "grad_norm": 1.4321177005767822, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8776322603225708, + "num_tokens": 249757586.0, + "step": 7114 + }, + { + "epoch": 1.3212627669452182, + "grad_norm": 1.4926068782806396, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8683999180793762, + "num_tokens": 249796063.0, + "step": 7115 + }, + { + "epoch": 1.321448467966574, + "grad_norm": 1.4346656799316406, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8946966528892517, + "num_tokens": 249827774.0, + "step": 7116 + }, + { + "epoch": 1.3216341689879294, + "grad_norm": 1.328622817993164, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8798894882202148, + "num_tokens": 249868970.0, + "step": 7117 + }, + { + "epoch": 1.3218198700092851, + "grad_norm": 1.3916268348693848, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8719850778579712, + "num_tokens": 249911835.0, + "step": 7118 + }, + { + "epoch": 1.3220055710306406, + "grad_norm": 1.4411500692367554, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8860746026039124, + "num_tokens": 249945532.0, + "step": 7119 + }, + { + "epoch": 1.3221912720519962, + "grad_norm": 1.3856431245803833, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8655853271484375, + "num_tokens": 249990705.0, + "step": 7120 + }, + { + "epoch": 1.322376973073352, + "grad_norm": 1.4902797937393188, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8768301606178284, + "num_tokens": 250027469.0, + "step": 7121 + }, + { + "epoch": 1.3225626740947076, + "grad_norm": 1.5644892454147339, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8712522983551025, + "num_tokens": 250060842.0, + "step": 7122 + }, + { + "epoch": 1.3227483751160631, + "grad_norm": 1.597066044807434, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8763554096221924, + "num_tokens": 250091834.0, + "step": 7123 + }, + { + "epoch": 1.3229340761374186, + "grad_norm": 1.5000050067901611, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8833971619606018, + "num_tokens": 250124494.0, + "step": 7124 + }, + { + "epoch": 1.3231197771587744, + "grad_norm": 1.4225958585739136, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8770723342895508, + "num_tokens": 250161284.0, + "step": 7125 + }, + { + "epoch": 1.3233054781801301, + "grad_norm": 1.342692255973816, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8800886869430542, + "num_tokens": 250203755.0, + "step": 7126 + }, + { + "epoch": 1.3234911792014856, + "grad_norm": 1.5326542854309082, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8728355765342712, + "num_tokens": 250237412.0, + "step": 7127 + }, + { + "epoch": 1.3236768802228411, + "grad_norm": 1.6353423595428467, + "learning_rate": 1e-06, + "loss": 0.4264, + "mean_token_accuracy": 0.8613842725753784, + "num_tokens": 250270857.0, + "step": 7128 + }, + { + "epoch": 1.3238625812441969, + "grad_norm": 1.4678031206130981, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8753933906555176, + "num_tokens": 250308034.0, + "step": 7129 + }, + { + "epoch": 1.3240482822655524, + "grad_norm": 1.703831672668457, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.870113730430603, + "num_tokens": 250335272.0, + "step": 7130 + }, + { + "epoch": 1.3242339832869081, + "grad_norm": 1.438347578048706, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8736056089401245, + "num_tokens": 250370586.0, + "step": 7131 + }, + { + "epoch": 1.3244196843082636, + "grad_norm": 1.560156226158142, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8811889886856079, + "num_tokens": 250403799.0, + "step": 7132 + }, + { + "epoch": 1.3246053853296194, + "grad_norm": 1.452521800994873, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8835310339927673, + "num_tokens": 250436683.0, + "step": 7133 + }, + { + "epoch": 1.3247910863509749, + "grad_norm": 1.4414112567901611, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.86609947681427, + "num_tokens": 250480616.0, + "step": 7134 + }, + { + "epoch": 1.3249767873723306, + "grad_norm": 1.4261784553527832, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8863841891288757, + "num_tokens": 250515620.0, + "step": 7135 + }, + { + "epoch": 1.325162488393686, + "grad_norm": 1.484983205795288, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8680694103240967, + "num_tokens": 250553963.0, + "step": 7136 + }, + { + "epoch": 1.3253481894150418, + "grad_norm": 1.5622352361679077, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8690639734268188, + "num_tokens": 250589190.0, + "step": 7137 + }, + { + "epoch": 1.3255338904363974, + "grad_norm": 1.6494783163070679, + "learning_rate": 1e-06, + "loss": 0.4443, + "mean_token_accuracy": 0.853345513343811, + "num_tokens": 250624506.0, + "step": 7138 + }, + { + "epoch": 1.325719591457753, + "grad_norm": 1.4531840085983276, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8809884786605835, + "num_tokens": 250663047.0, + "step": 7139 + }, + { + "epoch": 1.3259052924791086, + "grad_norm": 1.309667944908142, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8729551434516907, + "num_tokens": 250708837.0, + "step": 7140 + }, + { + "epoch": 1.3260909935004643, + "grad_norm": 1.6137654781341553, + "learning_rate": 1e-06, + "loss": 0.4425, + "mean_token_accuracy": 0.8507997393608093, + "num_tokens": 250743807.0, + "step": 7141 + }, + { + "epoch": 1.3262766945218198, + "grad_norm": 1.5327110290527344, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8789106011390686, + "num_tokens": 250780323.0, + "step": 7142 + }, + { + "epoch": 1.3264623955431754, + "grad_norm": 1.6955643892288208, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8755625486373901, + "num_tokens": 250809381.0, + "step": 7143 + }, + { + "epoch": 1.326648096564531, + "grad_norm": 1.55642569065094, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8753930926322937, + "num_tokens": 250844412.0, + "step": 7144 + }, + { + "epoch": 1.3268337975858868, + "grad_norm": 1.522563099861145, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.864081621170044, + "num_tokens": 250880002.0, + "step": 7145 + }, + { + "epoch": 1.3270194986072423, + "grad_norm": 1.4976463317871094, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8729167580604553, + "num_tokens": 250917806.0, + "step": 7146 + }, + { + "epoch": 1.3272051996285978, + "grad_norm": 1.5954583883285522, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8674769401550293, + "num_tokens": 250954290.0, + "step": 7147 + }, + { + "epoch": 1.3273909006499536, + "grad_norm": 1.4536906480789185, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8703802227973938, + "num_tokens": 250989846.0, + "step": 7148 + }, + { + "epoch": 1.3275766016713093, + "grad_norm": 1.5737905502319336, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8797593116760254, + "num_tokens": 251019286.0, + "step": 7149 + }, + { + "epoch": 1.3277623026926648, + "grad_norm": 1.514492392539978, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8693467378616333, + "num_tokens": 251052839.0, + "step": 7150 + }, + { + "epoch": 1.3279480037140203, + "grad_norm": 1.5210238695144653, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8725528717041016, + "num_tokens": 251085405.0, + "step": 7151 + }, + { + "epoch": 1.328133704735376, + "grad_norm": 1.6980481147766113, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8764504194259644, + "num_tokens": 251114322.0, + "step": 7152 + }, + { + "epoch": 1.3283194057567318, + "grad_norm": 1.7665588855743408, + "learning_rate": 1e-06, + "loss": 0.4364, + "mean_token_accuracy": 0.8599816560745239, + "num_tokens": 251142890.0, + "step": 7153 + }, + { + "epoch": 1.3285051067780873, + "grad_norm": 1.3991138935089111, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8816810846328735, + "num_tokens": 251179901.0, + "step": 7154 + }, + { + "epoch": 1.3286908077994428, + "grad_norm": 1.4905833005905151, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8792814612388611, + "num_tokens": 251212995.0, + "step": 7155 + }, + { + "epoch": 1.3288765088207986, + "grad_norm": 1.568971037864685, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8753848075866699, + "num_tokens": 251245048.0, + "step": 7156 + }, + { + "epoch": 1.329062209842154, + "grad_norm": 1.4967472553253174, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8726805448532104, + "num_tokens": 251281256.0, + "step": 7157 + }, + { + "epoch": 1.3292479108635098, + "grad_norm": 1.4377431869506836, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8713415861129761, + "num_tokens": 251318487.0, + "step": 7158 + }, + { + "epoch": 1.3294336118848653, + "grad_norm": 1.5425798892974854, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.87595534324646, + "num_tokens": 251354370.0, + "step": 7159 + }, + { + "epoch": 1.329619312906221, + "grad_norm": 1.512705683708191, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8782387971878052, + "num_tokens": 251385220.0, + "step": 7160 + }, + { + "epoch": 1.3298050139275766, + "grad_norm": 1.424428105354309, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8789966106414795, + "num_tokens": 251424587.0, + "step": 7161 + }, + { + "epoch": 1.3299907149489323, + "grad_norm": 1.6233702898025513, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.8534522652626038, + "num_tokens": 251462173.0, + "step": 7162 + }, + { + "epoch": 1.3301764159702878, + "grad_norm": 1.3772783279418945, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8910260200500488, + "num_tokens": 251498251.0, + "step": 7163 + }, + { + "epoch": 1.3303621169916435, + "grad_norm": 1.3436079025268555, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8849936127662659, + "num_tokens": 251537019.0, + "step": 7164 + }, + { + "epoch": 1.330547818012999, + "grad_norm": 1.6521592140197754, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8654811382293701, + "num_tokens": 251567018.0, + "step": 7165 + }, + { + "epoch": 1.3307335190343546, + "grad_norm": 1.4738215208053589, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8753846883773804, + "num_tokens": 251603202.0, + "step": 7166 + }, + { + "epoch": 1.3309192200557103, + "grad_norm": 1.4186869859695435, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8779910206794739, + "num_tokens": 251641040.0, + "step": 7167 + }, + { + "epoch": 1.331104921077066, + "grad_norm": 1.3048813343048096, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8948867321014404, + "num_tokens": 251681947.0, + "step": 7168 + }, + { + "epoch": 1.3312906220984215, + "grad_norm": 1.5348613262176514, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8714184761047363, + "num_tokens": 251715560.0, + "step": 7169 + }, + { + "epoch": 1.331476323119777, + "grad_norm": 1.4578781127929688, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8762729167938232, + "num_tokens": 251753210.0, + "step": 7170 + }, + { + "epoch": 1.3316620241411328, + "grad_norm": 1.4919449090957642, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8630099296569824, + "num_tokens": 251789686.0, + "step": 7171 + }, + { + "epoch": 1.3318477251624885, + "grad_norm": 1.4853315353393555, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8913091421127319, + "num_tokens": 251826549.0, + "step": 7172 + }, + { + "epoch": 1.332033426183844, + "grad_norm": 1.6181280612945557, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8711932897567749, + "num_tokens": 251857955.0, + "step": 7173 + }, + { + "epoch": 1.3322191272051995, + "grad_norm": 1.5704963207244873, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8679417371749878, + "num_tokens": 251889782.0, + "step": 7174 + }, + { + "epoch": 1.3324048282265553, + "grad_norm": 1.3842756748199463, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8892253041267395, + "num_tokens": 251927570.0, + "step": 7175 + }, + { + "epoch": 1.332590529247911, + "grad_norm": 1.7951849699020386, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8742172718048096, + "num_tokens": 251953786.0, + "step": 7176 + }, + { + "epoch": 1.3327762302692665, + "grad_norm": 1.699365258216858, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8763167858123779, + "num_tokens": 251980550.0, + "step": 7177 + }, + { + "epoch": 1.332961931290622, + "grad_norm": 1.531033992767334, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8681042194366455, + "num_tokens": 252013528.0, + "step": 7178 + }, + { + "epoch": 1.3331476323119777, + "grad_norm": 1.3599207401275635, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.8885172009468079, + "num_tokens": 252054761.0, + "step": 7179 + }, + { + "epoch": 1.3333333333333333, + "grad_norm": 1.412481665611267, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8781695365905762, + "num_tokens": 252092840.0, + "step": 7180 + }, + { + "epoch": 1.333519034354689, + "grad_norm": 1.435455560684204, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8778653740882874, + "num_tokens": 252130751.0, + "step": 7181 + }, + { + "epoch": 1.3337047353760445, + "grad_norm": 1.4144495725631714, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8786271214485168, + "num_tokens": 252170995.0, + "step": 7182 + }, + { + "epoch": 1.3338904363974002, + "grad_norm": 1.618027925491333, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8710330724716187, + "num_tokens": 252201216.0, + "step": 7183 + }, + { + "epoch": 1.3340761374187557, + "grad_norm": 1.5087754726409912, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8819267749786377, + "num_tokens": 252236758.0, + "step": 7184 + }, + { + "epoch": 1.3342618384401115, + "grad_norm": 1.5441577434539795, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8859432935714722, + "num_tokens": 252267941.0, + "step": 7185 + }, + { + "epoch": 1.334447539461467, + "grad_norm": 1.4933297634124756, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8859760165214539, + "num_tokens": 252301824.0, + "step": 7186 + }, + { + "epoch": 1.3346332404828227, + "grad_norm": 1.5125867128372192, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8787887096405029, + "num_tokens": 252335602.0, + "step": 7187 + }, + { + "epoch": 1.3348189415041782, + "grad_norm": 1.5697238445281982, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8638023734092712, + "num_tokens": 252371441.0, + "step": 7188 + }, + { + "epoch": 1.335004642525534, + "grad_norm": 1.6210670471191406, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8676321506500244, + "num_tokens": 252405379.0, + "step": 7189 + }, + { + "epoch": 1.3351903435468895, + "grad_norm": 1.4982614517211914, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8748787045478821, + "num_tokens": 252440488.0, + "step": 7190 + }, + { + "epoch": 1.3353760445682452, + "grad_norm": 1.5497559309005737, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8826889991760254, + "num_tokens": 252475614.0, + "step": 7191 + }, + { + "epoch": 1.3355617455896007, + "grad_norm": 1.253659725189209, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8952287435531616, + "num_tokens": 252516542.0, + "step": 7192 + }, + { + "epoch": 1.3357474466109562, + "grad_norm": 1.5200124979019165, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8715603947639465, + "num_tokens": 252551407.0, + "step": 7193 + }, + { + "epoch": 1.335933147632312, + "grad_norm": 1.63595449924469, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8689150214195251, + "num_tokens": 252583762.0, + "step": 7194 + }, + { + "epoch": 1.3361188486536677, + "grad_norm": 1.590946078300476, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8624619245529175, + "num_tokens": 252619030.0, + "step": 7195 + }, + { + "epoch": 1.3363045496750232, + "grad_norm": 1.4975557327270508, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8828825950622559, + "num_tokens": 252651559.0, + "step": 7196 + }, + { + "epoch": 1.3364902506963787, + "grad_norm": 1.6689778566360474, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8753042221069336, + "num_tokens": 252681477.0, + "step": 7197 + }, + { + "epoch": 1.3366759517177345, + "grad_norm": 1.6614582538604736, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8664316534996033, + "num_tokens": 252711650.0, + "step": 7198 + }, + { + "epoch": 1.3368616527390902, + "grad_norm": 1.5933563709259033, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8781402111053467, + "num_tokens": 252744187.0, + "step": 7199 + }, + { + "epoch": 1.3370473537604457, + "grad_norm": 1.5418449640274048, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8736587762832642, + "num_tokens": 252779481.0, + "step": 7200 + }, + { + "epoch": 1.3372330547818012, + "grad_norm": 1.457678198814392, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8836889266967773, + "num_tokens": 252813056.0, + "step": 7201 + }, + { + "epoch": 1.337418755803157, + "grad_norm": 1.6461085081100464, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8631398677825928, + "num_tokens": 252843080.0, + "step": 7202 + }, + { + "epoch": 1.3376044568245125, + "grad_norm": 1.6267341375350952, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8611242175102234, + "num_tokens": 252877799.0, + "step": 7203 + }, + { + "epoch": 1.3377901578458682, + "grad_norm": 1.5101828575134277, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8767862319946289, + "num_tokens": 252908430.0, + "step": 7204 + }, + { + "epoch": 1.3379758588672237, + "grad_norm": 1.5272064208984375, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8643306493759155, + "num_tokens": 252943187.0, + "step": 7205 + }, + { + "epoch": 1.3381615598885794, + "grad_norm": 1.5522931814193726, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8806660175323486, + "num_tokens": 252978365.0, + "step": 7206 + }, + { + "epoch": 1.338347260909935, + "grad_norm": 1.520595908164978, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8852880597114563, + "num_tokens": 253009082.0, + "step": 7207 + }, + { + "epoch": 1.3385329619312907, + "grad_norm": 1.7049063444137573, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8663070201873779, + "num_tokens": 253040574.0, + "step": 7208 + }, + { + "epoch": 1.3387186629526462, + "grad_norm": 1.5588151216506958, + "learning_rate": 1e-06, + "loss": 0.4388, + "mean_token_accuracy": 0.8604758977890015, + "num_tokens": 253078246.0, + "step": 7209 + }, + { + "epoch": 1.338904363974002, + "grad_norm": 1.431269884109497, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.869765043258667, + "num_tokens": 253119936.0, + "step": 7210 + }, + { + "epoch": 1.3390900649953574, + "grad_norm": 1.5598374605178833, + "learning_rate": 1e-06, + "loss": 0.4033, + "mean_token_accuracy": 0.8608965277671814, + "num_tokens": 253157840.0, + "step": 7211 + }, + { + "epoch": 1.3392757660167132, + "grad_norm": 1.7121648788452148, + "learning_rate": 1e-06, + "loss": 0.432, + "mean_token_accuracy": 0.8565800189971924, + "num_tokens": 253192631.0, + "step": 7212 + }, + { + "epoch": 1.3394614670380687, + "grad_norm": 1.4329603910446167, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8859139084815979, + "num_tokens": 253227196.0, + "step": 7213 + }, + { + "epoch": 1.3396471680594244, + "grad_norm": 1.6616950035095215, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8807183504104614, + "num_tokens": 253254536.0, + "step": 7214 + }, + { + "epoch": 1.33983286908078, + "grad_norm": 1.6737207174301147, + "learning_rate": 1e-06, + "loss": 0.3951, + "mean_token_accuracy": 0.8668291568756104, + "num_tokens": 253284833.0, + "step": 7215 + }, + { + "epoch": 1.3400185701021354, + "grad_norm": 1.6602845191955566, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8834600448608398, + "num_tokens": 253313471.0, + "step": 7216 + }, + { + "epoch": 1.3402042711234912, + "grad_norm": 1.5535228252410889, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8817297220230103, + "num_tokens": 253344710.0, + "step": 7217 + }, + { + "epoch": 1.340389972144847, + "grad_norm": 1.5149317979812622, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8662853240966797, + "num_tokens": 253382040.0, + "step": 7218 + }, + { + "epoch": 1.3405756731662024, + "grad_norm": 1.6258763074874878, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8731772899627686, + "num_tokens": 253413262.0, + "step": 7219 + }, + { + "epoch": 1.340761374187558, + "grad_norm": 1.7001198530197144, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8736318349838257, + "num_tokens": 253444408.0, + "step": 7220 + }, + { + "epoch": 1.3409470752089137, + "grad_norm": 1.5317429304122925, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.896091878414154, + "num_tokens": 253474777.0, + "step": 7221 + }, + { + "epoch": 1.3411327762302694, + "grad_norm": 1.4824151992797852, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8819456100463867, + "num_tokens": 253510932.0, + "step": 7222 + }, + { + "epoch": 1.341318477251625, + "grad_norm": 1.6925805807113647, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8762309551239014, + "num_tokens": 253539284.0, + "step": 7223 + }, + { + "epoch": 1.3415041782729804, + "grad_norm": 1.6534473896026611, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8898220062255859, + "num_tokens": 253573842.0, + "step": 7224 + }, + { + "epoch": 1.3416898792943361, + "grad_norm": 1.5055699348449707, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8720232844352722, + "num_tokens": 253611128.0, + "step": 7225 + }, + { + "epoch": 1.3418755803156919, + "grad_norm": 1.5935245752334595, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8747446537017822, + "num_tokens": 253650870.0, + "step": 7226 + }, + { + "epoch": 1.3420612813370474, + "grad_norm": 1.431441307067871, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8816210031509399, + "num_tokens": 253689234.0, + "step": 7227 + }, + { + "epoch": 1.342246982358403, + "grad_norm": 1.3826123476028442, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8900383710861206, + "num_tokens": 253727482.0, + "step": 7228 + }, + { + "epoch": 1.3424326833797586, + "grad_norm": 1.4939303398132324, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8786494135856628, + "num_tokens": 253763161.0, + "step": 7229 + }, + { + "epoch": 1.3426183844011141, + "grad_norm": 1.4755299091339111, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8731647729873657, + "num_tokens": 253799136.0, + "step": 7230 + }, + { + "epoch": 1.3428040854224699, + "grad_norm": 1.3967957496643066, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8851768970489502, + "num_tokens": 253839179.0, + "step": 7231 + }, + { + "epoch": 1.3429897864438254, + "grad_norm": 1.454541802406311, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8803005218505859, + "num_tokens": 253878445.0, + "step": 7232 + }, + { + "epoch": 1.3431754874651811, + "grad_norm": 1.3617013692855835, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8838185667991638, + "num_tokens": 253921274.0, + "step": 7233 + }, + { + "epoch": 1.3433611884865366, + "grad_norm": 1.4184372425079346, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8905360102653503, + "num_tokens": 253953298.0, + "step": 7234 + }, + { + "epoch": 1.3435468895078924, + "grad_norm": 1.4569913148880005, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8823227882385254, + "num_tokens": 253989996.0, + "step": 7235 + }, + { + "epoch": 1.3437325905292479, + "grad_norm": 1.5367342233657837, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8781949281692505, + "num_tokens": 254025429.0, + "step": 7236 + }, + { + "epoch": 1.3439182915506036, + "grad_norm": 1.4314876794815063, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8757920861244202, + "num_tokens": 254062204.0, + "step": 7237 + }, + { + "epoch": 1.3441039925719591, + "grad_norm": 1.5021986961364746, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8778518438339233, + "num_tokens": 254096856.0, + "step": 7238 + }, + { + "epoch": 1.3442896935933146, + "grad_norm": 1.4859763383865356, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8818607330322266, + "num_tokens": 254132715.0, + "step": 7239 + }, + { + "epoch": 1.3444753946146704, + "grad_norm": 1.478225827217102, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8790076971054077, + "num_tokens": 254169115.0, + "step": 7240 + }, + { + "epoch": 1.344661095636026, + "grad_norm": 1.5693116188049316, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8745436668395996, + "num_tokens": 254203772.0, + "step": 7241 + }, + { + "epoch": 1.3448467966573816, + "grad_norm": 1.563069462776184, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8637414574623108, + "num_tokens": 254235068.0, + "step": 7242 + }, + { + "epoch": 1.3450324976787371, + "grad_norm": 1.4537774324417114, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8758912682533264, + "num_tokens": 254272169.0, + "step": 7243 + }, + { + "epoch": 1.3452181987000928, + "grad_norm": 1.538487434387207, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8745273351669312, + "num_tokens": 254308867.0, + "step": 7244 + }, + { + "epoch": 1.3454038997214486, + "grad_norm": 1.5124058723449707, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8853742480278015, + "num_tokens": 254342794.0, + "step": 7245 + }, + { + "epoch": 1.345589600742804, + "grad_norm": 1.385867714881897, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8807839155197144, + "num_tokens": 254384284.0, + "step": 7246 + }, + { + "epoch": 1.3457753017641596, + "grad_norm": 1.5050060749053955, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8639166951179504, + "num_tokens": 254424791.0, + "step": 7247 + }, + { + "epoch": 1.3459610027855153, + "grad_norm": 1.5635403394699097, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8818258047103882, + "num_tokens": 254454569.0, + "step": 7248 + }, + { + "epoch": 1.346146703806871, + "grad_norm": 1.7187490463256836, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8700590133666992, + "num_tokens": 254484526.0, + "step": 7249 + }, + { + "epoch": 1.3463324048282266, + "grad_norm": 1.439744234085083, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8776677846908569, + "num_tokens": 254521252.0, + "step": 7250 + }, + { + "epoch": 1.346518105849582, + "grad_norm": 1.4822592735290527, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8806017637252808, + "num_tokens": 254560556.0, + "step": 7251 + }, + { + "epoch": 1.3467038068709378, + "grad_norm": 1.5482847690582275, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8856463432312012, + "num_tokens": 254594307.0, + "step": 7252 + }, + { + "epoch": 1.3468895078922933, + "grad_norm": 1.5212583541870117, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8805927038192749, + "num_tokens": 254630939.0, + "step": 7253 + }, + { + "epoch": 1.347075208913649, + "grad_norm": 1.453231930732727, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8696626424789429, + "num_tokens": 254670433.0, + "step": 7254 + }, + { + "epoch": 1.3472609099350046, + "grad_norm": 1.4048433303833008, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8758524656295776, + "num_tokens": 254709313.0, + "step": 7255 + }, + { + "epoch": 1.3474466109563603, + "grad_norm": 1.4614527225494385, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8791524171829224, + "num_tokens": 254744541.0, + "step": 7256 + }, + { + "epoch": 1.3476323119777158, + "grad_norm": 1.4758650064468384, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8900539875030518, + "num_tokens": 254778195.0, + "step": 7257 + }, + { + "epoch": 1.3478180129990716, + "grad_norm": 1.484681248664856, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8814783096313477, + "num_tokens": 254813288.0, + "step": 7258 + }, + { + "epoch": 1.348003714020427, + "grad_norm": 1.3963831663131714, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8745565414428711, + "num_tokens": 254851340.0, + "step": 7259 + }, + { + "epoch": 1.3481894150417828, + "grad_norm": 1.5651860237121582, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8707869052886963, + "num_tokens": 254884303.0, + "step": 7260 + }, + { + "epoch": 1.3483751160631383, + "grad_norm": 1.4227218627929688, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8675850033760071, + "num_tokens": 254925752.0, + "step": 7261 + }, + { + "epoch": 1.348560817084494, + "grad_norm": 1.416749119758606, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8818137049674988, + "num_tokens": 254963049.0, + "step": 7262 + }, + { + "epoch": 1.3487465181058496, + "grad_norm": 1.4520374536514282, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8672672510147095, + "num_tokens": 254999749.0, + "step": 7263 + }, + { + "epoch": 1.3489322191272053, + "grad_norm": 1.3859187364578247, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8916199207305908, + "num_tokens": 255034313.0, + "step": 7264 + }, + { + "epoch": 1.3491179201485608, + "grad_norm": 1.4136316776275635, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.8661671280860901, + "num_tokens": 255075042.0, + "step": 7265 + }, + { + "epoch": 1.3493036211699163, + "grad_norm": 1.5688673257827759, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8786418437957764, + "num_tokens": 255108377.0, + "step": 7266 + }, + { + "epoch": 1.349489322191272, + "grad_norm": 1.4882187843322754, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8754358291625977, + "num_tokens": 255144848.0, + "step": 7267 + }, + { + "epoch": 1.3496750232126278, + "grad_norm": 1.4362709522247314, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8758948445320129, + "num_tokens": 255182905.0, + "step": 7268 + }, + { + "epoch": 1.3498607242339833, + "grad_norm": 1.6244648694992065, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.871131181716919, + "num_tokens": 255212514.0, + "step": 7269 + }, + { + "epoch": 1.3500464252553388, + "grad_norm": 1.6221895217895508, + "learning_rate": 1e-06, + "loss": 0.4431, + "mean_token_accuracy": 0.8510705828666687, + "num_tokens": 255247852.0, + "step": 7270 + }, + { + "epoch": 1.3502321262766945, + "grad_norm": 1.3693817853927612, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8844016790390015, + "num_tokens": 255287355.0, + "step": 7271 + }, + { + "epoch": 1.3504178272980503, + "grad_norm": 1.5751372575759888, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8773097395896912, + "num_tokens": 255318904.0, + "step": 7272 + }, + { + "epoch": 1.3506035283194058, + "grad_norm": 1.4155360460281372, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8687448501586914, + "num_tokens": 255358765.0, + "step": 7273 + }, + { + "epoch": 1.3507892293407613, + "grad_norm": 1.516321063041687, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8844496607780457, + "num_tokens": 255388892.0, + "step": 7274 + }, + { + "epoch": 1.350974930362117, + "grad_norm": 1.5195090770721436, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.869247317314148, + "num_tokens": 255428272.0, + "step": 7275 + }, + { + "epoch": 1.3511606313834725, + "grad_norm": 1.541457176208496, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8761044144630432, + "num_tokens": 255461958.0, + "step": 7276 + }, + { + "epoch": 1.3513463324048283, + "grad_norm": 1.4512760639190674, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8689287900924683, + "num_tokens": 255501571.0, + "step": 7277 + }, + { + "epoch": 1.3515320334261838, + "grad_norm": 1.3912253379821777, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8856569528579712, + "num_tokens": 255538439.0, + "step": 7278 + }, + { + "epoch": 1.3517177344475395, + "grad_norm": 1.4508589506149292, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8762575387954712, + "num_tokens": 255574894.0, + "step": 7279 + }, + { + "epoch": 1.351903435468895, + "grad_norm": 1.4453284740447998, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8959691524505615, + "num_tokens": 255608640.0, + "step": 7280 + }, + { + "epoch": 1.3520891364902508, + "grad_norm": 1.5368942022323608, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.879275918006897, + "num_tokens": 255638375.0, + "step": 7281 + }, + { + "epoch": 1.3522748375116063, + "grad_norm": 1.5812351703643799, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8726374506950378, + "num_tokens": 255670155.0, + "step": 7282 + }, + { + "epoch": 1.352460538532962, + "grad_norm": 1.5165501832962036, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.8836642503738403, + "num_tokens": 255701100.0, + "step": 7283 + }, + { + "epoch": 1.3526462395543175, + "grad_norm": 1.438380479812622, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8706877827644348, + "num_tokens": 255739705.0, + "step": 7284 + }, + { + "epoch": 1.3528319405756732, + "grad_norm": 1.4939563274383545, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8774557113647461, + "num_tokens": 255773101.0, + "step": 7285 + }, + { + "epoch": 1.3530176415970288, + "grad_norm": 1.5530027151107788, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.8563804030418396, + "num_tokens": 255811847.0, + "step": 7286 + }, + { + "epoch": 1.3532033426183845, + "grad_norm": 1.579406499862671, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8794868588447571, + "num_tokens": 255844451.0, + "step": 7287 + }, + { + "epoch": 1.35338904363974, + "grad_norm": 1.4559814929962158, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.867151141166687, + "num_tokens": 255883403.0, + "step": 7288 + }, + { + "epoch": 1.3535747446610955, + "grad_norm": 1.3756664991378784, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8874104022979736, + "num_tokens": 255922089.0, + "step": 7289 + }, + { + "epoch": 1.3537604456824512, + "grad_norm": 1.454006552696228, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8867722153663635, + "num_tokens": 255961628.0, + "step": 7290 + }, + { + "epoch": 1.353946146703807, + "grad_norm": 1.3968569040298462, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8806928396224976, + "num_tokens": 256005376.0, + "step": 7291 + }, + { + "epoch": 1.3541318477251625, + "grad_norm": 1.5085700750350952, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8896225094795227, + "num_tokens": 256037072.0, + "step": 7292 + }, + { + "epoch": 1.354317548746518, + "grad_norm": 1.457044243812561, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8780725002288818, + "num_tokens": 256071760.0, + "step": 7293 + }, + { + "epoch": 1.3545032497678737, + "grad_norm": 1.4977893829345703, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8867154121398926, + "num_tokens": 256104889.0, + "step": 7294 + }, + { + "epoch": 1.3546889507892295, + "grad_norm": 1.5608662366867065, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8708062171936035, + "num_tokens": 256138589.0, + "step": 7295 + }, + { + "epoch": 1.354874651810585, + "grad_norm": 1.535607933998108, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.875288188457489, + "num_tokens": 256170857.0, + "step": 7296 + }, + { + "epoch": 1.3550603528319405, + "grad_norm": 1.546367883682251, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8667728900909424, + "num_tokens": 256206462.0, + "step": 7297 + }, + { + "epoch": 1.3552460538532962, + "grad_norm": 1.4108737707138062, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8753082156181335, + "num_tokens": 256242403.0, + "step": 7298 + }, + { + "epoch": 1.3554317548746517, + "grad_norm": 1.5498424768447876, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8702456951141357, + "num_tokens": 256275111.0, + "step": 7299 + }, + { + "epoch": 1.3556174558960075, + "grad_norm": 1.392470359802246, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8772459030151367, + "num_tokens": 256313527.0, + "step": 7300 + }, + { + "epoch": 1.355803156917363, + "grad_norm": 1.4922314882278442, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.8656456470489502, + "num_tokens": 256349320.0, + "step": 7301 + }, + { + "epoch": 1.3559888579387187, + "grad_norm": 1.455889105796814, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8715298175811768, + "num_tokens": 256385277.0, + "step": 7302 + }, + { + "epoch": 1.3561745589600742, + "grad_norm": 1.4497885704040527, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8830796480178833, + "num_tokens": 256417900.0, + "step": 7303 + }, + { + "epoch": 1.35636025998143, + "grad_norm": 1.6560505628585815, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8744029998779297, + "num_tokens": 256444629.0, + "step": 7304 + }, + { + "epoch": 1.3565459610027855, + "grad_norm": 1.4351621866226196, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8787753582000732, + "num_tokens": 256482126.0, + "step": 7305 + }, + { + "epoch": 1.3567316620241412, + "grad_norm": 1.3457945585250854, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8804640769958496, + "num_tokens": 256521079.0, + "step": 7306 + }, + { + "epoch": 1.3569173630454967, + "grad_norm": 1.498731255531311, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8794824481010437, + "num_tokens": 256556124.0, + "step": 7307 + }, + { + "epoch": 1.3571030640668524, + "grad_norm": 1.7008897066116333, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8731467723846436, + "num_tokens": 256585796.0, + "step": 7308 + }, + { + "epoch": 1.357288765088208, + "grad_norm": 1.6821175813674927, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8727740049362183, + "num_tokens": 256617191.0, + "step": 7309 + }, + { + "epoch": 1.3574744661095637, + "grad_norm": 1.470767855644226, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8848735094070435, + "num_tokens": 256653860.0, + "step": 7310 + }, + { + "epoch": 1.3576601671309192, + "grad_norm": 1.6568044424057007, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8683145046234131, + "num_tokens": 256688124.0, + "step": 7311 + }, + { + "epoch": 1.3578458681522747, + "grad_norm": 1.4573105573654175, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8563324809074402, + "num_tokens": 256729661.0, + "step": 7312 + }, + { + "epoch": 1.3580315691736304, + "grad_norm": 1.3956458568572998, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8803451657295227, + "num_tokens": 256768854.0, + "step": 7313 + }, + { + "epoch": 1.3582172701949862, + "grad_norm": 1.5716837644577026, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8730922341346741, + "num_tokens": 256803941.0, + "step": 7314 + }, + { + "epoch": 1.3584029712163417, + "grad_norm": 1.6226824522018433, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.866499662399292, + "num_tokens": 256837178.0, + "step": 7315 + }, + { + "epoch": 1.3585886722376972, + "grad_norm": 1.6262354850769043, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8675671219825745, + "num_tokens": 256872718.0, + "step": 7316 + }, + { + "epoch": 1.358774373259053, + "grad_norm": 1.3745596408843994, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8838511109352112, + "num_tokens": 256913309.0, + "step": 7317 + }, + { + "epoch": 1.3589600742804087, + "grad_norm": 1.4267247915267944, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.878521740436554, + "num_tokens": 256950629.0, + "step": 7318 + }, + { + "epoch": 1.3591457753017642, + "grad_norm": 1.4661139249801636, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8688204288482666, + "num_tokens": 256987736.0, + "step": 7319 + }, + { + "epoch": 1.3593314763231197, + "grad_norm": 1.3865766525268555, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8815435171127319, + "num_tokens": 257027164.0, + "step": 7320 + }, + { + "epoch": 1.3595171773444754, + "grad_norm": 1.6410579681396484, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8647241592407227, + "num_tokens": 257056203.0, + "step": 7321 + }, + { + "epoch": 1.3597028783658311, + "grad_norm": 1.5471892356872559, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8809967041015625, + "num_tokens": 257088648.0, + "step": 7322 + }, + { + "epoch": 1.3598885793871867, + "grad_norm": 1.4692540168762207, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8888108730316162, + "num_tokens": 257119528.0, + "step": 7323 + }, + { + "epoch": 1.3600742804085422, + "grad_norm": 1.5893322229385376, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8569690585136414, + "num_tokens": 257154147.0, + "step": 7324 + }, + { + "epoch": 1.360259981429898, + "grad_norm": 1.4271410703659058, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8769224882125854, + "num_tokens": 257190272.0, + "step": 7325 + }, + { + "epoch": 1.3604456824512534, + "grad_norm": 1.3092033863067627, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8884193897247314, + "num_tokens": 257230304.0, + "step": 7326 + }, + { + "epoch": 1.3606313834726091, + "grad_norm": 1.5409144163131714, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8773468136787415, + "num_tokens": 257264278.0, + "step": 7327 + }, + { + "epoch": 1.3608170844939647, + "grad_norm": 1.420263648033142, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8778343200683594, + "num_tokens": 257303428.0, + "step": 7328 + }, + { + "epoch": 1.3610027855153204, + "grad_norm": 1.4764505624771118, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8886709809303284, + "num_tokens": 257336002.0, + "step": 7329 + }, + { + "epoch": 1.361188486536676, + "grad_norm": 1.4146896600723267, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.883505642414093, + "num_tokens": 257370163.0, + "step": 7330 + }, + { + "epoch": 1.3613741875580316, + "grad_norm": 1.5314823389053345, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8831634521484375, + "num_tokens": 257404566.0, + "step": 7331 + }, + { + "epoch": 1.3615598885793871, + "grad_norm": 1.4810823202133179, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8846219182014465, + "num_tokens": 257441120.0, + "step": 7332 + }, + { + "epoch": 1.3617455896007429, + "grad_norm": 1.5531660318374634, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8755171298980713, + "num_tokens": 257475845.0, + "step": 7333 + }, + { + "epoch": 1.3619312906220984, + "grad_norm": 1.4423824548721313, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8764837980270386, + "num_tokens": 257516986.0, + "step": 7334 + }, + { + "epoch": 1.362116991643454, + "grad_norm": 1.3501946926116943, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8907471895217896, + "num_tokens": 257554703.0, + "step": 7335 + }, + { + "epoch": 1.3623026926648096, + "grad_norm": 1.532049298286438, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.874693751335144, + "num_tokens": 257585657.0, + "step": 7336 + }, + { + "epoch": 1.3624883936861654, + "grad_norm": 1.4835832118988037, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8818711042404175, + "num_tokens": 257618699.0, + "step": 7337 + }, + { + "epoch": 1.3626740947075209, + "grad_norm": 1.5748558044433594, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8837183713912964, + "num_tokens": 257647683.0, + "step": 7338 + }, + { + "epoch": 1.3628597957288764, + "grad_norm": 1.419419765472412, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8823375701904297, + "num_tokens": 257685306.0, + "step": 7339 + }, + { + "epoch": 1.3630454967502321, + "grad_norm": 1.539981722831726, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8749960660934448, + "num_tokens": 257718349.0, + "step": 7340 + }, + { + "epoch": 1.3632311977715879, + "grad_norm": 1.6287634372711182, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8803264498710632, + "num_tokens": 257747899.0, + "step": 7341 + }, + { + "epoch": 1.3634168987929434, + "grad_norm": 1.533454418182373, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8776273727416992, + "num_tokens": 257780960.0, + "step": 7342 + }, + { + "epoch": 1.3636025998142989, + "grad_norm": 1.5456972122192383, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.891629159450531, + "num_tokens": 257809387.0, + "step": 7343 + }, + { + "epoch": 1.3637883008356546, + "grad_norm": 1.5930910110473633, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8709037899971008, + "num_tokens": 257842572.0, + "step": 7344 + }, + { + "epoch": 1.3639740018570103, + "grad_norm": 1.385185956954956, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8820892572402954, + "num_tokens": 257880303.0, + "step": 7345 + }, + { + "epoch": 1.3641597028783659, + "grad_norm": 1.4378937482833862, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8848294019699097, + "num_tokens": 257917336.0, + "step": 7346 + }, + { + "epoch": 1.3643454038997214, + "grad_norm": 1.3943607807159424, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8856488466262817, + "num_tokens": 257953219.0, + "step": 7347 + }, + { + "epoch": 1.364531104921077, + "grad_norm": 1.5300660133361816, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8772217035293579, + "num_tokens": 257987328.0, + "step": 7348 + }, + { + "epoch": 1.3647168059424326, + "grad_norm": 1.5004502534866333, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.88056480884552, + "num_tokens": 258020244.0, + "step": 7349 + }, + { + "epoch": 1.3649025069637883, + "grad_norm": 1.362084984779358, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8930234313011169, + "num_tokens": 258060135.0, + "step": 7350 + }, + { + "epoch": 1.3650882079851439, + "grad_norm": 1.3960564136505127, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8727391362190247, + "num_tokens": 258098641.0, + "step": 7351 + }, + { + "epoch": 1.3652739090064996, + "grad_norm": 1.4439349174499512, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8682637214660645, + "num_tokens": 258139620.0, + "step": 7352 + }, + { + "epoch": 1.365459610027855, + "grad_norm": 1.5835789442062378, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.870013415813446, + "num_tokens": 258170514.0, + "step": 7353 + }, + { + "epoch": 1.3656453110492108, + "grad_norm": 1.500240683555603, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.8575676679611206, + "num_tokens": 258204994.0, + "step": 7354 + }, + { + "epoch": 1.3658310120705663, + "grad_norm": 1.634508490562439, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8709229230880737, + "num_tokens": 258236999.0, + "step": 7355 + }, + { + "epoch": 1.366016713091922, + "grad_norm": 1.351014494895935, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8774850964546204, + "num_tokens": 258278660.0, + "step": 7356 + }, + { + "epoch": 1.3662024141132776, + "grad_norm": 1.551681399345398, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.8631513118743896, + "num_tokens": 258317922.0, + "step": 7357 + }, + { + "epoch": 1.3663881151346333, + "grad_norm": 1.473770022392273, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8852179050445557, + "num_tokens": 258356152.0, + "step": 7358 + }, + { + "epoch": 1.3665738161559888, + "grad_norm": 1.5860780477523804, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8863385319709778, + "num_tokens": 258391240.0, + "step": 7359 + }, + { + "epoch": 1.3667595171773446, + "grad_norm": 1.5842710733413696, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8781119585037231, + "num_tokens": 258424518.0, + "step": 7360 + }, + { + "epoch": 1.3669452181987, + "grad_norm": 1.6176893711090088, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8766332268714905, + "num_tokens": 258459438.0, + "step": 7361 + }, + { + "epoch": 1.3671309192200556, + "grad_norm": 1.4527453184127808, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8674860000610352, + "num_tokens": 258499136.0, + "step": 7362 + }, + { + "epoch": 1.3673166202414113, + "grad_norm": 1.5708184242248535, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.867929995059967, + "num_tokens": 258534484.0, + "step": 7363 + }, + { + "epoch": 1.367502321262767, + "grad_norm": 1.423248529434204, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8788022994995117, + "num_tokens": 258569950.0, + "step": 7364 + }, + { + "epoch": 1.3676880222841226, + "grad_norm": 1.4996135234832764, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8664478659629822, + "num_tokens": 258606426.0, + "step": 7365 + }, + { + "epoch": 1.367873723305478, + "grad_norm": 1.525967001914978, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8719484806060791, + "num_tokens": 258643921.0, + "step": 7366 + }, + { + "epoch": 1.3680594243268338, + "grad_norm": 1.502419114112854, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8769041299819946, + "num_tokens": 258681279.0, + "step": 7367 + }, + { + "epoch": 1.3682451253481895, + "grad_norm": 1.5704489946365356, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.888952374458313, + "num_tokens": 258710736.0, + "step": 7368 + }, + { + "epoch": 1.368430826369545, + "grad_norm": 1.3700460195541382, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8657313585281372, + "num_tokens": 258757472.0, + "step": 7369 + }, + { + "epoch": 1.3686165273909006, + "grad_norm": 1.4468071460723877, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8771688938140869, + "num_tokens": 258791839.0, + "step": 7370 + }, + { + "epoch": 1.3688022284122563, + "grad_norm": 1.4964720010757446, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8751416802406311, + "num_tokens": 258825719.0, + "step": 7371 + }, + { + "epoch": 1.3689879294336118, + "grad_norm": 1.4845353364944458, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8800254464149475, + "num_tokens": 258858637.0, + "step": 7372 + }, + { + "epoch": 1.3691736304549675, + "grad_norm": 1.5466612577438354, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8784633874893188, + "num_tokens": 258891385.0, + "step": 7373 + }, + { + "epoch": 1.369359331476323, + "grad_norm": 1.6116818189620972, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8879743814468384, + "num_tokens": 258917464.0, + "step": 7374 + }, + { + "epoch": 1.3695450324976788, + "grad_norm": 1.488113284111023, + "learning_rate": 1e-06, + "loss": 0.4027, + "mean_token_accuracy": 0.8617396950721741, + "num_tokens": 258956156.0, + "step": 7375 + }, + { + "epoch": 1.3697307335190343, + "grad_norm": 1.5157060623168945, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8737475872039795, + "num_tokens": 258988972.0, + "step": 7376 + }, + { + "epoch": 1.36991643454039, + "grad_norm": 1.4141372442245483, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8696320056915283, + "num_tokens": 259031532.0, + "step": 7377 + }, + { + "epoch": 1.3701021355617455, + "grad_norm": 1.4291038513183594, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8775180578231812, + "num_tokens": 259070607.0, + "step": 7378 + }, + { + "epoch": 1.3702878365831013, + "grad_norm": 1.5223917961120605, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8735194206237793, + "num_tokens": 259104080.0, + "step": 7379 + }, + { + "epoch": 1.3704735376044568, + "grad_norm": 1.4149699211120605, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8660025596618652, + "num_tokens": 259144618.0, + "step": 7380 + }, + { + "epoch": 1.3706592386258125, + "grad_norm": 1.5395389795303345, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8755065202713013, + "num_tokens": 259181048.0, + "step": 7381 + }, + { + "epoch": 1.370844939647168, + "grad_norm": 1.4879257678985596, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.894393801689148, + "num_tokens": 259212482.0, + "step": 7382 + }, + { + "epoch": 1.3710306406685238, + "grad_norm": 1.536219835281372, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8665462732315063, + "num_tokens": 259247403.0, + "step": 7383 + }, + { + "epoch": 1.3712163416898793, + "grad_norm": 1.4943914413452148, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8699911236763, + "num_tokens": 259278617.0, + "step": 7384 + }, + { + "epoch": 1.3714020427112348, + "grad_norm": 1.4403331279754639, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8759336471557617, + "num_tokens": 259314944.0, + "step": 7385 + }, + { + "epoch": 1.3715877437325905, + "grad_norm": 1.5024031400680542, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8665893077850342, + "num_tokens": 259353863.0, + "step": 7386 + }, + { + "epoch": 1.3717734447539462, + "grad_norm": 1.5274455547332764, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8586942553520203, + "num_tokens": 259390480.0, + "step": 7387 + }, + { + "epoch": 1.3719591457753018, + "grad_norm": 1.42965567111969, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8863822817802429, + "num_tokens": 259425950.0, + "step": 7388 + }, + { + "epoch": 1.3721448467966573, + "grad_norm": 1.6380527019500732, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.8597238063812256, + "num_tokens": 259461713.0, + "step": 7389 + }, + { + "epoch": 1.372330547818013, + "grad_norm": 1.6299115419387817, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8772886991500854, + "num_tokens": 259494640.0, + "step": 7390 + }, + { + "epoch": 1.3725162488393687, + "grad_norm": 1.4726823568344116, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8785529732704163, + "num_tokens": 259532080.0, + "step": 7391 + }, + { + "epoch": 1.3727019498607242, + "grad_norm": 1.4410768747329712, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8775452375411987, + "num_tokens": 259566090.0, + "step": 7392 + }, + { + "epoch": 1.3728876508820798, + "grad_norm": 1.442510962486267, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8841847777366638, + "num_tokens": 259602642.0, + "step": 7393 + }, + { + "epoch": 1.3730733519034355, + "grad_norm": 1.5166643857955933, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8812710046768188, + "num_tokens": 259636566.0, + "step": 7394 + }, + { + "epoch": 1.3732590529247912, + "grad_norm": 1.4706549644470215, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8849614858627319, + "num_tokens": 259672655.0, + "step": 7395 + }, + { + "epoch": 1.3734447539461467, + "grad_norm": 1.4383660554885864, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8743218183517456, + "num_tokens": 259707694.0, + "step": 7396 + }, + { + "epoch": 1.3736304549675022, + "grad_norm": 1.5312871932983398, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8667839765548706, + "num_tokens": 259741964.0, + "step": 7397 + }, + { + "epoch": 1.373816155988858, + "grad_norm": 1.535910725593567, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8889271020889282, + "num_tokens": 259774395.0, + "step": 7398 + }, + { + "epoch": 1.3740018570102135, + "grad_norm": 1.5490185022354126, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8785574436187744, + "num_tokens": 259805296.0, + "step": 7399 + }, + { + "epoch": 1.3741875580315692, + "grad_norm": 1.3845176696777344, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8928165435791016, + "num_tokens": 259843797.0, + "step": 7400 + }, + { + "epoch": 1.3743732590529247, + "grad_norm": 1.5105998516082764, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8757439851760864, + "num_tokens": 259881205.0, + "step": 7401 + }, + { + "epoch": 1.3745589600742805, + "grad_norm": 1.5394021272659302, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8753054141998291, + "num_tokens": 259914806.0, + "step": 7402 + }, + { + "epoch": 1.374744661095636, + "grad_norm": 1.3558729887008667, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8881397247314453, + "num_tokens": 259953015.0, + "step": 7403 + }, + { + "epoch": 1.3749303621169917, + "grad_norm": 1.3981904983520508, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8833802342414856, + "num_tokens": 259990467.0, + "step": 7404 + }, + { + "epoch": 1.3751160631383472, + "grad_norm": 1.4192935228347778, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8882254362106323, + "num_tokens": 260025758.0, + "step": 7405 + }, + { + "epoch": 1.375301764159703, + "grad_norm": 1.4091050624847412, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8785828351974487, + "num_tokens": 260062499.0, + "step": 7406 + }, + { + "epoch": 1.3754874651810585, + "grad_norm": 1.613632321357727, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8761860728263855, + "num_tokens": 260096711.0, + "step": 7407 + }, + { + "epoch": 1.375673166202414, + "grad_norm": 1.6136871576309204, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8658510446548462, + "num_tokens": 260131092.0, + "step": 7408 + }, + { + "epoch": 1.3758588672237697, + "grad_norm": 1.4253849983215332, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8956833481788635, + "num_tokens": 260166094.0, + "step": 7409 + }, + { + "epoch": 1.3760445682451254, + "grad_norm": 1.4306265115737915, + "learning_rate": 1e-06, + "loss": 0.3698, + "mean_token_accuracy": 0.8753926753997803, + "num_tokens": 260204404.0, + "step": 7410 + }, + { + "epoch": 1.376230269266481, + "grad_norm": 1.4937479496002197, + "learning_rate": 1e-06, + "loss": 0.4139, + "mean_token_accuracy": 0.8592125177383423, + "num_tokens": 260243717.0, + "step": 7411 + }, + { + "epoch": 1.3764159702878365, + "grad_norm": 1.4372011423110962, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.883004367351532, + "num_tokens": 260276716.0, + "step": 7412 + }, + { + "epoch": 1.3766016713091922, + "grad_norm": 1.6117428541183472, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8782122135162354, + "num_tokens": 260308343.0, + "step": 7413 + }, + { + "epoch": 1.376787372330548, + "grad_norm": 1.4769880771636963, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8730241060256958, + "num_tokens": 260345587.0, + "step": 7414 + }, + { + "epoch": 1.3769730733519034, + "grad_norm": 1.6934105157852173, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8689900040626526, + "num_tokens": 260375408.0, + "step": 7415 + }, + { + "epoch": 1.377158774373259, + "grad_norm": 1.442620038986206, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8860480785369873, + "num_tokens": 260410449.0, + "step": 7416 + }, + { + "epoch": 1.3773444753946147, + "grad_norm": 1.4104012250900269, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8842869997024536, + "num_tokens": 260450608.0, + "step": 7417 + }, + { + "epoch": 1.3775301764159704, + "grad_norm": 1.436166524887085, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8834031820297241, + "num_tokens": 260488862.0, + "step": 7418 + }, + { + "epoch": 1.377715877437326, + "grad_norm": 1.6456482410430908, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8768166899681091, + "num_tokens": 260517328.0, + "step": 7419 + }, + { + "epoch": 1.3779015784586814, + "grad_norm": 1.4456923007965088, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8737489581108093, + "num_tokens": 260552603.0, + "step": 7420 + }, + { + "epoch": 1.3780872794800372, + "grad_norm": 1.6717363595962524, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8744519948959351, + "num_tokens": 260582026.0, + "step": 7421 + }, + { + "epoch": 1.3782729805013927, + "grad_norm": 1.4376155138015747, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8872649669647217, + "num_tokens": 260617741.0, + "step": 7422 + }, + { + "epoch": 1.3784586815227484, + "grad_norm": 1.499977946281433, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8851549625396729, + "num_tokens": 260651264.0, + "step": 7423 + }, + { + "epoch": 1.378644382544104, + "grad_norm": 1.4330999851226807, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.871608316898346, + "num_tokens": 260689794.0, + "step": 7424 + }, + { + "epoch": 1.3788300835654597, + "grad_norm": 1.333597183227539, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8870912790298462, + "num_tokens": 260727774.0, + "step": 7425 + }, + { + "epoch": 1.3790157845868152, + "grad_norm": 1.5298808813095093, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8678609728813171, + "num_tokens": 260765162.0, + "step": 7426 + }, + { + "epoch": 1.379201485608171, + "grad_norm": 1.530369758605957, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.877692461013794, + "num_tokens": 260796844.0, + "step": 7427 + }, + { + "epoch": 1.3793871866295264, + "grad_norm": 1.5873790979385376, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8739727735519409, + "num_tokens": 260832436.0, + "step": 7428 + }, + { + "epoch": 1.3795728876508822, + "grad_norm": 1.4919265508651733, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8692489266395569, + "num_tokens": 260869197.0, + "step": 7429 + }, + { + "epoch": 1.3797585886722377, + "grad_norm": 1.585549235343933, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8797839283943176, + "num_tokens": 260899309.0, + "step": 7430 + }, + { + "epoch": 1.3799442896935934, + "grad_norm": 1.3179854154586792, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8719900846481323, + "num_tokens": 260940979.0, + "step": 7431 + }, + { + "epoch": 1.380129990714949, + "grad_norm": 1.6901103258132935, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8697046041488647, + "num_tokens": 260972722.0, + "step": 7432 + }, + { + "epoch": 1.3803156917363046, + "grad_norm": 1.4478570222854614, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8810541033744812, + "num_tokens": 261006855.0, + "step": 7433 + }, + { + "epoch": 1.3805013927576602, + "grad_norm": 1.481832504272461, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8660807609558105, + "num_tokens": 261045089.0, + "step": 7434 + }, + { + "epoch": 1.3806870937790157, + "grad_norm": 1.4740644693374634, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8766745924949646, + "num_tokens": 261080512.0, + "step": 7435 + }, + { + "epoch": 1.3808727948003714, + "grad_norm": 1.5748686790466309, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.88277667760849, + "num_tokens": 261108974.0, + "step": 7436 + }, + { + "epoch": 1.3810584958217271, + "grad_norm": 1.6139646768569946, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8748265504837036, + "num_tokens": 261141275.0, + "step": 7437 + }, + { + "epoch": 1.3812441968430826, + "grad_norm": 1.4342905282974243, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8743903636932373, + "num_tokens": 261181657.0, + "step": 7438 + }, + { + "epoch": 1.3814298978644381, + "grad_norm": 1.4678611755371094, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8831546306610107, + "num_tokens": 261216645.0, + "step": 7439 + }, + { + "epoch": 1.3816155988857939, + "grad_norm": 1.5637651681900024, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8836808800697327, + "num_tokens": 261249180.0, + "step": 7440 + }, + { + "epoch": 1.3818012999071496, + "grad_norm": 1.3943840265274048, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8955644369125366, + "num_tokens": 261283813.0, + "step": 7441 + }, + { + "epoch": 1.3819870009285051, + "grad_norm": 1.7471917867660522, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.8587369918823242, + "num_tokens": 261319393.0, + "step": 7442 + }, + { + "epoch": 1.3821727019498606, + "grad_norm": 1.6906986236572266, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8729047179222107, + "num_tokens": 261348683.0, + "step": 7443 + }, + { + "epoch": 1.3823584029712164, + "grad_norm": 1.5899392366409302, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8774413466453552, + "num_tokens": 261381920.0, + "step": 7444 + }, + { + "epoch": 1.3825441039925719, + "grad_norm": 1.4487662315368652, + "learning_rate": 1e-06, + "loss": 0.4131, + "mean_token_accuracy": 0.859387993812561, + "num_tokens": 261421302.0, + "step": 7445 + }, + { + "epoch": 1.3827298050139276, + "grad_norm": 1.4612089395523071, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8734485507011414, + "num_tokens": 261454937.0, + "step": 7446 + }, + { + "epoch": 1.3829155060352831, + "grad_norm": 1.4230678081512451, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8846094012260437, + "num_tokens": 261489099.0, + "step": 7447 + }, + { + "epoch": 1.3831012070566389, + "grad_norm": 1.5313812494277954, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8845582604408264, + "num_tokens": 261522694.0, + "step": 7448 + }, + { + "epoch": 1.3832869080779944, + "grad_norm": 1.4902877807617188, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8764405250549316, + "num_tokens": 261561073.0, + "step": 7449 + }, + { + "epoch": 1.38347260909935, + "grad_norm": 1.6346650123596191, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8668742775917053, + "num_tokens": 261592299.0, + "step": 7450 + }, + { + "epoch": 1.3836583101207056, + "grad_norm": 1.6181780099868774, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8710314035415649, + "num_tokens": 261621591.0, + "step": 7451 + }, + { + "epoch": 1.3838440111420613, + "grad_norm": 2.2527034282684326, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8811225891113281, + "num_tokens": 261652908.0, + "step": 7452 + }, + { + "epoch": 1.3840297121634169, + "grad_norm": 1.6518409252166748, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8723071217536926, + "num_tokens": 261683817.0, + "step": 7453 + }, + { + "epoch": 1.3842154131847726, + "grad_norm": 1.5328389406204224, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8728643655776978, + "num_tokens": 261716635.0, + "step": 7454 + }, + { + "epoch": 1.384401114206128, + "grad_norm": 1.4874112606048584, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8909686803817749, + "num_tokens": 261748412.0, + "step": 7455 + }, + { + "epoch": 1.3845868152274838, + "grad_norm": 1.4265806674957275, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8873459100723267, + "num_tokens": 261783446.0, + "step": 7456 + }, + { + "epoch": 1.3847725162488393, + "grad_norm": 1.6412169933319092, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8714538216590881, + "num_tokens": 261813306.0, + "step": 7457 + }, + { + "epoch": 1.3849582172701949, + "grad_norm": 1.5986006259918213, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.868372917175293, + "num_tokens": 261845601.0, + "step": 7458 + }, + { + "epoch": 1.3851439182915506, + "grad_norm": 1.5058948993682861, + "learning_rate": 1e-06, + "loss": 0.4081, + "mean_token_accuracy": 0.8637168407440186, + "num_tokens": 261882278.0, + "step": 7459 + }, + { + "epoch": 1.3853296193129063, + "grad_norm": 1.4155024290084839, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8685320615768433, + "num_tokens": 261924155.0, + "step": 7460 + }, + { + "epoch": 1.3855153203342618, + "grad_norm": 1.4946651458740234, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8763364553451538, + "num_tokens": 261955765.0, + "step": 7461 + }, + { + "epoch": 1.3857010213556173, + "grad_norm": 1.4936604499816895, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.875783383846283, + "num_tokens": 261990169.0, + "step": 7462 + }, + { + "epoch": 1.385886722376973, + "grad_norm": 1.5540449619293213, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8631210923194885, + "num_tokens": 262024567.0, + "step": 7463 + }, + { + "epoch": 1.3860724233983288, + "grad_norm": 1.5629713535308838, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8753451704978943, + "num_tokens": 262056579.0, + "step": 7464 + }, + { + "epoch": 1.3862581244196843, + "grad_norm": 1.5115444660186768, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8737168908119202, + "num_tokens": 262090833.0, + "step": 7465 + }, + { + "epoch": 1.3864438254410398, + "grad_norm": 1.5095363855361938, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8798269629478455, + "num_tokens": 262126010.0, + "step": 7466 + }, + { + "epoch": 1.3866295264623956, + "grad_norm": 1.407887578010559, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8843244314193726, + "num_tokens": 262163023.0, + "step": 7467 + }, + { + "epoch": 1.386815227483751, + "grad_norm": 1.4406685829162598, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8827831745147705, + "num_tokens": 262199772.0, + "step": 7468 + }, + { + "epoch": 1.3870009285051068, + "grad_norm": 1.4791465997695923, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8777544498443604, + "num_tokens": 262234813.0, + "step": 7469 + }, + { + "epoch": 1.3871866295264623, + "grad_norm": 1.4854297637939453, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8915231227874756, + "num_tokens": 262268367.0, + "step": 7470 + }, + { + "epoch": 1.387372330547818, + "grad_norm": 1.6305148601531982, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.868345320224762, + "num_tokens": 262303672.0, + "step": 7471 + }, + { + "epoch": 1.3875580315691736, + "grad_norm": 1.503833293914795, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8723132610321045, + "num_tokens": 262339934.0, + "step": 7472 + }, + { + "epoch": 1.3877437325905293, + "grad_norm": 1.455631136894226, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8731112480163574, + "num_tokens": 262375999.0, + "step": 7473 + }, + { + "epoch": 1.3879294336118848, + "grad_norm": 1.501875877380371, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8797847032546997, + "num_tokens": 262410764.0, + "step": 7474 + }, + { + "epoch": 1.3881151346332405, + "grad_norm": 1.5490310192108154, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8877274990081787, + "num_tokens": 262443790.0, + "step": 7475 + }, + { + "epoch": 1.388300835654596, + "grad_norm": 1.6116783618927002, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8778990507125854, + "num_tokens": 262476314.0, + "step": 7476 + }, + { + "epoch": 1.3884865366759518, + "grad_norm": 1.4520633220672607, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8682695627212524, + "num_tokens": 262516270.0, + "step": 7477 + }, + { + "epoch": 1.3886722376973073, + "grad_norm": 1.4032070636749268, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8757002949714661, + "num_tokens": 262556858.0, + "step": 7478 + }, + { + "epoch": 1.388857938718663, + "grad_norm": 1.4665690660476685, + "learning_rate": 1e-06, + "loss": 0.386, + "mean_token_accuracy": 0.8694848418235779, + "num_tokens": 262595608.0, + "step": 7479 + }, + { + "epoch": 1.3890436397400185, + "grad_norm": 1.4887731075286865, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.884734034538269, + "num_tokens": 262629095.0, + "step": 7480 + }, + { + "epoch": 1.389229340761374, + "grad_norm": 1.3477935791015625, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.887654185295105, + "num_tokens": 262665846.0, + "step": 7481 + }, + { + "epoch": 1.3894150417827298, + "grad_norm": 1.3850407600402832, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8754664659500122, + "num_tokens": 262710791.0, + "step": 7482 + }, + { + "epoch": 1.3896007428040855, + "grad_norm": 1.4918326139450073, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.8606674671173096, + "num_tokens": 262747909.0, + "step": 7483 + }, + { + "epoch": 1.389786443825441, + "grad_norm": 1.6130123138427734, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8705428838729858, + "num_tokens": 262778902.0, + "step": 7484 + }, + { + "epoch": 1.3899721448467965, + "grad_norm": 1.5162678956985474, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8732872009277344, + "num_tokens": 262812725.0, + "step": 7485 + }, + { + "epoch": 1.3901578458681523, + "grad_norm": 1.5942858457565308, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8735442161560059, + "num_tokens": 262845878.0, + "step": 7486 + }, + { + "epoch": 1.390343546889508, + "grad_norm": 1.7081423997879028, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8815404176712036, + "num_tokens": 262876125.0, + "step": 7487 + }, + { + "epoch": 1.3905292479108635, + "grad_norm": 1.6049360036849976, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8764476776123047, + "num_tokens": 262909745.0, + "step": 7488 + }, + { + "epoch": 1.390714948932219, + "grad_norm": 1.5340518951416016, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8825594782829285, + "num_tokens": 262944546.0, + "step": 7489 + }, + { + "epoch": 1.3909006499535748, + "grad_norm": 1.520839810371399, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8694249987602234, + "num_tokens": 262984591.0, + "step": 7490 + }, + { + "epoch": 1.3910863509749305, + "grad_norm": 1.436023473739624, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8823422193527222, + "num_tokens": 263019679.0, + "step": 7491 + }, + { + "epoch": 1.391272051996286, + "grad_norm": 1.5646158456802368, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8618297576904297, + "num_tokens": 263055614.0, + "step": 7492 + }, + { + "epoch": 1.3914577530176415, + "grad_norm": 1.387347936630249, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8805856704711914, + "num_tokens": 263094804.0, + "step": 7493 + }, + { + "epoch": 1.3916434540389973, + "grad_norm": 1.3969085216522217, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8851889371871948, + "num_tokens": 263134894.0, + "step": 7494 + }, + { + "epoch": 1.3918291550603528, + "grad_norm": 1.5094324350357056, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8683779239654541, + "num_tokens": 263173129.0, + "step": 7495 + }, + { + "epoch": 1.3920148560817085, + "grad_norm": 1.52604079246521, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.8894425630569458, + "num_tokens": 263202365.0, + "step": 7496 + }, + { + "epoch": 1.392200557103064, + "grad_norm": 1.4590473175048828, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8761811256408691, + "num_tokens": 263237633.0, + "step": 7497 + }, + { + "epoch": 1.3923862581244197, + "grad_norm": 1.5581830739974976, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8753741979598999, + "num_tokens": 263270220.0, + "step": 7498 + }, + { + "epoch": 1.3925719591457753, + "grad_norm": 1.369765281677246, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8895789384841919, + "num_tokens": 263307892.0, + "step": 7499 + }, + { + "epoch": 1.392757660167131, + "grad_norm": 1.5640243291854858, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8738341331481934, + "num_tokens": 263342437.0, + "step": 7500 + }, + { + "epoch": 1.3929433611884865, + "grad_norm": 1.4485952854156494, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8708933591842651, + "num_tokens": 263380080.0, + "step": 7501 + }, + { + "epoch": 1.3931290622098422, + "grad_norm": 1.4280283451080322, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8725396394729614, + "num_tokens": 263417780.0, + "step": 7502 + }, + { + "epoch": 1.3933147632311977, + "grad_norm": 1.3628641366958618, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8737423419952393, + "num_tokens": 263458571.0, + "step": 7503 + }, + { + "epoch": 1.3935004642525533, + "grad_norm": 1.454035997390747, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8784311413764954, + "num_tokens": 263495108.0, + "step": 7504 + }, + { + "epoch": 1.393686165273909, + "grad_norm": 1.395695686340332, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8823683857917786, + "num_tokens": 263534359.0, + "step": 7505 + }, + { + "epoch": 1.3938718662952647, + "grad_norm": 1.519870638847351, + "learning_rate": 1e-06, + "loss": 0.4223, + "mean_token_accuracy": 0.8566979169845581, + "num_tokens": 263571526.0, + "step": 7506 + }, + { + "epoch": 1.3940575673166202, + "grad_norm": 1.5381678342819214, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8797707557678223, + "num_tokens": 263600059.0, + "step": 7507 + }, + { + "epoch": 1.3942432683379757, + "grad_norm": 1.4071053266525269, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8819308280944824, + "num_tokens": 263634263.0, + "step": 7508 + }, + { + "epoch": 1.3944289693593315, + "grad_norm": 1.6919701099395752, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8725324273109436, + "num_tokens": 263666380.0, + "step": 7509 + }, + { + "epoch": 1.3946146703806872, + "grad_norm": 1.3953378200531006, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8818489909172058, + "num_tokens": 263704966.0, + "step": 7510 + }, + { + "epoch": 1.3948003714020427, + "grad_norm": 1.4277540445327759, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8722642660140991, + "num_tokens": 263742622.0, + "step": 7511 + }, + { + "epoch": 1.3949860724233982, + "grad_norm": 1.484966516494751, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8759746551513672, + "num_tokens": 263781432.0, + "step": 7512 + }, + { + "epoch": 1.395171773444754, + "grad_norm": 1.6694825887680054, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8821408748626709, + "num_tokens": 263808226.0, + "step": 7513 + }, + { + "epoch": 1.3953574744661097, + "grad_norm": 1.6115388870239258, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8643817901611328, + "num_tokens": 263842219.0, + "step": 7514 + }, + { + "epoch": 1.3955431754874652, + "grad_norm": 1.3284953832626343, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8883764743804932, + "num_tokens": 263884568.0, + "step": 7515 + }, + { + "epoch": 1.3957288765088207, + "grad_norm": 1.3986103534698486, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8754653930664062, + "num_tokens": 263927557.0, + "step": 7516 + }, + { + "epoch": 1.3959145775301764, + "grad_norm": 1.4977409839630127, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8861355185508728, + "num_tokens": 263958083.0, + "step": 7517 + }, + { + "epoch": 1.396100278551532, + "grad_norm": 1.538217306137085, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8799409866333008, + "num_tokens": 263992909.0, + "step": 7518 + }, + { + "epoch": 1.3962859795728877, + "grad_norm": 1.5487860441207886, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8812863826751709, + "num_tokens": 264027391.0, + "step": 7519 + }, + { + "epoch": 1.3964716805942432, + "grad_norm": 1.4465265274047852, + "learning_rate": 1e-06, + "loss": 0.4157, + "mean_token_accuracy": 0.8581526279449463, + "num_tokens": 264065904.0, + "step": 7520 + }, + { + "epoch": 1.396657381615599, + "grad_norm": 1.3890622854232788, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8958984613418579, + "num_tokens": 264099254.0, + "step": 7521 + }, + { + "epoch": 1.3968430826369544, + "grad_norm": 1.4325891733169556, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8805328607559204, + "num_tokens": 264133588.0, + "step": 7522 + }, + { + "epoch": 1.3970287836583102, + "grad_norm": 1.4749552011489868, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8701344728469849, + "num_tokens": 264171632.0, + "step": 7523 + }, + { + "epoch": 1.3972144846796657, + "grad_norm": 1.5635502338409424, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8864423632621765, + "num_tokens": 264202109.0, + "step": 7524 + }, + { + "epoch": 1.3974001857010214, + "grad_norm": 1.42499840259552, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8931768536567688, + "num_tokens": 264236182.0, + "step": 7525 + }, + { + "epoch": 1.397585886722377, + "grad_norm": 1.539725661277771, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8755176067352295, + "num_tokens": 264270748.0, + "step": 7526 + }, + { + "epoch": 1.3977715877437327, + "grad_norm": 1.401038646697998, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8767731785774231, + "num_tokens": 264307282.0, + "step": 7527 + }, + { + "epoch": 1.3979572887650882, + "grad_norm": 1.452049732208252, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8818567991256714, + "num_tokens": 264343351.0, + "step": 7528 + }, + { + "epoch": 1.398142989786444, + "grad_norm": 1.5566102266311646, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.860680103302002, + "num_tokens": 264375943.0, + "step": 7529 + }, + { + "epoch": 1.3983286908077994, + "grad_norm": 1.5293220281600952, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.880870521068573, + "num_tokens": 264406459.0, + "step": 7530 + }, + { + "epoch": 1.398514391829155, + "grad_norm": 1.52287757396698, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8735253214836121, + "num_tokens": 264442318.0, + "step": 7531 + }, + { + "epoch": 1.3987000928505107, + "grad_norm": 1.3799153566360474, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8888761401176453, + "num_tokens": 264479816.0, + "step": 7532 + }, + { + "epoch": 1.3988857938718664, + "grad_norm": 1.5175665616989136, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8649206161499023, + "num_tokens": 264515199.0, + "step": 7533 + }, + { + "epoch": 1.399071494893222, + "grad_norm": 1.6261967420578003, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8631899356842041, + "num_tokens": 264550223.0, + "step": 7534 + }, + { + "epoch": 1.3992571959145774, + "grad_norm": 1.5425182580947876, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8852131366729736, + "num_tokens": 264581281.0, + "step": 7535 + }, + { + "epoch": 1.3994428969359332, + "grad_norm": 1.6756796836853027, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.883282482624054, + "num_tokens": 264605980.0, + "step": 7536 + }, + { + "epoch": 1.399628597957289, + "grad_norm": 1.5228376388549805, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8719585537910461, + "num_tokens": 264639844.0, + "step": 7537 + }, + { + "epoch": 1.3998142989786444, + "grad_norm": 1.4876354932785034, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8919283151626587, + "num_tokens": 264674202.0, + "step": 7538 + }, + { + "epoch": 1.4, + "grad_norm": 1.5288408994674683, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8774250149726868, + "num_tokens": 264706804.0, + "step": 7539 + }, + { + "epoch": 1.4001857010213556, + "grad_norm": 1.3885078430175781, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8885660171508789, + "num_tokens": 264742951.0, + "step": 7540 + }, + { + "epoch": 1.4003714020427112, + "grad_norm": 1.4053364992141724, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8765897750854492, + "num_tokens": 264779326.0, + "step": 7541 + }, + { + "epoch": 1.4005571030640669, + "grad_norm": 1.5078412294387817, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8801059722900391, + "num_tokens": 264810918.0, + "step": 7542 + }, + { + "epoch": 1.4007428040854224, + "grad_norm": 1.619089126586914, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8767368793487549, + "num_tokens": 264845142.0, + "step": 7543 + }, + { + "epoch": 1.4009285051067781, + "grad_norm": 1.4885680675506592, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8676714301109314, + "num_tokens": 264881330.0, + "step": 7544 + }, + { + "epoch": 1.4011142061281336, + "grad_norm": 1.5603102445602417, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8818730115890503, + "num_tokens": 264914180.0, + "step": 7545 + }, + { + "epoch": 1.4012999071494894, + "grad_norm": 1.3885358572006226, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8848965764045715, + "num_tokens": 264953943.0, + "step": 7546 + }, + { + "epoch": 1.4014856081708449, + "grad_norm": 1.6610645055770874, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8724033832550049, + "num_tokens": 264984946.0, + "step": 7547 + }, + { + "epoch": 1.4016713091922006, + "grad_norm": 1.525159478187561, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8840349912643433, + "num_tokens": 265015357.0, + "step": 7548 + }, + { + "epoch": 1.4018570102135561, + "grad_norm": 1.619277000427246, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8666062355041504, + "num_tokens": 265049724.0, + "step": 7549 + }, + { + "epoch": 1.4020427112349119, + "grad_norm": 1.4742815494537354, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8656927347183228, + "num_tokens": 265085853.0, + "step": 7550 + }, + { + "epoch": 1.4022284122562674, + "grad_norm": 1.5368436574935913, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8844053745269775, + "num_tokens": 265117057.0, + "step": 7551 + }, + { + "epoch": 1.402414113277623, + "grad_norm": 1.3640432357788086, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8714529871940613, + "num_tokens": 265159218.0, + "step": 7552 + }, + { + "epoch": 1.4025998142989786, + "grad_norm": 1.4886616468429565, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8701147437095642, + "num_tokens": 265195699.0, + "step": 7553 + }, + { + "epoch": 1.4027855153203341, + "grad_norm": 1.325021743774414, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8868393898010254, + "num_tokens": 265235191.0, + "step": 7554 + }, + { + "epoch": 1.4029712163416899, + "grad_norm": 1.3934773206710815, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8815566301345825, + "num_tokens": 265275571.0, + "step": 7555 + }, + { + "epoch": 1.4031569173630456, + "grad_norm": 1.486396074295044, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8628157377243042, + "num_tokens": 265312476.0, + "step": 7556 + }, + { + "epoch": 1.403342618384401, + "grad_norm": 1.5187695026397705, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8653030395507812, + "num_tokens": 265350884.0, + "step": 7557 + }, + { + "epoch": 1.4035283194057566, + "grad_norm": 1.4581443071365356, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8814738392829895, + "num_tokens": 265392068.0, + "step": 7558 + }, + { + "epoch": 1.4037140204271124, + "grad_norm": 1.4805675745010376, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8648391962051392, + "num_tokens": 265429348.0, + "step": 7559 + }, + { + "epoch": 1.403899721448468, + "grad_norm": 1.9315249919891357, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.889039158821106, + "num_tokens": 265466123.0, + "step": 7560 + }, + { + "epoch": 1.4040854224698236, + "grad_norm": 1.4953068494796753, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8851772546768188, + "num_tokens": 265497916.0, + "step": 7561 + }, + { + "epoch": 1.404271123491179, + "grad_norm": 1.578315019607544, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.888079047203064, + "num_tokens": 265529262.0, + "step": 7562 + }, + { + "epoch": 1.4044568245125348, + "grad_norm": 1.465023159980774, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8873566389083862, + "num_tokens": 265563738.0, + "step": 7563 + }, + { + "epoch": 1.4046425255338906, + "grad_norm": 1.5013667345046997, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8858309984207153, + "num_tokens": 265596903.0, + "step": 7564 + }, + { + "epoch": 1.404828226555246, + "grad_norm": 1.4951541423797607, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8762348890304565, + "num_tokens": 265630945.0, + "step": 7565 + }, + { + "epoch": 1.4050139275766016, + "grad_norm": 1.474574327468872, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8759381175041199, + "num_tokens": 265667051.0, + "step": 7566 + }, + { + "epoch": 1.4051996285979573, + "grad_norm": 1.266737937927246, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8867325782775879, + "num_tokens": 265713861.0, + "step": 7567 + }, + { + "epoch": 1.4053853296193128, + "grad_norm": 1.68331778049469, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8717016577720642, + "num_tokens": 265745410.0, + "step": 7568 + }, + { + "epoch": 1.4055710306406686, + "grad_norm": 1.5292949676513672, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8903167843818665, + "num_tokens": 265777966.0, + "step": 7569 + }, + { + "epoch": 1.405756731662024, + "grad_norm": 1.568605899810791, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8725827932357788, + "num_tokens": 265812515.0, + "step": 7570 + }, + { + "epoch": 1.4059424326833798, + "grad_norm": 1.5468025207519531, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8739715814590454, + "num_tokens": 265849143.0, + "step": 7571 + }, + { + "epoch": 1.4061281337047353, + "grad_norm": 1.5854871273040771, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8742327690124512, + "num_tokens": 265879236.0, + "step": 7572 + }, + { + "epoch": 1.406313834726091, + "grad_norm": 1.5578240156173706, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8753475546836853, + "num_tokens": 265912172.0, + "step": 7573 + }, + { + "epoch": 1.4064995357474466, + "grad_norm": 1.6259022951126099, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.876316487789154, + "num_tokens": 265943845.0, + "step": 7574 + }, + { + "epoch": 1.4066852367688023, + "grad_norm": 1.3724735975265503, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8802396059036255, + "num_tokens": 265982920.0, + "step": 7575 + }, + { + "epoch": 1.4068709377901578, + "grad_norm": 1.6233989000320435, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.872299313545227, + "num_tokens": 266014955.0, + "step": 7576 + }, + { + "epoch": 1.4070566388115133, + "grad_norm": 1.4529755115509033, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8812152147293091, + "num_tokens": 266049030.0, + "step": 7577 + }, + { + "epoch": 1.407242339832869, + "grad_norm": 1.5147558450698853, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8745139837265015, + "num_tokens": 266083743.0, + "step": 7578 + }, + { + "epoch": 1.4074280408542248, + "grad_norm": 1.5205936431884766, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8800035715103149, + "num_tokens": 266118103.0, + "step": 7579 + }, + { + "epoch": 1.4076137418755803, + "grad_norm": 1.3859740495681763, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8860087990760803, + "num_tokens": 266157471.0, + "step": 7580 + }, + { + "epoch": 1.4077994428969358, + "grad_norm": 1.5337382555007935, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8731132745742798, + "num_tokens": 266189373.0, + "step": 7581 + }, + { + "epoch": 1.4079851439182915, + "grad_norm": 1.4960827827453613, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8695962429046631, + "num_tokens": 266229771.0, + "step": 7582 + }, + { + "epoch": 1.4081708449396473, + "grad_norm": 1.678999662399292, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8778105974197388, + "num_tokens": 266260329.0, + "step": 7583 + }, + { + "epoch": 1.4083565459610028, + "grad_norm": 1.4220064878463745, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8726546168327332, + "num_tokens": 266299662.0, + "step": 7584 + }, + { + "epoch": 1.4085422469823583, + "grad_norm": 1.5830678939819336, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8714194297790527, + "num_tokens": 266332477.0, + "step": 7585 + }, + { + "epoch": 1.408727948003714, + "grad_norm": 1.5254451036453247, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8910104036331177, + "num_tokens": 266365961.0, + "step": 7586 + }, + { + "epoch": 1.4089136490250698, + "grad_norm": 1.6502091884613037, + "learning_rate": 1e-06, + "loss": 0.4507, + "mean_token_accuracy": 0.8520156145095825, + "num_tokens": 266400593.0, + "step": 7587 + }, + { + "epoch": 1.4090993500464253, + "grad_norm": 1.3967047929763794, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.879864513874054, + "num_tokens": 266439448.0, + "step": 7588 + }, + { + "epoch": 1.4092850510677808, + "grad_norm": 1.565908670425415, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8787035942077637, + "num_tokens": 266472350.0, + "step": 7589 + }, + { + "epoch": 1.4094707520891365, + "grad_norm": 1.5033681392669678, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.874634861946106, + "num_tokens": 266509380.0, + "step": 7590 + }, + { + "epoch": 1.409656453110492, + "grad_norm": 1.283769130706787, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8770546913146973, + "num_tokens": 266551950.0, + "step": 7591 + }, + { + "epoch": 1.4098421541318478, + "grad_norm": 1.5559017658233643, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8803632259368896, + "num_tokens": 266583336.0, + "step": 7592 + }, + { + "epoch": 1.4100278551532033, + "grad_norm": 1.5775985717773438, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8600639700889587, + "num_tokens": 266620915.0, + "step": 7593 + }, + { + "epoch": 1.410213556174559, + "grad_norm": 1.7219456434249878, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8773842453956604, + "num_tokens": 266649443.0, + "step": 7594 + }, + { + "epoch": 1.4103992571959145, + "grad_norm": 1.41108238697052, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8830256462097168, + "num_tokens": 266686710.0, + "step": 7595 + }, + { + "epoch": 1.4105849582172703, + "grad_norm": 1.52544105052948, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8798925876617432, + "num_tokens": 266720662.0, + "step": 7596 + }, + { + "epoch": 1.4107706592386258, + "grad_norm": 1.3661863803863525, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8828986287117004, + "num_tokens": 266761831.0, + "step": 7597 + }, + { + "epoch": 1.4109563602599815, + "grad_norm": 1.443435788154602, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8736166954040527, + "num_tokens": 266797817.0, + "step": 7598 + }, + { + "epoch": 1.411142061281337, + "grad_norm": 1.6319631338119507, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8606106042861938, + "num_tokens": 266827226.0, + "step": 7599 + }, + { + "epoch": 1.4113277623026927, + "grad_norm": 1.5505218505859375, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8594180345535278, + "num_tokens": 266863759.0, + "step": 7600 + }, + { + "epoch": 1.4115134633240483, + "grad_norm": 1.4718921184539795, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8702303171157837, + "num_tokens": 266902051.0, + "step": 7601 + }, + { + "epoch": 1.411699164345404, + "grad_norm": 1.62909734249115, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8808744549751282, + "num_tokens": 266930201.0, + "step": 7602 + }, + { + "epoch": 1.4118848653667595, + "grad_norm": 1.474816083908081, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8769447207450867, + "num_tokens": 266964735.0, + "step": 7603 + }, + { + "epoch": 1.412070566388115, + "grad_norm": 1.5140571594238281, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8757971525192261, + "num_tokens": 266998857.0, + "step": 7604 + }, + { + "epoch": 1.4122562674094707, + "grad_norm": 1.576692819595337, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.883165180683136, + "num_tokens": 267030969.0, + "step": 7605 + }, + { + "epoch": 1.4124419684308265, + "grad_norm": 1.5497102737426758, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8768221139907837, + "num_tokens": 267065934.0, + "step": 7606 + }, + { + "epoch": 1.412627669452182, + "grad_norm": 1.6295853853225708, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8682524561882019, + "num_tokens": 267098297.0, + "step": 7607 + }, + { + "epoch": 1.4128133704735375, + "grad_norm": 1.3659377098083496, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8965840935707092, + "num_tokens": 267135283.0, + "step": 7608 + }, + { + "epoch": 1.4129990714948932, + "grad_norm": 1.5628929138183594, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8622815608978271, + "num_tokens": 267167958.0, + "step": 7609 + }, + { + "epoch": 1.413184772516249, + "grad_norm": 1.3792270421981812, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8704497814178467, + "num_tokens": 267210768.0, + "step": 7610 + }, + { + "epoch": 1.4133704735376045, + "grad_norm": 1.5075047016143799, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.879209041595459, + "num_tokens": 267245894.0, + "step": 7611 + }, + { + "epoch": 1.41355617455896, + "grad_norm": 1.523620843887329, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.872367262840271, + "num_tokens": 267282456.0, + "step": 7612 + }, + { + "epoch": 1.4137418755803157, + "grad_norm": 1.355169653892517, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.879507839679718, + "num_tokens": 267324333.0, + "step": 7613 + }, + { + "epoch": 1.4139275766016712, + "grad_norm": 1.453554391860962, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8826261758804321, + "num_tokens": 267359972.0, + "step": 7614 + }, + { + "epoch": 1.414113277623027, + "grad_norm": 1.5095027685165405, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8751025199890137, + "num_tokens": 267392552.0, + "step": 7615 + }, + { + "epoch": 1.4142989786443825, + "grad_norm": 1.4125072956085205, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8749122619628906, + "num_tokens": 267431844.0, + "step": 7616 + }, + { + "epoch": 1.4144846796657382, + "grad_norm": 1.648512601852417, + "learning_rate": 1e-06, + "loss": 0.4179, + "mean_token_accuracy": 0.8578978180885315, + "num_tokens": 267464198.0, + "step": 7617 + }, + { + "epoch": 1.4146703806870937, + "grad_norm": 1.6334465742111206, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8730882406234741, + "num_tokens": 267494958.0, + "step": 7618 + }, + { + "epoch": 1.4148560817084495, + "grad_norm": 1.5496622323989868, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8799084424972534, + "num_tokens": 267529314.0, + "step": 7619 + }, + { + "epoch": 1.415041782729805, + "grad_norm": 1.4806020259857178, + "learning_rate": 1e-06, + "loss": 0.4038, + "mean_token_accuracy": 0.8627737760543823, + "num_tokens": 267565807.0, + "step": 7620 + }, + { + "epoch": 1.4152274837511607, + "grad_norm": 1.5665268898010254, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.870361328125, + "num_tokens": 267599218.0, + "step": 7621 + }, + { + "epoch": 1.4154131847725162, + "grad_norm": 1.478904128074646, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8837367296218872, + "num_tokens": 267630266.0, + "step": 7622 + }, + { + "epoch": 1.415598885793872, + "grad_norm": 1.645576000213623, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8799902200698853, + "num_tokens": 267659059.0, + "step": 7623 + }, + { + "epoch": 1.4157845868152275, + "grad_norm": 1.4730303287506104, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8771584033966064, + "num_tokens": 267698105.0, + "step": 7624 + }, + { + "epoch": 1.4159702878365832, + "grad_norm": 1.4299818277359009, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8736318349838257, + "num_tokens": 267738001.0, + "step": 7625 + }, + { + "epoch": 1.4161559888579387, + "grad_norm": 1.5205258131027222, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8750878572463989, + "num_tokens": 267770562.0, + "step": 7626 + }, + { + "epoch": 1.4163416898792942, + "grad_norm": 1.4493228197097778, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.878480076789856, + "num_tokens": 267805867.0, + "step": 7627 + }, + { + "epoch": 1.41652739090065, + "grad_norm": 1.6741527318954468, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8741980195045471, + "num_tokens": 267837790.0, + "step": 7628 + }, + { + "epoch": 1.4167130919220057, + "grad_norm": 1.560045599937439, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8792141079902649, + "num_tokens": 267869784.0, + "step": 7629 + }, + { + "epoch": 1.4168987929433612, + "grad_norm": 1.7173734903335571, + "learning_rate": 1e-06, + "loss": 0.386, + "mean_token_accuracy": 0.8736538887023926, + "num_tokens": 267900727.0, + "step": 7630 + }, + { + "epoch": 1.4170844939647167, + "grad_norm": 1.471544623374939, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8920489549636841, + "num_tokens": 267930565.0, + "step": 7631 + }, + { + "epoch": 1.4172701949860724, + "grad_norm": 1.5218257904052734, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8727339506149292, + "num_tokens": 267968002.0, + "step": 7632 + }, + { + "epoch": 1.4174558960074282, + "grad_norm": 1.609009861946106, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8808162212371826, + "num_tokens": 267997846.0, + "step": 7633 + }, + { + "epoch": 1.4176415970287837, + "grad_norm": 1.6504443883895874, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.865929126739502, + "num_tokens": 268031612.0, + "step": 7634 + }, + { + "epoch": 1.4178272980501392, + "grad_norm": 1.4882044792175293, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8837296962738037, + "num_tokens": 268064675.0, + "step": 7635 + }, + { + "epoch": 1.418012999071495, + "grad_norm": 1.4611083269119263, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.8727437257766724, + "num_tokens": 268100990.0, + "step": 7636 + }, + { + "epoch": 1.4181987000928504, + "grad_norm": 1.4240506887435913, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.881665825843811, + "num_tokens": 268138101.0, + "step": 7637 + }, + { + "epoch": 1.4183844011142062, + "grad_norm": 1.5357027053833008, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8843895196914673, + "num_tokens": 268171410.0, + "step": 7638 + }, + { + "epoch": 1.4185701021355617, + "grad_norm": 1.41767156124115, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.892546534538269, + "num_tokens": 268203409.0, + "step": 7639 + }, + { + "epoch": 1.4187558031569174, + "grad_norm": 1.4821268320083618, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8708658218383789, + "num_tokens": 268240056.0, + "step": 7640 + }, + { + "epoch": 1.418941504178273, + "grad_norm": 1.5151951313018799, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8713625073432922, + "num_tokens": 268278559.0, + "step": 7641 + }, + { + "epoch": 1.4191272051996286, + "grad_norm": 1.3928850889205933, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8792219161987305, + "num_tokens": 268319237.0, + "step": 7642 + }, + { + "epoch": 1.4193129062209842, + "grad_norm": 1.5626933574676514, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8742803335189819, + "num_tokens": 268350419.0, + "step": 7643 + }, + { + "epoch": 1.41949860724234, + "grad_norm": 1.4821727275848389, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8818815350532532, + "num_tokens": 268390208.0, + "step": 7644 + }, + { + "epoch": 1.4196843082636954, + "grad_norm": 1.4574906826019287, + "learning_rate": 1e-06, + "loss": 0.3986, + "mean_token_accuracy": 0.8652970790863037, + "num_tokens": 268426310.0, + "step": 7645 + }, + { + "epoch": 1.4198700092850511, + "grad_norm": 1.4582602977752686, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8866398930549622, + "num_tokens": 268457589.0, + "step": 7646 + }, + { + "epoch": 1.4200557103064066, + "grad_norm": 1.3482989072799683, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.879393994808197, + "num_tokens": 268497241.0, + "step": 7647 + }, + { + "epoch": 1.4202414113277624, + "grad_norm": 1.3399831056594849, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8774757385253906, + "num_tokens": 268541668.0, + "step": 7648 + }, + { + "epoch": 1.420427112349118, + "grad_norm": 1.5429877042770386, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8824813365936279, + "num_tokens": 268577276.0, + "step": 7649 + }, + { + "epoch": 1.4206128133704734, + "grad_norm": 1.557213544845581, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8787336349487305, + "num_tokens": 268607086.0, + "step": 7650 + }, + { + "epoch": 1.4207985143918291, + "grad_norm": 1.3881746530532837, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.897704005241394, + "num_tokens": 268642523.0, + "step": 7651 + }, + { + "epoch": 1.4209842154131849, + "grad_norm": 1.6997153759002686, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8751239776611328, + "num_tokens": 268670297.0, + "step": 7652 + }, + { + "epoch": 1.4211699164345404, + "grad_norm": 1.5494942665100098, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8795021772384644, + "num_tokens": 268704255.0, + "step": 7653 + }, + { + "epoch": 1.421355617455896, + "grad_norm": 1.5539058446884155, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8700703978538513, + "num_tokens": 268735097.0, + "step": 7654 + }, + { + "epoch": 1.4215413184772516, + "grad_norm": 1.5600477457046509, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8738424181938171, + "num_tokens": 268770319.0, + "step": 7655 + }, + { + "epoch": 1.4217270194986074, + "grad_norm": 1.556470274925232, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8842713236808777, + "num_tokens": 268805349.0, + "step": 7656 + }, + { + "epoch": 1.4219127205199629, + "grad_norm": 1.5142712593078613, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.876563310623169, + "num_tokens": 268839497.0, + "step": 7657 + }, + { + "epoch": 1.4220984215413184, + "grad_norm": 1.4433451890945435, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8722288012504578, + "num_tokens": 268877256.0, + "step": 7658 + }, + { + "epoch": 1.4222841225626741, + "grad_norm": 1.638689398765564, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8845407962799072, + "num_tokens": 268904798.0, + "step": 7659 + }, + { + "epoch": 1.4224698235840298, + "grad_norm": 1.6821489334106445, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8748583793640137, + "num_tokens": 268933352.0, + "step": 7660 + }, + { + "epoch": 1.4226555246053854, + "grad_norm": 1.5757097005844116, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8790098428726196, + "num_tokens": 268965081.0, + "step": 7661 + }, + { + "epoch": 1.4228412256267409, + "grad_norm": 1.475464940071106, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8706884384155273, + "num_tokens": 269009495.0, + "step": 7662 + }, + { + "epoch": 1.4230269266480966, + "grad_norm": 1.4863280057907104, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8715296387672424, + "num_tokens": 269050561.0, + "step": 7663 + }, + { + "epoch": 1.4232126276694521, + "grad_norm": 1.4648761749267578, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8898559212684631, + "num_tokens": 269084131.0, + "step": 7664 + }, + { + "epoch": 1.4233983286908078, + "grad_norm": 1.324554443359375, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8872484564781189, + "num_tokens": 269124740.0, + "step": 7665 + }, + { + "epoch": 1.4235840297121634, + "grad_norm": 1.504988670349121, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8894020318984985, + "num_tokens": 269159856.0, + "step": 7666 + }, + { + "epoch": 1.423769730733519, + "grad_norm": 1.519242286682129, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8760671615600586, + "num_tokens": 269196115.0, + "step": 7667 + }, + { + "epoch": 1.4239554317548746, + "grad_norm": 1.5104707479476929, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8759743571281433, + "num_tokens": 269234882.0, + "step": 7668 + }, + { + "epoch": 1.4241411327762303, + "grad_norm": 1.3696752786636353, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8830716609954834, + "num_tokens": 269277739.0, + "step": 7669 + }, + { + "epoch": 1.4243268337975858, + "grad_norm": 1.5841426849365234, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8708645105361938, + "num_tokens": 269312936.0, + "step": 7670 + }, + { + "epoch": 1.4245125348189416, + "grad_norm": 1.5631275177001953, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8794643878936768, + "num_tokens": 269346515.0, + "step": 7671 + }, + { + "epoch": 1.424698235840297, + "grad_norm": 1.4412791728973389, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8869819641113281, + "num_tokens": 269381644.0, + "step": 7672 + }, + { + "epoch": 1.4248839368616528, + "grad_norm": 1.5348505973815918, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8758199214935303, + "num_tokens": 269417194.0, + "step": 7673 + }, + { + "epoch": 1.4250696378830083, + "grad_norm": 1.6641793251037598, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.8687124252319336, + "num_tokens": 269447818.0, + "step": 7674 + }, + { + "epoch": 1.425255338904364, + "grad_norm": 1.5063087940216064, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.880069375038147, + "num_tokens": 269483688.0, + "step": 7675 + }, + { + "epoch": 1.4254410399257196, + "grad_norm": 1.5080817937850952, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8675899505615234, + "num_tokens": 269518819.0, + "step": 7676 + }, + { + "epoch": 1.425626740947075, + "grad_norm": 1.5577948093414307, + "learning_rate": 1e-06, + "loss": 0.3991, + "mean_token_accuracy": 0.8697459101676941, + "num_tokens": 269554012.0, + "step": 7677 + }, + { + "epoch": 1.4258124419684308, + "grad_norm": 1.6384986639022827, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8823015093803406, + "num_tokens": 269583402.0, + "step": 7678 + }, + { + "epoch": 1.4259981429897866, + "grad_norm": 1.616803526878357, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8728736042976379, + "num_tokens": 269617599.0, + "step": 7679 + }, + { + "epoch": 1.426183844011142, + "grad_norm": 1.3906419277191162, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8778247833251953, + "num_tokens": 269654043.0, + "step": 7680 + }, + { + "epoch": 1.4263695450324976, + "grad_norm": 1.5826950073242188, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8846649527549744, + "num_tokens": 269683948.0, + "step": 7681 + }, + { + "epoch": 1.4265552460538533, + "grad_norm": 1.4095780849456787, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.886414647102356, + "num_tokens": 269722419.0, + "step": 7682 + }, + { + "epoch": 1.426740947075209, + "grad_norm": 1.56141197681427, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8826894760131836, + "num_tokens": 269754562.0, + "step": 7683 + }, + { + "epoch": 1.4269266480965646, + "grad_norm": 1.7422477006912231, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8803277015686035, + "num_tokens": 269788198.0, + "step": 7684 + }, + { + "epoch": 1.42711234911792, + "grad_norm": 1.4902080297470093, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8619493842124939, + "num_tokens": 269829661.0, + "step": 7685 + }, + { + "epoch": 1.4272980501392758, + "grad_norm": 1.3945649862289429, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8835881948471069, + "num_tokens": 269868172.0, + "step": 7686 + }, + { + "epoch": 1.4274837511606313, + "grad_norm": 1.5199270248413086, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8805420994758606, + "num_tokens": 269905776.0, + "step": 7687 + }, + { + "epoch": 1.427669452181987, + "grad_norm": 1.4580919742584229, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8710688948631287, + "num_tokens": 269944055.0, + "step": 7688 + }, + { + "epoch": 1.4278551532033426, + "grad_norm": 1.3764744997024536, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.874081015586853, + "num_tokens": 269984476.0, + "step": 7689 + }, + { + "epoch": 1.4280408542246983, + "grad_norm": 1.559107780456543, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.8697224855422974, + "num_tokens": 270019756.0, + "step": 7690 + }, + { + "epoch": 1.4282265552460538, + "grad_norm": 1.605408787727356, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8662843704223633, + "num_tokens": 270051710.0, + "step": 7691 + }, + { + "epoch": 1.4284122562674095, + "grad_norm": 1.4599840641021729, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8767917156219482, + "num_tokens": 270092118.0, + "step": 7692 + }, + { + "epoch": 1.428597957288765, + "grad_norm": 1.3495081663131714, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8812308311462402, + "num_tokens": 270134659.0, + "step": 7693 + }, + { + "epoch": 1.4287836583101208, + "grad_norm": 1.561838984489441, + "learning_rate": 1e-06, + "loss": 0.294, + "mean_token_accuracy": 0.8983855843544006, + "num_tokens": 270159361.0, + "step": 7694 + }, + { + "epoch": 1.4289693593314763, + "grad_norm": 1.6460559368133545, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8730265498161316, + "num_tokens": 270192449.0, + "step": 7695 + }, + { + "epoch": 1.429155060352832, + "grad_norm": 1.3749966621398926, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8892017006874084, + "num_tokens": 270232250.0, + "step": 7696 + }, + { + "epoch": 1.4293407613741875, + "grad_norm": 1.4815682172775269, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8794822692871094, + "num_tokens": 270265482.0, + "step": 7697 + }, + { + "epoch": 1.4295264623955433, + "grad_norm": 1.432244062423706, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8732132911682129, + "num_tokens": 270300143.0, + "step": 7698 + }, + { + "epoch": 1.4297121634168988, + "grad_norm": 1.5369138717651367, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8829166293144226, + "num_tokens": 270330960.0, + "step": 7699 + }, + { + "epoch": 1.4298978644382543, + "grad_norm": 1.5272853374481201, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8703079223632812, + "num_tokens": 270369023.0, + "step": 7700 + }, + { + "epoch": 1.43008356545961, + "grad_norm": 1.4946097135543823, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8800040483474731, + "num_tokens": 270405346.0, + "step": 7701 + }, + { + "epoch": 1.4302692664809658, + "grad_norm": 1.5202795267105103, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8633126020431519, + "num_tokens": 270444235.0, + "step": 7702 + }, + { + "epoch": 1.4304549675023213, + "grad_norm": 1.608791470527649, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8744779825210571, + "num_tokens": 270475694.0, + "step": 7703 + }, + { + "epoch": 1.4306406685236768, + "grad_norm": 1.5467394590377808, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.870977520942688, + "num_tokens": 270510504.0, + "step": 7704 + }, + { + "epoch": 1.4308263695450325, + "grad_norm": 1.403624176979065, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8779428005218506, + "num_tokens": 270547187.0, + "step": 7705 + }, + { + "epoch": 1.4310120705663882, + "grad_norm": 1.6846369504928589, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8580185174942017, + "num_tokens": 270580908.0, + "step": 7706 + }, + { + "epoch": 1.4311977715877438, + "grad_norm": 1.415229082107544, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8864694237709045, + "num_tokens": 270615478.0, + "step": 7707 + }, + { + "epoch": 1.4313834726090993, + "grad_norm": 1.6635711193084717, + "learning_rate": 1e-06, + "loss": 0.4061, + "mean_token_accuracy": 0.8629851341247559, + "num_tokens": 270645161.0, + "step": 7708 + }, + { + "epoch": 1.431569173630455, + "grad_norm": 1.5382848978042603, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.865375280380249, + "num_tokens": 270682883.0, + "step": 7709 + }, + { + "epoch": 1.4317548746518105, + "grad_norm": 1.4955592155456543, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8894312381744385, + "num_tokens": 270714532.0, + "step": 7710 + }, + { + "epoch": 1.4319405756731662, + "grad_norm": 1.5267421007156372, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8679295778274536, + "num_tokens": 270749487.0, + "step": 7711 + }, + { + "epoch": 1.4321262766945217, + "grad_norm": 1.4872121810913086, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8735131025314331, + "num_tokens": 270784999.0, + "step": 7712 + }, + { + "epoch": 1.4323119777158775, + "grad_norm": 1.382510781288147, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8781927824020386, + "num_tokens": 270824599.0, + "step": 7713 + }, + { + "epoch": 1.432497678737233, + "grad_norm": 1.6624866724014282, + "learning_rate": 1e-06, + "loss": 0.3986, + "mean_token_accuracy": 0.8655664324760437, + "num_tokens": 270856043.0, + "step": 7714 + }, + { + "epoch": 1.4326833797585887, + "grad_norm": 1.4749326705932617, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8824465274810791, + "num_tokens": 270890265.0, + "step": 7715 + }, + { + "epoch": 1.4328690807799442, + "grad_norm": 1.6351261138916016, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8548867702484131, + "num_tokens": 270921763.0, + "step": 7716 + }, + { + "epoch": 1.4330547818013, + "grad_norm": 1.3317011594772339, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8928152322769165, + "num_tokens": 270961735.0, + "step": 7717 + }, + { + "epoch": 1.4332404828226555, + "grad_norm": 1.5819041728973389, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8819964528083801, + "num_tokens": 270994356.0, + "step": 7718 + }, + { + "epoch": 1.4334261838440112, + "grad_norm": 1.514344573020935, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8751965761184692, + "num_tokens": 271029910.0, + "step": 7719 + }, + { + "epoch": 1.4336118848653667, + "grad_norm": 1.3984180688858032, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8866023421287537, + "num_tokens": 271067135.0, + "step": 7720 + }, + { + "epoch": 1.4337975858867225, + "grad_norm": 1.5427552461624146, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8666256666183472, + "num_tokens": 271102997.0, + "step": 7721 + }, + { + "epoch": 1.433983286908078, + "grad_norm": 1.3357861042022705, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8852874636650085, + "num_tokens": 271142488.0, + "step": 7722 + }, + { + "epoch": 1.4341689879294335, + "grad_norm": 1.3952381610870361, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8894164562225342, + "num_tokens": 271177485.0, + "step": 7723 + }, + { + "epoch": 1.4343546889507892, + "grad_norm": 1.4951064586639404, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8725846409797668, + "num_tokens": 271212033.0, + "step": 7724 + }, + { + "epoch": 1.434540389972145, + "grad_norm": 1.478616714477539, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8709205389022827, + "num_tokens": 271249491.0, + "step": 7725 + }, + { + "epoch": 1.4347260909935005, + "grad_norm": 1.65109121799469, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8771011829376221, + "num_tokens": 271279148.0, + "step": 7726 + }, + { + "epoch": 1.434911792014856, + "grad_norm": 1.544666051864624, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.875504195690155, + "num_tokens": 271314790.0, + "step": 7727 + }, + { + "epoch": 1.4350974930362117, + "grad_norm": 1.4420124292373657, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8711681365966797, + "num_tokens": 271350644.0, + "step": 7728 + }, + { + "epoch": 1.4352831940575674, + "grad_norm": 1.4639008045196533, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8748271465301514, + "num_tokens": 271386822.0, + "step": 7729 + }, + { + "epoch": 1.435468895078923, + "grad_norm": 1.4541215896606445, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8831593990325928, + "num_tokens": 271425186.0, + "step": 7730 + }, + { + "epoch": 1.4356545961002785, + "grad_norm": 1.7689383029937744, + "learning_rate": 1e-06, + "loss": 0.4273, + "mean_token_accuracy": 0.8658864498138428, + "num_tokens": 271451162.0, + "step": 7731 + }, + { + "epoch": 1.4358402971216342, + "grad_norm": 1.4597196578979492, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8814316987991333, + "num_tokens": 271487228.0, + "step": 7732 + }, + { + "epoch": 1.43602599814299, + "grad_norm": 1.5578418970108032, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.861915111541748, + "num_tokens": 271525219.0, + "step": 7733 + }, + { + "epoch": 1.4362116991643454, + "grad_norm": 1.4813107252120972, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8676499128341675, + "num_tokens": 271565408.0, + "step": 7734 + }, + { + "epoch": 1.436397400185701, + "grad_norm": 1.7082914113998413, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8762590885162354, + "num_tokens": 271592601.0, + "step": 7735 + }, + { + "epoch": 1.4365831012070567, + "grad_norm": 1.5691436529159546, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8650685548782349, + "num_tokens": 271628157.0, + "step": 7736 + }, + { + "epoch": 1.4367688022284122, + "grad_norm": 1.4112884998321533, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8769407272338867, + "num_tokens": 271666191.0, + "step": 7737 + }, + { + "epoch": 1.436954503249768, + "grad_norm": 1.4324755668640137, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8805232048034668, + "num_tokens": 271705572.0, + "step": 7738 + }, + { + "epoch": 1.4371402042711234, + "grad_norm": 1.5492525100708008, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8708980083465576, + "num_tokens": 271742529.0, + "step": 7739 + }, + { + "epoch": 1.4373259052924792, + "grad_norm": 1.3911060094833374, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8885684013366699, + "num_tokens": 271779291.0, + "step": 7740 + }, + { + "epoch": 1.4375116063138347, + "grad_norm": 1.3783670663833618, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8751071095466614, + "num_tokens": 271818856.0, + "step": 7741 + }, + { + "epoch": 1.4376973073351904, + "grad_norm": 1.4519882202148438, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8644418120384216, + "num_tokens": 271858761.0, + "step": 7742 + }, + { + "epoch": 1.437883008356546, + "grad_norm": 1.5731358528137207, + "learning_rate": 1e-06, + "loss": 0.4205, + "mean_token_accuracy": 0.8596631288528442, + "num_tokens": 271893623.0, + "step": 7743 + }, + { + "epoch": 1.4380687093779017, + "grad_norm": 1.6193106174468994, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8696866035461426, + "num_tokens": 271925327.0, + "step": 7744 + }, + { + "epoch": 1.4382544103992572, + "grad_norm": 1.5877456665039062, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8746131658554077, + "num_tokens": 271958998.0, + "step": 7745 + }, + { + "epoch": 1.4384401114206127, + "grad_norm": 1.388381838798523, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8704007863998413, + "num_tokens": 271999659.0, + "step": 7746 + }, + { + "epoch": 1.4386258124419684, + "grad_norm": 1.607442021369934, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8823803663253784, + "num_tokens": 272031581.0, + "step": 7747 + }, + { + "epoch": 1.4388115134633241, + "grad_norm": 1.5995959043502808, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8748326301574707, + "num_tokens": 272061379.0, + "step": 7748 + }, + { + "epoch": 1.4389972144846797, + "grad_norm": 1.3991907835006714, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8826327919960022, + "num_tokens": 272098212.0, + "step": 7749 + }, + { + "epoch": 1.4391829155060352, + "grad_norm": 1.6968154907226562, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8722822666168213, + "num_tokens": 272130334.0, + "step": 7750 + }, + { + "epoch": 1.439368616527391, + "grad_norm": 1.5591973066329956, + "learning_rate": 1e-06, + "loss": 0.2996, + "mean_token_accuracy": 0.8941080570220947, + "num_tokens": 272161126.0, + "step": 7751 + }, + { + "epoch": 1.4395543175487466, + "grad_norm": 1.5882841348648071, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.8754261136054993, + "num_tokens": 272195578.0, + "step": 7752 + }, + { + "epoch": 1.4397400185701021, + "grad_norm": 1.5636334419250488, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8702868223190308, + "num_tokens": 272232440.0, + "step": 7753 + }, + { + "epoch": 1.4399257195914577, + "grad_norm": 1.3993703126907349, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.875572919845581, + "num_tokens": 272267668.0, + "step": 7754 + }, + { + "epoch": 1.4401114206128134, + "grad_norm": 1.4421179294586182, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8781371116638184, + "num_tokens": 272311876.0, + "step": 7755 + }, + { + "epoch": 1.4402971216341691, + "grad_norm": 1.4037647247314453, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8816689252853394, + "num_tokens": 272351181.0, + "step": 7756 + }, + { + "epoch": 1.4404828226555246, + "grad_norm": 1.4162462949752808, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8835886716842651, + "num_tokens": 272389719.0, + "step": 7757 + }, + { + "epoch": 1.4406685236768801, + "grad_norm": 1.7462974786758423, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.866480827331543, + "num_tokens": 272420312.0, + "step": 7758 + }, + { + "epoch": 1.4408542246982359, + "grad_norm": 1.4557815790176392, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8825797438621521, + "num_tokens": 272456522.0, + "step": 7759 + }, + { + "epoch": 1.4410399257195914, + "grad_norm": 1.7082951068878174, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8847178220748901, + "num_tokens": 272483823.0, + "step": 7760 + }, + { + "epoch": 1.4412256267409471, + "grad_norm": 1.4801509380340576, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8869487047195435, + "num_tokens": 272520002.0, + "step": 7761 + }, + { + "epoch": 1.4414113277623026, + "grad_norm": 1.7311371564865112, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8709216117858887, + "num_tokens": 272547914.0, + "step": 7762 + }, + { + "epoch": 1.4415970287836584, + "grad_norm": 1.58528733253479, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8718025088310242, + "num_tokens": 272582295.0, + "step": 7763 + }, + { + "epoch": 1.4417827298050139, + "grad_norm": 1.5193700790405273, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8819440603256226, + "num_tokens": 272616049.0, + "step": 7764 + }, + { + "epoch": 1.4419684308263696, + "grad_norm": 1.5256375074386597, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8730218410491943, + "num_tokens": 272653008.0, + "step": 7765 + }, + { + "epoch": 1.4421541318477251, + "grad_norm": 1.5567418336868286, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.879205584526062, + "num_tokens": 272683530.0, + "step": 7766 + }, + { + "epoch": 1.4423398328690809, + "grad_norm": 1.4020603895187378, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8813353776931763, + "num_tokens": 272724193.0, + "step": 7767 + }, + { + "epoch": 1.4425255338904364, + "grad_norm": 1.5877811908721924, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.8904932141304016, + "num_tokens": 272752908.0, + "step": 7768 + }, + { + "epoch": 1.442711234911792, + "grad_norm": 1.4959590435028076, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8790103793144226, + "num_tokens": 272785473.0, + "step": 7769 + }, + { + "epoch": 1.4428969359331476, + "grad_norm": 1.4535942077636719, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.886845588684082, + "num_tokens": 272820287.0, + "step": 7770 + }, + { + "epoch": 1.4430826369545033, + "grad_norm": 1.4644371271133423, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.873939037322998, + "num_tokens": 272859057.0, + "step": 7771 + }, + { + "epoch": 1.4432683379758589, + "grad_norm": 1.6095352172851562, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8762800097465515, + "num_tokens": 272888740.0, + "step": 7772 + }, + { + "epoch": 1.4434540389972144, + "grad_norm": 1.3507899045944214, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8779361248016357, + "num_tokens": 272930946.0, + "step": 7773 + }, + { + "epoch": 1.44363974001857, + "grad_norm": 1.5612879991531372, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.872433066368103, + "num_tokens": 272964016.0, + "step": 7774 + }, + { + "epoch": 1.4438254410399258, + "grad_norm": 1.5096981525421143, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8823016881942749, + "num_tokens": 272997941.0, + "step": 7775 + }, + { + "epoch": 1.4440111420612813, + "grad_norm": 1.2981629371643066, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8922037482261658, + "num_tokens": 273043478.0, + "step": 7776 + }, + { + "epoch": 1.4441968430826368, + "grad_norm": 1.5005756616592407, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8861439228057861, + "num_tokens": 273078812.0, + "step": 7777 + }, + { + "epoch": 1.4443825441039926, + "grad_norm": 1.3930059671401978, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8829852342605591, + "num_tokens": 273115361.0, + "step": 7778 + }, + { + "epoch": 1.4445682451253483, + "grad_norm": 1.4919596910476685, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8853211402893066, + "num_tokens": 273147479.0, + "step": 7779 + }, + { + "epoch": 1.4447539461467038, + "grad_norm": 1.3524178266525269, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8778343200683594, + "num_tokens": 273191035.0, + "step": 7780 + }, + { + "epoch": 1.4449396471680593, + "grad_norm": 1.4636374711990356, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8837665319442749, + "num_tokens": 273224454.0, + "step": 7781 + }, + { + "epoch": 1.445125348189415, + "grad_norm": 1.5656059980392456, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8705699443817139, + "num_tokens": 273260911.0, + "step": 7782 + }, + { + "epoch": 1.4453110492107706, + "grad_norm": 1.6419501304626465, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8764299154281616, + "num_tokens": 273291148.0, + "step": 7783 + }, + { + "epoch": 1.4454967502321263, + "grad_norm": 1.414720058441162, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8928858041763306, + "num_tokens": 273327017.0, + "step": 7784 + }, + { + "epoch": 1.4456824512534818, + "grad_norm": 1.3822165727615356, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8924728631973267, + "num_tokens": 273363097.0, + "step": 7785 + }, + { + "epoch": 1.4458681522748376, + "grad_norm": 1.5402108430862427, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8726767897605896, + "num_tokens": 273397720.0, + "step": 7786 + }, + { + "epoch": 1.446053853296193, + "grad_norm": 1.4113057851791382, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8896373510360718, + "num_tokens": 273434816.0, + "step": 7787 + }, + { + "epoch": 1.4462395543175488, + "grad_norm": 1.5676076412200928, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.873885989189148, + "num_tokens": 273469968.0, + "step": 7788 + }, + { + "epoch": 1.4464252553389043, + "grad_norm": 1.5586103200912476, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8840320110321045, + "num_tokens": 273501087.0, + "step": 7789 + }, + { + "epoch": 1.44661095636026, + "grad_norm": 1.5053589344024658, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8782312273979187, + "num_tokens": 273535321.0, + "step": 7790 + }, + { + "epoch": 1.4467966573816156, + "grad_norm": 1.4303131103515625, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8814147710800171, + "num_tokens": 273571978.0, + "step": 7791 + }, + { + "epoch": 1.4469823584029713, + "grad_norm": 1.5772104263305664, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8832249641418457, + "num_tokens": 273601328.0, + "step": 7792 + }, + { + "epoch": 1.4471680594243268, + "grad_norm": 1.5494900941848755, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8693749904632568, + "num_tokens": 273639465.0, + "step": 7793 + }, + { + "epoch": 1.4473537604456825, + "grad_norm": 1.4738515615463257, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8838306665420532, + "num_tokens": 273674017.0, + "step": 7794 + }, + { + "epoch": 1.447539461467038, + "grad_norm": 1.5356069803237915, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8735624551773071, + "num_tokens": 273708505.0, + "step": 7795 + }, + { + "epoch": 1.4477251624883936, + "grad_norm": 1.3790302276611328, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.873642086982727, + "num_tokens": 273751400.0, + "step": 7796 + }, + { + "epoch": 1.4479108635097493, + "grad_norm": 1.5457656383514404, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8901833295822144, + "num_tokens": 273779956.0, + "step": 7797 + }, + { + "epoch": 1.448096564531105, + "grad_norm": 1.5155583620071411, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8776547908782959, + "num_tokens": 273813353.0, + "step": 7798 + }, + { + "epoch": 1.4482822655524605, + "grad_norm": 1.5340150594711304, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8909578323364258, + "num_tokens": 273843751.0, + "step": 7799 + }, + { + "epoch": 1.448467966573816, + "grad_norm": 1.719474196434021, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8549597859382629, + "num_tokens": 273873393.0, + "step": 7800 + }, + { + "epoch": 1.4486536675951718, + "grad_norm": 1.5572935342788696, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8743153810501099, + "num_tokens": 273906501.0, + "step": 7801 + }, + { + "epoch": 1.4488393686165275, + "grad_norm": 1.6683964729309082, + "learning_rate": 1e-06, + "loss": 0.4051, + "mean_token_accuracy": 0.866195797920227, + "num_tokens": 273937173.0, + "step": 7802 + }, + { + "epoch": 1.449025069637883, + "grad_norm": 1.4547477960586548, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8732755780220032, + "num_tokens": 273975672.0, + "step": 7803 + }, + { + "epoch": 1.4492107706592385, + "grad_norm": 1.5490161180496216, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8762320876121521, + "num_tokens": 274009586.0, + "step": 7804 + }, + { + "epoch": 1.4493964716805943, + "grad_norm": 1.5215036869049072, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8781461715698242, + "num_tokens": 274042563.0, + "step": 7805 + }, + { + "epoch": 1.4495821727019498, + "grad_norm": 1.418962001800537, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8727816343307495, + "num_tokens": 274080111.0, + "step": 7806 + }, + { + "epoch": 1.4497678737233055, + "grad_norm": 1.6201930046081543, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8752308487892151, + "num_tokens": 274113836.0, + "step": 7807 + }, + { + "epoch": 1.449953574744661, + "grad_norm": 1.570918083190918, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8654584884643555, + "num_tokens": 274150651.0, + "step": 7808 + }, + { + "epoch": 1.4501392757660168, + "grad_norm": 1.4662580490112305, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8789350986480713, + "num_tokens": 274185646.0, + "step": 7809 + }, + { + "epoch": 1.4503249767873723, + "grad_norm": 1.6518206596374512, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.87929368019104, + "num_tokens": 274212409.0, + "step": 7810 + }, + { + "epoch": 1.450510677808728, + "grad_norm": 1.4912927150726318, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8781989812850952, + "num_tokens": 274246425.0, + "step": 7811 + }, + { + "epoch": 1.4506963788300835, + "grad_norm": 1.5502512454986572, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8788583278656006, + "num_tokens": 274279679.0, + "step": 7812 + }, + { + "epoch": 1.4508820798514392, + "grad_norm": 1.4755653142929077, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8812952041625977, + "num_tokens": 274314618.0, + "step": 7813 + }, + { + "epoch": 1.4510677808727948, + "grad_norm": 1.4475693702697754, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8795831203460693, + "num_tokens": 274350361.0, + "step": 7814 + }, + { + "epoch": 1.4512534818941505, + "grad_norm": 1.4274169206619263, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8769143223762512, + "num_tokens": 274387340.0, + "step": 7815 + }, + { + "epoch": 1.451439182915506, + "grad_norm": 1.5599806308746338, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8784863352775574, + "num_tokens": 274419899.0, + "step": 7816 + }, + { + "epoch": 1.4516248839368617, + "grad_norm": 1.437962532043457, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8753399848937988, + "num_tokens": 274457297.0, + "step": 7817 + }, + { + "epoch": 1.4518105849582172, + "grad_norm": 1.5179778337478638, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8667585849761963, + "num_tokens": 274491619.0, + "step": 7818 + }, + { + "epoch": 1.4519962859795728, + "grad_norm": 1.4432930946350098, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8729631304740906, + "num_tokens": 274531562.0, + "step": 7819 + }, + { + "epoch": 1.4521819870009285, + "grad_norm": 1.6461533308029175, + "learning_rate": 1e-06, + "loss": 0.4043, + "mean_token_accuracy": 0.863676130771637, + "num_tokens": 274562311.0, + "step": 7820 + }, + { + "epoch": 1.4523676880222842, + "grad_norm": 1.541426420211792, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8736208081245422, + "num_tokens": 274597303.0, + "step": 7821 + }, + { + "epoch": 1.4525533890436397, + "grad_norm": 1.4824191331863403, + "learning_rate": 1e-06, + "loss": 0.4084, + "mean_token_accuracy": 0.8603078722953796, + "num_tokens": 274634483.0, + "step": 7822 + }, + { + "epoch": 1.4527390900649952, + "grad_norm": 1.649731159210205, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8727843761444092, + "num_tokens": 274673183.0, + "step": 7823 + }, + { + "epoch": 1.452924791086351, + "grad_norm": 1.5041722059249878, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8816639184951782, + "num_tokens": 274710137.0, + "step": 7824 + }, + { + "epoch": 1.4531104921077067, + "grad_norm": 1.4813414812088013, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8954302072525024, + "num_tokens": 274740592.0, + "step": 7825 + }, + { + "epoch": 1.4532961931290622, + "grad_norm": 1.4962754249572754, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8830679655075073, + "num_tokens": 274773395.0, + "step": 7826 + }, + { + "epoch": 1.4534818941504177, + "grad_norm": 1.4974048137664795, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8796710968017578, + "num_tokens": 274808560.0, + "step": 7827 + }, + { + "epoch": 1.4536675951717735, + "grad_norm": 1.4755778312683105, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8786550760269165, + "num_tokens": 274843652.0, + "step": 7828 + }, + { + "epoch": 1.4538532961931292, + "grad_norm": 1.5777583122253418, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.879185140132904, + "num_tokens": 274877689.0, + "step": 7829 + }, + { + "epoch": 1.4540389972144847, + "grad_norm": 1.6674408912658691, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.879727840423584, + "num_tokens": 274910364.0, + "step": 7830 + }, + { + "epoch": 1.4542246982358402, + "grad_norm": 1.5891320705413818, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8688645362854004, + "num_tokens": 274943937.0, + "step": 7831 + }, + { + "epoch": 1.454410399257196, + "grad_norm": 1.5364694595336914, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8838702440261841, + "num_tokens": 274975098.0, + "step": 7832 + }, + { + "epoch": 1.4545961002785515, + "grad_norm": 1.4570269584655762, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8866767883300781, + "num_tokens": 275009099.0, + "step": 7833 + }, + { + "epoch": 1.4547818012999072, + "grad_norm": 1.529144525527954, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8772046566009521, + "num_tokens": 275045412.0, + "step": 7834 + }, + { + "epoch": 1.4549675023212627, + "grad_norm": 1.6994876861572266, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8609414100646973, + "num_tokens": 275075717.0, + "step": 7835 + }, + { + "epoch": 1.4551532033426184, + "grad_norm": 1.76446533203125, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8824838399887085, + "num_tokens": 275101374.0, + "step": 7836 + }, + { + "epoch": 1.455338904363974, + "grad_norm": 1.6391181945800781, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8725346326828003, + "num_tokens": 275134267.0, + "step": 7837 + }, + { + "epoch": 1.4555246053853297, + "grad_norm": 1.5321128368377686, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8834387063980103, + "num_tokens": 275164425.0, + "step": 7838 + }, + { + "epoch": 1.4557103064066852, + "grad_norm": 1.4972718954086304, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8740103840827942, + "num_tokens": 275201933.0, + "step": 7839 + }, + { + "epoch": 1.455896007428041, + "grad_norm": 1.522745966911316, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8749252557754517, + "num_tokens": 275237649.0, + "step": 7840 + }, + { + "epoch": 1.4560817084493964, + "grad_norm": 1.5414438247680664, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8682816624641418, + "num_tokens": 275273522.0, + "step": 7841 + }, + { + "epoch": 1.4562674094707522, + "grad_norm": 1.4988884925842285, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8718451261520386, + "num_tokens": 275309882.0, + "step": 7842 + }, + { + "epoch": 1.4564531104921077, + "grad_norm": 1.4385831356048584, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8730261325836182, + "num_tokens": 275348118.0, + "step": 7843 + }, + { + "epoch": 1.4566388115134634, + "grad_norm": 1.613892674446106, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8707386255264282, + "num_tokens": 275379889.0, + "step": 7844 + }, + { + "epoch": 1.456824512534819, + "grad_norm": 1.5602383613586426, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8810663223266602, + "num_tokens": 275410406.0, + "step": 7845 + }, + { + "epoch": 1.4570102135561744, + "grad_norm": 1.4941610097885132, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8803943395614624, + "num_tokens": 275444670.0, + "step": 7846 + }, + { + "epoch": 1.4571959145775302, + "grad_norm": 1.5891813039779663, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.869646430015564, + "num_tokens": 275480805.0, + "step": 7847 + }, + { + "epoch": 1.457381615598886, + "grad_norm": 1.5238032341003418, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8673035502433777, + "num_tokens": 275517584.0, + "step": 7848 + }, + { + "epoch": 1.4575673166202414, + "grad_norm": 1.4351948499679565, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8779258728027344, + "num_tokens": 275555661.0, + "step": 7849 + }, + { + "epoch": 1.457753017641597, + "grad_norm": 1.5317623615264893, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8919103145599365, + "num_tokens": 275587508.0, + "step": 7850 + }, + { + "epoch": 1.4579387186629527, + "grad_norm": 1.5234134197235107, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.875455915927887, + "num_tokens": 275619559.0, + "step": 7851 + }, + { + "epoch": 1.4581244196843084, + "grad_norm": 1.6079145669937134, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8791863918304443, + "num_tokens": 275655299.0, + "step": 7852 + }, + { + "epoch": 1.458310120705664, + "grad_norm": 1.4973373413085938, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8727270364761353, + "num_tokens": 275690071.0, + "step": 7853 + }, + { + "epoch": 1.4584958217270194, + "grad_norm": 1.5855679512023926, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8716968297958374, + "num_tokens": 275722116.0, + "step": 7854 + }, + { + "epoch": 1.4586815227483751, + "grad_norm": 1.5985357761383057, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8720935583114624, + "num_tokens": 275754443.0, + "step": 7855 + }, + { + "epoch": 1.4588672237697307, + "grad_norm": 1.5057551860809326, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8801752924919128, + "num_tokens": 275786969.0, + "step": 7856 + }, + { + "epoch": 1.4590529247910864, + "grad_norm": 1.4357922077178955, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8693222999572754, + "num_tokens": 275825277.0, + "step": 7857 + }, + { + "epoch": 1.459238625812442, + "grad_norm": 1.3927245140075684, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8796408176422119, + "num_tokens": 275866710.0, + "step": 7858 + }, + { + "epoch": 1.4594243268337976, + "grad_norm": 1.4872876405715942, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8875862956047058, + "num_tokens": 275902564.0, + "step": 7859 + }, + { + "epoch": 1.4596100278551531, + "grad_norm": 1.4380779266357422, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8836141228675842, + "num_tokens": 275934287.0, + "step": 7860 + }, + { + "epoch": 1.4597957288765089, + "grad_norm": 1.7144222259521484, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8598193526268005, + "num_tokens": 275963866.0, + "step": 7861 + }, + { + "epoch": 1.4599814298978644, + "grad_norm": 1.53695809841156, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8930325508117676, + "num_tokens": 275996165.0, + "step": 7862 + }, + { + "epoch": 1.4601671309192201, + "grad_norm": 1.473890781402588, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8702653646469116, + "num_tokens": 276031115.0, + "step": 7863 + }, + { + "epoch": 1.4603528319405756, + "grad_norm": 1.5556718111038208, + "learning_rate": 1e-06, + "loss": 0.4354, + "mean_token_accuracy": 0.8558540940284729, + "num_tokens": 276069747.0, + "step": 7864 + }, + { + "epoch": 1.4605385329619314, + "grad_norm": 1.7423986196517944, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8924767374992371, + "num_tokens": 276092086.0, + "step": 7865 + }, + { + "epoch": 1.4607242339832869, + "grad_norm": 1.392209768295288, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8941413164138794, + "num_tokens": 276129214.0, + "step": 7866 + }, + { + "epoch": 1.4609099350046426, + "grad_norm": 1.4849841594696045, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8690855503082275, + "num_tokens": 276167916.0, + "step": 7867 + }, + { + "epoch": 1.4610956360259981, + "grad_norm": 1.4820209741592407, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8812927007675171, + "num_tokens": 276201735.0, + "step": 7868 + }, + { + "epoch": 1.4612813370473536, + "grad_norm": 1.6642035245895386, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8752735257148743, + "num_tokens": 276237213.0, + "step": 7869 + }, + { + "epoch": 1.4614670380687094, + "grad_norm": 1.5620468854904175, + "learning_rate": 1e-06, + "loss": 0.4284, + "mean_token_accuracy": 0.8580818176269531, + "num_tokens": 276273080.0, + "step": 7870 + }, + { + "epoch": 1.461652739090065, + "grad_norm": 1.5594102144241333, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.8700578808784485, + "num_tokens": 276306008.0, + "step": 7871 + }, + { + "epoch": 1.4618384401114206, + "grad_norm": 1.4230440855026245, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8741898536682129, + "num_tokens": 276343008.0, + "step": 7872 + }, + { + "epoch": 1.4620241411327761, + "grad_norm": 1.5864273309707642, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8698368072509766, + "num_tokens": 276374552.0, + "step": 7873 + }, + { + "epoch": 1.4622098421541319, + "grad_norm": 1.5480122566223145, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8749808669090271, + "num_tokens": 276408038.0, + "step": 7874 + }, + { + "epoch": 1.4623955431754876, + "grad_norm": 1.5012496709823608, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8833234906196594, + "num_tokens": 276441961.0, + "step": 7875 + }, + { + "epoch": 1.462581244196843, + "grad_norm": 1.5038739442825317, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8656971454620361, + "num_tokens": 276477131.0, + "step": 7876 + }, + { + "epoch": 1.4627669452181986, + "grad_norm": 1.4229222536087036, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8778560161590576, + "num_tokens": 276514947.0, + "step": 7877 + }, + { + "epoch": 1.4629526462395543, + "grad_norm": 1.472931981086731, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8818379044532776, + "num_tokens": 276550034.0, + "step": 7878 + }, + { + "epoch": 1.4631383472609099, + "grad_norm": 1.4697895050048828, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8808913826942444, + "num_tokens": 276591235.0, + "step": 7879 + }, + { + "epoch": 1.4633240482822656, + "grad_norm": 1.645704746246338, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8846192359924316, + "num_tokens": 276621247.0, + "step": 7880 + }, + { + "epoch": 1.463509749303621, + "grad_norm": 1.595481514930725, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8721718788146973, + "num_tokens": 276653333.0, + "step": 7881 + }, + { + "epoch": 1.4636954503249768, + "grad_norm": 1.5489650964736938, + "learning_rate": 1e-06, + "loss": 0.436, + "mean_token_accuracy": 0.8514092564582825, + "num_tokens": 276693897.0, + "step": 7882 + }, + { + "epoch": 1.4638811513463323, + "grad_norm": 1.4228235483169556, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8833154439926147, + "num_tokens": 276730234.0, + "step": 7883 + }, + { + "epoch": 1.464066852367688, + "grad_norm": 1.648777961730957, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8887474536895752, + "num_tokens": 276760651.0, + "step": 7884 + }, + { + "epoch": 1.4642525533890436, + "grad_norm": 1.5178723335266113, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8751053214073181, + "num_tokens": 276795900.0, + "step": 7885 + }, + { + "epoch": 1.4644382544103993, + "grad_norm": 1.504183053970337, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.876255214214325, + "num_tokens": 276830400.0, + "step": 7886 + }, + { + "epoch": 1.4646239554317548, + "grad_norm": 1.5780246257781982, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8792904019355774, + "num_tokens": 276862786.0, + "step": 7887 + }, + { + "epoch": 1.4648096564531106, + "grad_norm": 1.5026828050613403, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.867786169052124, + "num_tokens": 276901097.0, + "step": 7888 + }, + { + "epoch": 1.464995357474466, + "grad_norm": 1.4440677165985107, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8968247175216675, + "num_tokens": 276933092.0, + "step": 7889 + }, + { + "epoch": 1.4651810584958218, + "grad_norm": 1.3939039707183838, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8768374919891357, + "num_tokens": 276969733.0, + "step": 7890 + }, + { + "epoch": 1.4653667595171773, + "grad_norm": 1.5230799913406372, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8841164708137512, + "num_tokens": 277003964.0, + "step": 7891 + }, + { + "epoch": 1.4655524605385328, + "grad_norm": 1.5303022861480713, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8797510862350464, + "num_tokens": 277039336.0, + "step": 7892 + }, + { + "epoch": 1.4657381615598886, + "grad_norm": 1.618775486946106, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.860758900642395, + "num_tokens": 277073267.0, + "step": 7893 + }, + { + "epoch": 1.4659238625812443, + "grad_norm": 1.4640076160430908, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8856213688850403, + "num_tokens": 277106971.0, + "step": 7894 + }, + { + "epoch": 1.4661095636025998, + "grad_norm": 1.3837089538574219, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8873393535614014, + "num_tokens": 277144676.0, + "step": 7895 + }, + { + "epoch": 1.4662952646239553, + "grad_norm": 1.5552382469177246, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8823893666267395, + "num_tokens": 277177411.0, + "step": 7896 + }, + { + "epoch": 1.466480965645311, + "grad_norm": 1.532189965248108, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8666191697120667, + "num_tokens": 277211597.0, + "step": 7897 + }, + { + "epoch": 1.4666666666666668, + "grad_norm": 1.6796544790267944, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8641840219497681, + "num_tokens": 277243904.0, + "step": 7898 + }, + { + "epoch": 1.4668523676880223, + "grad_norm": 1.6115200519561768, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.880825400352478, + "num_tokens": 277275021.0, + "step": 7899 + }, + { + "epoch": 1.4670380687093778, + "grad_norm": 1.6226933002471924, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8724092245101929, + "num_tokens": 277305275.0, + "step": 7900 + }, + { + "epoch": 1.4672237697307335, + "grad_norm": 1.6429333686828613, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8805269002914429, + "num_tokens": 277333531.0, + "step": 7901 + }, + { + "epoch": 1.4674094707520893, + "grad_norm": 1.4397592544555664, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8738712072372437, + "num_tokens": 277371181.0, + "step": 7902 + }, + { + "epoch": 1.4675951717734448, + "grad_norm": 1.4055982828140259, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8861171007156372, + "num_tokens": 277407536.0, + "step": 7903 + }, + { + "epoch": 1.4677808727948003, + "grad_norm": 1.6253840923309326, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8620223999023438, + "num_tokens": 277440547.0, + "step": 7904 + }, + { + "epoch": 1.467966573816156, + "grad_norm": 1.5086365938186646, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8724415302276611, + "num_tokens": 277475292.0, + "step": 7905 + }, + { + "epoch": 1.4681522748375115, + "grad_norm": 1.4045814275741577, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8839250802993774, + "num_tokens": 277513063.0, + "step": 7906 + }, + { + "epoch": 1.4683379758588673, + "grad_norm": 1.402115821838379, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8850166201591492, + "num_tokens": 277552276.0, + "step": 7907 + }, + { + "epoch": 1.4685236768802228, + "grad_norm": 1.503843903541565, + "learning_rate": 1e-06, + "loss": 0.4173, + "mean_token_accuracy": 0.8648771047592163, + "num_tokens": 277589314.0, + "step": 7908 + }, + { + "epoch": 1.4687093779015785, + "grad_norm": 1.3038703203201294, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8830395936965942, + "num_tokens": 277632499.0, + "step": 7909 + }, + { + "epoch": 1.468895078922934, + "grad_norm": 1.527860403060913, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.876517653465271, + "num_tokens": 277666194.0, + "step": 7910 + }, + { + "epoch": 1.4690807799442898, + "grad_norm": 1.7470064163208008, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8813586831092834, + "num_tokens": 277701522.0, + "step": 7911 + }, + { + "epoch": 1.4692664809656453, + "grad_norm": 1.3785268068313599, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8689848780632019, + "num_tokens": 277742085.0, + "step": 7912 + }, + { + "epoch": 1.469452181987001, + "grad_norm": 1.5978063344955444, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8651405572891235, + "num_tokens": 277776755.0, + "step": 7913 + }, + { + "epoch": 1.4696378830083565, + "grad_norm": 1.4949060678482056, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8762675523757935, + "num_tokens": 277813540.0, + "step": 7914 + }, + { + "epoch": 1.469823584029712, + "grad_norm": 1.4876848459243774, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8762218952178955, + "num_tokens": 277846750.0, + "step": 7915 + }, + { + "epoch": 1.4700092850510678, + "grad_norm": 1.434262990951538, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8829306364059448, + "num_tokens": 277881573.0, + "step": 7916 + }, + { + "epoch": 1.4701949860724235, + "grad_norm": 1.4289008378982544, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8828491568565369, + "num_tokens": 277918984.0, + "step": 7917 + }, + { + "epoch": 1.470380687093779, + "grad_norm": 1.3997399806976318, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8862744569778442, + "num_tokens": 277955632.0, + "step": 7918 + }, + { + "epoch": 1.4705663881151345, + "grad_norm": 1.3989830017089844, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8701982498168945, + "num_tokens": 277996672.0, + "step": 7919 + }, + { + "epoch": 1.4707520891364902, + "grad_norm": 1.395559310913086, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.877780020236969, + "num_tokens": 278038231.0, + "step": 7920 + }, + { + "epoch": 1.470937790157846, + "grad_norm": 1.6606394052505493, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8677976131439209, + "num_tokens": 278071654.0, + "step": 7921 + }, + { + "epoch": 1.4711234911792015, + "grad_norm": 1.5630853176116943, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8780066967010498, + "num_tokens": 278100929.0, + "step": 7922 + }, + { + "epoch": 1.471309192200557, + "grad_norm": 1.5144602060317993, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.869855523109436, + "num_tokens": 278138569.0, + "step": 7923 + }, + { + "epoch": 1.4714948932219127, + "grad_norm": 1.4303244352340698, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8638073205947876, + "num_tokens": 278181194.0, + "step": 7924 + }, + { + "epoch": 1.4716805942432685, + "grad_norm": 1.585178256034851, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8696033954620361, + "num_tokens": 278213510.0, + "step": 7925 + }, + { + "epoch": 1.471866295264624, + "grad_norm": 1.7972520589828491, + "learning_rate": 1e-06, + "loss": 0.44, + "mean_token_accuracy": 0.8550556898117065, + "num_tokens": 278240227.0, + "step": 7926 + }, + { + "epoch": 1.4720519962859795, + "grad_norm": 1.4632506370544434, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8838323950767517, + "num_tokens": 278277384.0, + "step": 7927 + }, + { + "epoch": 1.4722376973073352, + "grad_norm": 1.535136342048645, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8704690933227539, + "num_tokens": 278316154.0, + "step": 7928 + }, + { + "epoch": 1.4724233983286907, + "grad_norm": 1.4428086280822754, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8800982236862183, + "num_tokens": 278350768.0, + "step": 7929 + }, + { + "epoch": 1.4726090993500465, + "grad_norm": 1.4570894241333008, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8775284290313721, + "num_tokens": 278384142.0, + "step": 7930 + }, + { + "epoch": 1.472794800371402, + "grad_norm": 1.4835089445114136, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8782087564468384, + "num_tokens": 278417211.0, + "step": 7931 + }, + { + "epoch": 1.4729805013927577, + "grad_norm": 1.514457106590271, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8799870014190674, + "num_tokens": 278453468.0, + "step": 7932 + }, + { + "epoch": 1.4731662024141132, + "grad_norm": 1.4544986486434937, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8863446712493896, + "num_tokens": 278488521.0, + "step": 7933 + }, + { + "epoch": 1.473351903435469, + "grad_norm": 1.4553676843643188, + "learning_rate": 1e-06, + "loss": 0.3878, + "mean_token_accuracy": 0.8694159388542175, + "num_tokens": 278525034.0, + "step": 7934 + }, + { + "epoch": 1.4735376044568245, + "grad_norm": 1.4034870862960815, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8766799569129944, + "num_tokens": 278560447.0, + "step": 7935 + }, + { + "epoch": 1.4737233054781802, + "grad_norm": 1.3180665969848633, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8743141889572144, + "num_tokens": 278602109.0, + "step": 7936 + }, + { + "epoch": 1.4739090064995357, + "grad_norm": 1.4768199920654297, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.887000322341919, + "num_tokens": 278635043.0, + "step": 7937 + }, + { + "epoch": 1.4740947075208914, + "grad_norm": 1.4329904317855835, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.872072696685791, + "num_tokens": 278669669.0, + "step": 7938 + }, + { + "epoch": 1.474280408542247, + "grad_norm": 1.4497450590133667, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.877302348613739, + "num_tokens": 278704768.0, + "step": 7939 + }, + { + "epoch": 1.4744661095636027, + "grad_norm": 1.3915349245071411, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8772327303886414, + "num_tokens": 278743655.0, + "step": 7940 + }, + { + "epoch": 1.4746518105849582, + "grad_norm": 1.6679399013519287, + "learning_rate": 1e-06, + "loss": 0.4652, + "mean_token_accuracy": 0.8444164991378784, + "num_tokens": 278778279.0, + "step": 7941 + }, + { + "epoch": 1.4748375116063137, + "grad_norm": 1.515224814414978, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8719862699508667, + "num_tokens": 278813712.0, + "step": 7942 + }, + { + "epoch": 1.4750232126276694, + "grad_norm": 1.463987946510315, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.867525577545166, + "num_tokens": 278849595.0, + "step": 7943 + }, + { + "epoch": 1.4752089136490252, + "grad_norm": 1.422569990158081, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8799853324890137, + "num_tokens": 278884225.0, + "step": 7944 + }, + { + "epoch": 1.4753946146703807, + "grad_norm": 1.362614631652832, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.880517840385437, + "num_tokens": 278925474.0, + "step": 7945 + }, + { + "epoch": 1.4755803156917362, + "grad_norm": 1.4298343658447266, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8721956014633179, + "num_tokens": 278968027.0, + "step": 7946 + }, + { + "epoch": 1.475766016713092, + "grad_norm": 1.3856805562973022, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8893078565597534, + "num_tokens": 279008328.0, + "step": 7947 + }, + { + "epoch": 1.4759517177344477, + "grad_norm": 1.5255043506622314, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8878366947174072, + "num_tokens": 279041758.0, + "step": 7948 + }, + { + "epoch": 1.4761374187558032, + "grad_norm": 1.4682060480117798, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.885332465171814, + "num_tokens": 279077541.0, + "step": 7949 + }, + { + "epoch": 1.4763231197771587, + "grad_norm": 1.3505911827087402, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8826307654380798, + "num_tokens": 279116133.0, + "step": 7950 + }, + { + "epoch": 1.4765088207985144, + "grad_norm": 1.439100980758667, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8786860108375549, + "num_tokens": 279151396.0, + "step": 7951 + }, + { + "epoch": 1.47669452181987, + "grad_norm": 1.4769529104232788, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8757567405700684, + "num_tokens": 279185647.0, + "step": 7952 + }, + { + "epoch": 1.4768802228412257, + "grad_norm": 1.4404325485229492, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8772785067558289, + "num_tokens": 279221831.0, + "step": 7953 + }, + { + "epoch": 1.4770659238625812, + "grad_norm": 1.501215934753418, + "learning_rate": 1e-06, + "loss": 0.4057, + "mean_token_accuracy": 0.8638103604316711, + "num_tokens": 279264796.0, + "step": 7954 + }, + { + "epoch": 1.477251624883937, + "grad_norm": 1.559214472770691, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8730493187904358, + "num_tokens": 279299791.0, + "step": 7955 + }, + { + "epoch": 1.4774373259052924, + "grad_norm": 1.5955045223236084, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8698862791061401, + "num_tokens": 279333987.0, + "step": 7956 + }, + { + "epoch": 1.4776230269266482, + "grad_norm": 1.5959365367889404, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8707607984542847, + "num_tokens": 279364229.0, + "step": 7957 + }, + { + "epoch": 1.4778087279480037, + "grad_norm": 1.5379457473754883, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8700227737426758, + "num_tokens": 279400266.0, + "step": 7958 + }, + { + "epoch": 1.4779944289693594, + "grad_norm": 1.4729290008544922, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8826712965965271, + "num_tokens": 279433517.0, + "step": 7959 + }, + { + "epoch": 1.478180129990715, + "grad_norm": 1.384096384048462, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8932743072509766, + "num_tokens": 279469390.0, + "step": 7960 + }, + { + "epoch": 1.4783658310120706, + "grad_norm": 1.4891602993011475, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8817604184150696, + "num_tokens": 279504082.0, + "step": 7961 + }, + { + "epoch": 1.4785515320334262, + "grad_norm": 1.7117055654525757, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8706513047218323, + "num_tokens": 279532905.0, + "step": 7962 + }, + { + "epoch": 1.4787372330547819, + "grad_norm": 1.478279948234558, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8814574480056763, + "num_tokens": 279567311.0, + "step": 7963 + }, + { + "epoch": 1.4789229340761374, + "grad_norm": 1.5348777770996094, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8806196451187134, + "num_tokens": 279597489.0, + "step": 7964 + }, + { + "epoch": 1.479108635097493, + "grad_norm": 1.6742595434188843, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.869465708732605, + "num_tokens": 279629796.0, + "step": 7965 + }, + { + "epoch": 1.4792943361188486, + "grad_norm": 1.457756757736206, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8724143505096436, + "num_tokens": 279666375.0, + "step": 7966 + }, + { + "epoch": 1.4794800371402044, + "grad_norm": 1.501572608947754, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.895977258682251, + "num_tokens": 279707294.0, + "step": 7967 + }, + { + "epoch": 1.4796657381615599, + "grad_norm": 1.5200471878051758, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8697564601898193, + "num_tokens": 279741397.0, + "step": 7968 + }, + { + "epoch": 1.4798514391829154, + "grad_norm": 1.4789035320281982, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8681938052177429, + "num_tokens": 279776618.0, + "step": 7969 + }, + { + "epoch": 1.4800371402042711, + "grad_norm": 1.495389461517334, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.881543755531311, + "num_tokens": 279809293.0, + "step": 7970 + }, + { + "epoch": 1.4802228412256269, + "grad_norm": 1.532994031906128, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8959782719612122, + "num_tokens": 279839155.0, + "step": 7971 + }, + { + "epoch": 1.4804085422469824, + "grad_norm": 1.5333245992660522, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8661826848983765, + "num_tokens": 279875641.0, + "step": 7972 + }, + { + "epoch": 1.4805942432683379, + "grad_norm": 1.4956053495407104, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8778309226036072, + "num_tokens": 279907834.0, + "step": 7973 + }, + { + "epoch": 1.4807799442896936, + "grad_norm": 1.5165575742721558, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8764959573745728, + "num_tokens": 279944274.0, + "step": 7974 + }, + { + "epoch": 1.4809656453110491, + "grad_norm": 1.344569444656372, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8901277184486389, + "num_tokens": 279980694.0, + "step": 7975 + }, + { + "epoch": 1.4811513463324049, + "grad_norm": 1.5258779525756836, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.8607345819473267, + "num_tokens": 280018551.0, + "step": 7976 + }, + { + "epoch": 1.4813370473537604, + "grad_norm": 1.3784183263778687, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8781582117080688, + "num_tokens": 280058373.0, + "step": 7977 + }, + { + "epoch": 1.481522748375116, + "grad_norm": 1.6335619688034058, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.874596893787384, + "num_tokens": 280087163.0, + "step": 7978 + }, + { + "epoch": 1.4817084493964716, + "grad_norm": 1.5381190776824951, + "learning_rate": 1e-06, + "loss": 0.3937, + "mean_token_accuracy": 0.8682249188423157, + "num_tokens": 280122219.0, + "step": 7979 + }, + { + "epoch": 1.4818941504178273, + "grad_norm": 1.4259344339370728, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8785131573677063, + "num_tokens": 280159137.0, + "step": 7980 + }, + { + "epoch": 1.4820798514391829, + "grad_norm": 1.4589956998825073, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8712930679321289, + "num_tokens": 280199850.0, + "step": 7981 + }, + { + "epoch": 1.4822655524605386, + "grad_norm": 1.525458812713623, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8828086256980896, + "num_tokens": 280234107.0, + "step": 7982 + }, + { + "epoch": 1.482451253481894, + "grad_norm": 1.5131688117980957, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8861667513847351, + "num_tokens": 280267219.0, + "step": 7983 + }, + { + "epoch": 1.4826369545032498, + "grad_norm": 1.4972994327545166, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8882474303245544, + "num_tokens": 280298208.0, + "step": 7984 + }, + { + "epoch": 1.4828226555246053, + "grad_norm": 1.6154290437698364, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.879626452922821, + "num_tokens": 280328629.0, + "step": 7985 + }, + { + "epoch": 1.483008356545961, + "grad_norm": 1.4976190328598022, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8748864531517029, + "num_tokens": 280362602.0, + "step": 7986 + }, + { + "epoch": 1.4831940575673166, + "grad_norm": 1.4325017929077148, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8804031610488892, + "num_tokens": 280399963.0, + "step": 7987 + }, + { + "epoch": 1.483379758588672, + "grad_norm": 1.5206457376480103, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8729736804962158, + "num_tokens": 280433824.0, + "step": 7988 + }, + { + "epoch": 1.4835654596100278, + "grad_norm": 1.7154978513717651, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8691853880882263, + "num_tokens": 280463852.0, + "step": 7989 + }, + { + "epoch": 1.4837511606313836, + "grad_norm": 1.590636968612671, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8624662160873413, + "num_tokens": 280498393.0, + "step": 7990 + }, + { + "epoch": 1.483936861652739, + "grad_norm": 1.3521562814712524, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.877713680267334, + "num_tokens": 280537785.0, + "step": 7991 + }, + { + "epoch": 1.4841225626740946, + "grad_norm": 1.5418317317962646, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8758792281150818, + "num_tokens": 280569713.0, + "step": 7992 + }, + { + "epoch": 1.4843082636954503, + "grad_norm": 1.5143389701843262, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8766713738441467, + "num_tokens": 280604123.0, + "step": 7993 + }, + { + "epoch": 1.484493964716806, + "grad_norm": 1.6045739650726318, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8888666033744812, + "num_tokens": 280630760.0, + "step": 7994 + }, + { + "epoch": 1.4846796657381616, + "grad_norm": 1.5121023654937744, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8877238631248474, + "num_tokens": 280659349.0, + "step": 7995 + }, + { + "epoch": 1.484865366759517, + "grad_norm": 1.4852415323257446, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8833833932876587, + "num_tokens": 280696904.0, + "step": 7996 + }, + { + "epoch": 1.4850510677808728, + "grad_norm": 1.6651196479797363, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8657130002975464, + "num_tokens": 280727889.0, + "step": 7997 + }, + { + "epoch": 1.4852367688022285, + "grad_norm": 1.5975571870803833, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.881662130355835, + "num_tokens": 280758380.0, + "step": 7998 + }, + { + "epoch": 1.485422469823584, + "grad_norm": 1.4713577032089233, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.882179319858551, + "num_tokens": 280791986.0, + "step": 7999 + }, + { + "epoch": 1.4856081708449396, + "grad_norm": 1.6733121871948242, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.883708655834198, + "num_tokens": 280818957.0, + "step": 8000 + }, + { + "epoch": 1.4857938718662953, + "grad_norm": 1.6643588542938232, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8704366683959961, + "num_tokens": 280851286.0, + "step": 8001 + }, + { + "epoch": 1.4859795728876508, + "grad_norm": 1.4213167428970337, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8800718188285828, + "num_tokens": 280886993.0, + "step": 8002 + }, + { + "epoch": 1.4861652739090065, + "grad_norm": 1.3219051361083984, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8921434283256531, + "num_tokens": 280925736.0, + "step": 8003 + }, + { + "epoch": 1.486350974930362, + "grad_norm": 1.4872022867202759, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8700490593910217, + "num_tokens": 280960414.0, + "step": 8004 + }, + { + "epoch": 1.4865366759517178, + "grad_norm": 1.7664506435394287, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8640108108520508, + "num_tokens": 280988419.0, + "step": 8005 + }, + { + "epoch": 1.4867223769730733, + "grad_norm": 1.479862093925476, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8774268627166748, + "num_tokens": 281024469.0, + "step": 8006 + }, + { + "epoch": 1.486908077994429, + "grad_norm": 1.5730290412902832, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8718779683113098, + "num_tokens": 281058711.0, + "step": 8007 + }, + { + "epoch": 1.4870937790157845, + "grad_norm": 1.3385168313980103, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8822063207626343, + "num_tokens": 281102054.0, + "step": 8008 + }, + { + "epoch": 1.4872794800371403, + "grad_norm": 1.519297480583191, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8720695376396179, + "num_tokens": 281138067.0, + "step": 8009 + }, + { + "epoch": 1.4874651810584958, + "grad_norm": 1.5810352563858032, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8792083263397217, + "num_tokens": 281170367.0, + "step": 8010 + }, + { + "epoch": 1.4876508820798515, + "grad_norm": 1.4881709814071655, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8789803981781006, + "num_tokens": 281206124.0, + "step": 8011 + }, + { + "epoch": 1.487836583101207, + "grad_norm": 1.4988954067230225, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8827594518661499, + "num_tokens": 281241690.0, + "step": 8012 + }, + { + "epoch": 1.4880222841225628, + "grad_norm": 1.5809228420257568, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8831892609596252, + "num_tokens": 281273473.0, + "step": 8013 + }, + { + "epoch": 1.4882079851439183, + "grad_norm": 1.4749144315719604, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8749827146530151, + "num_tokens": 281306849.0, + "step": 8014 + }, + { + "epoch": 1.4883936861652738, + "grad_norm": 1.6239300966262817, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.875413715839386, + "num_tokens": 281335214.0, + "step": 8015 + }, + { + "epoch": 1.4885793871866295, + "grad_norm": 1.4317408800125122, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8784867525100708, + "num_tokens": 281370902.0, + "step": 8016 + }, + { + "epoch": 1.4887650882079853, + "grad_norm": 1.8475977182388306, + "learning_rate": 1e-06, + "loss": 0.4088, + "mean_token_accuracy": 0.8619781732559204, + "num_tokens": 281398001.0, + "step": 8017 + }, + { + "epoch": 1.4889507892293408, + "grad_norm": 1.4701125621795654, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8699530363082886, + "num_tokens": 281438309.0, + "step": 8018 + }, + { + "epoch": 1.4891364902506963, + "grad_norm": 1.5466129779815674, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8742161989212036, + "num_tokens": 281476282.0, + "step": 8019 + }, + { + "epoch": 1.489322191272052, + "grad_norm": 1.3583958148956299, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8645097017288208, + "num_tokens": 281524010.0, + "step": 8020 + }, + { + "epoch": 1.4895078922934077, + "grad_norm": 1.5530093908309937, + "learning_rate": 1e-06, + "loss": 0.4444, + "mean_token_accuracy": 0.859019935131073, + "num_tokens": 281564662.0, + "step": 8021 + }, + { + "epoch": 1.4896935933147633, + "grad_norm": 1.6401301622390747, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8791218996047974, + "num_tokens": 281596881.0, + "step": 8022 + }, + { + "epoch": 1.4898792943361188, + "grad_norm": 1.681972861289978, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8638303279876709, + "num_tokens": 281631146.0, + "step": 8023 + }, + { + "epoch": 1.4900649953574745, + "grad_norm": 1.3658368587493896, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.881291925907135, + "num_tokens": 281669584.0, + "step": 8024 + }, + { + "epoch": 1.49025069637883, + "grad_norm": 1.48722505569458, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8782428503036499, + "num_tokens": 281702939.0, + "step": 8025 + }, + { + "epoch": 1.4904363974001857, + "grad_norm": 1.4331767559051514, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8880549669265747, + "num_tokens": 281738497.0, + "step": 8026 + }, + { + "epoch": 1.4906220984215413, + "grad_norm": 1.5581281185150146, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8779749274253845, + "num_tokens": 281772150.0, + "step": 8027 + }, + { + "epoch": 1.490807799442897, + "grad_norm": 1.5790300369262695, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8778936266899109, + "num_tokens": 281803137.0, + "step": 8028 + }, + { + "epoch": 1.4909935004642525, + "grad_norm": 1.4899438619613647, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8671377301216125, + "num_tokens": 281841702.0, + "step": 8029 + }, + { + "epoch": 1.4911792014856082, + "grad_norm": 1.4202957153320312, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8651670217514038, + "num_tokens": 281881334.0, + "step": 8030 + }, + { + "epoch": 1.4913649025069637, + "grad_norm": 1.4637919664382935, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8618179559707642, + "num_tokens": 281923342.0, + "step": 8031 + }, + { + "epoch": 1.4915506035283195, + "grad_norm": 1.6183743476867676, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8861410617828369, + "num_tokens": 281952688.0, + "step": 8032 + }, + { + "epoch": 1.491736304549675, + "grad_norm": 1.4675873517990112, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8803713321685791, + "num_tokens": 281986365.0, + "step": 8033 + }, + { + "epoch": 1.4919220055710307, + "grad_norm": 1.52041494846344, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8741858601570129, + "num_tokens": 282020414.0, + "step": 8034 + }, + { + "epoch": 1.4921077065923862, + "grad_norm": 1.3896952867507935, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8843457698822021, + "num_tokens": 282060467.0, + "step": 8035 + }, + { + "epoch": 1.492293407613742, + "grad_norm": 1.523238182067871, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8857145309448242, + "num_tokens": 282092343.0, + "step": 8036 + }, + { + "epoch": 1.4924791086350975, + "grad_norm": 1.5082865953445435, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8565644025802612, + "num_tokens": 282130546.0, + "step": 8037 + }, + { + "epoch": 1.492664809656453, + "grad_norm": 1.4928587675094604, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8710253238677979, + "num_tokens": 282165793.0, + "step": 8038 + }, + { + "epoch": 1.4928505106778087, + "grad_norm": 1.5626261234283447, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8855090737342834, + "num_tokens": 282196381.0, + "step": 8039 + }, + { + "epoch": 1.4930362116991645, + "grad_norm": 1.3721067905426025, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8936755061149597, + "num_tokens": 282232268.0, + "step": 8040 + }, + { + "epoch": 1.49322191272052, + "grad_norm": 1.4146257638931274, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8746907711029053, + "num_tokens": 282271852.0, + "step": 8041 + }, + { + "epoch": 1.4934076137418755, + "grad_norm": 1.5020546913146973, + "learning_rate": 1e-06, + "loss": 0.4162, + "mean_token_accuracy": 0.860351026058197, + "num_tokens": 282310571.0, + "step": 8042 + }, + { + "epoch": 1.4935933147632312, + "grad_norm": 1.6064860820770264, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8799026608467102, + "num_tokens": 282347712.0, + "step": 8043 + }, + { + "epoch": 1.493779015784587, + "grad_norm": 1.3822098970413208, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.878097414970398, + "num_tokens": 282385427.0, + "step": 8044 + }, + { + "epoch": 1.4939647168059424, + "grad_norm": 1.5699717998504639, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.879676342010498, + "num_tokens": 282416587.0, + "step": 8045 + }, + { + "epoch": 1.494150417827298, + "grad_norm": 1.4354267120361328, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8904616832733154, + "num_tokens": 282453126.0, + "step": 8046 + }, + { + "epoch": 1.4943361188486537, + "grad_norm": 1.4936002492904663, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8724385499954224, + "num_tokens": 282487166.0, + "step": 8047 + }, + { + "epoch": 1.4945218198700092, + "grad_norm": 1.4297140836715698, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8788905143737793, + "num_tokens": 282526987.0, + "step": 8048 + }, + { + "epoch": 1.494707520891365, + "grad_norm": 1.5641182661056519, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8645040988922119, + "num_tokens": 282562521.0, + "step": 8049 + }, + { + "epoch": 1.4948932219127204, + "grad_norm": 1.4800163507461548, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8860604763031006, + "num_tokens": 282593324.0, + "step": 8050 + }, + { + "epoch": 1.4950789229340762, + "grad_norm": 1.396100401878357, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8699489831924438, + "num_tokens": 282635056.0, + "step": 8051 + }, + { + "epoch": 1.4952646239554317, + "grad_norm": 1.5011755228042603, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8735870122909546, + "num_tokens": 282669456.0, + "step": 8052 + }, + { + "epoch": 1.4954503249767874, + "grad_norm": 1.4589881896972656, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8838714361190796, + "num_tokens": 282704858.0, + "step": 8053 + }, + { + "epoch": 1.495636025998143, + "grad_norm": 1.477888822555542, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8796613812446594, + "num_tokens": 282743646.0, + "step": 8054 + }, + { + "epoch": 1.4958217270194987, + "grad_norm": 1.8939180374145508, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8779817819595337, + "num_tokens": 282767492.0, + "step": 8055 + }, + { + "epoch": 1.4960074280408542, + "grad_norm": 1.390673279762268, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8884544372558594, + "num_tokens": 282808451.0, + "step": 8056 + }, + { + "epoch": 1.49619312906221, + "grad_norm": 1.4445409774780273, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8776822090148926, + "num_tokens": 282844114.0, + "step": 8057 + }, + { + "epoch": 1.4963788300835654, + "grad_norm": 1.6357462406158447, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8769623041152954, + "num_tokens": 282872661.0, + "step": 8058 + }, + { + "epoch": 1.4965645311049212, + "grad_norm": 1.4368871450424194, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8780899047851562, + "num_tokens": 282910358.0, + "step": 8059 + }, + { + "epoch": 1.4967502321262767, + "grad_norm": 1.5101526975631714, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8913803100585938, + "num_tokens": 282943018.0, + "step": 8060 + }, + { + "epoch": 1.4969359331476322, + "grad_norm": 1.4797234535217285, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8877013921737671, + "num_tokens": 282976232.0, + "step": 8061 + }, + { + "epoch": 1.497121634168988, + "grad_norm": 1.5479921102523804, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8669391870498657, + "num_tokens": 283011664.0, + "step": 8062 + }, + { + "epoch": 1.4973073351903436, + "grad_norm": 1.479775309562683, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8845343589782715, + "num_tokens": 283046802.0, + "step": 8063 + }, + { + "epoch": 1.4974930362116992, + "grad_norm": 1.4219030141830444, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8675123453140259, + "num_tokens": 283085145.0, + "step": 8064 + }, + { + "epoch": 1.4976787372330547, + "grad_norm": 1.5514389276504517, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8739447593688965, + "num_tokens": 283117308.0, + "step": 8065 + }, + { + "epoch": 1.4978644382544104, + "grad_norm": 1.4937598705291748, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8726825714111328, + "num_tokens": 283150894.0, + "step": 8066 + }, + { + "epoch": 1.4980501392757661, + "grad_norm": 1.3687859773635864, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8774603009223938, + "num_tokens": 283191868.0, + "step": 8067 + }, + { + "epoch": 1.4982358402971216, + "grad_norm": 1.3178812265396118, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8837295770645142, + "num_tokens": 283236518.0, + "step": 8068 + }, + { + "epoch": 1.4984215413184772, + "grad_norm": 1.58284592628479, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8678058981895447, + "num_tokens": 283270452.0, + "step": 8069 + }, + { + "epoch": 1.498607242339833, + "grad_norm": 1.629971981048584, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8782628774642944, + "num_tokens": 283302522.0, + "step": 8070 + }, + { + "epoch": 1.4987929433611886, + "grad_norm": 1.5579065084457397, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8684886693954468, + "num_tokens": 283338635.0, + "step": 8071 + }, + { + "epoch": 1.4989786443825441, + "grad_norm": 1.5271683931350708, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.866532564163208, + "num_tokens": 283374148.0, + "step": 8072 + }, + { + "epoch": 1.4991643454038996, + "grad_norm": 1.7617173194885254, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8769757747650146, + "num_tokens": 283401362.0, + "step": 8073 + }, + { + "epoch": 1.4993500464252554, + "grad_norm": 1.4078161716461182, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8905037641525269, + "num_tokens": 283439651.0, + "step": 8074 + }, + { + "epoch": 1.499535747446611, + "grad_norm": 1.4672986268997192, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8906885385513306, + "num_tokens": 283471611.0, + "step": 8075 + }, + { + "epoch": 1.4997214484679666, + "grad_norm": 1.4403995275497437, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8793531656265259, + "num_tokens": 283510094.0, + "step": 8076 + }, + { + "epoch": 1.4999071494893221, + "grad_norm": 1.4776562452316284, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8831095695495605, + "num_tokens": 283547125.0, + "step": 8077 + }, + { + "epoch": 1.5000928505106779, + "grad_norm": 1.5804253816604614, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.883466362953186, + "num_tokens": 283581451.0, + "step": 8078 + }, + { + "epoch": 1.5002785515320334, + "grad_norm": 1.6102555990219116, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8798420429229736, + "num_tokens": 283619489.0, + "step": 8079 + }, + { + "epoch": 1.5004642525533889, + "grad_norm": 1.3347415924072266, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8958209753036499, + "num_tokens": 283654557.0, + "step": 8080 + }, + { + "epoch": 1.5006499535747446, + "grad_norm": 1.4154362678527832, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8700643181800842, + "num_tokens": 283694782.0, + "step": 8081 + }, + { + "epoch": 1.5008356545961004, + "grad_norm": 1.5526669025421143, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8770370483398438, + "num_tokens": 283728803.0, + "step": 8082 + }, + { + "epoch": 1.5010213556174559, + "grad_norm": 1.4962660074234009, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8660865426063538, + "num_tokens": 283765716.0, + "step": 8083 + }, + { + "epoch": 1.5012070566388114, + "grad_norm": 1.3297532796859741, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8847815990447998, + "num_tokens": 283806481.0, + "step": 8084 + }, + { + "epoch": 1.501392757660167, + "grad_norm": 1.4595504999160767, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8904445171356201, + "num_tokens": 283841247.0, + "step": 8085 + }, + { + "epoch": 1.5015784586815228, + "grad_norm": 1.4890211820602417, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.872590184211731, + "num_tokens": 283874975.0, + "step": 8086 + }, + { + "epoch": 1.5017641597028784, + "grad_norm": 1.3943012952804565, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8722569942474365, + "num_tokens": 283917487.0, + "step": 8087 + }, + { + "epoch": 1.5019498607242339, + "grad_norm": 1.5630476474761963, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8772356510162354, + "num_tokens": 283949195.0, + "step": 8088 + }, + { + "epoch": 1.5021355617455896, + "grad_norm": 1.4627546072006226, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8868294954299927, + "num_tokens": 283984969.0, + "step": 8089 + }, + { + "epoch": 1.5023212627669453, + "grad_norm": 1.533437967300415, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8762847185134888, + "num_tokens": 284018142.0, + "step": 8090 + }, + { + "epoch": 1.5025069637883008, + "grad_norm": 1.6193196773529053, + "learning_rate": 1e-06, + "loss": 0.3862, + "mean_token_accuracy": 0.8685743808746338, + "num_tokens": 284050869.0, + "step": 8091 + }, + { + "epoch": 1.5026926648096564, + "grad_norm": 1.4890092611312866, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8821020126342773, + "num_tokens": 284085234.0, + "step": 8092 + }, + { + "epoch": 1.502878365831012, + "grad_norm": 1.5065032243728638, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8750406503677368, + "num_tokens": 284124939.0, + "step": 8093 + }, + { + "epoch": 1.5030640668523678, + "grad_norm": 1.4705744981765747, + "learning_rate": 1e-06, + "loss": 0.3966, + "mean_token_accuracy": 0.8645550012588501, + "num_tokens": 284164453.0, + "step": 8094 + }, + { + "epoch": 1.5032497678737233, + "grad_norm": 1.6404805183410645, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8801632523536682, + "num_tokens": 284197587.0, + "step": 8095 + }, + { + "epoch": 1.5034354688950788, + "grad_norm": 1.4769281148910522, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8778008222579956, + "num_tokens": 284232455.0, + "step": 8096 + }, + { + "epoch": 1.5036211699164346, + "grad_norm": 1.6068140268325806, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8628265857696533, + "num_tokens": 284264423.0, + "step": 8097 + }, + { + "epoch": 1.5038068709377903, + "grad_norm": 1.5174036026000977, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8915987610816956, + "num_tokens": 284298706.0, + "step": 8098 + }, + { + "epoch": 1.5039925719591458, + "grad_norm": 1.4880272150039673, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8828657865524292, + "num_tokens": 284332652.0, + "step": 8099 + }, + { + "epoch": 1.5041782729805013, + "grad_norm": 1.4491745233535767, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8758472204208374, + "num_tokens": 284366609.0, + "step": 8100 + }, + { + "epoch": 1.504363974001857, + "grad_norm": 1.5850552320480347, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.871910810470581, + "num_tokens": 284397758.0, + "step": 8101 + }, + { + "epoch": 1.5045496750232126, + "grad_norm": 1.5448038578033447, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8764897584915161, + "num_tokens": 284430198.0, + "step": 8102 + }, + { + "epoch": 1.504735376044568, + "grad_norm": 1.527576208114624, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8759212493896484, + "num_tokens": 284463075.0, + "step": 8103 + }, + { + "epoch": 1.5049210770659238, + "grad_norm": 1.6479171514511108, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8734182715415955, + "num_tokens": 284492485.0, + "step": 8104 + }, + { + "epoch": 1.5051067780872796, + "grad_norm": 1.523959994316101, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8687233924865723, + "num_tokens": 284528116.0, + "step": 8105 + }, + { + "epoch": 1.505292479108635, + "grad_norm": 1.4533277750015259, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8763989210128784, + "num_tokens": 284565679.0, + "step": 8106 + }, + { + "epoch": 1.5054781801299906, + "grad_norm": 1.675304651260376, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8890358209609985, + "num_tokens": 284592582.0, + "step": 8107 + }, + { + "epoch": 1.5056638811513463, + "grad_norm": 1.5117151737213135, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.8585160970687866, + "num_tokens": 284632336.0, + "step": 8108 + }, + { + "epoch": 1.505849582172702, + "grad_norm": 1.393699049949646, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8813809752464294, + "num_tokens": 284674681.0, + "step": 8109 + }, + { + "epoch": 1.5060352831940576, + "grad_norm": 1.497881531715393, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8774021863937378, + "num_tokens": 284710494.0, + "step": 8110 + }, + { + "epoch": 1.506220984215413, + "grad_norm": 1.4881473779678345, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8786065578460693, + "num_tokens": 284746000.0, + "step": 8111 + }, + { + "epoch": 1.5064066852367688, + "grad_norm": 1.4754996299743652, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8732644319534302, + "num_tokens": 284781998.0, + "step": 8112 + }, + { + "epoch": 1.5065923862581245, + "grad_norm": 1.37160325050354, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8766834735870361, + "num_tokens": 284823945.0, + "step": 8113 + }, + { + "epoch": 1.50677808727948, + "grad_norm": 1.511098027229309, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8710645437240601, + "num_tokens": 284860579.0, + "step": 8114 + }, + { + "epoch": 1.5069637883008355, + "grad_norm": 1.5264692306518555, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8813865780830383, + "num_tokens": 284892176.0, + "step": 8115 + }, + { + "epoch": 1.5071494893221913, + "grad_norm": 1.4669907093048096, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8815707564353943, + "num_tokens": 284929041.0, + "step": 8116 + }, + { + "epoch": 1.507335190343547, + "grad_norm": 1.3920142650604248, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8755085468292236, + "num_tokens": 284970982.0, + "step": 8117 + }, + { + "epoch": 1.5075208913649025, + "grad_norm": 1.568268895149231, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8734139800071716, + "num_tokens": 285003594.0, + "step": 8118 + }, + { + "epoch": 1.507706592386258, + "grad_norm": 1.4553241729736328, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8713418245315552, + "num_tokens": 285041207.0, + "step": 8119 + }, + { + "epoch": 1.5078922934076138, + "grad_norm": 1.6252260208129883, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8733119964599609, + "num_tokens": 285073839.0, + "step": 8120 + }, + { + "epoch": 1.5080779944289695, + "grad_norm": 1.485907793045044, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8750239610671997, + "num_tokens": 285109052.0, + "step": 8121 + }, + { + "epoch": 1.508263695450325, + "grad_norm": 1.2960482835769653, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8967373967170715, + "num_tokens": 285146474.0, + "step": 8122 + }, + { + "epoch": 1.5084493964716805, + "grad_norm": 1.5494189262390137, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8865014314651489, + "num_tokens": 285176158.0, + "step": 8123 + }, + { + "epoch": 1.5086350974930363, + "grad_norm": 1.4494420289993286, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8873609304428101, + "num_tokens": 285209979.0, + "step": 8124 + }, + { + "epoch": 1.508820798514392, + "grad_norm": 1.4626222848892212, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8717848658561707, + "num_tokens": 285250371.0, + "step": 8125 + }, + { + "epoch": 1.5090064995357473, + "grad_norm": 1.4264984130859375, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.88332599401474, + "num_tokens": 285288485.0, + "step": 8126 + }, + { + "epoch": 1.509192200557103, + "grad_norm": 1.5261077880859375, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8696020245552063, + "num_tokens": 285324272.0, + "step": 8127 + }, + { + "epoch": 1.5093779015784587, + "grad_norm": 1.4710513353347778, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8804301619529724, + "num_tokens": 285357195.0, + "step": 8128 + }, + { + "epoch": 1.5095636025998143, + "grad_norm": 1.489287257194519, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.888919472694397, + "num_tokens": 285387742.0, + "step": 8129 + }, + { + "epoch": 1.5097493036211698, + "grad_norm": 1.509644865989685, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8849923610687256, + "num_tokens": 285423693.0, + "step": 8130 + }, + { + "epoch": 1.5099350046425255, + "grad_norm": 1.4176948070526123, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8843985795974731, + "num_tokens": 285459576.0, + "step": 8131 + }, + { + "epoch": 1.5101207056638812, + "grad_norm": 1.5669618844985962, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8670285940170288, + "num_tokens": 285493057.0, + "step": 8132 + }, + { + "epoch": 1.5103064066852367, + "grad_norm": 1.401352047920227, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8730607032775879, + "num_tokens": 285537169.0, + "step": 8133 + }, + { + "epoch": 1.5104921077065923, + "grad_norm": 1.4046982526779175, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8664303421974182, + "num_tokens": 285578759.0, + "step": 8134 + }, + { + "epoch": 1.510677808727948, + "grad_norm": 1.4726240634918213, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8842515349388123, + "num_tokens": 285611297.0, + "step": 8135 + }, + { + "epoch": 1.5108635097493037, + "grad_norm": 1.349291443824768, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8855903148651123, + "num_tokens": 285648432.0, + "step": 8136 + }, + { + "epoch": 1.5110492107706592, + "grad_norm": 1.5611850023269653, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.875885009765625, + "num_tokens": 285681208.0, + "step": 8137 + }, + { + "epoch": 1.5112349117920147, + "grad_norm": 1.5487351417541504, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8624453544616699, + "num_tokens": 285718564.0, + "step": 8138 + }, + { + "epoch": 1.5114206128133705, + "grad_norm": 1.4490010738372803, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8735753297805786, + "num_tokens": 285756024.0, + "step": 8139 + }, + { + "epoch": 1.5116063138347262, + "grad_norm": 1.580899715423584, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8831461668014526, + "num_tokens": 285786669.0, + "step": 8140 + }, + { + "epoch": 1.5117920148560817, + "grad_norm": 1.5398300886154175, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8641691207885742, + "num_tokens": 285821016.0, + "step": 8141 + }, + { + "epoch": 1.5119777158774372, + "grad_norm": 1.5818183422088623, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8753253817558289, + "num_tokens": 285852056.0, + "step": 8142 + }, + { + "epoch": 1.512163416898793, + "grad_norm": 1.5229616165161133, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8637213110923767, + "num_tokens": 285888204.0, + "step": 8143 + }, + { + "epoch": 1.5123491179201487, + "grad_norm": 1.5190678834915161, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.8685310482978821, + "num_tokens": 285924123.0, + "step": 8144 + }, + { + "epoch": 1.5125348189415042, + "grad_norm": 1.3747186660766602, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8833715915679932, + "num_tokens": 285965061.0, + "step": 8145 + }, + { + "epoch": 1.5127205199628597, + "grad_norm": 1.5922856330871582, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8849118947982788, + "num_tokens": 285994502.0, + "step": 8146 + }, + { + "epoch": 1.5129062209842155, + "grad_norm": 1.4979743957519531, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8708063364028931, + "num_tokens": 286030834.0, + "step": 8147 + }, + { + "epoch": 1.5130919220055712, + "grad_norm": 1.591380000114441, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8766340613365173, + "num_tokens": 286062447.0, + "step": 8148 + }, + { + "epoch": 1.5132776230269267, + "grad_norm": 1.5150203704833984, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8611000776290894, + "num_tokens": 286099284.0, + "step": 8149 + }, + { + "epoch": 1.5134633240482822, + "grad_norm": 1.595460295677185, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8761428594589233, + "num_tokens": 286131756.0, + "step": 8150 + }, + { + "epoch": 1.513649025069638, + "grad_norm": 1.4326531887054443, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.8653543591499329, + "num_tokens": 286173955.0, + "step": 8151 + }, + { + "epoch": 1.5138347260909935, + "grad_norm": 1.435451626777649, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.889448881149292, + "num_tokens": 286209429.0, + "step": 8152 + }, + { + "epoch": 1.514020427112349, + "grad_norm": 1.7340096235275269, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8741515874862671, + "num_tokens": 286235589.0, + "step": 8153 + }, + { + "epoch": 1.5142061281337047, + "grad_norm": 1.584037184715271, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8728315234184265, + "num_tokens": 286265468.0, + "step": 8154 + }, + { + "epoch": 1.5143918291550604, + "grad_norm": 1.4842731952667236, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8622777462005615, + "num_tokens": 286304726.0, + "step": 8155 + }, + { + "epoch": 1.514577530176416, + "grad_norm": 1.505624532699585, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8838292956352234, + "num_tokens": 286337585.0, + "step": 8156 + }, + { + "epoch": 1.5147632311977715, + "grad_norm": 1.5609757900238037, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8822382688522339, + "num_tokens": 286371963.0, + "step": 8157 + }, + { + "epoch": 1.5149489322191272, + "grad_norm": 1.5449696779251099, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8786190152168274, + "num_tokens": 286406232.0, + "step": 8158 + }, + { + "epoch": 1.515134633240483, + "grad_norm": 1.5343838930130005, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8760769367218018, + "num_tokens": 286439760.0, + "step": 8159 + }, + { + "epoch": 1.5153203342618384, + "grad_norm": 1.3597378730773926, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8883073329925537, + "num_tokens": 286479980.0, + "step": 8160 + }, + { + "epoch": 1.515506035283194, + "grad_norm": 1.4414232969284058, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8644678592681885, + "num_tokens": 286522717.0, + "step": 8161 + }, + { + "epoch": 1.5156917363045497, + "grad_norm": 1.722834587097168, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8641144037246704, + "num_tokens": 286551765.0, + "step": 8162 + }, + { + "epoch": 1.5158774373259054, + "grad_norm": 1.3943325281143188, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8864272832870483, + "num_tokens": 286587405.0, + "step": 8163 + }, + { + "epoch": 1.516063138347261, + "grad_norm": 1.5772740840911865, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8754405975341797, + "num_tokens": 286623200.0, + "step": 8164 + }, + { + "epoch": 1.5162488393686164, + "grad_norm": 1.611512303352356, + "learning_rate": 1e-06, + "loss": 0.4329, + "mean_token_accuracy": 0.8572832345962524, + "num_tokens": 286656551.0, + "step": 8165 + }, + { + "epoch": 1.5164345403899722, + "grad_norm": 1.5960297584533691, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8751249313354492, + "num_tokens": 286688385.0, + "step": 8166 + }, + { + "epoch": 1.516620241411328, + "grad_norm": 1.3971444368362427, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.869101881980896, + "num_tokens": 286727501.0, + "step": 8167 + }, + { + "epoch": 1.5168059424326834, + "grad_norm": 1.530020833015442, + "learning_rate": 1e-06, + "loss": 0.4104, + "mean_token_accuracy": 0.857615053653717, + "num_tokens": 286766692.0, + "step": 8168 + }, + { + "epoch": 1.516991643454039, + "grad_norm": 1.533968210220337, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8696819543838501, + "num_tokens": 286802953.0, + "step": 8169 + }, + { + "epoch": 1.5171773444753947, + "grad_norm": 1.4245322942733765, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8674079775810242, + "num_tokens": 286842673.0, + "step": 8170 + }, + { + "epoch": 1.5173630454967504, + "grad_norm": 1.547166347503662, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8659458756446838, + "num_tokens": 286878549.0, + "step": 8171 + }, + { + "epoch": 1.517548746518106, + "grad_norm": 1.6023656129837036, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.872084379196167, + "num_tokens": 286910092.0, + "step": 8172 + }, + { + "epoch": 1.5177344475394614, + "grad_norm": 1.4546840190887451, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8699804544448853, + "num_tokens": 286947941.0, + "step": 8173 + }, + { + "epoch": 1.5179201485608171, + "grad_norm": 1.4925706386566162, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8868100643157959, + "num_tokens": 286980818.0, + "step": 8174 + }, + { + "epoch": 1.5181058495821727, + "grad_norm": 1.621745228767395, + "learning_rate": 1e-06, + "loss": 0.4225, + "mean_token_accuracy": 0.8565468192100525, + "num_tokens": 287013258.0, + "step": 8175 + }, + { + "epoch": 1.5182915506035282, + "grad_norm": 1.557361364364624, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8734058141708374, + "num_tokens": 287045294.0, + "step": 8176 + }, + { + "epoch": 1.518477251624884, + "grad_norm": 1.4590729475021362, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8724437355995178, + "num_tokens": 287084100.0, + "step": 8177 + }, + { + "epoch": 1.5186629526462396, + "grad_norm": 1.5049412250518799, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.878896951675415, + "num_tokens": 287117527.0, + "step": 8178 + }, + { + "epoch": 1.5188486536675951, + "grad_norm": 1.5063751935958862, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.875681459903717, + "num_tokens": 287150126.0, + "step": 8179 + }, + { + "epoch": 1.5190343546889506, + "grad_norm": 1.5011705160140991, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8824723958969116, + "num_tokens": 287182474.0, + "step": 8180 + }, + { + "epoch": 1.5192200557103064, + "grad_norm": 1.5741044282913208, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.8669419288635254, + "num_tokens": 287216370.0, + "step": 8181 + }, + { + "epoch": 1.5194057567316621, + "grad_norm": 1.5914944410324097, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8697655200958252, + "num_tokens": 287252179.0, + "step": 8182 + }, + { + "epoch": 1.5195914577530176, + "grad_norm": 1.642840027809143, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8786472678184509, + "num_tokens": 287282827.0, + "step": 8183 + }, + { + "epoch": 1.5197771587743731, + "grad_norm": 1.4772005081176758, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8835779428482056, + "num_tokens": 287319020.0, + "step": 8184 + }, + { + "epoch": 1.5199628597957289, + "grad_norm": 1.4386988878250122, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8697341680526733, + "num_tokens": 287357108.0, + "step": 8185 + }, + { + "epoch": 1.5201485608170846, + "grad_norm": 1.6578752994537354, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8743864893913269, + "num_tokens": 287384228.0, + "step": 8186 + }, + { + "epoch": 1.5203342618384401, + "grad_norm": 1.3694528341293335, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8891322612762451, + "num_tokens": 287424401.0, + "step": 8187 + }, + { + "epoch": 1.5205199628597956, + "grad_norm": 1.5400968790054321, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8792954087257385, + "num_tokens": 287461819.0, + "step": 8188 + }, + { + "epoch": 1.5207056638811514, + "grad_norm": 1.5141185522079468, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8748866319656372, + "num_tokens": 287497460.0, + "step": 8189 + }, + { + "epoch": 1.520891364902507, + "grad_norm": 1.4783406257629395, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8650798201560974, + "num_tokens": 287534069.0, + "step": 8190 + }, + { + "epoch": 1.5210770659238626, + "grad_norm": 1.5603865385055542, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8724879026412964, + "num_tokens": 287568875.0, + "step": 8191 + }, + { + "epoch": 1.5212627669452181, + "grad_norm": 1.4615947008132935, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8852648138999939, + "num_tokens": 287605706.0, + "step": 8192 + }, + { + "epoch": 1.5214484679665738, + "grad_norm": 1.6829808950424194, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8749942183494568, + "num_tokens": 287635490.0, + "step": 8193 + }, + { + "epoch": 1.5216341689879296, + "grad_norm": 1.4974408149719238, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8711876273155212, + "num_tokens": 287672716.0, + "step": 8194 + }, + { + "epoch": 1.521819870009285, + "grad_norm": 1.5414516925811768, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8732625842094421, + "num_tokens": 287705509.0, + "step": 8195 + }, + { + "epoch": 1.5220055710306406, + "grad_norm": 1.4219051599502563, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8710711598396301, + "num_tokens": 287744501.0, + "step": 8196 + }, + { + "epoch": 1.5221912720519963, + "grad_norm": 1.474829077720642, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8762227296829224, + "num_tokens": 287785432.0, + "step": 8197 + }, + { + "epoch": 1.522376973073352, + "grad_norm": 1.6126779317855835, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8820092082023621, + "num_tokens": 287815062.0, + "step": 8198 + }, + { + "epoch": 1.5225626740947074, + "grad_norm": 1.478440284729004, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8794945478439331, + "num_tokens": 287849573.0, + "step": 8199 + }, + { + "epoch": 1.522748375116063, + "grad_norm": 1.4653089046478271, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8823416233062744, + "num_tokens": 287885875.0, + "step": 8200 + }, + { + "epoch": 1.5229340761374188, + "grad_norm": 1.4729015827178955, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8715846538543701, + "num_tokens": 287919989.0, + "step": 8201 + }, + { + "epoch": 1.5231197771587743, + "grad_norm": 1.5604548454284668, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8750779032707214, + "num_tokens": 287953919.0, + "step": 8202 + }, + { + "epoch": 1.5233054781801298, + "grad_norm": 1.525644063949585, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8713468313217163, + "num_tokens": 287989639.0, + "step": 8203 + }, + { + "epoch": 1.5234911792014856, + "grad_norm": 1.51052987575531, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8808377385139465, + "num_tokens": 288023436.0, + "step": 8204 + }, + { + "epoch": 1.5236768802228413, + "grad_norm": 1.3795673847198486, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8910089135169983, + "num_tokens": 288062958.0, + "step": 8205 + }, + { + "epoch": 1.5238625812441968, + "grad_norm": 1.4912989139556885, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.880640983581543, + "num_tokens": 288095528.0, + "step": 8206 + }, + { + "epoch": 1.5240482822655523, + "grad_norm": 1.5691535472869873, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8765257596969604, + "num_tokens": 288129205.0, + "step": 8207 + }, + { + "epoch": 1.524233983286908, + "grad_norm": 1.4457613229751587, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8777450919151306, + "num_tokens": 288168170.0, + "step": 8208 + }, + { + "epoch": 1.5244196843082638, + "grad_norm": 1.5791743993759155, + "learning_rate": 1e-06, + "loss": 0.3994, + "mean_token_accuracy": 0.8608429431915283, + "num_tokens": 288202427.0, + "step": 8209 + }, + { + "epoch": 1.5246053853296193, + "grad_norm": 1.6376631259918213, + "learning_rate": 1e-06, + "loss": 0.4009, + "mean_token_accuracy": 0.862755537033081, + "num_tokens": 288234459.0, + "step": 8210 + }, + { + "epoch": 1.5247910863509748, + "grad_norm": 1.4882750511169434, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8749717473983765, + "num_tokens": 288271627.0, + "step": 8211 + }, + { + "epoch": 1.5249767873723306, + "grad_norm": 1.4543273448944092, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8743264675140381, + "num_tokens": 288311271.0, + "step": 8212 + }, + { + "epoch": 1.5251624883936863, + "grad_norm": 1.5784592628479004, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.87070631980896, + "num_tokens": 288344444.0, + "step": 8213 + }, + { + "epoch": 1.5253481894150418, + "grad_norm": 1.4533544778823853, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8721734285354614, + "num_tokens": 288380774.0, + "step": 8214 + }, + { + "epoch": 1.5255338904363973, + "grad_norm": 1.5509127378463745, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8676068782806396, + "num_tokens": 288416080.0, + "step": 8215 + }, + { + "epoch": 1.525719591457753, + "grad_norm": 1.559874176979065, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8721666932106018, + "num_tokens": 288447020.0, + "step": 8216 + }, + { + "epoch": 1.5259052924791088, + "grad_norm": 1.4806482791900635, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8822576403617859, + "num_tokens": 288481300.0, + "step": 8217 + }, + { + "epoch": 1.5260909935004643, + "grad_norm": 1.6225665807724, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8648059368133545, + "num_tokens": 288513563.0, + "step": 8218 + }, + { + "epoch": 1.5262766945218198, + "grad_norm": 1.5104626417160034, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8779696226119995, + "num_tokens": 288546556.0, + "step": 8219 + }, + { + "epoch": 1.5264623955431755, + "grad_norm": 1.476572871208191, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8752439022064209, + "num_tokens": 288585475.0, + "step": 8220 + }, + { + "epoch": 1.5266480965645313, + "grad_norm": 1.5122634172439575, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8753151893615723, + "num_tokens": 288620622.0, + "step": 8221 + }, + { + "epoch": 1.5268337975858868, + "grad_norm": 1.421501874923706, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8922672271728516, + "num_tokens": 288658628.0, + "step": 8222 + }, + { + "epoch": 1.5270194986072423, + "grad_norm": 1.450788140296936, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8886966705322266, + "num_tokens": 288689239.0, + "step": 8223 + }, + { + "epoch": 1.527205199628598, + "grad_norm": 1.677999496459961, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8802839517593384, + "num_tokens": 288719261.0, + "step": 8224 + }, + { + "epoch": 1.5273909006499535, + "grad_norm": 1.7882522344589233, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.867789626121521, + "num_tokens": 288751455.0, + "step": 8225 + }, + { + "epoch": 1.527576601671309, + "grad_norm": 1.6064502000808716, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8664098978042603, + "num_tokens": 288783511.0, + "step": 8226 + }, + { + "epoch": 1.5277623026926648, + "grad_norm": 1.5187435150146484, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8715421557426453, + "num_tokens": 288817444.0, + "step": 8227 + }, + { + "epoch": 1.5279480037140205, + "grad_norm": 1.5437339544296265, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8777428865432739, + "num_tokens": 288851818.0, + "step": 8228 + }, + { + "epoch": 1.528133704735376, + "grad_norm": 1.44564688205719, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8818598985671997, + "num_tokens": 288886217.0, + "step": 8229 + }, + { + "epoch": 1.5283194057567315, + "grad_norm": 1.4661452770233154, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8807393312454224, + "num_tokens": 288920721.0, + "step": 8230 + }, + { + "epoch": 1.5285051067780873, + "grad_norm": 1.5107771158218384, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8874920010566711, + "num_tokens": 288954965.0, + "step": 8231 + }, + { + "epoch": 1.528690807799443, + "grad_norm": 1.5967990159988403, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8873228430747986, + "num_tokens": 288988639.0, + "step": 8232 + }, + { + "epoch": 1.5288765088207985, + "grad_norm": 1.5523021221160889, + "learning_rate": 1e-06, + "loss": 0.4164, + "mean_token_accuracy": 0.8622322678565979, + "num_tokens": 289020096.0, + "step": 8233 + }, + { + "epoch": 1.529062209842154, + "grad_norm": 1.5173871517181396, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8641800284385681, + "num_tokens": 289057032.0, + "step": 8234 + }, + { + "epoch": 1.5292479108635098, + "grad_norm": 1.6824703216552734, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8745155334472656, + "num_tokens": 289084984.0, + "step": 8235 + }, + { + "epoch": 1.5294336118848655, + "grad_norm": 1.5923285484313965, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8648576736450195, + "num_tokens": 289117051.0, + "step": 8236 + }, + { + "epoch": 1.529619312906221, + "grad_norm": 1.5160512924194336, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8740140199661255, + "num_tokens": 289151688.0, + "step": 8237 + }, + { + "epoch": 1.5298050139275765, + "grad_norm": 1.6548347473144531, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8686047792434692, + "num_tokens": 289180403.0, + "step": 8238 + }, + { + "epoch": 1.5299907149489322, + "grad_norm": 1.4773342609405518, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8828070163726807, + "num_tokens": 289218364.0, + "step": 8239 + }, + { + "epoch": 1.530176415970288, + "grad_norm": 1.5459380149841309, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8660114407539368, + "num_tokens": 289253279.0, + "step": 8240 + }, + { + "epoch": 1.5303621169916435, + "grad_norm": 1.6817498207092285, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8724426031112671, + "num_tokens": 289283844.0, + "step": 8241 + }, + { + "epoch": 1.530547818012999, + "grad_norm": 1.406723141670227, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8910341262817383, + "num_tokens": 289319888.0, + "step": 8242 + }, + { + "epoch": 1.5307335190343547, + "grad_norm": 1.5880236625671387, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8767993450164795, + "num_tokens": 289349737.0, + "step": 8243 + }, + { + "epoch": 1.5309192200557105, + "grad_norm": 1.5181328058242798, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8810492753982544, + "num_tokens": 289381633.0, + "step": 8244 + }, + { + "epoch": 1.531104921077066, + "grad_norm": 1.6193041801452637, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8721067905426025, + "num_tokens": 289413649.0, + "step": 8245 + }, + { + "epoch": 1.5312906220984215, + "grad_norm": 1.5523604154586792, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8713592886924744, + "num_tokens": 289447273.0, + "step": 8246 + }, + { + "epoch": 1.5314763231197772, + "grad_norm": 1.6647582054138184, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8678421974182129, + "num_tokens": 289476961.0, + "step": 8247 + }, + { + "epoch": 1.5316620241411327, + "grad_norm": 1.443747639656067, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8771487474441528, + "num_tokens": 289516069.0, + "step": 8248 + }, + { + "epoch": 1.5318477251624882, + "grad_norm": 1.5432074069976807, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8801939487457275, + "num_tokens": 289548529.0, + "step": 8249 + }, + { + "epoch": 1.532033426183844, + "grad_norm": 1.8359750509262085, + "learning_rate": 1e-06, + "loss": 0.4086, + "mean_token_accuracy": 0.8619487285614014, + "num_tokens": 289575616.0, + "step": 8250 + }, + { + "epoch": 1.5322191272051997, + "grad_norm": 1.5725172758102417, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8795772790908813, + "num_tokens": 289604470.0, + "step": 8251 + }, + { + "epoch": 1.5324048282265552, + "grad_norm": 1.4730063676834106, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8656057119369507, + "num_tokens": 289642000.0, + "step": 8252 + }, + { + "epoch": 1.5325905292479107, + "grad_norm": 1.6865054368972778, + "learning_rate": 1e-06, + "loss": 0.4291, + "mean_token_accuracy": 0.851798415184021, + "num_tokens": 289673614.0, + "step": 8253 + }, + { + "epoch": 1.5327762302692665, + "grad_norm": 1.6413073539733887, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8856559991836548, + "num_tokens": 289708013.0, + "step": 8254 + }, + { + "epoch": 1.5329619312906222, + "grad_norm": 1.5800909996032715, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8751957416534424, + "num_tokens": 289740822.0, + "step": 8255 + }, + { + "epoch": 1.5331476323119777, + "grad_norm": 1.4717849493026733, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.8685762882232666, + "num_tokens": 289779171.0, + "step": 8256 + }, + { + "epoch": 1.5333333333333332, + "grad_norm": 1.5118263959884644, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.866741955280304, + "num_tokens": 289817537.0, + "step": 8257 + }, + { + "epoch": 1.533519034354689, + "grad_norm": 1.5859678983688354, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8891613483428955, + "num_tokens": 289847670.0, + "step": 8258 + }, + { + "epoch": 1.5337047353760447, + "grad_norm": 1.5095274448394775, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8757574558258057, + "num_tokens": 289879855.0, + "step": 8259 + }, + { + "epoch": 1.5338904363974002, + "grad_norm": 1.3797999620437622, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8954096436500549, + "num_tokens": 289914361.0, + "step": 8260 + }, + { + "epoch": 1.5340761374187557, + "grad_norm": 1.5571337938308716, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8798339366912842, + "num_tokens": 289944751.0, + "step": 8261 + }, + { + "epoch": 1.5342618384401114, + "grad_norm": 1.3148057460784912, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8770110607147217, + "num_tokens": 289990130.0, + "step": 8262 + }, + { + "epoch": 1.5344475394614672, + "grad_norm": 1.3821475505828857, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8876196146011353, + "num_tokens": 290026036.0, + "step": 8263 + }, + { + "epoch": 1.5346332404828227, + "grad_norm": 1.4183331727981567, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8724855184555054, + "num_tokens": 290062487.0, + "step": 8264 + }, + { + "epoch": 1.5348189415041782, + "grad_norm": 1.549765944480896, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8674895167350769, + "num_tokens": 290094582.0, + "step": 8265 + }, + { + "epoch": 1.535004642525534, + "grad_norm": 1.4421991109848022, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8871314525604248, + "num_tokens": 290130173.0, + "step": 8266 + }, + { + "epoch": 1.5351903435468897, + "grad_norm": 1.583932638168335, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8821289539337158, + "num_tokens": 290161959.0, + "step": 8267 + }, + { + "epoch": 1.5353760445682452, + "grad_norm": 1.4294027090072632, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8820683360099792, + "num_tokens": 290199416.0, + "step": 8268 + }, + { + "epoch": 1.5355617455896007, + "grad_norm": 1.46769118309021, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8746029138565063, + "num_tokens": 290234291.0, + "step": 8269 + }, + { + "epoch": 1.5357474466109564, + "grad_norm": 1.4513630867004395, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8789411783218384, + "num_tokens": 290270645.0, + "step": 8270 + }, + { + "epoch": 1.535933147632312, + "grad_norm": 1.4779465198516846, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8813729882240295, + "num_tokens": 290306164.0, + "step": 8271 + }, + { + "epoch": 1.5361188486536674, + "grad_norm": 1.4941526651382446, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8881043791770935, + "num_tokens": 290340657.0, + "step": 8272 + }, + { + "epoch": 1.5363045496750232, + "grad_norm": 1.5707831382751465, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8752175569534302, + "num_tokens": 290373742.0, + "step": 8273 + }, + { + "epoch": 1.536490250696379, + "grad_norm": 1.581971287727356, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8649677038192749, + "num_tokens": 290408520.0, + "step": 8274 + }, + { + "epoch": 1.5366759517177344, + "grad_norm": 1.624546766281128, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8650431632995605, + "num_tokens": 290441452.0, + "step": 8275 + }, + { + "epoch": 1.53686165273909, + "grad_norm": 1.4913132190704346, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8772362470626831, + "num_tokens": 290474305.0, + "step": 8276 + }, + { + "epoch": 1.5370473537604457, + "grad_norm": 1.5909581184387207, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8716232776641846, + "num_tokens": 290507169.0, + "step": 8277 + }, + { + "epoch": 1.5372330547818014, + "grad_norm": 1.5194010734558105, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8733595609664917, + "num_tokens": 290540075.0, + "step": 8278 + }, + { + "epoch": 1.537418755803157, + "grad_norm": 1.4676315784454346, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8703022003173828, + "num_tokens": 290579327.0, + "step": 8279 + }, + { + "epoch": 1.5376044568245124, + "grad_norm": 1.662945032119751, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8797404766082764, + "num_tokens": 290614243.0, + "step": 8280 + }, + { + "epoch": 1.5377901578458681, + "grad_norm": 1.4528290033340454, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8651694059371948, + "num_tokens": 290653155.0, + "step": 8281 + }, + { + "epoch": 1.5379758588672239, + "grad_norm": 1.4104840755462646, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8897676467895508, + "num_tokens": 290693114.0, + "step": 8282 + }, + { + "epoch": 1.5381615598885794, + "grad_norm": 1.5084245204925537, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8671665191650391, + "num_tokens": 290727346.0, + "step": 8283 + }, + { + "epoch": 1.538347260909935, + "grad_norm": 1.5441559553146362, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8778474926948547, + "num_tokens": 290760325.0, + "step": 8284 + }, + { + "epoch": 1.5385329619312906, + "grad_norm": 1.40705406665802, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.8957467079162598, + "num_tokens": 290797893.0, + "step": 8285 + }, + { + "epoch": 1.5387186629526464, + "grad_norm": 1.796321988105774, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8701428771018982, + "num_tokens": 290836070.0, + "step": 8286 + }, + { + "epoch": 1.5389043639740019, + "grad_norm": 1.627045750617981, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8751847743988037, + "num_tokens": 290865687.0, + "step": 8287 + }, + { + "epoch": 1.5390900649953574, + "grad_norm": 1.3694384098052979, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8835949897766113, + "num_tokens": 290906560.0, + "step": 8288 + }, + { + "epoch": 1.5392757660167131, + "grad_norm": 1.7280514240264893, + "learning_rate": 1e-06, + "loss": 0.4036, + "mean_token_accuracy": 0.8610733151435852, + "num_tokens": 290944525.0, + "step": 8289 + }, + { + "epoch": 1.5394614670380689, + "grad_norm": 1.495348334312439, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8692266941070557, + "num_tokens": 290981722.0, + "step": 8290 + }, + { + "epoch": 1.5396471680594244, + "grad_norm": 1.432395339012146, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8846516013145447, + "num_tokens": 291018040.0, + "step": 8291 + }, + { + "epoch": 1.5398328690807799, + "grad_norm": 1.4423794746398926, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8617613315582275, + "num_tokens": 291059226.0, + "step": 8292 + }, + { + "epoch": 1.5400185701021356, + "grad_norm": 1.507192850112915, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8781967163085938, + "num_tokens": 291091775.0, + "step": 8293 + }, + { + "epoch": 1.5402042711234913, + "grad_norm": 1.4075852632522583, + "learning_rate": 1e-06, + "loss": 0.2929, + "mean_token_accuracy": 0.9035366177558899, + "num_tokens": 291127322.0, + "step": 8294 + }, + { + "epoch": 1.5403899721448466, + "grad_norm": 1.6013184785842896, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8753082752227783, + "num_tokens": 291159859.0, + "step": 8295 + }, + { + "epoch": 1.5405756731662024, + "grad_norm": 1.4957131147384644, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8762514591217041, + "num_tokens": 291198697.0, + "step": 8296 + }, + { + "epoch": 1.540761374187558, + "grad_norm": 1.5371038913726807, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8692283034324646, + "num_tokens": 291233800.0, + "step": 8297 + }, + { + "epoch": 1.5409470752089136, + "grad_norm": 1.5552780628204346, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8722290992736816, + "num_tokens": 291270441.0, + "step": 8298 + }, + { + "epoch": 1.5411327762302691, + "grad_norm": 1.3847873210906982, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8837895393371582, + "num_tokens": 291309413.0, + "step": 8299 + }, + { + "epoch": 1.5413184772516249, + "grad_norm": 1.4090967178344727, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.861446738243103, + "num_tokens": 291348202.0, + "step": 8300 + }, + { + "epoch": 1.5415041782729806, + "grad_norm": 1.6130458116531372, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8704092502593994, + "num_tokens": 291383815.0, + "step": 8301 + }, + { + "epoch": 1.541689879294336, + "grad_norm": 1.5776628255844116, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8758058547973633, + "num_tokens": 291413947.0, + "step": 8302 + }, + { + "epoch": 1.5418755803156916, + "grad_norm": 1.5662810802459717, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8833431601524353, + "num_tokens": 291450277.0, + "step": 8303 + }, + { + "epoch": 1.5420612813370473, + "grad_norm": 1.6200357675552368, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8833072185516357, + "num_tokens": 291480181.0, + "step": 8304 + }, + { + "epoch": 1.542246982358403, + "grad_norm": 1.6734355688095093, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8763633966445923, + "num_tokens": 291510892.0, + "step": 8305 + }, + { + "epoch": 1.5424326833797586, + "grad_norm": 1.3975284099578857, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8815950155258179, + "num_tokens": 291546094.0, + "step": 8306 + }, + { + "epoch": 1.542618384401114, + "grad_norm": 1.3809040784835815, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8961688876152039, + "num_tokens": 291581643.0, + "step": 8307 + }, + { + "epoch": 1.5428040854224698, + "grad_norm": 1.6180386543273926, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.869127631187439, + "num_tokens": 291619488.0, + "step": 8308 + }, + { + "epoch": 1.5429897864438256, + "grad_norm": 1.5101572275161743, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8841037750244141, + "num_tokens": 291655748.0, + "step": 8309 + }, + { + "epoch": 1.543175487465181, + "grad_norm": 1.4564054012298584, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8911420702934265, + "num_tokens": 291689578.0, + "step": 8310 + }, + { + "epoch": 1.5433611884865366, + "grad_norm": 1.5936723947525024, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8781147599220276, + "num_tokens": 291719394.0, + "step": 8311 + }, + { + "epoch": 1.5435468895078923, + "grad_norm": 1.5067546367645264, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8784257173538208, + "num_tokens": 291751302.0, + "step": 8312 + }, + { + "epoch": 1.543732590529248, + "grad_norm": 1.585871696472168, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8751015663146973, + "num_tokens": 291785527.0, + "step": 8313 + }, + { + "epoch": 1.5439182915506036, + "grad_norm": 1.5520011186599731, + "learning_rate": 1e-06, + "loss": 0.399, + "mean_token_accuracy": 0.8629655838012695, + "num_tokens": 291822371.0, + "step": 8314 + }, + { + "epoch": 1.544103992571959, + "grad_norm": 1.6819754838943481, + "learning_rate": 1e-06, + "loss": 0.3978, + "mean_token_accuracy": 0.8653319478034973, + "num_tokens": 291854691.0, + "step": 8315 + }, + { + "epoch": 1.5442896935933148, + "grad_norm": 1.5878371000289917, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8715073466300964, + "num_tokens": 291891822.0, + "step": 8316 + }, + { + "epoch": 1.5444753946146705, + "grad_norm": 1.3710440397262573, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8916144371032715, + "num_tokens": 291930811.0, + "step": 8317 + }, + { + "epoch": 1.544661095636026, + "grad_norm": 1.533861517906189, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8635664582252502, + "num_tokens": 291965128.0, + "step": 8318 + }, + { + "epoch": 1.5448467966573816, + "grad_norm": 1.5257282257080078, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8712666034698486, + "num_tokens": 291997339.0, + "step": 8319 + }, + { + "epoch": 1.5450324976787373, + "grad_norm": 1.4323877096176147, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8821729421615601, + "num_tokens": 292033152.0, + "step": 8320 + }, + { + "epoch": 1.5452181987000928, + "grad_norm": 1.496280312538147, + "learning_rate": 1e-06, + "loss": 0.4092, + "mean_token_accuracy": 0.8603780269622803, + "num_tokens": 292071652.0, + "step": 8321 + }, + { + "epoch": 1.5454038997214483, + "grad_norm": 1.5220140218734741, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8828310966491699, + "num_tokens": 292103254.0, + "step": 8322 + }, + { + "epoch": 1.545589600742804, + "grad_norm": 1.6043434143066406, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.879856288433075, + "num_tokens": 292133101.0, + "step": 8323 + }, + { + "epoch": 1.5457753017641598, + "grad_norm": 1.4103960990905762, + "learning_rate": 1e-06, + "loss": 0.2984, + "mean_token_accuracy": 0.8948128819465637, + "num_tokens": 292165717.0, + "step": 8324 + }, + { + "epoch": 1.5459610027855153, + "grad_norm": 1.4276880025863647, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8780069351196289, + "num_tokens": 292201940.0, + "step": 8325 + }, + { + "epoch": 1.5461467038068708, + "grad_norm": 1.4518873691558838, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8891010284423828, + "num_tokens": 292236921.0, + "step": 8326 + }, + { + "epoch": 1.5463324048282265, + "grad_norm": 1.5020020008087158, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8949878811836243, + "num_tokens": 292269114.0, + "step": 8327 + }, + { + "epoch": 1.5465181058495823, + "grad_norm": 1.5590111017227173, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.8681784272193909, + "num_tokens": 292303938.0, + "step": 8328 + }, + { + "epoch": 1.5467038068709378, + "grad_norm": 1.515141487121582, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8916516304016113, + "num_tokens": 292338691.0, + "step": 8329 + }, + { + "epoch": 1.5468895078922933, + "grad_norm": 1.6235817670822144, + "learning_rate": 1e-06, + "loss": 0.4218, + "mean_token_accuracy": 0.8572827577590942, + "num_tokens": 292374479.0, + "step": 8330 + }, + { + "epoch": 1.547075208913649, + "grad_norm": 1.5242820978164673, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8788115978240967, + "num_tokens": 292407834.0, + "step": 8331 + }, + { + "epoch": 1.5472609099350048, + "grad_norm": 1.537624478340149, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8819540143013, + "num_tokens": 292440911.0, + "step": 8332 + }, + { + "epoch": 1.5474466109563603, + "grad_norm": 1.5020967721939087, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8832886219024658, + "num_tokens": 292476324.0, + "step": 8333 + }, + { + "epoch": 1.5476323119777158, + "grad_norm": 1.4536943435668945, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8741230964660645, + "num_tokens": 292521663.0, + "step": 8334 + }, + { + "epoch": 1.5478180129990715, + "grad_norm": 1.3401274681091309, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.8718380331993103, + "num_tokens": 292568990.0, + "step": 8335 + }, + { + "epoch": 1.5480037140204272, + "grad_norm": 1.4010941982269287, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8864476680755615, + "num_tokens": 292607543.0, + "step": 8336 + }, + { + "epoch": 1.5481894150417828, + "grad_norm": 1.494515299797058, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8638353943824768, + "num_tokens": 292647112.0, + "step": 8337 + }, + { + "epoch": 1.5483751160631383, + "grad_norm": 1.4174154996871948, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8870998620986938, + "num_tokens": 292684914.0, + "step": 8338 + }, + { + "epoch": 1.548560817084494, + "grad_norm": 1.4371833801269531, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8941566944122314, + "num_tokens": 292723418.0, + "step": 8339 + }, + { + "epoch": 1.5487465181058497, + "grad_norm": 1.5723406076431274, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8802168369293213, + "num_tokens": 292757533.0, + "step": 8340 + }, + { + "epoch": 1.5489322191272052, + "grad_norm": 1.528407335281372, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8821331262588501, + "num_tokens": 292790034.0, + "step": 8341 + }, + { + "epoch": 1.5491179201485608, + "grad_norm": 1.4636820554733276, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8776909112930298, + "num_tokens": 292827136.0, + "step": 8342 + }, + { + "epoch": 1.5493036211699165, + "grad_norm": 1.7730270624160767, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8856731653213501, + "num_tokens": 292865237.0, + "step": 8343 + }, + { + "epoch": 1.549489322191272, + "grad_norm": 1.432750940322876, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8860034942626953, + "num_tokens": 292899202.0, + "step": 8344 + }, + { + "epoch": 1.5496750232126275, + "grad_norm": 1.463186264038086, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8728126883506775, + "num_tokens": 292939806.0, + "step": 8345 + }, + { + "epoch": 1.5498607242339832, + "grad_norm": 1.6326587200164795, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8564696311950684, + "num_tokens": 292976555.0, + "step": 8346 + }, + { + "epoch": 1.550046425255339, + "grad_norm": 1.4247666597366333, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8756449818611145, + "num_tokens": 293016510.0, + "step": 8347 + }, + { + "epoch": 1.5502321262766945, + "grad_norm": 1.6396209001541138, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8800684213638306, + "num_tokens": 293048576.0, + "step": 8348 + }, + { + "epoch": 1.55041782729805, + "grad_norm": 1.536911964416504, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8681221008300781, + "num_tokens": 293083806.0, + "step": 8349 + }, + { + "epoch": 1.5506035283194057, + "grad_norm": 1.6287362575531006, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8730008006095886, + "num_tokens": 293113998.0, + "step": 8350 + }, + { + "epoch": 1.5507892293407615, + "grad_norm": 1.5977061986923218, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8809860348701477, + "num_tokens": 293147256.0, + "step": 8351 + }, + { + "epoch": 1.550974930362117, + "grad_norm": 1.6540051698684692, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8863507509231567, + "num_tokens": 293182663.0, + "step": 8352 + }, + { + "epoch": 1.5511606313834725, + "grad_norm": 1.6963794231414795, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8784776329994202, + "num_tokens": 293212776.0, + "step": 8353 + }, + { + "epoch": 1.5513463324048282, + "grad_norm": 1.528574824333191, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8610768914222717, + "num_tokens": 293251378.0, + "step": 8354 + }, + { + "epoch": 1.551532033426184, + "grad_norm": 1.5550205707550049, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.8611234426498413, + "num_tokens": 293287912.0, + "step": 8355 + }, + { + "epoch": 1.5517177344475395, + "grad_norm": 1.5192798376083374, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8748106360435486, + "num_tokens": 293324081.0, + "step": 8356 + }, + { + "epoch": 1.551903435468895, + "grad_norm": 1.6330682039260864, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8696914911270142, + "num_tokens": 293356943.0, + "step": 8357 + }, + { + "epoch": 1.5520891364902507, + "grad_norm": 1.5637012720108032, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8896269798278809, + "num_tokens": 293386802.0, + "step": 8358 + }, + { + "epoch": 1.5522748375116064, + "grad_norm": 1.5595029592514038, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8746119737625122, + "num_tokens": 293419082.0, + "step": 8359 + }, + { + "epoch": 1.552460538532962, + "grad_norm": 1.4241291284561157, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8730320334434509, + "num_tokens": 293458134.0, + "step": 8360 + }, + { + "epoch": 1.5526462395543175, + "grad_norm": 1.450636625289917, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8658466339111328, + "num_tokens": 293498092.0, + "step": 8361 + }, + { + "epoch": 1.5528319405756732, + "grad_norm": 1.42941153049469, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.88243168592453, + "num_tokens": 293531663.0, + "step": 8362 + }, + { + "epoch": 1.553017641597029, + "grad_norm": 1.4751594066619873, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8802756071090698, + "num_tokens": 293565246.0, + "step": 8363 + }, + { + "epoch": 1.5532033426183844, + "grad_norm": 1.3820281028747559, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8751558065414429, + "num_tokens": 293607138.0, + "step": 8364 + }, + { + "epoch": 1.55338904363974, + "grad_norm": 1.5208394527435303, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8816068768501282, + "num_tokens": 293641074.0, + "step": 8365 + }, + { + "epoch": 1.5535747446610957, + "grad_norm": 1.4844828844070435, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8793630599975586, + "num_tokens": 293674287.0, + "step": 8366 + }, + { + "epoch": 1.5537604456824514, + "grad_norm": 1.632779598236084, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8651008009910583, + "num_tokens": 293709806.0, + "step": 8367 + }, + { + "epoch": 1.5539461467038067, + "grad_norm": 1.568036437034607, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8703646659851074, + "num_tokens": 293743503.0, + "step": 8368 + }, + { + "epoch": 1.5541318477251624, + "grad_norm": 1.493133544921875, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8886591792106628, + "num_tokens": 293777407.0, + "step": 8369 + }, + { + "epoch": 1.5543175487465182, + "grad_norm": 1.5430059432983398, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8801696300506592, + "num_tokens": 293811584.0, + "step": 8370 + }, + { + "epoch": 1.5545032497678737, + "grad_norm": 1.5115406513214111, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8836105465888977, + "num_tokens": 293846812.0, + "step": 8371 + }, + { + "epoch": 1.5546889507892292, + "grad_norm": 1.5099018812179565, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8759914636611938, + "num_tokens": 293883572.0, + "step": 8372 + }, + { + "epoch": 1.554874651810585, + "grad_norm": 1.5475913286209106, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8726385831832886, + "num_tokens": 293917227.0, + "step": 8373 + }, + { + "epoch": 1.5550603528319407, + "grad_norm": 1.379759669303894, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.879546046257019, + "num_tokens": 293956855.0, + "step": 8374 + }, + { + "epoch": 1.5552460538532962, + "grad_norm": 1.4887489080429077, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8751069903373718, + "num_tokens": 293991536.0, + "step": 8375 + }, + { + "epoch": 1.5554317548746517, + "grad_norm": 1.4221158027648926, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.880976140499115, + "num_tokens": 294029720.0, + "step": 8376 + }, + { + "epoch": 1.5556174558960074, + "grad_norm": 1.5143879652023315, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8662317991256714, + "num_tokens": 294065271.0, + "step": 8377 + }, + { + "epoch": 1.5558031569173632, + "grad_norm": 1.5991029739379883, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8790429830551147, + "num_tokens": 294094478.0, + "step": 8378 + }, + { + "epoch": 1.5559888579387187, + "grad_norm": 1.7041908502578735, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8773657083511353, + "num_tokens": 294121352.0, + "step": 8379 + }, + { + "epoch": 1.5561745589600742, + "grad_norm": 1.515450119972229, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8761062026023865, + "num_tokens": 294156757.0, + "step": 8380 + }, + { + "epoch": 1.55636025998143, + "grad_norm": 1.4974720478057861, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8808808326721191, + "num_tokens": 294190263.0, + "step": 8381 + }, + { + "epoch": 1.5565459610027856, + "grad_norm": 1.533027172088623, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8701551556587219, + "num_tokens": 294226388.0, + "step": 8382 + }, + { + "epoch": 1.5567316620241411, + "grad_norm": 1.4545952081680298, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8743911981582642, + "num_tokens": 294261331.0, + "step": 8383 + }, + { + "epoch": 1.5569173630454967, + "grad_norm": 1.3707374334335327, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8874750137329102, + "num_tokens": 294300380.0, + "step": 8384 + }, + { + "epoch": 1.5571030640668524, + "grad_norm": 1.5471060276031494, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8676374554634094, + "num_tokens": 294336833.0, + "step": 8385 + }, + { + "epoch": 1.5572887650882081, + "grad_norm": 1.5939626693725586, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8832123279571533, + "num_tokens": 294368828.0, + "step": 8386 + }, + { + "epoch": 1.5574744661095636, + "grad_norm": 1.5912057161331177, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8826320171356201, + "num_tokens": 294398072.0, + "step": 8387 + }, + { + "epoch": 1.5576601671309191, + "grad_norm": 1.6723576784133911, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8629550337791443, + "num_tokens": 294429629.0, + "step": 8388 + }, + { + "epoch": 1.5578458681522749, + "grad_norm": 1.4704560041427612, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8802903294563293, + "num_tokens": 294466001.0, + "step": 8389 + }, + { + "epoch": 1.5580315691736306, + "grad_norm": 1.359900951385498, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8810251951217651, + "num_tokens": 294506450.0, + "step": 8390 + }, + { + "epoch": 1.5582172701949861, + "grad_norm": 1.6218948364257812, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8681445717811584, + "num_tokens": 294536286.0, + "step": 8391 + }, + { + "epoch": 1.5584029712163416, + "grad_norm": 1.4872530698776245, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.873996376991272, + "num_tokens": 294572170.0, + "step": 8392 + }, + { + "epoch": 1.5585886722376974, + "grad_norm": 1.3813170194625854, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.886228084564209, + "num_tokens": 294611066.0, + "step": 8393 + }, + { + "epoch": 1.5587743732590529, + "grad_norm": 1.5093255043029785, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8713487982749939, + "num_tokens": 294648212.0, + "step": 8394 + }, + { + "epoch": 1.5589600742804084, + "grad_norm": 1.4501585960388184, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8846748471260071, + "num_tokens": 294683682.0, + "step": 8395 + }, + { + "epoch": 1.5591457753017641, + "grad_norm": 1.4363279342651367, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8703458309173584, + "num_tokens": 294720097.0, + "step": 8396 + }, + { + "epoch": 1.5593314763231199, + "grad_norm": 1.5461307764053345, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8704447746276855, + "num_tokens": 294749757.0, + "step": 8397 + }, + { + "epoch": 1.5595171773444754, + "grad_norm": 1.480605125427246, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8759647011756897, + "num_tokens": 294786251.0, + "step": 8398 + }, + { + "epoch": 1.5597028783658309, + "grad_norm": 1.6402698755264282, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8754059076309204, + "num_tokens": 294813981.0, + "step": 8399 + }, + { + "epoch": 1.5598885793871866, + "grad_norm": 1.5054222345352173, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8862678408622742, + "num_tokens": 294848546.0, + "step": 8400 + }, + { + "epoch": 1.5600742804085423, + "grad_norm": 1.5906739234924316, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8791600465774536, + "num_tokens": 294878425.0, + "step": 8401 + }, + { + "epoch": 1.5602599814298979, + "grad_norm": 1.520282506942749, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8660160899162292, + "num_tokens": 294914710.0, + "step": 8402 + }, + { + "epoch": 1.5604456824512534, + "grad_norm": 1.8160738945007324, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8658398985862732, + "num_tokens": 294938051.0, + "step": 8403 + }, + { + "epoch": 1.560631383472609, + "grad_norm": 1.4914096593856812, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8694400191307068, + "num_tokens": 294975695.0, + "step": 8404 + }, + { + "epoch": 1.5608170844939648, + "grad_norm": 1.5614252090454102, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8847020864486694, + "num_tokens": 295006149.0, + "step": 8405 + }, + { + "epoch": 1.5610027855153203, + "grad_norm": 1.4820290803909302, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8824995756149292, + "num_tokens": 295041328.0, + "step": 8406 + }, + { + "epoch": 1.5611884865366759, + "grad_norm": 1.5586155652999878, + "learning_rate": 1e-06, + "loss": 0.4439, + "mean_token_accuracy": 0.8508391976356506, + "num_tokens": 295078807.0, + "step": 8407 + }, + { + "epoch": 1.5613741875580316, + "grad_norm": 1.476049780845642, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8742376565933228, + "num_tokens": 295113662.0, + "step": 8408 + }, + { + "epoch": 1.5615598885793873, + "grad_norm": 1.505932331085205, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.880523681640625, + "num_tokens": 295146933.0, + "step": 8409 + }, + { + "epoch": 1.5617455896007428, + "grad_norm": 1.558003544807434, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8871856927871704, + "num_tokens": 295178779.0, + "step": 8410 + }, + { + "epoch": 1.5619312906220983, + "grad_norm": 1.5394742488861084, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8752413988113403, + "num_tokens": 295214166.0, + "step": 8411 + }, + { + "epoch": 1.562116991643454, + "grad_norm": 1.3788237571716309, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8910378217697144, + "num_tokens": 295247678.0, + "step": 8412 + }, + { + "epoch": 1.5623026926648098, + "grad_norm": 1.53294837474823, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8721718192100525, + "num_tokens": 295284978.0, + "step": 8413 + }, + { + "epoch": 1.5624883936861653, + "grad_norm": 1.3725570440292358, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8736503720283508, + "num_tokens": 295327262.0, + "step": 8414 + }, + { + "epoch": 1.5626740947075208, + "grad_norm": 1.5840152502059937, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8720262050628662, + "num_tokens": 295365164.0, + "step": 8415 + }, + { + "epoch": 1.5628597957288766, + "grad_norm": 1.5369336605072021, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8804553747177124, + "num_tokens": 295395490.0, + "step": 8416 + }, + { + "epoch": 1.563045496750232, + "grad_norm": 1.4488880634307861, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8767213821411133, + "num_tokens": 295432344.0, + "step": 8417 + }, + { + "epoch": 1.5632311977715876, + "grad_norm": 1.5779780149459839, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8745729923248291, + "num_tokens": 295465182.0, + "step": 8418 + }, + { + "epoch": 1.5634168987929433, + "grad_norm": 1.3674201965332031, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8857969045639038, + "num_tokens": 295504187.0, + "step": 8419 + }, + { + "epoch": 1.563602599814299, + "grad_norm": 1.6041666269302368, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8798532485961914, + "num_tokens": 295536521.0, + "step": 8420 + }, + { + "epoch": 1.5637883008356546, + "grad_norm": 1.4269598722457886, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8670823574066162, + "num_tokens": 295576358.0, + "step": 8421 + }, + { + "epoch": 1.56397400185701, + "grad_norm": 1.4203813076019287, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8959234952926636, + "num_tokens": 295610054.0, + "step": 8422 + }, + { + "epoch": 1.5641597028783658, + "grad_norm": 1.3533395528793335, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8835957050323486, + "num_tokens": 295652840.0, + "step": 8423 + }, + { + "epoch": 1.5643454038997215, + "grad_norm": 1.4679476022720337, + "learning_rate": 1e-06, + "loss": 0.3699, + "mean_token_accuracy": 0.8701938390731812, + "num_tokens": 295693966.0, + "step": 8424 + }, + { + "epoch": 1.564531104921077, + "grad_norm": 1.4664523601531982, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8730344176292419, + "num_tokens": 295730857.0, + "step": 8425 + }, + { + "epoch": 1.5647168059424326, + "grad_norm": 1.4378068447113037, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8871560096740723, + "num_tokens": 295766329.0, + "step": 8426 + }, + { + "epoch": 1.5649025069637883, + "grad_norm": 1.442751169204712, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8771653771400452, + "num_tokens": 295802237.0, + "step": 8427 + }, + { + "epoch": 1.565088207985144, + "grad_norm": 1.3329989910125732, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8892958760261536, + "num_tokens": 295840416.0, + "step": 8428 + }, + { + "epoch": 1.5652739090064995, + "grad_norm": 1.5646206140518188, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8828938007354736, + "num_tokens": 295871382.0, + "step": 8429 + }, + { + "epoch": 1.565459610027855, + "grad_norm": 1.4949135780334473, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8693573474884033, + "num_tokens": 295911038.0, + "step": 8430 + }, + { + "epoch": 1.5656453110492108, + "grad_norm": 1.4763089418411255, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8778897523880005, + "num_tokens": 295945610.0, + "step": 8431 + }, + { + "epoch": 1.5658310120705665, + "grad_norm": 1.4552860260009766, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8763566017150879, + "num_tokens": 295980553.0, + "step": 8432 + }, + { + "epoch": 1.566016713091922, + "grad_norm": 1.4695215225219727, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8768758773803711, + "num_tokens": 296015318.0, + "step": 8433 + }, + { + "epoch": 1.5662024141132775, + "grad_norm": 1.5341774225234985, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8685415983200073, + "num_tokens": 296051430.0, + "step": 8434 + }, + { + "epoch": 1.5663881151346333, + "grad_norm": 1.5471417903900146, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8633137941360474, + "num_tokens": 296083906.0, + "step": 8435 + }, + { + "epoch": 1.566573816155989, + "grad_norm": 1.3505579233169556, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8775452375411987, + "num_tokens": 296128377.0, + "step": 8436 + }, + { + "epoch": 1.5667595171773445, + "grad_norm": 1.466942548751831, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8671820163726807, + "num_tokens": 296166301.0, + "step": 8437 + }, + { + "epoch": 1.5669452181987, + "grad_norm": 1.5387879610061646, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8733687400817871, + "num_tokens": 296199350.0, + "step": 8438 + }, + { + "epoch": 1.5671309192200558, + "grad_norm": 1.4759905338287354, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8635010719299316, + "num_tokens": 296236876.0, + "step": 8439 + }, + { + "epoch": 1.5673166202414113, + "grad_norm": 1.6183271408081055, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8760628700256348, + "num_tokens": 296266563.0, + "step": 8440 + }, + { + "epoch": 1.5675023212627668, + "grad_norm": 1.7355122566223145, + "learning_rate": 1e-06, + "loss": 0.2814, + "mean_token_accuracy": 0.9037660360336304, + "num_tokens": 296291844.0, + "step": 8441 + }, + { + "epoch": 1.5676880222841225, + "grad_norm": 1.4671145677566528, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8741927742958069, + "num_tokens": 296328545.0, + "step": 8442 + }, + { + "epoch": 1.5678737233054783, + "grad_norm": 1.5434712171554565, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.881986677646637, + "num_tokens": 296361020.0, + "step": 8443 + }, + { + "epoch": 1.5680594243268338, + "grad_norm": 1.6642413139343262, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8737846612930298, + "num_tokens": 296387937.0, + "step": 8444 + }, + { + "epoch": 1.5682451253481893, + "grad_norm": 1.7483495473861694, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8754086494445801, + "num_tokens": 296417397.0, + "step": 8445 + }, + { + "epoch": 1.568430826369545, + "grad_norm": 1.526432991027832, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8819407224655151, + "num_tokens": 296447696.0, + "step": 8446 + }, + { + "epoch": 1.5686165273909007, + "grad_norm": 1.5294831991195679, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8861278891563416, + "num_tokens": 296483780.0, + "step": 8447 + }, + { + "epoch": 1.5688022284122562, + "grad_norm": 1.7019128799438477, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8862040042877197, + "num_tokens": 296509485.0, + "step": 8448 + }, + { + "epoch": 1.5689879294336118, + "grad_norm": 1.6498191356658936, + "learning_rate": 1e-06, + "loss": 0.416, + "mean_token_accuracy": 0.8574756383895874, + "num_tokens": 296542639.0, + "step": 8449 + }, + { + "epoch": 1.5691736304549675, + "grad_norm": 1.518309235572815, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.884863555431366, + "num_tokens": 296573016.0, + "step": 8450 + }, + { + "epoch": 1.5693593314763232, + "grad_norm": 1.4943574666976929, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8785754442214966, + "num_tokens": 296606529.0, + "step": 8451 + }, + { + "epoch": 1.5695450324976787, + "grad_norm": 1.5176959037780762, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8795607686042786, + "num_tokens": 296640692.0, + "step": 8452 + }, + { + "epoch": 1.5697307335190342, + "grad_norm": 1.5374811887741089, + "learning_rate": 1e-06, + "loss": 0.41, + "mean_token_accuracy": 0.8619861006736755, + "num_tokens": 296678538.0, + "step": 8453 + }, + { + "epoch": 1.56991643454039, + "grad_norm": 1.6254966259002686, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8678480386734009, + "num_tokens": 296711936.0, + "step": 8454 + }, + { + "epoch": 1.5701021355617457, + "grad_norm": 1.5846226215362549, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.885290265083313, + "num_tokens": 296744927.0, + "step": 8455 + }, + { + "epoch": 1.5702878365831012, + "grad_norm": 1.7116868495941162, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.866604745388031, + "num_tokens": 296774703.0, + "step": 8456 + }, + { + "epoch": 1.5704735376044567, + "grad_norm": 1.4543020725250244, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8756739497184753, + "num_tokens": 296809055.0, + "step": 8457 + }, + { + "epoch": 1.5706592386258125, + "grad_norm": 1.4793412685394287, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8814492225646973, + "num_tokens": 296842702.0, + "step": 8458 + }, + { + "epoch": 1.5708449396471682, + "grad_norm": 1.8050618171691895, + "learning_rate": 1e-06, + "loss": 0.4678, + "mean_token_accuracy": 0.8511345982551575, + "num_tokens": 296872273.0, + "step": 8459 + }, + { + "epoch": 1.5710306406685237, + "grad_norm": 1.5435446500778198, + "learning_rate": 1e-06, + "loss": 0.4166, + "mean_token_accuracy": 0.8593522310256958, + "num_tokens": 296909803.0, + "step": 8460 + }, + { + "epoch": 1.5712163416898792, + "grad_norm": 1.4320228099822998, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8727967739105225, + "num_tokens": 296949779.0, + "step": 8461 + }, + { + "epoch": 1.571402042711235, + "grad_norm": 1.5143489837646484, + "learning_rate": 1e-06, + "loss": 0.3023, + "mean_token_accuracy": 0.8955763578414917, + "num_tokens": 296979831.0, + "step": 8462 + }, + { + "epoch": 1.5715877437325907, + "grad_norm": 1.4550517797470093, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8893781304359436, + "num_tokens": 297010727.0, + "step": 8463 + }, + { + "epoch": 1.571773444753946, + "grad_norm": 1.4184144735336304, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8712401390075684, + "num_tokens": 297053604.0, + "step": 8464 + }, + { + "epoch": 1.5719591457753017, + "grad_norm": 1.3977783918380737, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8838164806365967, + "num_tokens": 297091903.0, + "step": 8465 + }, + { + "epoch": 1.5721448467966574, + "grad_norm": 1.4244951009750366, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8641994595527649, + "num_tokens": 297133829.0, + "step": 8466 + }, + { + "epoch": 1.572330547818013, + "grad_norm": 1.5119770765304565, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.872381329536438, + "num_tokens": 297168443.0, + "step": 8467 + }, + { + "epoch": 1.5725162488393685, + "grad_norm": 1.453141212463379, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.877510666847229, + "num_tokens": 297206328.0, + "step": 8468 + }, + { + "epoch": 1.5727019498607242, + "grad_norm": 1.3951958417892456, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8835296630859375, + "num_tokens": 297247508.0, + "step": 8469 + }, + { + "epoch": 1.57288765088208, + "grad_norm": 1.4452190399169922, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8717430830001831, + "num_tokens": 297286370.0, + "step": 8470 + }, + { + "epoch": 1.5730733519034354, + "grad_norm": 1.5212299823760986, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8824225068092346, + "num_tokens": 297326038.0, + "step": 8471 + }, + { + "epoch": 1.573259052924791, + "grad_norm": 1.6579750776290894, + "learning_rate": 1e-06, + "loss": 0.3878, + "mean_token_accuracy": 0.8672541975975037, + "num_tokens": 297358892.0, + "step": 8472 + }, + { + "epoch": 1.5734447539461467, + "grad_norm": 1.495719313621521, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.866951584815979, + "num_tokens": 297394213.0, + "step": 8473 + }, + { + "epoch": 1.5736304549675024, + "grad_norm": 1.6176717281341553, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8726299405097961, + "num_tokens": 297430953.0, + "step": 8474 + }, + { + "epoch": 1.573816155988858, + "grad_norm": 1.3825955390930176, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8860581517219543, + "num_tokens": 297470174.0, + "step": 8475 + }, + { + "epoch": 1.5740018570102134, + "grad_norm": 1.4701427221298218, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8695403337478638, + "num_tokens": 297505435.0, + "step": 8476 + }, + { + "epoch": 1.5741875580315692, + "grad_norm": 1.5657110214233398, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8846288919448853, + "num_tokens": 297539143.0, + "step": 8477 + }, + { + "epoch": 1.574373259052925, + "grad_norm": 1.5077680349349976, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8783150911331177, + "num_tokens": 297576126.0, + "step": 8478 + }, + { + "epoch": 1.5745589600742804, + "grad_norm": 1.529229760169983, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8696100115776062, + "num_tokens": 297611694.0, + "step": 8479 + }, + { + "epoch": 1.574744661095636, + "grad_norm": 1.5720584392547607, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8818058967590332, + "num_tokens": 297648159.0, + "step": 8480 + }, + { + "epoch": 1.5749303621169917, + "grad_norm": 1.4749095439910889, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8887014389038086, + "num_tokens": 297683123.0, + "step": 8481 + }, + { + "epoch": 1.5751160631383474, + "grad_norm": 1.4356712102890015, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8762422204017639, + "num_tokens": 297717001.0, + "step": 8482 + }, + { + "epoch": 1.575301764159703, + "grad_norm": 1.7024123668670654, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8883816599845886, + "num_tokens": 297741935.0, + "step": 8483 + }, + { + "epoch": 1.5754874651810584, + "grad_norm": 1.7026185989379883, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8685790300369263, + "num_tokens": 297770995.0, + "step": 8484 + }, + { + "epoch": 1.5756731662024142, + "grad_norm": 1.5764458179473877, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8721103668212891, + "num_tokens": 297800738.0, + "step": 8485 + }, + { + "epoch": 1.5758588672237699, + "grad_norm": 1.5888268947601318, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8773004412651062, + "num_tokens": 297835704.0, + "step": 8486 + }, + { + "epoch": 1.5760445682451254, + "grad_norm": 1.401995062828064, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8768364787101746, + "num_tokens": 297874405.0, + "step": 8487 + }, + { + "epoch": 1.576230269266481, + "grad_norm": 1.5915344953536987, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8732370138168335, + "num_tokens": 297907017.0, + "step": 8488 + }, + { + "epoch": 1.5764159702878366, + "grad_norm": 1.4633227586746216, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8805751204490662, + "num_tokens": 297943186.0, + "step": 8489 + }, + { + "epoch": 1.5766016713091922, + "grad_norm": 1.4440487623214722, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8777674436569214, + "num_tokens": 297978891.0, + "step": 8490 + }, + { + "epoch": 1.5767873723305477, + "grad_norm": 1.3891568183898926, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8793797492980957, + "num_tokens": 298017780.0, + "step": 8491 + }, + { + "epoch": 1.5769730733519034, + "grad_norm": 1.5673199892044067, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8632660508155823, + "num_tokens": 298050483.0, + "step": 8492 + }, + { + "epoch": 1.5771587743732591, + "grad_norm": 1.4535725116729736, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.880703866481781, + "num_tokens": 298084201.0, + "step": 8493 + }, + { + "epoch": 1.5773444753946146, + "grad_norm": 1.3855278491973877, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8788235187530518, + "num_tokens": 298124370.0, + "step": 8494 + }, + { + "epoch": 1.5775301764159702, + "grad_norm": 1.7148433923721313, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8795070648193359, + "num_tokens": 298153023.0, + "step": 8495 + }, + { + "epoch": 1.5777158774373259, + "grad_norm": 1.3653794527053833, + "learning_rate": 1e-06, + "loss": 0.3049, + "mean_token_accuracy": 0.8931737542152405, + "num_tokens": 298187308.0, + "step": 8496 + }, + { + "epoch": 1.5779015784586816, + "grad_norm": 1.5591785907745361, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8753066062927246, + "num_tokens": 298223700.0, + "step": 8497 + }, + { + "epoch": 1.5780872794800371, + "grad_norm": 1.5158542394638062, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.875806450843811, + "num_tokens": 298258091.0, + "step": 8498 + }, + { + "epoch": 1.5782729805013926, + "grad_norm": 1.4965343475341797, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8841044902801514, + "num_tokens": 298293008.0, + "step": 8499 + }, + { + "epoch": 1.5784586815227484, + "grad_norm": 1.4920880794525146, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8631672859191895, + "num_tokens": 298332026.0, + "step": 8500 + }, + { + "epoch": 1.578644382544104, + "grad_norm": 1.35837984085083, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8843910694122314, + "num_tokens": 298372234.0, + "step": 8501 + }, + { + "epoch": 1.5788300835654596, + "grad_norm": 1.5933966636657715, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8762716054916382, + "num_tokens": 298401294.0, + "step": 8502 + }, + { + "epoch": 1.5790157845868151, + "grad_norm": 1.4323055744171143, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8713972568511963, + "num_tokens": 298438577.0, + "step": 8503 + }, + { + "epoch": 1.5792014856081709, + "grad_norm": 1.5702341794967651, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8640940189361572, + "num_tokens": 298471019.0, + "step": 8504 + }, + { + "epoch": 1.5793871866295266, + "grad_norm": 1.6005357503890991, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8798689842224121, + "num_tokens": 298502913.0, + "step": 8505 + }, + { + "epoch": 1.579572887650882, + "grad_norm": 1.5088400840759277, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.877034604549408, + "num_tokens": 298535555.0, + "step": 8506 + }, + { + "epoch": 1.5797585886722376, + "grad_norm": 1.5017224550247192, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.882422924041748, + "num_tokens": 298564434.0, + "step": 8507 + }, + { + "epoch": 1.5799442896935934, + "grad_norm": 1.6265684366226196, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8700495958328247, + "num_tokens": 298593077.0, + "step": 8508 + }, + { + "epoch": 1.580129990714949, + "grad_norm": 1.470321536064148, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8785787224769592, + "num_tokens": 298626270.0, + "step": 8509 + }, + { + "epoch": 1.5803156917363046, + "grad_norm": 1.2716033458709717, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8786998987197876, + "num_tokens": 298670498.0, + "step": 8510 + }, + { + "epoch": 1.58050139275766, + "grad_norm": 1.3986387252807617, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8786401748657227, + "num_tokens": 298713985.0, + "step": 8511 + }, + { + "epoch": 1.5806870937790158, + "grad_norm": 1.4687085151672363, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8892845511436462, + "num_tokens": 298749598.0, + "step": 8512 + }, + { + "epoch": 1.5808727948003714, + "grad_norm": 1.6680563688278198, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8730530142784119, + "num_tokens": 298784828.0, + "step": 8513 + }, + { + "epoch": 1.5810584958217269, + "grad_norm": 1.428907871246338, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8611217737197876, + "num_tokens": 298826614.0, + "step": 8514 + }, + { + "epoch": 1.5812441968430826, + "grad_norm": 1.465872883796692, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8838667869567871, + "num_tokens": 298862607.0, + "step": 8515 + }, + { + "epoch": 1.5814298978644383, + "grad_norm": 1.5904781818389893, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8741683959960938, + "num_tokens": 298894838.0, + "step": 8516 + }, + { + "epoch": 1.5816155988857938, + "grad_norm": 1.4163625240325928, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8944907784461975, + "num_tokens": 298929710.0, + "step": 8517 + }, + { + "epoch": 1.5818012999071493, + "grad_norm": 1.4578828811645508, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8836721181869507, + "num_tokens": 298965768.0, + "step": 8518 + }, + { + "epoch": 1.581987000928505, + "grad_norm": 1.4349285364151, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.884899377822876, + "num_tokens": 299000103.0, + "step": 8519 + }, + { + "epoch": 1.5821727019498608, + "grad_norm": 1.452463984489441, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8861477375030518, + "num_tokens": 299035203.0, + "step": 8520 + }, + { + "epoch": 1.5823584029712163, + "grad_norm": 1.6604957580566406, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8728036880493164, + "num_tokens": 299064165.0, + "step": 8521 + }, + { + "epoch": 1.5825441039925718, + "grad_norm": 1.4518376588821411, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8790944814682007, + "num_tokens": 299098467.0, + "step": 8522 + }, + { + "epoch": 1.5827298050139276, + "grad_norm": 1.4632248878479004, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8807533979415894, + "num_tokens": 299133946.0, + "step": 8523 + }, + { + "epoch": 1.5829155060352833, + "grad_norm": 1.509146809577942, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8754512071609497, + "num_tokens": 299169929.0, + "step": 8524 + }, + { + "epoch": 1.5831012070566388, + "grad_norm": 1.5141187906265259, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.880866289138794, + "num_tokens": 299208272.0, + "step": 8525 + }, + { + "epoch": 1.5832869080779943, + "grad_norm": 1.5108760595321655, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8894425630569458, + "num_tokens": 299239568.0, + "step": 8526 + }, + { + "epoch": 1.58347260909935, + "grad_norm": 1.5689119100570679, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8766042590141296, + "num_tokens": 299270166.0, + "step": 8527 + }, + { + "epoch": 1.5836583101207058, + "grad_norm": 1.7851749658584595, + "learning_rate": 1e-06, + "loss": 0.4059, + "mean_token_accuracy": 0.8626205921173096, + "num_tokens": 299299424.0, + "step": 8528 + }, + { + "epoch": 1.5838440111420613, + "grad_norm": 1.6323649883270264, + "learning_rate": 1e-06, + "loss": 0.4016, + "mean_token_accuracy": 0.8672484755516052, + "num_tokens": 299329884.0, + "step": 8529 + }, + { + "epoch": 1.5840297121634168, + "grad_norm": 1.4448552131652832, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.870637059211731, + "num_tokens": 299368185.0, + "step": 8530 + }, + { + "epoch": 1.5842154131847725, + "grad_norm": 1.4811277389526367, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8813846111297607, + "num_tokens": 299405627.0, + "step": 8531 + }, + { + "epoch": 1.5844011142061283, + "grad_norm": 1.4597173929214478, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8821817636489868, + "num_tokens": 299441599.0, + "step": 8532 + }, + { + "epoch": 1.5845868152274838, + "grad_norm": 1.4396967887878418, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8817403316497803, + "num_tokens": 299476284.0, + "step": 8533 + }, + { + "epoch": 1.5847725162488393, + "grad_norm": 1.6634089946746826, + "learning_rate": 1e-06, + "loss": 0.4253, + "mean_token_accuracy": 0.8600536584854126, + "num_tokens": 299508557.0, + "step": 8534 + }, + { + "epoch": 1.584958217270195, + "grad_norm": 1.685615062713623, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8728139400482178, + "num_tokens": 299538537.0, + "step": 8535 + }, + { + "epoch": 1.5851439182915508, + "grad_norm": 1.5094146728515625, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8721330165863037, + "num_tokens": 299574292.0, + "step": 8536 + }, + { + "epoch": 1.585329619312906, + "grad_norm": 1.709052562713623, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.8732443451881409, + "num_tokens": 299605175.0, + "step": 8537 + }, + { + "epoch": 1.5855153203342618, + "grad_norm": 1.4683279991149902, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8840188980102539, + "num_tokens": 299643942.0, + "step": 8538 + }, + { + "epoch": 1.5857010213556175, + "grad_norm": 1.5801918506622314, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8815720677375793, + "num_tokens": 299674866.0, + "step": 8539 + }, + { + "epoch": 1.585886722376973, + "grad_norm": 1.326147437095642, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8873405456542969, + "num_tokens": 299712884.0, + "step": 8540 + }, + { + "epoch": 1.5860724233983285, + "grad_norm": 1.4438738822937012, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8913545608520508, + "num_tokens": 299744756.0, + "step": 8541 + }, + { + "epoch": 1.5862581244196843, + "grad_norm": 1.4732874631881714, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8813206553459167, + "num_tokens": 299775879.0, + "step": 8542 + }, + { + "epoch": 1.58644382544104, + "grad_norm": 1.5037851333618164, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8824092149734497, + "num_tokens": 299807690.0, + "step": 8543 + }, + { + "epoch": 1.5866295264623955, + "grad_norm": 1.611649751663208, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.880066990852356, + "num_tokens": 299841242.0, + "step": 8544 + }, + { + "epoch": 1.586815227483751, + "grad_norm": 1.3931372165679932, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8911333084106445, + "num_tokens": 299876572.0, + "step": 8545 + }, + { + "epoch": 1.5870009285051068, + "grad_norm": 1.4170880317687988, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8765896558761597, + "num_tokens": 299913630.0, + "step": 8546 + }, + { + "epoch": 1.5871866295264625, + "grad_norm": 1.4951008558273315, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8838638663291931, + "num_tokens": 299948674.0, + "step": 8547 + }, + { + "epoch": 1.587372330547818, + "grad_norm": 1.571982741355896, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8696802854537964, + "num_tokens": 299984125.0, + "step": 8548 + }, + { + "epoch": 1.5875580315691735, + "grad_norm": 1.4825679063796997, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8755241632461548, + "num_tokens": 300018612.0, + "step": 8549 + }, + { + "epoch": 1.5877437325905293, + "grad_norm": 1.4995709657669067, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8693736791610718, + "num_tokens": 300054323.0, + "step": 8550 + }, + { + "epoch": 1.587929433611885, + "grad_norm": 1.4507262706756592, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8748082518577576, + "num_tokens": 300087835.0, + "step": 8551 + }, + { + "epoch": 1.5881151346332405, + "grad_norm": 1.5325605869293213, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8674965500831604, + "num_tokens": 300124440.0, + "step": 8552 + }, + { + "epoch": 1.588300835654596, + "grad_norm": 1.4683703184127808, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8865955471992493, + "num_tokens": 300159285.0, + "step": 8553 + }, + { + "epoch": 1.5884865366759517, + "grad_norm": 1.5471305847167969, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8630350828170776, + "num_tokens": 300193333.0, + "step": 8554 + }, + { + "epoch": 1.5886722376973075, + "grad_norm": 1.4070485830307007, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8628562688827515, + "num_tokens": 300234081.0, + "step": 8555 + }, + { + "epoch": 1.588857938718663, + "grad_norm": 1.5618391036987305, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8715227842330933, + "num_tokens": 300269070.0, + "step": 8556 + }, + { + "epoch": 1.5890436397400185, + "grad_norm": 1.5249230861663818, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8735698461532593, + "num_tokens": 300302343.0, + "step": 8557 + }, + { + "epoch": 1.5892293407613742, + "grad_norm": 1.547330379486084, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8736072182655334, + "num_tokens": 300336126.0, + "step": 8558 + }, + { + "epoch": 1.58941504178273, + "grad_norm": 1.4850560426712036, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8660008907318115, + "num_tokens": 300371112.0, + "step": 8559 + }, + { + "epoch": 1.5896007428040855, + "grad_norm": 1.3613989353179932, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.874291181564331, + "num_tokens": 300411575.0, + "step": 8560 + }, + { + "epoch": 1.589786443825441, + "grad_norm": 1.5105869770050049, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8729212284088135, + "num_tokens": 300444508.0, + "step": 8561 + }, + { + "epoch": 1.5899721448467967, + "grad_norm": 1.4562948942184448, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8817764520645142, + "num_tokens": 300478308.0, + "step": 8562 + }, + { + "epoch": 1.5901578458681522, + "grad_norm": 1.4049052000045776, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8842747211456299, + "num_tokens": 300514525.0, + "step": 8563 + }, + { + "epoch": 1.5903435468895077, + "grad_norm": 1.3755278587341309, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8813583254814148, + "num_tokens": 300555474.0, + "step": 8564 + }, + { + "epoch": 1.5905292479108635, + "grad_norm": 1.5574333667755127, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8624817132949829, + "num_tokens": 300590684.0, + "step": 8565 + }, + { + "epoch": 1.5907149489322192, + "grad_norm": 1.4007980823516846, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8779239058494568, + "num_tokens": 300630264.0, + "step": 8566 + }, + { + "epoch": 1.5909006499535747, + "grad_norm": 1.4389711618423462, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8882417678833008, + "num_tokens": 300667753.0, + "step": 8567 + }, + { + "epoch": 1.5910863509749302, + "grad_norm": 1.49411141872406, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8742048740386963, + "num_tokens": 300706416.0, + "step": 8568 + }, + { + "epoch": 1.591272051996286, + "grad_norm": 1.4744824171066284, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8725731372833252, + "num_tokens": 300740224.0, + "step": 8569 + }, + { + "epoch": 1.5914577530176417, + "grad_norm": 1.4619518518447876, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8802356719970703, + "num_tokens": 300777638.0, + "step": 8570 + }, + { + "epoch": 1.5916434540389972, + "grad_norm": 1.6180702447891235, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8801430463790894, + "num_tokens": 300808138.0, + "step": 8571 + }, + { + "epoch": 1.5918291550603527, + "grad_norm": 1.4411048889160156, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8758945465087891, + "num_tokens": 300845192.0, + "step": 8572 + }, + { + "epoch": 1.5920148560817085, + "grad_norm": 1.6377062797546387, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8789156079292297, + "num_tokens": 300874866.0, + "step": 8573 + }, + { + "epoch": 1.5922005571030642, + "grad_norm": 1.4302351474761963, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.8686717748641968, + "num_tokens": 300917837.0, + "step": 8574 + }, + { + "epoch": 1.5923862581244197, + "grad_norm": 1.6388424634933472, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8689752221107483, + "num_tokens": 300948848.0, + "step": 8575 + }, + { + "epoch": 1.5925719591457752, + "grad_norm": 1.4262193441390991, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8673208951950073, + "num_tokens": 300987818.0, + "step": 8576 + }, + { + "epoch": 1.592757660167131, + "grad_norm": 1.443882703781128, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8746705055236816, + "num_tokens": 301024966.0, + "step": 8577 + }, + { + "epoch": 1.5929433611884867, + "grad_norm": 1.468004822731018, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8726729154586792, + "num_tokens": 301066366.0, + "step": 8578 + }, + { + "epoch": 1.5931290622098422, + "grad_norm": 1.6108646392822266, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8753721117973328, + "num_tokens": 301095989.0, + "step": 8579 + }, + { + "epoch": 1.5933147632311977, + "grad_norm": 1.57562255859375, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8898670673370361, + "num_tokens": 301124009.0, + "step": 8580 + }, + { + "epoch": 1.5935004642525534, + "grad_norm": 1.417440414428711, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8716685771942139, + "num_tokens": 301167273.0, + "step": 8581 + }, + { + "epoch": 1.5936861652739092, + "grad_norm": 1.4819133281707764, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8665043711662292, + "num_tokens": 301203761.0, + "step": 8582 + }, + { + "epoch": 1.5938718662952647, + "grad_norm": 1.5430262088775635, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8833752870559692, + "num_tokens": 301236685.0, + "step": 8583 + }, + { + "epoch": 1.5940575673166202, + "grad_norm": 1.4187378883361816, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8856842517852783, + "num_tokens": 301273738.0, + "step": 8584 + }, + { + "epoch": 1.594243268337976, + "grad_norm": 1.5078500509262085, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8897686004638672, + "num_tokens": 301303951.0, + "step": 8585 + }, + { + "epoch": 1.5944289693593314, + "grad_norm": 1.5526341199874878, + "learning_rate": 1e-06, + "loss": 0.4127, + "mean_token_accuracy": 0.8653388023376465, + "num_tokens": 301339625.0, + "step": 8586 + }, + { + "epoch": 1.594614670380687, + "grad_norm": 1.476460576057434, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8823177814483643, + "num_tokens": 301371549.0, + "step": 8587 + }, + { + "epoch": 1.5948003714020427, + "grad_norm": 1.397229552268982, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8757265210151672, + "num_tokens": 301412572.0, + "step": 8588 + }, + { + "epoch": 1.5949860724233984, + "grad_norm": 1.4734257459640503, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8695605993270874, + "num_tokens": 301449250.0, + "step": 8589 + }, + { + "epoch": 1.595171773444754, + "grad_norm": 1.465118646621704, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8780604600906372, + "num_tokens": 301486669.0, + "step": 8590 + }, + { + "epoch": 1.5953574744661094, + "grad_norm": 1.3940221071243286, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8787972331047058, + "num_tokens": 301527131.0, + "step": 8591 + }, + { + "epoch": 1.5955431754874652, + "grad_norm": 1.3736295700073242, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8943012356758118, + "num_tokens": 301562470.0, + "step": 8592 + }, + { + "epoch": 1.595728876508821, + "grad_norm": 1.4367848634719849, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8830488920211792, + "num_tokens": 301597334.0, + "step": 8593 + }, + { + "epoch": 1.5959145775301764, + "grad_norm": 1.736958622932434, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8702730536460876, + "num_tokens": 301629322.0, + "step": 8594 + }, + { + "epoch": 1.596100278551532, + "grad_norm": 1.3947277069091797, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8825539350509644, + "num_tokens": 301665746.0, + "step": 8595 + }, + { + "epoch": 1.5962859795728876, + "grad_norm": 1.358978509902954, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8823212385177612, + "num_tokens": 301703391.0, + "step": 8596 + }, + { + "epoch": 1.5964716805942434, + "grad_norm": 1.4232515096664429, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8686960935592651, + "num_tokens": 301741925.0, + "step": 8597 + }, + { + "epoch": 1.596657381615599, + "grad_norm": 1.4222956895828247, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8757778406143188, + "num_tokens": 301780788.0, + "step": 8598 + }, + { + "epoch": 1.5968430826369544, + "grad_norm": 1.46065092086792, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8846087455749512, + "num_tokens": 301812546.0, + "step": 8599 + }, + { + "epoch": 1.5970287836583101, + "grad_norm": 1.4850034713745117, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8774111270904541, + "num_tokens": 301852489.0, + "step": 8600 + }, + { + "epoch": 1.5972144846796659, + "grad_norm": 1.4243485927581787, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8669979572296143, + "num_tokens": 301893341.0, + "step": 8601 + }, + { + "epoch": 1.5974001857010214, + "grad_norm": 1.5646181106567383, + "learning_rate": 1e-06, + "loss": 0.4137, + "mean_token_accuracy": 0.8577799797058105, + "num_tokens": 301928145.0, + "step": 8602 + }, + { + "epoch": 1.597585886722377, + "grad_norm": 1.4572356939315796, + "learning_rate": 1e-06, + "loss": 0.4106, + "mean_token_accuracy": 0.8639878630638123, + "num_tokens": 301969162.0, + "step": 8603 + }, + { + "epoch": 1.5977715877437326, + "grad_norm": 1.386313796043396, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8822664022445679, + "num_tokens": 302008419.0, + "step": 8604 + }, + { + "epoch": 1.5979572887650884, + "grad_norm": 1.4403053522109985, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8784908056259155, + "num_tokens": 302042934.0, + "step": 8605 + }, + { + "epoch": 1.5981429897864439, + "grad_norm": 1.3866935968399048, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8925522565841675, + "num_tokens": 302080109.0, + "step": 8606 + }, + { + "epoch": 1.5983286908077994, + "grad_norm": 1.6391240358352661, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8657695055007935, + "num_tokens": 302112596.0, + "step": 8607 + }, + { + "epoch": 1.5985143918291551, + "grad_norm": 1.485228180885315, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.867882490158081, + "num_tokens": 302148360.0, + "step": 8608 + }, + { + "epoch": 1.5987000928505106, + "grad_norm": 1.507947325706482, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8834264278411865, + "num_tokens": 302182671.0, + "step": 8609 + }, + { + "epoch": 1.5988857938718661, + "grad_norm": 1.5565444231033325, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8719891905784607, + "num_tokens": 302216235.0, + "step": 8610 + }, + { + "epoch": 1.5990714948932219, + "grad_norm": 1.3898509740829468, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8883335590362549, + "num_tokens": 302255335.0, + "step": 8611 + }, + { + "epoch": 1.5992571959145776, + "grad_norm": 1.6714779138565063, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8706291317939758, + "num_tokens": 302287032.0, + "step": 8612 + }, + { + "epoch": 1.5994428969359331, + "grad_norm": 1.5836318731307983, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8659031391143799, + "num_tokens": 302321894.0, + "step": 8613 + }, + { + "epoch": 1.5996285979572886, + "grad_norm": 1.4814239740371704, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8803699016571045, + "num_tokens": 302355122.0, + "step": 8614 + }, + { + "epoch": 1.5998142989786444, + "grad_norm": 1.564221739768982, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8799893856048584, + "num_tokens": 302386888.0, + "step": 8615 + }, + { + "epoch": 1.6, + "grad_norm": 1.5796191692352295, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8651236295700073, + "num_tokens": 302424570.0, + "step": 8616 + }, + { + "epoch": 1.6001857010213556, + "grad_norm": 1.604599952697754, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.884665310382843, + "num_tokens": 302455276.0, + "step": 8617 + }, + { + "epoch": 1.600371402042711, + "grad_norm": 1.4565600156784058, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8805992603302002, + "num_tokens": 302491159.0, + "step": 8618 + }, + { + "epoch": 1.6005571030640668, + "grad_norm": 1.4195008277893066, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8875037431716919, + "num_tokens": 302530968.0, + "step": 8619 + }, + { + "epoch": 1.6007428040854226, + "grad_norm": 1.3481287956237793, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8878686428070068, + "num_tokens": 302568400.0, + "step": 8620 + }, + { + "epoch": 1.600928505106778, + "grad_norm": 1.5450894832611084, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8783884048461914, + "num_tokens": 302603996.0, + "step": 8621 + }, + { + "epoch": 1.6011142061281336, + "grad_norm": 1.6327664852142334, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8737339973449707, + "num_tokens": 302635566.0, + "step": 8622 + }, + { + "epoch": 1.6012999071494893, + "grad_norm": 1.4809801578521729, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8837476372718811, + "num_tokens": 302671815.0, + "step": 8623 + }, + { + "epoch": 1.601485608170845, + "grad_norm": 1.4673552513122559, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8769695162773132, + "num_tokens": 302713582.0, + "step": 8624 + }, + { + "epoch": 1.6016713091922006, + "grad_norm": 1.4713090658187866, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8705569505691528, + "num_tokens": 302749288.0, + "step": 8625 + }, + { + "epoch": 1.601857010213556, + "grad_norm": 1.4165786504745483, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8769035935401917, + "num_tokens": 302786433.0, + "step": 8626 + }, + { + "epoch": 1.6020427112349118, + "grad_norm": 1.4297308921813965, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.880942702293396, + "num_tokens": 302822252.0, + "step": 8627 + }, + { + "epoch": 1.6022284122562676, + "grad_norm": 1.479364037513733, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8811379075050354, + "num_tokens": 302859764.0, + "step": 8628 + }, + { + "epoch": 1.602414113277623, + "grad_norm": 1.5755908489227295, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8843398690223694, + "num_tokens": 302891625.0, + "step": 8629 + }, + { + "epoch": 1.6025998142989786, + "grad_norm": 1.5160139799118042, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8877857327461243, + "num_tokens": 302929245.0, + "step": 8630 + }, + { + "epoch": 1.6027855153203343, + "grad_norm": 1.7122186422348022, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8834847807884216, + "num_tokens": 302956429.0, + "step": 8631 + }, + { + "epoch": 1.60297121634169, + "grad_norm": 1.6253349781036377, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.879921555519104, + "num_tokens": 302988385.0, + "step": 8632 + }, + { + "epoch": 1.6031569173630453, + "grad_norm": 1.355232834815979, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8841647505760193, + "num_tokens": 303028787.0, + "step": 8633 + }, + { + "epoch": 1.603342618384401, + "grad_norm": 1.5980998277664185, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8873264789581299, + "num_tokens": 303062478.0, + "step": 8634 + }, + { + "epoch": 1.6035283194057568, + "grad_norm": 1.7450066804885864, + "learning_rate": 1e-06, + "loss": 0.4362, + "mean_token_accuracy": 0.855399489402771, + "num_tokens": 303092627.0, + "step": 8635 + }, + { + "epoch": 1.6037140204271123, + "grad_norm": 1.5201228857040405, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.885986864566803, + "num_tokens": 303124763.0, + "step": 8636 + }, + { + "epoch": 1.6038997214484678, + "grad_norm": 1.402514100074768, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8692458271980286, + "num_tokens": 303165702.0, + "step": 8637 + }, + { + "epoch": 1.6040854224698236, + "grad_norm": 1.5030585527420044, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8790280818939209, + "num_tokens": 303202602.0, + "step": 8638 + }, + { + "epoch": 1.6042711234911793, + "grad_norm": 1.5860258340835571, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8747576475143433, + "num_tokens": 303232011.0, + "step": 8639 + }, + { + "epoch": 1.6044568245125348, + "grad_norm": 1.592974066734314, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8842132687568665, + "num_tokens": 303260254.0, + "step": 8640 + }, + { + "epoch": 1.6046425255338903, + "grad_norm": 1.5185840129852295, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8675191402435303, + "num_tokens": 303294634.0, + "step": 8641 + }, + { + "epoch": 1.604828226555246, + "grad_norm": 1.571911096572876, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8797376155853271, + "num_tokens": 303328127.0, + "step": 8642 + }, + { + "epoch": 1.6050139275766018, + "grad_norm": 1.4565813541412354, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8767073750495911, + "num_tokens": 303365641.0, + "step": 8643 + }, + { + "epoch": 1.6051996285979573, + "grad_norm": 1.4470667839050293, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8761793375015259, + "num_tokens": 303404607.0, + "step": 8644 + }, + { + "epoch": 1.6053853296193128, + "grad_norm": 1.4617562294006348, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8727469444274902, + "num_tokens": 303444123.0, + "step": 8645 + }, + { + "epoch": 1.6055710306406685, + "grad_norm": 1.4659079313278198, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8808298110961914, + "num_tokens": 303477124.0, + "step": 8646 + }, + { + "epoch": 1.6057567316620243, + "grad_norm": 1.5831379890441895, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8643879294395447, + "num_tokens": 303509093.0, + "step": 8647 + }, + { + "epoch": 1.6059424326833798, + "grad_norm": 1.577566385269165, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8843064308166504, + "num_tokens": 303542479.0, + "step": 8648 + }, + { + "epoch": 1.6061281337047353, + "grad_norm": 1.564078450202942, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8661361932754517, + "num_tokens": 303577075.0, + "step": 8649 + }, + { + "epoch": 1.606313834726091, + "grad_norm": 1.4768052101135254, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8683639168739319, + "num_tokens": 303617012.0, + "step": 8650 + }, + { + "epoch": 1.6064995357474467, + "grad_norm": 1.4810786247253418, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8822818398475647, + "num_tokens": 303652187.0, + "step": 8651 + }, + { + "epoch": 1.6066852367688023, + "grad_norm": 1.5248034000396729, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8908596634864807, + "num_tokens": 303680402.0, + "step": 8652 + }, + { + "epoch": 1.6068709377901578, + "grad_norm": 1.4832098484039307, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.892601728439331, + "num_tokens": 303710535.0, + "step": 8653 + }, + { + "epoch": 1.6070566388115135, + "grad_norm": 1.5580028295516968, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8852225542068481, + "num_tokens": 303741758.0, + "step": 8654 + }, + { + "epoch": 1.6072423398328692, + "grad_norm": 1.691349983215332, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8601583242416382, + "num_tokens": 303774425.0, + "step": 8655 + }, + { + "epoch": 1.6074280408542247, + "grad_norm": 1.3342605829238892, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8929621577262878, + "num_tokens": 303820191.0, + "step": 8656 + }, + { + "epoch": 1.6076137418755803, + "grad_norm": 1.382869005203247, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8812426328659058, + "num_tokens": 303859441.0, + "step": 8657 + }, + { + "epoch": 1.607799442896936, + "grad_norm": 1.8695498704910278, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8764501810073853, + "num_tokens": 303886690.0, + "step": 8658 + }, + { + "epoch": 1.6079851439182915, + "grad_norm": 1.49898362159729, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8826941847801208, + "num_tokens": 303923537.0, + "step": 8659 + }, + { + "epoch": 1.608170844939647, + "grad_norm": 1.6255830526351929, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.867475152015686, + "num_tokens": 303955188.0, + "step": 8660 + }, + { + "epoch": 1.6083565459610027, + "grad_norm": 1.5517035722732544, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8870565295219421, + "num_tokens": 303990358.0, + "step": 8661 + }, + { + "epoch": 1.6085422469823585, + "grad_norm": 1.364833950996399, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8764212131500244, + "num_tokens": 304035103.0, + "step": 8662 + }, + { + "epoch": 1.608727948003714, + "grad_norm": 1.6805388927459717, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.86026531457901, + "num_tokens": 304065798.0, + "step": 8663 + }, + { + "epoch": 1.6089136490250695, + "grad_norm": 1.6605457067489624, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8804674744606018, + "num_tokens": 304094664.0, + "step": 8664 + }, + { + "epoch": 1.6090993500464252, + "grad_norm": 1.365863561630249, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8996173739433289, + "num_tokens": 304130237.0, + "step": 8665 + }, + { + "epoch": 1.609285051067781, + "grad_norm": 1.578798532485962, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8808488845825195, + "num_tokens": 304163079.0, + "step": 8666 + }, + { + "epoch": 1.6094707520891365, + "grad_norm": 1.4224168062210083, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8751341700553894, + "num_tokens": 304201172.0, + "step": 8667 + }, + { + "epoch": 1.609656453110492, + "grad_norm": 1.4720947742462158, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8794476985931396, + "num_tokens": 304235952.0, + "step": 8668 + }, + { + "epoch": 1.6098421541318477, + "grad_norm": 1.5684974193572998, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8783180713653564, + "num_tokens": 304271284.0, + "step": 8669 + }, + { + "epoch": 1.6100278551532035, + "grad_norm": 1.390541911125183, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8861373662948608, + "num_tokens": 304309009.0, + "step": 8670 + }, + { + "epoch": 1.610213556174559, + "grad_norm": 1.5264538526535034, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8772478699684143, + "num_tokens": 304343878.0, + "step": 8671 + }, + { + "epoch": 1.6103992571959145, + "grad_norm": 1.4023878574371338, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8883217573165894, + "num_tokens": 304379967.0, + "step": 8672 + }, + { + "epoch": 1.6105849582172702, + "grad_norm": 1.5144702196121216, + "learning_rate": 1e-06, + "loss": 0.3957, + "mean_token_accuracy": 0.8678634166717529, + "num_tokens": 304413564.0, + "step": 8673 + }, + { + "epoch": 1.610770659238626, + "grad_norm": 1.5041192770004272, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8747696876525879, + "num_tokens": 304450313.0, + "step": 8674 + }, + { + "epoch": 1.6109563602599815, + "grad_norm": 1.5144734382629395, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8744767904281616, + "num_tokens": 304485601.0, + "step": 8675 + }, + { + "epoch": 1.611142061281337, + "grad_norm": 1.5974193811416626, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8711270689964294, + "num_tokens": 304520785.0, + "step": 8676 + }, + { + "epoch": 1.6113277623026927, + "grad_norm": 1.43065345287323, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8840346336364746, + "num_tokens": 304559405.0, + "step": 8677 + }, + { + "epoch": 1.6115134633240484, + "grad_norm": 1.5042251348495483, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8773189187049866, + "num_tokens": 304594080.0, + "step": 8678 + }, + { + "epoch": 1.611699164345404, + "grad_norm": 1.4106810092926025, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.88685142993927, + "num_tokens": 304635133.0, + "step": 8679 + }, + { + "epoch": 1.6118848653667595, + "grad_norm": 1.6764671802520752, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.868034839630127, + "num_tokens": 304665249.0, + "step": 8680 + }, + { + "epoch": 1.6120705663881152, + "grad_norm": 1.5834828615188599, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8764488101005554, + "num_tokens": 304697759.0, + "step": 8681 + }, + { + "epoch": 1.6122562674094707, + "grad_norm": 1.3617640733718872, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8823290467262268, + "num_tokens": 304736562.0, + "step": 8682 + }, + { + "epoch": 1.6124419684308262, + "grad_norm": 1.5565518140792847, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8756011724472046, + "num_tokens": 304769295.0, + "step": 8683 + }, + { + "epoch": 1.612627669452182, + "grad_norm": 1.4756547212600708, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8714494705200195, + "num_tokens": 304810816.0, + "step": 8684 + }, + { + "epoch": 1.6128133704735377, + "grad_norm": 1.6713759899139404, + "learning_rate": 1e-06, + "loss": 0.4127, + "mean_token_accuracy": 0.861907422542572, + "num_tokens": 304846990.0, + "step": 8685 + }, + { + "epoch": 1.6129990714948932, + "grad_norm": 1.6081074476242065, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8662509918212891, + "num_tokens": 304882202.0, + "step": 8686 + }, + { + "epoch": 1.6131847725162487, + "grad_norm": 1.4065319299697876, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8860177397727966, + "num_tokens": 304919557.0, + "step": 8687 + }, + { + "epoch": 1.6133704735376044, + "grad_norm": 1.659658670425415, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8803903460502625, + "num_tokens": 304952836.0, + "step": 8688 + }, + { + "epoch": 1.6135561745589602, + "grad_norm": 1.6550031900405884, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8720766305923462, + "num_tokens": 304981830.0, + "step": 8689 + }, + { + "epoch": 1.6137418755803157, + "grad_norm": 1.536708116531372, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8655641078948975, + "num_tokens": 305017660.0, + "step": 8690 + }, + { + "epoch": 1.6139275766016712, + "grad_norm": 1.4925668239593506, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8684197664260864, + "num_tokens": 305052942.0, + "step": 8691 + }, + { + "epoch": 1.614113277623027, + "grad_norm": 1.3687317371368408, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8783403635025024, + "num_tokens": 305092479.0, + "step": 8692 + }, + { + "epoch": 1.6142989786443827, + "grad_norm": 1.452431321144104, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8702877759933472, + "num_tokens": 305128510.0, + "step": 8693 + }, + { + "epoch": 1.6144846796657382, + "grad_norm": 1.5685608386993408, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8815866708755493, + "num_tokens": 305159953.0, + "step": 8694 + }, + { + "epoch": 1.6146703806870937, + "grad_norm": 1.651180386543274, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8765655755996704, + "num_tokens": 305189690.0, + "step": 8695 + }, + { + "epoch": 1.6148560817084494, + "grad_norm": 1.4845157861709595, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8671001195907593, + "num_tokens": 305225280.0, + "step": 8696 + }, + { + "epoch": 1.6150417827298051, + "grad_norm": 1.6279489994049072, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8805264830589294, + "num_tokens": 305255791.0, + "step": 8697 + }, + { + "epoch": 1.6152274837511607, + "grad_norm": 1.4140280485153198, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8768787980079651, + "num_tokens": 305296094.0, + "step": 8698 + }, + { + "epoch": 1.6154131847725162, + "grad_norm": 1.4506679773330688, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8882778882980347, + "num_tokens": 305332751.0, + "step": 8699 + }, + { + "epoch": 1.615598885793872, + "grad_norm": 1.406745195388794, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8830748796463013, + "num_tokens": 305369360.0, + "step": 8700 + }, + { + "epoch": 1.6157845868152276, + "grad_norm": 1.4287720918655396, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8778994679450989, + "num_tokens": 305408072.0, + "step": 8701 + }, + { + "epoch": 1.6159702878365831, + "grad_norm": 1.3997539281845093, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8906287550926208, + "num_tokens": 305445009.0, + "step": 8702 + }, + { + "epoch": 1.6161559888579387, + "grad_norm": 1.5959357023239136, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8745639324188232, + "num_tokens": 305479669.0, + "step": 8703 + }, + { + "epoch": 1.6163416898792944, + "grad_norm": 1.4948408603668213, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8932453989982605, + "num_tokens": 305512474.0, + "step": 8704 + }, + { + "epoch": 1.6165273909006501, + "grad_norm": 1.5467257499694824, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8798109292984009, + "num_tokens": 305543880.0, + "step": 8705 + }, + { + "epoch": 1.6167130919220054, + "grad_norm": 1.3047274351119995, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8820558190345764, + "num_tokens": 305586436.0, + "step": 8706 + }, + { + "epoch": 1.6168987929433611, + "grad_norm": 1.5195002555847168, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8842570781707764, + "num_tokens": 305619416.0, + "step": 8707 + }, + { + "epoch": 1.6170844939647169, + "grad_norm": 1.6787757873535156, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8729267120361328, + "num_tokens": 305649767.0, + "step": 8708 + }, + { + "epoch": 1.6172701949860724, + "grad_norm": 1.4061808586120605, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8813339471817017, + "num_tokens": 305687252.0, + "step": 8709 + }, + { + "epoch": 1.617455896007428, + "grad_norm": 1.4195038080215454, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.886960506439209, + "num_tokens": 305725241.0, + "step": 8710 + }, + { + "epoch": 1.6176415970287836, + "grad_norm": 1.6364080905914307, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8824055194854736, + "num_tokens": 305756038.0, + "step": 8711 + }, + { + "epoch": 1.6178272980501394, + "grad_norm": 1.4752508401870728, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.874705970287323, + "num_tokens": 305794578.0, + "step": 8712 + }, + { + "epoch": 1.6180129990714949, + "grad_norm": 1.375638484954834, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8693130612373352, + "num_tokens": 305834884.0, + "step": 8713 + }, + { + "epoch": 1.6181987000928504, + "grad_norm": 1.6998589038848877, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8694379925727844, + "num_tokens": 305865383.0, + "step": 8714 + }, + { + "epoch": 1.6183844011142061, + "grad_norm": 1.597496747970581, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.869655430316925, + "num_tokens": 305897789.0, + "step": 8715 + }, + { + "epoch": 1.6185701021355619, + "grad_norm": 1.6143461465835571, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8690913915634155, + "num_tokens": 305931407.0, + "step": 8716 + }, + { + "epoch": 1.6187558031569174, + "grad_norm": 1.679085612297058, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8667205572128296, + "num_tokens": 305967095.0, + "step": 8717 + }, + { + "epoch": 1.6189415041782729, + "grad_norm": 1.5259267091751099, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8761764764785767, + "num_tokens": 306003372.0, + "step": 8718 + }, + { + "epoch": 1.6191272051996286, + "grad_norm": 1.4589874744415283, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8797821998596191, + "num_tokens": 306042218.0, + "step": 8719 + }, + { + "epoch": 1.6193129062209843, + "grad_norm": 1.4640742540359497, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8705742955207825, + "num_tokens": 306078434.0, + "step": 8720 + }, + { + "epoch": 1.6194986072423398, + "grad_norm": 1.7491344213485718, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8778506517410278, + "num_tokens": 306107619.0, + "step": 8721 + }, + { + "epoch": 1.6196843082636954, + "grad_norm": 1.5851788520812988, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.878434956073761, + "num_tokens": 306145475.0, + "step": 8722 + }, + { + "epoch": 1.619870009285051, + "grad_norm": 1.4632056951522827, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8773585557937622, + "num_tokens": 306182636.0, + "step": 8723 + }, + { + "epoch": 1.6200557103064068, + "grad_norm": 1.4702199697494507, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8804879784584045, + "num_tokens": 306216227.0, + "step": 8724 + }, + { + "epoch": 1.6202414113277623, + "grad_norm": 1.3943432569503784, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8783169984817505, + "num_tokens": 306255584.0, + "step": 8725 + }, + { + "epoch": 1.6204271123491178, + "grad_norm": 1.3491299152374268, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8735235929489136, + "num_tokens": 306296749.0, + "step": 8726 + }, + { + "epoch": 1.6206128133704736, + "grad_norm": 1.501503348350525, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8812845945358276, + "num_tokens": 306329059.0, + "step": 8727 + }, + { + "epoch": 1.6207985143918293, + "grad_norm": 1.3936340808868408, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8652371168136597, + "num_tokens": 306370190.0, + "step": 8728 + }, + { + "epoch": 1.6209842154131848, + "grad_norm": 1.5598809719085693, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8757041692733765, + "num_tokens": 306401848.0, + "step": 8729 + }, + { + "epoch": 1.6211699164345403, + "grad_norm": 1.4672404527664185, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8750290870666504, + "num_tokens": 306441910.0, + "step": 8730 + }, + { + "epoch": 1.621355617455896, + "grad_norm": 1.4114313125610352, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8801937699317932, + "num_tokens": 306482304.0, + "step": 8731 + }, + { + "epoch": 1.6215413184772516, + "grad_norm": 1.6649271249771118, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8728946447372437, + "num_tokens": 306511204.0, + "step": 8732 + }, + { + "epoch": 1.621727019498607, + "grad_norm": 1.4344637393951416, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8825497627258301, + "num_tokens": 306548756.0, + "step": 8733 + }, + { + "epoch": 1.6219127205199628, + "grad_norm": 1.4652959108352661, + "learning_rate": 1e-06, + "loss": 0.4398, + "mean_token_accuracy": 0.8478121757507324, + "num_tokens": 306589508.0, + "step": 8734 + }, + { + "epoch": 1.6220984215413186, + "grad_norm": 1.5007725954055786, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8764089345932007, + "num_tokens": 306622750.0, + "step": 8735 + }, + { + "epoch": 1.622284122562674, + "grad_norm": 1.6759421825408936, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.874485433101654, + "num_tokens": 306651217.0, + "step": 8736 + }, + { + "epoch": 1.6224698235840296, + "grad_norm": 1.4404492378234863, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8841179609298706, + "num_tokens": 306693222.0, + "step": 8737 + }, + { + "epoch": 1.6226555246053853, + "grad_norm": 1.4970009326934814, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8699041604995728, + "num_tokens": 306730108.0, + "step": 8738 + }, + { + "epoch": 1.622841225626741, + "grad_norm": 1.6293103694915771, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8763409852981567, + "num_tokens": 306761457.0, + "step": 8739 + }, + { + "epoch": 1.6230269266480966, + "grad_norm": 1.7536689043045044, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8675086498260498, + "num_tokens": 306792302.0, + "step": 8740 + }, + { + "epoch": 1.623212627669452, + "grad_norm": 1.5174241065979004, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8745543956756592, + "num_tokens": 306827639.0, + "step": 8741 + }, + { + "epoch": 1.6233983286908078, + "grad_norm": 1.5220268964767456, + "learning_rate": 1e-06, + "loss": 0.4182, + "mean_token_accuracy": 0.8585789203643799, + "num_tokens": 306862645.0, + "step": 8742 + }, + { + "epoch": 1.6235840297121635, + "grad_norm": 1.5249418020248413, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8747638463973999, + "num_tokens": 306899276.0, + "step": 8743 + }, + { + "epoch": 1.623769730733519, + "grad_norm": 1.50497567653656, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8754664659500122, + "num_tokens": 306930376.0, + "step": 8744 + }, + { + "epoch": 1.6239554317548746, + "grad_norm": 1.4634649753570557, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8761186599731445, + "num_tokens": 306966496.0, + "step": 8745 + }, + { + "epoch": 1.6241411327762303, + "grad_norm": 1.3840909004211426, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.877822756767273, + "num_tokens": 307005593.0, + "step": 8746 + }, + { + "epoch": 1.624326833797586, + "grad_norm": 1.6565399169921875, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8749035596847534, + "num_tokens": 307030086.0, + "step": 8747 + }, + { + "epoch": 1.6245125348189415, + "grad_norm": 1.5201406478881836, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8667408227920532, + "num_tokens": 307065124.0, + "step": 8748 + }, + { + "epoch": 1.624698235840297, + "grad_norm": 1.4840967655181885, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8842877745628357, + "num_tokens": 307100169.0, + "step": 8749 + }, + { + "epoch": 1.6248839368616528, + "grad_norm": 1.4848341941833496, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8776321411132812, + "num_tokens": 307134406.0, + "step": 8750 + }, + { + "epoch": 1.6250696378830085, + "grad_norm": 1.660075068473816, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8838348388671875, + "num_tokens": 307163519.0, + "step": 8751 + }, + { + "epoch": 1.625255338904364, + "grad_norm": 1.4854177236557007, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8742350935935974, + "num_tokens": 307201017.0, + "step": 8752 + }, + { + "epoch": 1.6254410399257195, + "grad_norm": 1.4537285566329956, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8653137683868408, + "num_tokens": 307240629.0, + "step": 8753 + }, + { + "epoch": 1.6256267409470753, + "grad_norm": 1.5789217948913574, + "learning_rate": 1e-06, + "loss": 0.396, + "mean_token_accuracy": 0.8687289953231812, + "num_tokens": 307274871.0, + "step": 8754 + }, + { + "epoch": 1.6258124419684308, + "grad_norm": 1.5646930932998657, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8758558034896851, + "num_tokens": 307307950.0, + "step": 8755 + }, + { + "epoch": 1.6259981429897863, + "grad_norm": 1.3943865299224854, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8894426822662354, + "num_tokens": 307346477.0, + "step": 8756 + }, + { + "epoch": 1.626183844011142, + "grad_norm": 1.4669229984283447, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8832692503929138, + "num_tokens": 307382870.0, + "step": 8757 + }, + { + "epoch": 1.6263695450324978, + "grad_norm": 1.5771912336349487, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8723326325416565, + "num_tokens": 307417611.0, + "step": 8758 + }, + { + "epoch": 1.6265552460538533, + "grad_norm": 1.5983731746673584, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8692153692245483, + "num_tokens": 307449424.0, + "step": 8759 + }, + { + "epoch": 1.6267409470752088, + "grad_norm": 1.434687852859497, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8807917237281799, + "num_tokens": 307482595.0, + "step": 8760 + }, + { + "epoch": 1.6269266480965645, + "grad_norm": 1.4245076179504395, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8805575966835022, + "num_tokens": 307520548.0, + "step": 8761 + }, + { + "epoch": 1.6271123491179202, + "grad_norm": 1.5866235494613647, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8663567304611206, + "num_tokens": 307555109.0, + "step": 8762 + }, + { + "epoch": 1.6272980501392758, + "grad_norm": 1.473534345626831, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8799179196357727, + "num_tokens": 307590944.0, + "step": 8763 + }, + { + "epoch": 1.6274837511606313, + "grad_norm": 1.5540928840637207, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8743789792060852, + "num_tokens": 307628145.0, + "step": 8764 + }, + { + "epoch": 1.627669452181987, + "grad_norm": 1.592418909072876, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8807533979415894, + "num_tokens": 307656229.0, + "step": 8765 + }, + { + "epoch": 1.6278551532033427, + "grad_norm": 1.4167540073394775, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8827605843544006, + "num_tokens": 307691475.0, + "step": 8766 + }, + { + "epoch": 1.6280408542246982, + "grad_norm": 1.3142675161361694, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8898992538452148, + "num_tokens": 307729625.0, + "step": 8767 + }, + { + "epoch": 1.6282265552460538, + "grad_norm": 1.5412684679031372, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8771457672119141, + "num_tokens": 307761508.0, + "step": 8768 + }, + { + "epoch": 1.6284122562674095, + "grad_norm": 1.4379920959472656, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8731703758239746, + "num_tokens": 307800309.0, + "step": 8769 + }, + { + "epoch": 1.6285979572887652, + "grad_norm": 1.4354006052017212, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8758375644683838, + "num_tokens": 307837878.0, + "step": 8770 + }, + { + "epoch": 1.6287836583101207, + "grad_norm": 1.4986417293548584, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8792598247528076, + "num_tokens": 307874139.0, + "step": 8771 + }, + { + "epoch": 1.6289693593314762, + "grad_norm": 1.442514181137085, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8950084447860718, + "num_tokens": 307904202.0, + "step": 8772 + }, + { + "epoch": 1.629155060352832, + "grad_norm": 1.497694969177246, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8754478096961975, + "num_tokens": 307936612.0, + "step": 8773 + }, + { + "epoch": 1.6293407613741877, + "grad_norm": 1.556930422782898, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8744391202926636, + "num_tokens": 307972063.0, + "step": 8774 + }, + { + "epoch": 1.6295264623955432, + "grad_norm": 1.4880216121673584, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8757921457290649, + "num_tokens": 308008649.0, + "step": 8775 + }, + { + "epoch": 1.6297121634168987, + "grad_norm": 1.4032464027404785, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8780128955841064, + "num_tokens": 308047212.0, + "step": 8776 + }, + { + "epoch": 1.6298978644382545, + "grad_norm": 1.6344482898712158, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8821244835853577, + "num_tokens": 308075595.0, + "step": 8777 + }, + { + "epoch": 1.63008356545961, + "grad_norm": 1.5348190069198608, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8705311417579651, + "num_tokens": 308107498.0, + "step": 8778 + }, + { + "epoch": 1.6302692664809655, + "grad_norm": 1.3867136240005493, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8775243759155273, + "num_tokens": 308149448.0, + "step": 8779 + }, + { + "epoch": 1.6304549675023212, + "grad_norm": 1.5465947389602661, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8711714148521423, + "num_tokens": 308183591.0, + "step": 8780 + }, + { + "epoch": 1.630640668523677, + "grad_norm": 1.4757568836212158, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8833496570587158, + "num_tokens": 308218615.0, + "step": 8781 + }, + { + "epoch": 1.6308263695450325, + "grad_norm": 1.4068989753723145, + "learning_rate": 1e-06, + "loss": 0.3898, + "mean_token_accuracy": 0.8647124171257019, + "num_tokens": 308259416.0, + "step": 8782 + }, + { + "epoch": 1.631012070566388, + "grad_norm": 1.6037265062332153, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8876352310180664, + "num_tokens": 308288989.0, + "step": 8783 + }, + { + "epoch": 1.6311977715877437, + "grad_norm": 1.447264552116394, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8808102607727051, + "num_tokens": 308326045.0, + "step": 8784 + }, + { + "epoch": 1.6313834726090994, + "grad_norm": 1.5189186334609985, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8615312576293945, + "num_tokens": 308362184.0, + "step": 8785 + }, + { + "epoch": 1.631569173630455, + "grad_norm": 1.4772027730941772, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8766642808914185, + "num_tokens": 308396195.0, + "step": 8786 + }, + { + "epoch": 1.6317548746518105, + "grad_norm": 1.5007781982421875, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8865938186645508, + "num_tokens": 308430222.0, + "step": 8787 + }, + { + "epoch": 1.6319405756731662, + "grad_norm": 1.5558912754058838, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8836768865585327, + "num_tokens": 308464184.0, + "step": 8788 + }, + { + "epoch": 1.632126276694522, + "grad_norm": 1.6307247877120972, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.867243766784668, + "num_tokens": 308498817.0, + "step": 8789 + }, + { + "epoch": 1.6323119777158774, + "grad_norm": 1.5598108768463135, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8807798624038696, + "num_tokens": 308527962.0, + "step": 8790 + }, + { + "epoch": 1.632497678737233, + "grad_norm": 1.4567112922668457, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.88678377866745, + "num_tokens": 308559381.0, + "step": 8791 + }, + { + "epoch": 1.6326833797585887, + "grad_norm": 1.4463914632797241, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8756959438323975, + "num_tokens": 308594621.0, + "step": 8792 + }, + { + "epoch": 1.6328690807799444, + "grad_norm": 1.466325283050537, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8774162530899048, + "num_tokens": 308630701.0, + "step": 8793 + }, + { + "epoch": 1.6330547818013, + "grad_norm": 1.4016757011413574, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8833144307136536, + "num_tokens": 308668106.0, + "step": 8794 + }, + { + "epoch": 1.6332404828226554, + "grad_norm": 1.5025616884231567, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.8715728521347046, + "num_tokens": 308705786.0, + "step": 8795 + }, + { + "epoch": 1.6334261838440112, + "grad_norm": 1.4519577026367188, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8778420090675354, + "num_tokens": 308741771.0, + "step": 8796 + }, + { + "epoch": 1.633611884865367, + "grad_norm": 1.4087580442428589, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.897010326385498, + "num_tokens": 308775035.0, + "step": 8797 + }, + { + "epoch": 1.6337975858867224, + "grad_norm": 1.5945243835449219, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8802605867385864, + "num_tokens": 308806425.0, + "step": 8798 + }, + { + "epoch": 1.633983286908078, + "grad_norm": 1.3737229108810425, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8853780627250671, + "num_tokens": 308845975.0, + "step": 8799 + }, + { + "epoch": 1.6341689879294337, + "grad_norm": 1.3632408380508423, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8733023405075073, + "num_tokens": 308888459.0, + "step": 8800 + }, + { + "epoch": 1.6343546889507894, + "grad_norm": 1.5310003757476807, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8710340857505798, + "num_tokens": 308923266.0, + "step": 8801 + }, + { + "epoch": 1.6345403899721447, + "grad_norm": 1.5288628339767456, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8884814977645874, + "num_tokens": 308954272.0, + "step": 8802 + }, + { + "epoch": 1.6347260909935004, + "grad_norm": 1.511094093322754, + "learning_rate": 1e-06, + "loss": 0.3912, + "mean_token_accuracy": 0.866833508014679, + "num_tokens": 308988677.0, + "step": 8803 + }, + { + "epoch": 1.6349117920148561, + "grad_norm": 1.432759404182434, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.878625750541687, + "num_tokens": 309025709.0, + "step": 8804 + }, + { + "epoch": 1.6350974930362117, + "grad_norm": 1.5850740671157837, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8811576962471008, + "num_tokens": 309056264.0, + "step": 8805 + }, + { + "epoch": 1.6352831940575672, + "grad_norm": 1.4017839431762695, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8781734108924866, + "num_tokens": 309093172.0, + "step": 8806 + }, + { + "epoch": 1.635468895078923, + "grad_norm": 1.5310770273208618, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8824511170387268, + "num_tokens": 309126738.0, + "step": 8807 + }, + { + "epoch": 1.6356545961002786, + "grad_norm": 1.363408088684082, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8856353759765625, + "num_tokens": 309164633.0, + "step": 8808 + }, + { + "epoch": 1.6358402971216341, + "grad_norm": 1.6239843368530273, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8895658254623413, + "num_tokens": 309196289.0, + "step": 8809 + }, + { + "epoch": 1.6360259981429897, + "grad_norm": 1.4996860027313232, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8782535791397095, + "num_tokens": 309233236.0, + "step": 8810 + }, + { + "epoch": 1.6362116991643454, + "grad_norm": 1.4274225234985352, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8712470531463623, + "num_tokens": 309271774.0, + "step": 8811 + }, + { + "epoch": 1.6363974001857011, + "grad_norm": 1.5128093957901, + "learning_rate": 1e-06, + "loss": 0.4042, + "mean_token_accuracy": 0.8663289546966553, + "num_tokens": 309308750.0, + "step": 8812 + }, + { + "epoch": 1.6365831012070566, + "grad_norm": 1.53651762008667, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.865760087966919, + "num_tokens": 309344092.0, + "step": 8813 + }, + { + "epoch": 1.6367688022284121, + "grad_norm": 1.6628198623657227, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8629043698310852, + "num_tokens": 309376334.0, + "step": 8814 + }, + { + "epoch": 1.6369545032497679, + "grad_norm": 1.5079712867736816, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8867231607437134, + "num_tokens": 309406816.0, + "step": 8815 + }, + { + "epoch": 1.6371402042711236, + "grad_norm": 1.5168379545211792, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8839998245239258, + "num_tokens": 309439429.0, + "step": 8816 + }, + { + "epoch": 1.6373259052924791, + "grad_norm": 1.4663163423538208, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8878685235977173, + "num_tokens": 309471170.0, + "step": 8817 + }, + { + "epoch": 1.6375116063138346, + "grad_norm": 1.4187413454055786, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8791185021400452, + "num_tokens": 309508099.0, + "step": 8818 + }, + { + "epoch": 1.6376973073351904, + "grad_norm": 1.4594498872756958, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8922228813171387, + "num_tokens": 309540555.0, + "step": 8819 + }, + { + "epoch": 1.637883008356546, + "grad_norm": 1.3749327659606934, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8793330788612366, + "num_tokens": 309583947.0, + "step": 8820 + }, + { + "epoch": 1.6380687093779016, + "grad_norm": 1.6074861288070679, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8774174451828003, + "num_tokens": 309612753.0, + "step": 8821 + }, + { + "epoch": 1.6382544103992571, + "grad_norm": 1.3940857648849487, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8880269527435303, + "num_tokens": 309648312.0, + "step": 8822 + }, + { + "epoch": 1.6384401114206129, + "grad_norm": 1.4204806089401245, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8843574523925781, + "num_tokens": 309685078.0, + "step": 8823 + }, + { + "epoch": 1.6386258124419686, + "grad_norm": 1.5077236890792847, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8775948286056519, + "num_tokens": 309720573.0, + "step": 8824 + }, + { + "epoch": 1.638811513463324, + "grad_norm": 1.4946953058242798, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8772401213645935, + "num_tokens": 309754227.0, + "step": 8825 + }, + { + "epoch": 1.6389972144846796, + "grad_norm": 1.545601725578308, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8753557205200195, + "num_tokens": 309791059.0, + "step": 8826 + }, + { + "epoch": 1.6391829155060353, + "grad_norm": 1.4983574151992798, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8839691281318665, + "num_tokens": 309824063.0, + "step": 8827 + }, + { + "epoch": 1.6393686165273909, + "grad_norm": 1.2805207967758179, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8864437937736511, + "num_tokens": 309868546.0, + "step": 8828 + }, + { + "epoch": 1.6395543175487464, + "grad_norm": 1.3213863372802734, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8755791187286377, + "num_tokens": 309911301.0, + "step": 8829 + }, + { + "epoch": 1.639740018570102, + "grad_norm": 1.3048515319824219, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8883644938468933, + "num_tokens": 309952850.0, + "step": 8830 + }, + { + "epoch": 1.6399257195914578, + "grad_norm": 1.4632625579833984, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8914116621017456, + "num_tokens": 309985603.0, + "step": 8831 + }, + { + "epoch": 1.6401114206128133, + "grad_norm": 1.411647081375122, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8695309162139893, + "num_tokens": 310028373.0, + "step": 8832 + }, + { + "epoch": 1.6402971216341689, + "grad_norm": 1.402010440826416, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8895637392997742, + "num_tokens": 310066137.0, + "step": 8833 + }, + { + "epoch": 1.6404828226555246, + "grad_norm": 1.4082362651824951, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8799095153808594, + "num_tokens": 310105484.0, + "step": 8834 + }, + { + "epoch": 1.6406685236768803, + "grad_norm": 1.36593496799469, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.889857828617096, + "num_tokens": 310142714.0, + "step": 8835 + }, + { + "epoch": 1.6408542246982358, + "grad_norm": 1.578908085823059, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8769838213920593, + "num_tokens": 310174657.0, + "step": 8836 + }, + { + "epoch": 1.6410399257195913, + "grad_norm": 1.350860357284546, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.892787516117096, + "num_tokens": 310211267.0, + "step": 8837 + }, + { + "epoch": 1.641225626740947, + "grad_norm": 1.5297051668167114, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8768810629844666, + "num_tokens": 310242197.0, + "step": 8838 + }, + { + "epoch": 1.6414113277623028, + "grad_norm": 1.3844653367996216, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8766921758651733, + "num_tokens": 310283894.0, + "step": 8839 + }, + { + "epoch": 1.6415970287836583, + "grad_norm": 1.6219422817230225, + "learning_rate": 1e-06, + "loss": 0.3985, + "mean_token_accuracy": 0.8643733263015747, + "num_tokens": 310316491.0, + "step": 8840 + }, + { + "epoch": 1.6417827298050138, + "grad_norm": 1.3059686422348022, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8888291120529175, + "num_tokens": 310356772.0, + "step": 8841 + }, + { + "epoch": 1.6419684308263696, + "grad_norm": 1.456917643547058, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8838257193565369, + "num_tokens": 310390496.0, + "step": 8842 + }, + { + "epoch": 1.6421541318477253, + "grad_norm": 1.541982650756836, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8716624975204468, + "num_tokens": 310426661.0, + "step": 8843 + }, + { + "epoch": 1.6423398328690808, + "grad_norm": 1.4936922788619995, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8770549893379211, + "num_tokens": 310459981.0, + "step": 8844 + }, + { + "epoch": 1.6425255338904363, + "grad_norm": 1.278973937034607, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8884695768356323, + "num_tokens": 310501576.0, + "step": 8845 + }, + { + "epoch": 1.642711234911792, + "grad_norm": 1.50526762008667, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8660356402397156, + "num_tokens": 310541200.0, + "step": 8846 + }, + { + "epoch": 1.6428969359331478, + "grad_norm": 1.6272704601287842, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8682695031166077, + "num_tokens": 310572561.0, + "step": 8847 + }, + { + "epoch": 1.6430826369545033, + "grad_norm": 1.4456580877304077, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8822522163391113, + "num_tokens": 310609409.0, + "step": 8848 + }, + { + "epoch": 1.6432683379758588, + "grad_norm": 1.5452362298965454, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8743082284927368, + "num_tokens": 310643142.0, + "step": 8849 + }, + { + "epoch": 1.6434540389972145, + "grad_norm": 1.4430928230285645, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8802481293678284, + "num_tokens": 310679412.0, + "step": 8850 + }, + { + "epoch": 1.64363974001857, + "grad_norm": 1.529344081878662, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8723039031028748, + "num_tokens": 310713601.0, + "step": 8851 + }, + { + "epoch": 1.6438254410399256, + "grad_norm": 1.604614496231079, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8754822015762329, + "num_tokens": 310743629.0, + "step": 8852 + }, + { + "epoch": 1.6440111420612813, + "grad_norm": 1.3580518960952759, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8893928527832031, + "num_tokens": 310782722.0, + "step": 8853 + }, + { + "epoch": 1.644196843082637, + "grad_norm": 1.4977353811264038, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8680915236473083, + "num_tokens": 310819302.0, + "step": 8854 + }, + { + "epoch": 1.6443825441039925, + "grad_norm": 1.458878755569458, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8715255260467529, + "num_tokens": 310853953.0, + "step": 8855 + }, + { + "epoch": 1.644568245125348, + "grad_norm": 1.5845290422439575, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8736267685890198, + "num_tokens": 310884545.0, + "step": 8856 + }, + { + "epoch": 1.6447539461467038, + "grad_norm": 1.4041110277175903, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.87961345911026, + "num_tokens": 310921697.0, + "step": 8857 + }, + { + "epoch": 1.6449396471680595, + "grad_norm": 1.498123049736023, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8716681003570557, + "num_tokens": 310955978.0, + "step": 8858 + }, + { + "epoch": 1.645125348189415, + "grad_norm": 1.5172655582427979, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8731529712677002, + "num_tokens": 310990908.0, + "step": 8859 + }, + { + "epoch": 1.6453110492107705, + "grad_norm": 1.4287794828414917, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8774913549423218, + "num_tokens": 311032535.0, + "step": 8860 + }, + { + "epoch": 1.6454967502321263, + "grad_norm": 1.4721314907073975, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8837524652481079, + "num_tokens": 311067283.0, + "step": 8861 + }, + { + "epoch": 1.645682451253482, + "grad_norm": 1.5101882219314575, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8720141053199768, + "num_tokens": 311100582.0, + "step": 8862 + }, + { + "epoch": 1.6458681522748375, + "grad_norm": 1.4064384698867798, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8824411034584045, + "num_tokens": 311134713.0, + "step": 8863 + }, + { + "epoch": 1.646053853296193, + "grad_norm": 1.5500668287277222, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8832617998123169, + "num_tokens": 311166975.0, + "step": 8864 + }, + { + "epoch": 1.6462395543175488, + "grad_norm": 1.5555071830749512, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8763458728790283, + "num_tokens": 311198135.0, + "step": 8865 + }, + { + "epoch": 1.6464252553389045, + "grad_norm": 1.693751573562622, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8649401068687439, + "num_tokens": 311229797.0, + "step": 8866 + }, + { + "epoch": 1.64661095636026, + "grad_norm": 1.4373862743377686, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8781421780586243, + "num_tokens": 311269515.0, + "step": 8867 + }, + { + "epoch": 1.6467966573816155, + "grad_norm": 1.4530004262924194, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8837056159973145, + "num_tokens": 311306494.0, + "step": 8868 + }, + { + "epoch": 1.6469823584029712, + "grad_norm": 1.673527479171753, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8685054779052734, + "num_tokens": 311336635.0, + "step": 8869 + }, + { + "epoch": 1.647168059424327, + "grad_norm": 1.4069550037384033, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8822675943374634, + "num_tokens": 311375565.0, + "step": 8870 + }, + { + "epoch": 1.6473537604456825, + "grad_norm": 1.6127345561981201, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8600842356681824, + "num_tokens": 311407827.0, + "step": 8871 + }, + { + "epoch": 1.647539461467038, + "grad_norm": 1.4788233041763306, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8787150382995605, + "num_tokens": 311445379.0, + "step": 8872 + }, + { + "epoch": 1.6477251624883937, + "grad_norm": 1.547415018081665, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8820879459381104, + "num_tokens": 311475762.0, + "step": 8873 + }, + { + "epoch": 1.6479108635097495, + "grad_norm": 1.4218932390213013, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8686239719390869, + "num_tokens": 311511998.0, + "step": 8874 + }, + { + "epoch": 1.6480965645311048, + "grad_norm": 1.5078322887420654, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8700742721557617, + "num_tokens": 311542302.0, + "step": 8875 + }, + { + "epoch": 1.6482822655524605, + "grad_norm": 1.4601716995239258, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8852750062942505, + "num_tokens": 311578932.0, + "step": 8876 + }, + { + "epoch": 1.6484679665738162, + "grad_norm": 1.4418424367904663, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8829767107963562, + "num_tokens": 311614700.0, + "step": 8877 + }, + { + "epoch": 1.6486536675951717, + "grad_norm": 1.5497798919677734, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8659979701042175, + "num_tokens": 311651367.0, + "step": 8878 + }, + { + "epoch": 1.6488393686165272, + "grad_norm": 1.4978878498077393, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8718371987342834, + "num_tokens": 311685410.0, + "step": 8879 + }, + { + "epoch": 1.649025069637883, + "grad_norm": 1.4779270887374878, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8919307589530945, + "num_tokens": 311716595.0, + "step": 8880 + }, + { + "epoch": 1.6492107706592387, + "grad_norm": 1.5030838251113892, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8751817941665649, + "num_tokens": 311751930.0, + "step": 8881 + }, + { + "epoch": 1.6493964716805942, + "grad_norm": 1.3794820308685303, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8843585252761841, + "num_tokens": 311787550.0, + "step": 8882 + }, + { + "epoch": 1.6495821727019497, + "grad_norm": 1.5367891788482666, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8759887218475342, + "num_tokens": 311818596.0, + "step": 8883 + }, + { + "epoch": 1.6497678737233055, + "grad_norm": 1.4309027194976807, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8853834867477417, + "num_tokens": 311854172.0, + "step": 8884 + }, + { + "epoch": 1.6499535747446612, + "grad_norm": 1.4439386129379272, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8749986886978149, + "num_tokens": 311889336.0, + "step": 8885 + }, + { + "epoch": 1.6501392757660167, + "grad_norm": 1.4589424133300781, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8698766231536865, + "num_tokens": 311929478.0, + "step": 8886 + }, + { + "epoch": 1.6503249767873722, + "grad_norm": 1.602218747138977, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8753880262374878, + "num_tokens": 311960451.0, + "step": 8887 + }, + { + "epoch": 1.650510677808728, + "grad_norm": 1.4992519617080688, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8879350423812866, + "num_tokens": 311993830.0, + "step": 8888 + }, + { + "epoch": 1.6506963788300837, + "grad_norm": 1.4460217952728271, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8760766386985779, + "num_tokens": 312035826.0, + "step": 8889 + }, + { + "epoch": 1.6508820798514392, + "grad_norm": 1.6405922174453735, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8715041875839233, + "num_tokens": 312070575.0, + "step": 8890 + }, + { + "epoch": 1.6510677808727947, + "grad_norm": 1.5438960790634155, + "learning_rate": 1e-06, + "loss": 0.3082, + "mean_token_accuracy": 0.8857293128967285, + "num_tokens": 312102343.0, + "step": 8891 + }, + { + "epoch": 1.6512534818941504, + "grad_norm": 1.565231204032898, + "learning_rate": 1e-06, + "loss": 0.3966, + "mean_token_accuracy": 0.8671566247940063, + "num_tokens": 312136654.0, + "step": 8892 + }, + { + "epoch": 1.6514391829155062, + "grad_norm": 1.5464839935302734, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.88274747133255, + "num_tokens": 312169518.0, + "step": 8893 + }, + { + "epoch": 1.6516248839368617, + "grad_norm": 1.5232517719268799, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8693109750747681, + "num_tokens": 312204307.0, + "step": 8894 + }, + { + "epoch": 1.6518105849582172, + "grad_norm": 1.4199012517929077, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8942059278488159, + "num_tokens": 312241318.0, + "step": 8895 + }, + { + "epoch": 1.651996285979573, + "grad_norm": 1.5746244192123413, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8774037957191467, + "num_tokens": 312272865.0, + "step": 8896 + }, + { + "epoch": 1.6521819870009287, + "grad_norm": 1.3962976932525635, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8877435922622681, + "num_tokens": 312308769.0, + "step": 8897 + }, + { + "epoch": 1.6523676880222842, + "grad_norm": 1.4503345489501953, + "learning_rate": 1e-06, + "loss": 0.3933, + "mean_token_accuracy": 0.8683817386627197, + "num_tokens": 312348813.0, + "step": 8898 + }, + { + "epoch": 1.6525533890436397, + "grad_norm": 1.4929317235946655, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8711650371551514, + "num_tokens": 312386072.0, + "step": 8899 + }, + { + "epoch": 1.6527390900649954, + "grad_norm": 1.541202187538147, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.8624224066734314, + "num_tokens": 312421330.0, + "step": 8900 + }, + { + "epoch": 1.652924791086351, + "grad_norm": 1.6579794883728027, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8664565682411194, + "num_tokens": 312450091.0, + "step": 8901 + }, + { + "epoch": 1.6531104921077064, + "grad_norm": 1.4600235223770142, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8770980834960938, + "num_tokens": 312487521.0, + "step": 8902 + }, + { + "epoch": 1.6532961931290622, + "grad_norm": 1.5353635549545288, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8778787851333618, + "num_tokens": 312523242.0, + "step": 8903 + }, + { + "epoch": 1.653481894150418, + "grad_norm": 1.5757137537002563, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8826462030410767, + "num_tokens": 312554970.0, + "step": 8904 + }, + { + "epoch": 1.6536675951717734, + "grad_norm": 1.6116771697998047, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8766695261001587, + "num_tokens": 312587653.0, + "step": 8905 + }, + { + "epoch": 1.653853296193129, + "grad_norm": 1.5805928707122803, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8824294805526733, + "num_tokens": 312619261.0, + "step": 8906 + }, + { + "epoch": 1.6540389972144847, + "grad_norm": 1.5317460298538208, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8783299922943115, + "num_tokens": 312652983.0, + "step": 8907 + }, + { + "epoch": 1.6542246982358404, + "grad_norm": 1.6111831665039062, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8791823983192444, + "num_tokens": 312682737.0, + "step": 8908 + }, + { + "epoch": 1.654410399257196, + "grad_norm": 1.4328930377960205, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8814191818237305, + "num_tokens": 312718430.0, + "step": 8909 + }, + { + "epoch": 1.6545961002785514, + "grad_norm": 1.4977713823318481, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8890573382377625, + "num_tokens": 312751107.0, + "step": 8910 + }, + { + "epoch": 1.6547818012999072, + "grad_norm": 1.5553786754608154, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8770425319671631, + "num_tokens": 312783954.0, + "step": 8911 + }, + { + "epoch": 1.6549675023212629, + "grad_norm": 1.3799360990524292, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8805739283561707, + "num_tokens": 312821373.0, + "step": 8912 + }, + { + "epoch": 1.6551532033426184, + "grad_norm": 1.491994023323059, + "learning_rate": 1e-06, + "loss": 0.409, + "mean_token_accuracy": 0.8617466688156128, + "num_tokens": 312859001.0, + "step": 8913 + }, + { + "epoch": 1.655338904363974, + "grad_norm": 1.5718905925750732, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8758643865585327, + "num_tokens": 312894998.0, + "step": 8914 + }, + { + "epoch": 1.6555246053853296, + "grad_norm": 1.4413161277770996, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8722396492958069, + "num_tokens": 312932359.0, + "step": 8915 + }, + { + "epoch": 1.6557103064066854, + "grad_norm": 1.4730380773544312, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8720489144325256, + "num_tokens": 312970539.0, + "step": 8916 + }, + { + "epoch": 1.6558960074280409, + "grad_norm": 1.5009304285049438, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8709856271743774, + "num_tokens": 313006042.0, + "step": 8917 + }, + { + "epoch": 1.6560817084493964, + "grad_norm": 1.4980642795562744, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8935078978538513, + "num_tokens": 313037488.0, + "step": 8918 + }, + { + "epoch": 1.6562674094707521, + "grad_norm": 1.518992304801941, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8785380721092224, + "num_tokens": 313077612.0, + "step": 8919 + }, + { + "epoch": 1.6564531104921079, + "grad_norm": 1.561333417892456, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8814961314201355, + "num_tokens": 313111020.0, + "step": 8920 + }, + { + "epoch": 1.6566388115134634, + "grad_norm": 1.428183674812317, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8848541975021362, + "num_tokens": 313150456.0, + "step": 8921 + }, + { + "epoch": 1.6568245125348189, + "grad_norm": 1.414833903312683, + "learning_rate": 1e-06, + "loss": 0.3847, + "mean_token_accuracy": 0.8699184656143188, + "num_tokens": 313191336.0, + "step": 8922 + }, + { + "epoch": 1.6570102135561746, + "grad_norm": 1.4563193321228027, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8842340111732483, + "num_tokens": 313224960.0, + "step": 8923 + }, + { + "epoch": 1.6571959145775301, + "grad_norm": 1.398996114730835, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8877568244934082, + "num_tokens": 313263143.0, + "step": 8924 + }, + { + "epoch": 1.6573816155988856, + "grad_norm": 1.5616075992584229, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8760997653007507, + "num_tokens": 313295511.0, + "step": 8925 + }, + { + "epoch": 1.6575673166202414, + "grad_norm": 1.6809061765670776, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8638133406639099, + "num_tokens": 313328399.0, + "step": 8926 + }, + { + "epoch": 1.657753017641597, + "grad_norm": 1.4010223150253296, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8820702433586121, + "num_tokens": 313369846.0, + "step": 8927 + }, + { + "epoch": 1.6579387186629526, + "grad_norm": 1.4902492761611938, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8693682551383972, + "num_tokens": 313409787.0, + "step": 8928 + }, + { + "epoch": 1.6581244196843081, + "grad_norm": 1.4307024478912354, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8842066526412964, + "num_tokens": 313442911.0, + "step": 8929 + }, + { + "epoch": 1.6583101207056639, + "grad_norm": 1.4153804779052734, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8870418071746826, + "num_tokens": 313480112.0, + "step": 8930 + }, + { + "epoch": 1.6584958217270196, + "grad_norm": 1.5887768268585205, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8796696662902832, + "num_tokens": 313512239.0, + "step": 8931 + }, + { + "epoch": 1.658681522748375, + "grad_norm": 1.6603704690933228, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8838609457015991, + "num_tokens": 313540983.0, + "step": 8932 + }, + { + "epoch": 1.6588672237697306, + "grad_norm": 1.7119016647338867, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8780423402786255, + "num_tokens": 313570836.0, + "step": 8933 + }, + { + "epoch": 1.6590529247910863, + "grad_norm": 1.4785358905792236, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8766493797302246, + "num_tokens": 313608008.0, + "step": 8934 + }, + { + "epoch": 1.659238625812442, + "grad_norm": 1.5587208271026611, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8617152571678162, + "num_tokens": 313643141.0, + "step": 8935 + }, + { + "epoch": 1.6594243268337976, + "grad_norm": 1.3356274366378784, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8847534656524658, + "num_tokens": 313682960.0, + "step": 8936 + }, + { + "epoch": 1.659610027855153, + "grad_norm": 1.4171867370605469, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8795843720436096, + "num_tokens": 313721657.0, + "step": 8937 + }, + { + "epoch": 1.6597957288765088, + "grad_norm": 1.583942174911499, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8736321926116943, + "num_tokens": 313753296.0, + "step": 8938 + }, + { + "epoch": 1.6599814298978646, + "grad_norm": 1.4786089658737183, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8795340061187744, + "num_tokens": 313785009.0, + "step": 8939 + }, + { + "epoch": 1.66016713091922, + "grad_norm": 1.6569262742996216, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8669551610946655, + "num_tokens": 313814861.0, + "step": 8940 + }, + { + "epoch": 1.6603528319405756, + "grad_norm": 1.3852895498275757, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8901100158691406, + "num_tokens": 313851948.0, + "step": 8941 + }, + { + "epoch": 1.6605385329619313, + "grad_norm": 1.5846341848373413, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8797866106033325, + "num_tokens": 313883308.0, + "step": 8942 + }, + { + "epoch": 1.660724233983287, + "grad_norm": 1.5607349872589111, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8935115337371826, + "num_tokens": 313912111.0, + "step": 8943 + }, + { + "epoch": 1.6609099350046426, + "grad_norm": 1.4154248237609863, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8678572177886963, + "num_tokens": 313959535.0, + "step": 8944 + }, + { + "epoch": 1.661095636025998, + "grad_norm": 1.619620680809021, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8816410303115845, + "num_tokens": 313988060.0, + "step": 8945 + }, + { + "epoch": 1.6612813370473538, + "grad_norm": 1.395810604095459, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8697311878204346, + "num_tokens": 314028669.0, + "step": 8946 + }, + { + "epoch": 1.6614670380687093, + "grad_norm": 1.687446117401123, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8705959320068359, + "num_tokens": 314055233.0, + "step": 8947 + }, + { + "epoch": 1.6616527390900648, + "grad_norm": 1.4888468980789185, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8866939544677734, + "num_tokens": 314092595.0, + "step": 8948 + }, + { + "epoch": 1.6618384401114206, + "grad_norm": 1.5771421194076538, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8816367387771606, + "num_tokens": 314123959.0, + "step": 8949 + }, + { + "epoch": 1.6620241411327763, + "grad_norm": 1.5175920724868774, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8790723085403442, + "num_tokens": 314157394.0, + "step": 8950 + }, + { + "epoch": 1.6622098421541318, + "grad_norm": 1.6635692119598389, + "learning_rate": 1e-06, + "loss": 0.3919, + "mean_token_accuracy": 0.8666136264801025, + "num_tokens": 314185724.0, + "step": 8951 + }, + { + "epoch": 1.6623955431754873, + "grad_norm": 1.528615951538086, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.881332516670227, + "num_tokens": 314217829.0, + "step": 8952 + }, + { + "epoch": 1.662581244196843, + "grad_norm": 1.417943000793457, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8745057582855225, + "num_tokens": 314257254.0, + "step": 8953 + }, + { + "epoch": 1.6627669452181988, + "grad_norm": 1.4616116285324097, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.883428692817688, + "num_tokens": 314295358.0, + "step": 8954 + }, + { + "epoch": 1.6629526462395543, + "grad_norm": 1.492841362953186, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8787522315979004, + "num_tokens": 314336763.0, + "step": 8955 + }, + { + "epoch": 1.6631383472609098, + "grad_norm": 1.4244499206542969, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8739511966705322, + "num_tokens": 314373425.0, + "step": 8956 + }, + { + "epoch": 1.6633240482822655, + "grad_norm": 1.3898066282272339, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8837448954582214, + "num_tokens": 314411694.0, + "step": 8957 + }, + { + "epoch": 1.6635097493036213, + "grad_norm": 1.299759030342102, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8854560256004333, + "num_tokens": 314453636.0, + "step": 8958 + }, + { + "epoch": 1.6636954503249768, + "grad_norm": 1.5241189002990723, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8728893995285034, + "num_tokens": 314487839.0, + "step": 8959 + }, + { + "epoch": 1.6638811513463323, + "grad_norm": 1.4180927276611328, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8683732748031616, + "num_tokens": 314530331.0, + "step": 8960 + }, + { + "epoch": 1.664066852367688, + "grad_norm": 1.3864901065826416, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8870178461074829, + "num_tokens": 314567379.0, + "step": 8961 + }, + { + "epoch": 1.6642525533890438, + "grad_norm": 1.4768270254135132, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8719997406005859, + "num_tokens": 314605834.0, + "step": 8962 + }, + { + "epoch": 1.6644382544103993, + "grad_norm": 1.4720933437347412, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8815141916275024, + "num_tokens": 314639339.0, + "step": 8963 + }, + { + "epoch": 1.6646239554317548, + "grad_norm": 1.5926569700241089, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8803694248199463, + "num_tokens": 314665943.0, + "step": 8964 + }, + { + "epoch": 1.6648096564531105, + "grad_norm": 1.4614050388336182, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8767109513282776, + "num_tokens": 314699768.0, + "step": 8965 + }, + { + "epoch": 1.6649953574744663, + "grad_norm": 1.5714119672775269, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8690395355224609, + "num_tokens": 314733267.0, + "step": 8966 + }, + { + "epoch": 1.6651810584958218, + "grad_norm": 1.5689194202423096, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8763726949691772, + "num_tokens": 314763944.0, + "step": 8967 + }, + { + "epoch": 1.6653667595171773, + "grad_norm": 1.4743478298187256, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8765724301338196, + "num_tokens": 314798210.0, + "step": 8968 + }, + { + "epoch": 1.665552460538533, + "grad_norm": 1.5938302278518677, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8772450685501099, + "num_tokens": 314827897.0, + "step": 8969 + }, + { + "epoch": 1.6657381615598887, + "grad_norm": 1.4186654090881348, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8791335821151733, + "num_tokens": 314865683.0, + "step": 8970 + }, + { + "epoch": 1.665923862581244, + "grad_norm": 1.3195950984954834, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8786659240722656, + "num_tokens": 314910717.0, + "step": 8971 + }, + { + "epoch": 1.6661095636025998, + "grad_norm": 1.543084979057312, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8783237934112549, + "num_tokens": 314941468.0, + "step": 8972 + }, + { + "epoch": 1.6662952646239555, + "grad_norm": 1.4684723615646362, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.8613355755805969, + "num_tokens": 314979189.0, + "step": 8973 + }, + { + "epoch": 1.666480965645311, + "grad_norm": 1.4819104671478271, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8822580575942993, + "num_tokens": 315014161.0, + "step": 8974 + }, + { + "epoch": 1.6666666666666665, + "grad_norm": 1.3451778888702393, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.883178174495697, + "num_tokens": 315054834.0, + "step": 8975 + }, + { + "epoch": 1.6668523676880223, + "grad_norm": 1.6439192295074463, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8709442615509033, + "num_tokens": 315085839.0, + "step": 8976 + }, + { + "epoch": 1.667038068709378, + "grad_norm": 1.4950464963912964, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8817818760871887, + "num_tokens": 315121091.0, + "step": 8977 + }, + { + "epoch": 1.6672237697307335, + "grad_norm": 1.4208166599273682, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8738401532173157, + "num_tokens": 315161153.0, + "step": 8978 + }, + { + "epoch": 1.667409470752089, + "grad_norm": 1.4734199047088623, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8846890330314636, + "num_tokens": 315195620.0, + "step": 8979 + }, + { + "epoch": 1.6675951717734447, + "grad_norm": 1.4665693044662476, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8843325972557068, + "num_tokens": 315230490.0, + "step": 8980 + }, + { + "epoch": 1.6677808727948005, + "grad_norm": 1.489055871963501, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8797163963317871, + "num_tokens": 315272687.0, + "step": 8981 + }, + { + "epoch": 1.667966573816156, + "grad_norm": 1.6828410625457764, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8663746118545532, + "num_tokens": 315301351.0, + "step": 8982 + }, + { + "epoch": 1.6681522748375115, + "grad_norm": 1.3619203567504883, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.88041090965271, + "num_tokens": 315343881.0, + "step": 8983 + }, + { + "epoch": 1.6683379758588672, + "grad_norm": 1.498069405555725, + "learning_rate": 1e-06, + "loss": 0.4022, + "mean_token_accuracy": 0.8632454872131348, + "num_tokens": 315381988.0, + "step": 8984 + }, + { + "epoch": 1.668523676880223, + "grad_norm": 1.4142370223999023, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8774124979972839, + "num_tokens": 315418499.0, + "step": 8985 + }, + { + "epoch": 1.6687093779015785, + "grad_norm": 1.4491097927093506, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8764479160308838, + "num_tokens": 315457617.0, + "step": 8986 + }, + { + "epoch": 1.668895078922934, + "grad_norm": 1.4553438425064087, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.872458815574646, + "num_tokens": 315498958.0, + "step": 8987 + }, + { + "epoch": 1.6690807799442897, + "grad_norm": 1.5813261270523071, + "learning_rate": 1e-06, + "loss": 0.4224, + "mean_token_accuracy": 0.859821081161499, + "num_tokens": 315536886.0, + "step": 8988 + }, + { + "epoch": 1.6692664809656454, + "grad_norm": 1.4006712436676025, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8786537647247314, + "num_tokens": 315575985.0, + "step": 8989 + }, + { + "epoch": 1.669452181987001, + "grad_norm": 1.539009928703308, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8774067163467407, + "num_tokens": 315609458.0, + "step": 8990 + }, + { + "epoch": 1.6696378830083565, + "grad_norm": 1.4516160488128662, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8775162100791931, + "num_tokens": 315643835.0, + "step": 8991 + }, + { + "epoch": 1.6698235840297122, + "grad_norm": 1.4653114080429077, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.872672438621521, + "num_tokens": 315680521.0, + "step": 8992 + }, + { + "epoch": 1.670009285051068, + "grad_norm": 1.5314147472381592, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8897472023963928, + "num_tokens": 315716888.0, + "step": 8993 + }, + { + "epoch": 1.6701949860724234, + "grad_norm": 1.551112174987793, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.870224118232727, + "num_tokens": 315750662.0, + "step": 8994 + }, + { + "epoch": 1.670380687093779, + "grad_norm": 1.438402533531189, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8847528100013733, + "num_tokens": 315787549.0, + "step": 8995 + }, + { + "epoch": 1.6705663881151347, + "grad_norm": 1.3597935438156128, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8870492577552795, + "num_tokens": 315823792.0, + "step": 8996 + }, + { + "epoch": 1.6707520891364902, + "grad_norm": 1.5024694204330444, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8844307065010071, + "num_tokens": 315852849.0, + "step": 8997 + }, + { + "epoch": 1.6709377901578457, + "grad_norm": 1.3797111511230469, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8831807374954224, + "num_tokens": 315895057.0, + "step": 8998 + }, + { + "epoch": 1.6711234911792014, + "grad_norm": 1.4570506811141968, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8771709203720093, + "num_tokens": 315931342.0, + "step": 8999 + }, + { + "epoch": 1.6713091922005572, + "grad_norm": 1.4546589851379395, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8872294425964355, + "num_tokens": 315963084.0, + "step": 9000 + }, + { + "epoch": 1.6714948932219127, + "grad_norm": 1.4664394855499268, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8677573204040527, + "num_tokens": 315999182.0, + "step": 9001 + }, + { + "epoch": 1.6716805942432682, + "grad_norm": 1.533925175666809, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8661251664161682, + "num_tokens": 316035163.0, + "step": 9002 + }, + { + "epoch": 1.671866295264624, + "grad_norm": 1.607791543006897, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8731854557991028, + "num_tokens": 316064469.0, + "step": 9003 + }, + { + "epoch": 1.6720519962859797, + "grad_norm": 1.4331886768341064, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8756078481674194, + "num_tokens": 316104455.0, + "step": 9004 + }, + { + "epoch": 1.6722376973073352, + "grad_norm": 1.3197314739227295, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8888187408447266, + "num_tokens": 316145451.0, + "step": 9005 + }, + { + "epoch": 1.6724233983286907, + "grad_norm": 1.5235648155212402, + "learning_rate": 1e-06, + "loss": 0.4004, + "mean_token_accuracy": 0.8646876811981201, + "num_tokens": 316184953.0, + "step": 9006 + }, + { + "epoch": 1.6726090993500464, + "grad_norm": 1.5450019836425781, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8858804702758789, + "num_tokens": 316216294.0, + "step": 9007 + }, + { + "epoch": 1.6727948003714022, + "grad_norm": 1.5891202688217163, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8719297051429749, + "num_tokens": 316246495.0, + "step": 9008 + }, + { + "epoch": 1.6729805013927577, + "grad_norm": 1.4366565942764282, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8875060677528381, + "num_tokens": 316281824.0, + "step": 9009 + }, + { + "epoch": 1.6731662024141132, + "grad_norm": 1.7056576013565063, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8762835264205933, + "num_tokens": 316310110.0, + "step": 9010 + }, + { + "epoch": 1.673351903435469, + "grad_norm": 1.4996144771575928, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8824400901794434, + "num_tokens": 316343515.0, + "step": 9011 + }, + { + "epoch": 1.6735376044568246, + "grad_norm": 1.7418900728225708, + "learning_rate": 1e-06, + "loss": 0.3692, + "mean_token_accuracy": 0.876838743686676, + "num_tokens": 316372555.0, + "step": 9012 + }, + { + "epoch": 1.6737233054781802, + "grad_norm": 1.4904927015304565, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8833284974098206, + "num_tokens": 316405474.0, + "step": 9013 + }, + { + "epoch": 1.6739090064995357, + "grad_norm": 1.3917081356048584, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8758511543273926, + "num_tokens": 316446317.0, + "step": 9014 + }, + { + "epoch": 1.6740947075208914, + "grad_norm": 1.5508310794830322, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8748484253883362, + "num_tokens": 316480066.0, + "step": 9015 + }, + { + "epoch": 1.6742804085422471, + "grad_norm": 1.4994292259216309, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.8873106837272644, + "num_tokens": 316513690.0, + "step": 9016 + }, + { + "epoch": 1.6744661095636026, + "grad_norm": 1.7422521114349365, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.869318962097168, + "num_tokens": 316545989.0, + "step": 9017 + }, + { + "epoch": 1.6746518105849582, + "grad_norm": 1.4327982664108276, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.875859797000885, + "num_tokens": 316584147.0, + "step": 9018 + }, + { + "epoch": 1.674837511606314, + "grad_norm": 1.5301992893218994, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8824827671051025, + "num_tokens": 316614314.0, + "step": 9019 + }, + { + "epoch": 1.6750232126276694, + "grad_norm": 1.5610707998275757, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.891675591468811, + "num_tokens": 316643143.0, + "step": 9020 + }, + { + "epoch": 1.675208913649025, + "grad_norm": 1.4765740633010864, + "learning_rate": 1e-06, + "loss": 0.4056, + "mean_token_accuracy": 0.86689692735672, + "num_tokens": 316681717.0, + "step": 9021 + }, + { + "epoch": 1.6753946146703806, + "grad_norm": 1.3461275100708008, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.9004707336425781, + "num_tokens": 316715711.0, + "step": 9022 + }, + { + "epoch": 1.6755803156917364, + "grad_norm": 1.384986400604248, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8768852353096008, + "num_tokens": 316755496.0, + "step": 9023 + }, + { + "epoch": 1.6757660167130919, + "grad_norm": 1.3872724771499634, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8897581100463867, + "num_tokens": 316794567.0, + "step": 9024 + }, + { + "epoch": 1.6759517177344474, + "grad_norm": 1.485870122909546, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.896624743938446, + "num_tokens": 316824722.0, + "step": 9025 + }, + { + "epoch": 1.6761374187558031, + "grad_norm": 1.6366963386535645, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8714727163314819, + "num_tokens": 316857119.0, + "step": 9026 + }, + { + "epoch": 1.6763231197771589, + "grad_norm": 1.4423328638076782, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8815599679946899, + "num_tokens": 316892516.0, + "step": 9027 + }, + { + "epoch": 1.6765088207985144, + "grad_norm": 1.3186331987380981, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8942216634750366, + "num_tokens": 316929957.0, + "step": 9028 + }, + { + "epoch": 1.6766945218198699, + "grad_norm": 1.6300251483917236, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8674957752227783, + "num_tokens": 316961782.0, + "step": 9029 + }, + { + "epoch": 1.6768802228412256, + "grad_norm": 1.657475471496582, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8793435096740723, + "num_tokens": 316990603.0, + "step": 9030 + }, + { + "epoch": 1.6770659238625814, + "grad_norm": 1.4541898965835571, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8802953362464905, + "num_tokens": 317026629.0, + "step": 9031 + }, + { + "epoch": 1.6772516248839369, + "grad_norm": 1.4535819292068481, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8791267275810242, + "num_tokens": 317064126.0, + "step": 9032 + }, + { + "epoch": 1.6774373259052924, + "grad_norm": 1.539636254310608, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8781434297561646, + "num_tokens": 317096118.0, + "step": 9033 + }, + { + "epoch": 1.677623026926648, + "grad_norm": 1.5415947437286377, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8608466386795044, + "num_tokens": 317135817.0, + "step": 9034 + }, + { + "epoch": 1.6778087279480038, + "grad_norm": 1.4595845937728882, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8751758337020874, + "num_tokens": 317171923.0, + "step": 9035 + }, + { + "epoch": 1.6779944289693594, + "grad_norm": 1.613466501235962, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8667392730712891, + "num_tokens": 317202976.0, + "step": 9036 + }, + { + "epoch": 1.6781801299907149, + "grad_norm": 1.6526473760604858, + "learning_rate": 1e-06, + "loss": 0.3944, + "mean_token_accuracy": 0.8640467524528503, + "num_tokens": 317237117.0, + "step": 9037 + }, + { + "epoch": 1.6783658310120706, + "grad_norm": 1.389039158821106, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8819458484649658, + "num_tokens": 317274929.0, + "step": 9038 + }, + { + "epoch": 1.6785515320334263, + "grad_norm": 1.5098527669906616, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8706680536270142, + "num_tokens": 317308748.0, + "step": 9039 + }, + { + "epoch": 1.6787372330547818, + "grad_norm": 1.4321624040603638, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8872352242469788, + "num_tokens": 317343259.0, + "step": 9040 + }, + { + "epoch": 1.6789229340761374, + "grad_norm": 1.5503164529800415, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8613461256027222, + "num_tokens": 317375262.0, + "step": 9041 + }, + { + "epoch": 1.679108635097493, + "grad_norm": 1.5579004287719727, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8834899067878723, + "num_tokens": 317409316.0, + "step": 9042 + }, + { + "epoch": 1.6792943361188488, + "grad_norm": 1.5094672441482544, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8718124032020569, + "num_tokens": 317445090.0, + "step": 9043 + }, + { + "epoch": 1.679480037140204, + "grad_norm": 1.4726313352584839, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8824191093444824, + "num_tokens": 317483034.0, + "step": 9044 + }, + { + "epoch": 1.6796657381615598, + "grad_norm": 1.4312282800674438, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8815149068832397, + "num_tokens": 317518301.0, + "step": 9045 + }, + { + "epoch": 1.6798514391829156, + "grad_norm": 1.4263434410095215, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8838934898376465, + "num_tokens": 317554646.0, + "step": 9046 + }, + { + "epoch": 1.680037140204271, + "grad_norm": 1.5525623559951782, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8808626532554626, + "num_tokens": 317588774.0, + "step": 9047 + }, + { + "epoch": 1.6802228412256266, + "grad_norm": 1.548115849494934, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8734739422798157, + "num_tokens": 317621850.0, + "step": 9048 + }, + { + "epoch": 1.6804085422469823, + "grad_norm": 1.554495096206665, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8828140497207642, + "num_tokens": 317656008.0, + "step": 9049 + }, + { + "epoch": 1.680594243268338, + "grad_norm": 1.5156327486038208, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8724480867385864, + "num_tokens": 317688904.0, + "step": 9050 + }, + { + "epoch": 1.6807799442896936, + "grad_norm": 1.387790322303772, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8817310929298401, + "num_tokens": 317726445.0, + "step": 9051 + }, + { + "epoch": 1.680965645311049, + "grad_norm": 1.4148404598236084, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8780148029327393, + "num_tokens": 317767813.0, + "step": 9052 + }, + { + "epoch": 1.6811513463324048, + "grad_norm": 1.3214462995529175, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8902824521064758, + "num_tokens": 317806891.0, + "step": 9053 + }, + { + "epoch": 1.6813370473537605, + "grad_norm": 1.589869499206543, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.875031590461731, + "num_tokens": 317837829.0, + "step": 9054 + }, + { + "epoch": 1.681522748375116, + "grad_norm": 1.3762346506118774, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8910642862319946, + "num_tokens": 317873821.0, + "step": 9055 + }, + { + "epoch": 1.6817084493964716, + "grad_norm": 1.5405691862106323, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8805490732192993, + "num_tokens": 317903525.0, + "step": 9056 + }, + { + "epoch": 1.6818941504178273, + "grad_norm": 1.3664180040359497, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8727485537528992, + "num_tokens": 317942931.0, + "step": 9057 + }, + { + "epoch": 1.682079851439183, + "grad_norm": 1.6780086755752563, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8832806944847107, + "num_tokens": 317971956.0, + "step": 9058 + }, + { + "epoch": 1.6822655524605385, + "grad_norm": 1.5008444786071777, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8808724880218506, + "num_tokens": 318007120.0, + "step": 9059 + }, + { + "epoch": 1.682451253481894, + "grad_norm": 1.4367789030075073, + "learning_rate": 1e-06, + "loss": 0.3958, + "mean_token_accuracy": 0.8642771244049072, + "num_tokens": 318047005.0, + "step": 9060 + }, + { + "epoch": 1.6826369545032498, + "grad_norm": 1.3927977085113525, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8797934055328369, + "num_tokens": 318086928.0, + "step": 9061 + }, + { + "epoch": 1.6828226555246055, + "grad_norm": 1.496133804321289, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8708446025848389, + "num_tokens": 318118764.0, + "step": 9062 + }, + { + "epoch": 1.683008356545961, + "grad_norm": 1.523267388343811, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8735101222991943, + "num_tokens": 318152158.0, + "step": 9063 + }, + { + "epoch": 1.6831940575673165, + "grad_norm": 1.5756758451461792, + "learning_rate": 1e-06, + "loss": 0.4068, + "mean_token_accuracy": 0.8698815703392029, + "num_tokens": 318188657.0, + "step": 9064 + }, + { + "epoch": 1.6833797585886723, + "grad_norm": 1.4046276807785034, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8697835803031921, + "num_tokens": 318231011.0, + "step": 9065 + }, + { + "epoch": 1.683565459610028, + "grad_norm": 1.4438045024871826, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8766493201255798, + "num_tokens": 318270906.0, + "step": 9066 + }, + { + "epoch": 1.6837511606313835, + "grad_norm": 1.3277757167816162, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8913685083389282, + "num_tokens": 318308847.0, + "step": 9067 + }, + { + "epoch": 1.683936861652739, + "grad_norm": 1.507662296295166, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8795866370201111, + "num_tokens": 318341958.0, + "step": 9068 + }, + { + "epoch": 1.6841225626740948, + "grad_norm": 1.490190863609314, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8903154730796814, + "num_tokens": 318373824.0, + "step": 9069 + }, + { + "epoch": 1.6843082636954503, + "grad_norm": 1.4700212478637695, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8774850368499756, + "num_tokens": 318413041.0, + "step": 9070 + }, + { + "epoch": 1.6844939647168058, + "grad_norm": 1.375577449798584, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8771722912788391, + "num_tokens": 318452688.0, + "step": 9071 + }, + { + "epoch": 1.6846796657381615, + "grad_norm": 1.420769453048706, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.8855655193328857, + "num_tokens": 318486101.0, + "step": 9072 + }, + { + "epoch": 1.6848653667595173, + "grad_norm": 1.5125503540039062, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8651129007339478, + "num_tokens": 318521158.0, + "step": 9073 + }, + { + "epoch": 1.6850510677808728, + "grad_norm": 1.5853984355926514, + "learning_rate": 1e-06, + "loss": 0.412, + "mean_token_accuracy": 0.8609927296638489, + "num_tokens": 318553602.0, + "step": 9074 + }, + { + "epoch": 1.6852367688022283, + "grad_norm": 1.5198085308074951, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8809319138526917, + "num_tokens": 318588863.0, + "step": 9075 + }, + { + "epoch": 1.685422469823584, + "grad_norm": 1.6814790964126587, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8770442008972168, + "num_tokens": 318620753.0, + "step": 9076 + }, + { + "epoch": 1.6856081708449397, + "grad_norm": 1.536743402481079, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8809530138969421, + "num_tokens": 318654883.0, + "step": 9077 + }, + { + "epoch": 1.6857938718662953, + "grad_norm": 1.4866994619369507, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8759051561355591, + "num_tokens": 318690850.0, + "step": 9078 + }, + { + "epoch": 1.6859795728876508, + "grad_norm": 1.4965698719024658, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8804538249969482, + "num_tokens": 318732181.0, + "step": 9079 + }, + { + "epoch": 1.6861652739090065, + "grad_norm": 1.5535168647766113, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8745994567871094, + "num_tokens": 318764702.0, + "step": 9080 + }, + { + "epoch": 1.6863509749303622, + "grad_norm": 1.4885549545288086, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.87154221534729, + "num_tokens": 318799671.0, + "step": 9081 + }, + { + "epoch": 1.6865366759517177, + "grad_norm": 1.7036312818527222, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8823970556259155, + "num_tokens": 318827740.0, + "step": 9082 + }, + { + "epoch": 1.6867223769730733, + "grad_norm": 1.3950402736663818, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8828330039978027, + "num_tokens": 318866641.0, + "step": 9083 + }, + { + "epoch": 1.686908077994429, + "grad_norm": 1.4023381471633911, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8769629597663879, + "num_tokens": 318904009.0, + "step": 9084 + }, + { + "epoch": 1.6870937790157847, + "grad_norm": 1.5032646656036377, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8761922717094421, + "num_tokens": 318939876.0, + "step": 9085 + }, + { + "epoch": 1.6872794800371402, + "grad_norm": 1.4681893587112427, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.873954176902771, + "num_tokens": 318973904.0, + "step": 9086 + }, + { + "epoch": 1.6874651810584957, + "grad_norm": 1.5087999105453491, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8767560720443726, + "num_tokens": 319006679.0, + "step": 9087 + }, + { + "epoch": 1.6876508820798515, + "grad_norm": 1.3274980783462524, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8881804943084717, + "num_tokens": 319049384.0, + "step": 9088 + }, + { + "epoch": 1.6878365831012072, + "grad_norm": 1.3921555280685425, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8776856660842896, + "num_tokens": 319089914.0, + "step": 9089 + }, + { + "epoch": 1.6880222841225627, + "grad_norm": 1.5111806392669678, + "learning_rate": 1e-06, + "loss": 0.3929, + "mean_token_accuracy": 0.8675142526626587, + "num_tokens": 319126651.0, + "step": 9090 + }, + { + "epoch": 1.6882079851439182, + "grad_norm": 1.402958869934082, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8847278952598572, + "num_tokens": 319164171.0, + "step": 9091 + }, + { + "epoch": 1.688393686165274, + "grad_norm": 1.5158474445343018, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8866020441055298, + "num_tokens": 319199119.0, + "step": 9092 + }, + { + "epoch": 1.6885793871866295, + "grad_norm": 1.5107333660125732, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8786295652389526, + "num_tokens": 319231878.0, + "step": 9093 + }, + { + "epoch": 1.688765088207985, + "grad_norm": 1.3190193176269531, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8907036781311035, + "num_tokens": 319271131.0, + "step": 9094 + }, + { + "epoch": 1.6889507892293407, + "grad_norm": 1.4885491132736206, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8754898309707642, + "num_tokens": 319309876.0, + "step": 9095 + }, + { + "epoch": 1.6891364902506965, + "grad_norm": 1.465329885482788, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8850125670433044, + "num_tokens": 319347238.0, + "step": 9096 + }, + { + "epoch": 1.689322191272052, + "grad_norm": 1.5418680906295776, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8847317099571228, + "num_tokens": 319383611.0, + "step": 9097 + }, + { + "epoch": 1.6895078922934075, + "grad_norm": 1.64774489402771, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8695889711380005, + "num_tokens": 319413939.0, + "step": 9098 + }, + { + "epoch": 1.6896935933147632, + "grad_norm": 1.5123136043548584, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8771221041679382, + "num_tokens": 319450293.0, + "step": 9099 + }, + { + "epoch": 1.689879294336119, + "grad_norm": 1.5226068496704102, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.87419193983078, + "num_tokens": 319485159.0, + "step": 9100 + }, + { + "epoch": 1.6900649953574745, + "grad_norm": 1.4584306478500366, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8813040256500244, + "num_tokens": 319523331.0, + "step": 9101 + }, + { + "epoch": 1.69025069637883, + "grad_norm": 1.6058744192123413, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8703460693359375, + "num_tokens": 319554902.0, + "step": 9102 + }, + { + "epoch": 1.6904363974001857, + "grad_norm": 1.681506633758545, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8800420165061951, + "num_tokens": 319585015.0, + "step": 9103 + }, + { + "epoch": 1.6906220984215414, + "grad_norm": 1.4960577487945557, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8770304918289185, + "num_tokens": 319617615.0, + "step": 9104 + }, + { + "epoch": 1.690807799442897, + "grad_norm": 1.5563774108886719, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8744170665740967, + "num_tokens": 319650201.0, + "step": 9105 + }, + { + "epoch": 1.6909935004642525, + "grad_norm": 1.706931710243225, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8955951929092407, + "num_tokens": 319675430.0, + "step": 9106 + }, + { + "epoch": 1.6911792014856082, + "grad_norm": 1.592403531074524, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8838199973106384, + "num_tokens": 319703970.0, + "step": 9107 + }, + { + "epoch": 1.691364902506964, + "grad_norm": 1.5001376867294312, + "learning_rate": 1e-06, + "loss": 0.4486, + "mean_token_accuracy": 0.8483613133430481, + "num_tokens": 319746110.0, + "step": 9108 + }, + { + "epoch": 1.6915506035283194, + "grad_norm": 1.3881474733352661, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8792049288749695, + "num_tokens": 319785041.0, + "step": 9109 + }, + { + "epoch": 1.691736304549675, + "grad_norm": 1.421013593673706, + "learning_rate": 1e-06, + "loss": 0.3887, + "mean_token_accuracy": 0.8683815002441406, + "num_tokens": 319826961.0, + "step": 9110 + }, + { + "epoch": 1.6919220055710307, + "grad_norm": 1.3498389720916748, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8837257623672485, + "num_tokens": 319869482.0, + "step": 9111 + }, + { + "epoch": 1.6921077065923864, + "grad_norm": 1.4978801012039185, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8806003332138062, + "num_tokens": 319904992.0, + "step": 9112 + }, + { + "epoch": 1.692293407613742, + "grad_norm": 1.451201319694519, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8649967908859253, + "num_tokens": 319941536.0, + "step": 9113 + }, + { + "epoch": 1.6924791086350974, + "grad_norm": 1.6820825338363647, + "learning_rate": 1e-06, + "loss": 0.4295, + "mean_token_accuracy": 0.8508113026618958, + "num_tokens": 319973726.0, + "step": 9114 + }, + { + "epoch": 1.6926648096564532, + "grad_norm": 1.5869868993759155, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8708869218826294, + "num_tokens": 320009950.0, + "step": 9115 + }, + { + "epoch": 1.6928505106778087, + "grad_norm": 1.6021920442581177, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8649247884750366, + "num_tokens": 320040724.0, + "step": 9116 + }, + { + "epoch": 1.6930362116991642, + "grad_norm": 1.594747543334961, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8891927599906921, + "num_tokens": 320076033.0, + "step": 9117 + }, + { + "epoch": 1.69322191272052, + "grad_norm": 1.5452120304107666, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8750877380371094, + "num_tokens": 320112022.0, + "step": 9118 + }, + { + "epoch": 1.6934076137418757, + "grad_norm": 1.533049464225769, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8765724897384644, + "num_tokens": 320142496.0, + "step": 9119 + }, + { + "epoch": 1.6935933147632312, + "grad_norm": 1.3642003536224365, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8839554786682129, + "num_tokens": 320180368.0, + "step": 9120 + }, + { + "epoch": 1.6937790157845867, + "grad_norm": 1.8074136972427368, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8815277814865112, + "num_tokens": 320206430.0, + "step": 9121 + }, + { + "epoch": 1.6939647168059424, + "grad_norm": 1.4076367616653442, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8830907344818115, + "num_tokens": 320243732.0, + "step": 9122 + }, + { + "epoch": 1.6941504178272981, + "grad_norm": 1.4596483707427979, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8681925535202026, + "num_tokens": 320280277.0, + "step": 9123 + }, + { + "epoch": 1.6943361188486536, + "grad_norm": 1.6262620687484741, + "learning_rate": 1e-06, + "loss": 0.4001, + "mean_token_accuracy": 0.8654100298881531, + "num_tokens": 320313697.0, + "step": 9124 + }, + { + "epoch": 1.6945218198700092, + "grad_norm": 1.4540042877197266, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8769733905792236, + "num_tokens": 320346261.0, + "step": 9125 + }, + { + "epoch": 1.694707520891365, + "grad_norm": 1.5805197954177856, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8662330508232117, + "num_tokens": 320383405.0, + "step": 9126 + }, + { + "epoch": 1.6948932219127206, + "grad_norm": 1.5904380083084106, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8788384199142456, + "num_tokens": 320414572.0, + "step": 9127 + }, + { + "epoch": 1.6950789229340761, + "grad_norm": 1.5006102323532104, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8831760883331299, + "num_tokens": 320447379.0, + "step": 9128 + }, + { + "epoch": 1.6952646239554316, + "grad_norm": 1.5436882972717285, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8770198225975037, + "num_tokens": 320485106.0, + "step": 9129 + }, + { + "epoch": 1.6954503249767874, + "grad_norm": 1.406243085861206, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8764612674713135, + "num_tokens": 320522291.0, + "step": 9130 + }, + { + "epoch": 1.6956360259981431, + "grad_norm": 1.4991735219955444, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.8607802987098694, + "num_tokens": 320560377.0, + "step": 9131 + }, + { + "epoch": 1.6958217270194986, + "grad_norm": 1.4735265970230103, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8798373937606812, + "num_tokens": 320593556.0, + "step": 9132 + }, + { + "epoch": 1.6960074280408541, + "grad_norm": 1.5522785186767578, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8717705011367798, + "num_tokens": 320625700.0, + "step": 9133 + }, + { + "epoch": 1.6961931290622099, + "grad_norm": 1.4426566362380981, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8896356821060181, + "num_tokens": 320663340.0, + "step": 9134 + }, + { + "epoch": 1.6963788300835656, + "grad_norm": 1.6052178144454956, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8740086555480957, + "num_tokens": 320698391.0, + "step": 9135 + }, + { + "epoch": 1.6965645311049211, + "grad_norm": 1.3684501647949219, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.878190279006958, + "num_tokens": 320742318.0, + "step": 9136 + }, + { + "epoch": 1.6967502321262766, + "grad_norm": 1.4061732292175293, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8726329207420349, + "num_tokens": 320781534.0, + "step": 9137 + }, + { + "epoch": 1.6969359331476324, + "grad_norm": 1.492871880531311, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8733310699462891, + "num_tokens": 320817954.0, + "step": 9138 + }, + { + "epoch": 1.697121634168988, + "grad_norm": 1.4759609699249268, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8836460709571838, + "num_tokens": 320853103.0, + "step": 9139 + }, + { + "epoch": 1.6973073351903436, + "grad_norm": 1.5672799348831177, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8788118362426758, + "num_tokens": 320885571.0, + "step": 9140 + }, + { + "epoch": 1.6974930362116991, + "grad_norm": 1.5433952808380127, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8955720067024231, + "num_tokens": 320917738.0, + "step": 9141 + }, + { + "epoch": 1.6976787372330548, + "grad_norm": 1.4250171184539795, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8821848630905151, + "num_tokens": 320953810.0, + "step": 9142 + }, + { + "epoch": 1.6978644382544104, + "grad_norm": 1.5427701473236084, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8840012550354004, + "num_tokens": 320989200.0, + "step": 9143 + }, + { + "epoch": 1.6980501392757659, + "grad_norm": 1.74136221408844, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8795384168624878, + "num_tokens": 321014511.0, + "step": 9144 + }, + { + "epoch": 1.6982358402971216, + "grad_norm": 1.5583184957504272, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8773833513259888, + "num_tokens": 321046569.0, + "step": 9145 + }, + { + "epoch": 1.6984215413184773, + "grad_norm": 1.576620101928711, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8741722106933594, + "num_tokens": 321079975.0, + "step": 9146 + }, + { + "epoch": 1.6986072423398328, + "grad_norm": 1.711828589439392, + "learning_rate": 1e-06, + "loss": 0.3993, + "mean_token_accuracy": 0.8696459531784058, + "num_tokens": 321109131.0, + "step": 9147 + }, + { + "epoch": 1.6987929433611884, + "grad_norm": 1.497975468635559, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8765223026275635, + "num_tokens": 321147602.0, + "step": 9148 + }, + { + "epoch": 1.698978644382544, + "grad_norm": 1.5372412204742432, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8793740272521973, + "num_tokens": 321183949.0, + "step": 9149 + }, + { + "epoch": 1.6991643454038998, + "grad_norm": 1.5115349292755127, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8791598677635193, + "num_tokens": 321220530.0, + "step": 9150 + }, + { + "epoch": 1.6993500464252553, + "grad_norm": 1.5330266952514648, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.869870662689209, + "num_tokens": 321257755.0, + "step": 9151 + }, + { + "epoch": 1.6995357474466108, + "grad_norm": 1.4382495880126953, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8676877021789551, + "num_tokens": 321296171.0, + "step": 9152 + }, + { + "epoch": 1.6997214484679666, + "grad_norm": 1.5711593627929688, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8687716126441956, + "num_tokens": 321328098.0, + "step": 9153 + }, + { + "epoch": 1.6999071494893223, + "grad_norm": 1.424290418624878, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8821673393249512, + "num_tokens": 321365848.0, + "step": 9154 + }, + { + "epoch": 1.7000928505106778, + "grad_norm": 1.445560097694397, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8916079998016357, + "num_tokens": 321403687.0, + "step": 9155 + }, + { + "epoch": 1.7002785515320333, + "grad_norm": 1.5065276622772217, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8811121582984924, + "num_tokens": 321438508.0, + "step": 9156 + }, + { + "epoch": 1.700464252553389, + "grad_norm": 1.4398984909057617, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8734002113342285, + "num_tokens": 321479620.0, + "step": 9157 + }, + { + "epoch": 1.7006499535747448, + "grad_norm": 1.392722487449646, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.877163290977478, + "num_tokens": 321517629.0, + "step": 9158 + }, + { + "epoch": 1.7008356545961003, + "grad_norm": 1.3514227867126465, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.885558009147644, + "num_tokens": 321560439.0, + "step": 9159 + }, + { + "epoch": 1.7010213556174558, + "grad_norm": 1.5402277708053589, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8672083616256714, + "num_tokens": 321596788.0, + "step": 9160 + }, + { + "epoch": 1.7012070566388116, + "grad_norm": 1.6930327415466309, + "learning_rate": 1e-06, + "loss": 0.3842, + "mean_token_accuracy": 0.8683645725250244, + "num_tokens": 321626821.0, + "step": 9161 + }, + { + "epoch": 1.7013927576601673, + "grad_norm": 1.5122427940368652, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8781673908233643, + "num_tokens": 321660511.0, + "step": 9162 + }, + { + "epoch": 1.7015784586815228, + "grad_norm": 1.5766257047653198, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8682591915130615, + "num_tokens": 321691638.0, + "step": 9163 + }, + { + "epoch": 1.7017641597028783, + "grad_norm": 1.577318549156189, + "learning_rate": 1e-06, + "loss": 0.4112, + "mean_token_accuracy": 0.8624585270881653, + "num_tokens": 321726149.0, + "step": 9164 + }, + { + "epoch": 1.701949860724234, + "grad_norm": 1.4481126070022583, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8724492788314819, + "num_tokens": 321761046.0, + "step": 9165 + }, + { + "epoch": 1.7021355617455896, + "grad_norm": 1.5698435306549072, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8743908405303955, + "num_tokens": 321792106.0, + "step": 9166 + }, + { + "epoch": 1.702321262766945, + "grad_norm": 1.353164792060852, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8852357864379883, + "num_tokens": 321830055.0, + "step": 9167 + }, + { + "epoch": 1.7025069637883008, + "grad_norm": 1.4929349422454834, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8722105026245117, + "num_tokens": 321866548.0, + "step": 9168 + }, + { + "epoch": 1.7026926648096565, + "grad_norm": 1.3899638652801514, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.870958149433136, + "num_tokens": 321906934.0, + "step": 9169 + }, + { + "epoch": 1.702878365831012, + "grad_norm": 1.4973636865615845, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8673893809318542, + "num_tokens": 321943889.0, + "step": 9170 + }, + { + "epoch": 1.7030640668523676, + "grad_norm": 1.5144988298416138, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8846749067306519, + "num_tokens": 321978568.0, + "step": 9171 + }, + { + "epoch": 1.7032497678737233, + "grad_norm": 1.5619487762451172, + "learning_rate": 1e-06, + "loss": 0.398, + "mean_token_accuracy": 0.8643175363540649, + "num_tokens": 322012017.0, + "step": 9172 + }, + { + "epoch": 1.703435468895079, + "grad_norm": 1.4104312658309937, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8736279010772705, + "num_tokens": 322049330.0, + "step": 9173 + }, + { + "epoch": 1.7036211699164345, + "grad_norm": 1.531930685043335, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8898840546607971, + "num_tokens": 322079610.0, + "step": 9174 + }, + { + "epoch": 1.70380687093779, + "grad_norm": 1.4900541305541992, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8806040287017822, + "num_tokens": 322111220.0, + "step": 9175 + }, + { + "epoch": 1.7039925719591458, + "grad_norm": 1.4672125577926636, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8827950954437256, + "num_tokens": 322146339.0, + "step": 9176 + }, + { + "epoch": 1.7041782729805015, + "grad_norm": 1.5603669881820679, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8708063364028931, + "num_tokens": 322178595.0, + "step": 9177 + }, + { + "epoch": 1.704363974001857, + "grad_norm": 1.4690985679626465, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.870625376701355, + "num_tokens": 322215881.0, + "step": 9178 + }, + { + "epoch": 1.7045496750232125, + "grad_norm": 1.4940788745880127, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.879814624786377, + "num_tokens": 322247316.0, + "step": 9179 + }, + { + "epoch": 1.7047353760445683, + "grad_norm": 1.5307953357696533, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.873843252658844, + "num_tokens": 322277931.0, + "step": 9180 + }, + { + "epoch": 1.704921077065924, + "grad_norm": 1.5445266962051392, + "learning_rate": 1e-06, + "loss": 0.3911, + "mean_token_accuracy": 0.8694667816162109, + "num_tokens": 322312211.0, + "step": 9181 + }, + { + "epoch": 1.7051067780872795, + "grad_norm": 1.3484212160110474, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8862352967262268, + "num_tokens": 322348813.0, + "step": 9182 + }, + { + "epoch": 1.705292479108635, + "grad_norm": 1.608436942100525, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8744953870773315, + "num_tokens": 322379925.0, + "step": 9183 + }, + { + "epoch": 1.7054781801299908, + "grad_norm": 1.553995132446289, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8757966756820679, + "num_tokens": 322412642.0, + "step": 9184 + }, + { + "epoch": 1.7056638811513465, + "grad_norm": 1.4320777654647827, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8797167539596558, + "num_tokens": 322450653.0, + "step": 9185 + }, + { + "epoch": 1.705849582172702, + "grad_norm": 1.4135640859603882, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8649060130119324, + "num_tokens": 322492005.0, + "step": 9186 + }, + { + "epoch": 1.7060352831940575, + "grad_norm": 1.474953293800354, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8783735632896423, + "num_tokens": 322524749.0, + "step": 9187 + }, + { + "epoch": 1.7062209842154132, + "grad_norm": 1.2897415161132812, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8919564485549927, + "num_tokens": 322563976.0, + "step": 9188 + }, + { + "epoch": 1.7064066852367687, + "grad_norm": 1.5659387111663818, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.873967707157135, + "num_tokens": 322597959.0, + "step": 9189 + }, + { + "epoch": 1.7065923862581243, + "grad_norm": 1.4418474435806274, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8764506578445435, + "num_tokens": 322634935.0, + "step": 9190 + }, + { + "epoch": 1.70677808727948, + "grad_norm": 1.6220183372497559, + "learning_rate": 1e-06, + "loss": 0.4018, + "mean_token_accuracy": 0.870595395565033, + "num_tokens": 322667333.0, + "step": 9191 + }, + { + "epoch": 1.7069637883008357, + "grad_norm": 1.4497599601745605, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8861728310585022, + "num_tokens": 322702066.0, + "step": 9192 + }, + { + "epoch": 1.7071494893221912, + "grad_norm": 1.5412849187850952, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8716400861740112, + "num_tokens": 322730475.0, + "step": 9193 + }, + { + "epoch": 1.7073351903435467, + "grad_norm": 1.5209120512008667, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8837698101997375, + "num_tokens": 322761732.0, + "step": 9194 + }, + { + "epoch": 1.7075208913649025, + "grad_norm": 1.4158411026000977, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8759255409240723, + "num_tokens": 322797904.0, + "step": 9195 + }, + { + "epoch": 1.7077065923862582, + "grad_norm": 1.327962875366211, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8940637707710266, + "num_tokens": 322832759.0, + "step": 9196 + }, + { + "epoch": 1.7078922934076137, + "grad_norm": 1.3831069469451904, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.875855565071106, + "num_tokens": 322870103.0, + "step": 9197 + }, + { + "epoch": 1.7080779944289692, + "grad_norm": 1.2975554466247559, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8788971304893494, + "num_tokens": 322911918.0, + "step": 9198 + }, + { + "epoch": 1.708263695450325, + "grad_norm": 1.395556926727295, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8696870803833008, + "num_tokens": 322953065.0, + "step": 9199 + }, + { + "epoch": 1.7084493964716807, + "grad_norm": 1.3941222429275513, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8819125890731812, + "num_tokens": 322993807.0, + "step": 9200 + }, + { + "epoch": 1.7086350974930362, + "grad_norm": 1.440078616142273, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.885418176651001, + "num_tokens": 323029182.0, + "step": 9201 + }, + { + "epoch": 1.7088207985143917, + "grad_norm": 1.571014165878296, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8712928295135498, + "num_tokens": 323064383.0, + "step": 9202 + }, + { + "epoch": 1.7090064995357475, + "grad_norm": 1.4954944849014282, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8896503448486328, + "num_tokens": 323098171.0, + "step": 9203 + }, + { + "epoch": 1.7091922005571032, + "grad_norm": 1.6373064517974854, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8736072778701782, + "num_tokens": 323128435.0, + "step": 9204 + }, + { + "epoch": 1.7093779015784587, + "grad_norm": 1.4522485733032227, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.884392261505127, + "num_tokens": 323161439.0, + "step": 9205 + }, + { + "epoch": 1.7095636025998142, + "grad_norm": 1.4284957647323608, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8777850866317749, + "num_tokens": 323201191.0, + "step": 9206 + }, + { + "epoch": 1.70974930362117, + "grad_norm": 1.6083903312683105, + "learning_rate": 1e-06, + "loss": 0.4264, + "mean_token_accuracy": 0.8584646582603455, + "num_tokens": 323232550.0, + "step": 9207 + }, + { + "epoch": 1.7099350046425257, + "grad_norm": 1.6186740398406982, + "learning_rate": 1e-06, + "loss": 0.4303, + "mean_token_accuracy": 0.8540225625038147, + "num_tokens": 323273536.0, + "step": 9208 + }, + { + "epoch": 1.7101207056638812, + "grad_norm": 1.5061763525009155, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8708674907684326, + "num_tokens": 323310031.0, + "step": 9209 + }, + { + "epoch": 1.7103064066852367, + "grad_norm": 1.420382022857666, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8912608623504639, + "num_tokens": 323345110.0, + "step": 9210 + }, + { + "epoch": 1.7104921077065924, + "grad_norm": 1.469096064567566, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8701832890510559, + "num_tokens": 323380642.0, + "step": 9211 + }, + { + "epoch": 1.7106778087279482, + "grad_norm": 1.5617834329605103, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8690230846405029, + "num_tokens": 323412566.0, + "step": 9212 + }, + { + "epoch": 1.7108635097493035, + "grad_norm": 1.40988028049469, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8945879340171814, + "num_tokens": 323444854.0, + "step": 9213 + }, + { + "epoch": 1.7110492107706592, + "grad_norm": 1.3691902160644531, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8853038549423218, + "num_tokens": 323482589.0, + "step": 9214 + }, + { + "epoch": 1.711234911792015, + "grad_norm": 1.4181703329086304, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8670412302017212, + "num_tokens": 323521799.0, + "step": 9215 + }, + { + "epoch": 1.7114206128133704, + "grad_norm": 1.5606039762496948, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8684631586074829, + "num_tokens": 323555331.0, + "step": 9216 + }, + { + "epoch": 1.711606313834726, + "grad_norm": 1.703293800354004, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8914209008216858, + "num_tokens": 323579825.0, + "step": 9217 + }, + { + "epoch": 1.7117920148560817, + "grad_norm": 1.563876748085022, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8713903427124023, + "num_tokens": 323616504.0, + "step": 9218 + }, + { + "epoch": 1.7119777158774374, + "grad_norm": 1.565605640411377, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8666360378265381, + "num_tokens": 323652463.0, + "step": 9219 + }, + { + "epoch": 1.712163416898793, + "grad_norm": 1.419018030166626, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8723050355911255, + "num_tokens": 323692252.0, + "step": 9220 + }, + { + "epoch": 1.7123491179201484, + "grad_norm": 1.4618959426879883, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8618927001953125, + "num_tokens": 323732983.0, + "step": 9221 + }, + { + "epoch": 1.7125348189415042, + "grad_norm": 1.4168462753295898, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8784031271934509, + "num_tokens": 323770082.0, + "step": 9222 + }, + { + "epoch": 1.71272051996286, + "grad_norm": 1.679870367050171, + "learning_rate": 1e-06, + "loss": 0.4222, + "mean_token_accuracy": 0.8619539737701416, + "num_tokens": 323798959.0, + "step": 9223 + }, + { + "epoch": 1.7129062209842154, + "grad_norm": 1.5203086137771606, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8785848617553711, + "num_tokens": 323834810.0, + "step": 9224 + }, + { + "epoch": 1.713091922005571, + "grad_norm": 1.6044046878814697, + "learning_rate": 1e-06, + "loss": 0.3891, + "mean_token_accuracy": 0.8658021092414856, + "num_tokens": 323871234.0, + "step": 9225 + }, + { + "epoch": 1.7132776230269267, + "grad_norm": 1.5409965515136719, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8759112358093262, + "num_tokens": 323906690.0, + "step": 9226 + }, + { + "epoch": 1.7134633240482824, + "grad_norm": 1.5626827478408813, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8707010746002197, + "num_tokens": 323938390.0, + "step": 9227 + }, + { + "epoch": 1.713649025069638, + "grad_norm": 1.584571361541748, + "learning_rate": 1e-06, + "loss": 0.4459, + "mean_token_accuracy": 0.8531196117401123, + "num_tokens": 323974057.0, + "step": 9228 + }, + { + "epoch": 1.7138347260909934, + "grad_norm": 1.534100890159607, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8802391290664673, + "num_tokens": 324005233.0, + "step": 9229 + }, + { + "epoch": 1.7140204271123491, + "grad_norm": 1.3943465948104858, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8818599581718445, + "num_tokens": 324044723.0, + "step": 9230 + }, + { + "epoch": 1.7142061281337049, + "grad_norm": 1.5010390281677246, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8720343112945557, + "num_tokens": 324079427.0, + "step": 9231 + }, + { + "epoch": 1.7143918291550604, + "grad_norm": 1.696366786956787, + "learning_rate": 1e-06, + "loss": 0.4797, + "mean_token_accuracy": 0.8416969776153564, + "num_tokens": 324111021.0, + "step": 9232 + }, + { + "epoch": 1.714577530176416, + "grad_norm": 1.5827021598815918, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8846448659896851, + "num_tokens": 324141963.0, + "step": 9233 + }, + { + "epoch": 1.7147632311977716, + "grad_norm": 1.587497591972351, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.876757025718689, + "num_tokens": 324174341.0, + "step": 9234 + }, + { + "epoch": 1.7149489322191274, + "grad_norm": 1.4735745191574097, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8812907934188843, + "num_tokens": 324206438.0, + "step": 9235 + }, + { + "epoch": 1.7151346332404829, + "grad_norm": 1.4887309074401855, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8796853423118591, + "num_tokens": 324238874.0, + "step": 9236 + }, + { + "epoch": 1.7153203342618384, + "grad_norm": 1.3996556997299194, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8823820352554321, + "num_tokens": 324276035.0, + "step": 9237 + }, + { + "epoch": 1.7155060352831941, + "grad_norm": 1.5013267993927002, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8873707056045532, + "num_tokens": 324312930.0, + "step": 9238 + }, + { + "epoch": 1.7156917363045496, + "grad_norm": 1.5464197397232056, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8736592531204224, + "num_tokens": 324346546.0, + "step": 9239 + }, + { + "epoch": 1.7158774373259051, + "grad_norm": 1.5189015865325928, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.865924596786499, + "num_tokens": 324385938.0, + "step": 9240 + }, + { + "epoch": 1.7160631383472609, + "grad_norm": 1.5842571258544922, + "learning_rate": 1e-06, + "loss": 0.4151, + "mean_token_accuracy": 0.8604347705841064, + "num_tokens": 324424539.0, + "step": 9241 + }, + { + "epoch": 1.7162488393686166, + "grad_norm": 1.433048129081726, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8728190660476685, + "num_tokens": 324463573.0, + "step": 9242 + }, + { + "epoch": 1.7164345403899721, + "grad_norm": 1.5205844640731812, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8696833848953247, + "num_tokens": 324498570.0, + "step": 9243 + }, + { + "epoch": 1.7166202414113276, + "grad_norm": 1.5216519832611084, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8846869468688965, + "num_tokens": 324532238.0, + "step": 9244 + }, + { + "epoch": 1.7168059424326834, + "grad_norm": 1.4734158515930176, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8807019591331482, + "num_tokens": 324568942.0, + "step": 9245 + }, + { + "epoch": 1.716991643454039, + "grad_norm": 1.3303744792938232, + "learning_rate": 1e-06, + "loss": 0.2942, + "mean_token_accuracy": 0.8952926397323608, + "num_tokens": 324607876.0, + "step": 9246 + }, + { + "epoch": 1.7171773444753946, + "grad_norm": 1.4774349927902222, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8793794512748718, + "num_tokens": 324642632.0, + "step": 9247 + }, + { + "epoch": 1.7173630454967501, + "grad_norm": 1.6633692979812622, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.8754199147224426, + "num_tokens": 324669910.0, + "step": 9248 + }, + { + "epoch": 1.7175487465181059, + "grad_norm": 1.487331748008728, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8864444494247437, + "num_tokens": 324704524.0, + "step": 9249 + }, + { + "epoch": 1.7177344475394616, + "grad_norm": 1.6195769309997559, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8862822651863098, + "num_tokens": 324731207.0, + "step": 9250 + }, + { + "epoch": 1.717920148560817, + "grad_norm": 1.4665319919586182, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8729099631309509, + "num_tokens": 324768172.0, + "step": 9251 + }, + { + "epoch": 1.7181058495821726, + "grad_norm": 1.5206475257873535, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8788326978683472, + "num_tokens": 324807760.0, + "step": 9252 + }, + { + "epoch": 1.7182915506035283, + "grad_norm": 1.4192882776260376, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8674089312553406, + "num_tokens": 324846654.0, + "step": 9253 + }, + { + "epoch": 1.718477251624884, + "grad_norm": 1.3717691898345947, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8839622735977173, + "num_tokens": 324886707.0, + "step": 9254 + }, + { + "epoch": 1.7186629526462396, + "grad_norm": 1.5183457136154175, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8814256191253662, + "num_tokens": 324918157.0, + "step": 9255 + }, + { + "epoch": 1.718848653667595, + "grad_norm": 1.4496731758117676, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8717674016952515, + "num_tokens": 324956629.0, + "step": 9256 + }, + { + "epoch": 1.7190343546889508, + "grad_norm": 1.426418662071228, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8768907189369202, + "num_tokens": 324996600.0, + "step": 9257 + }, + { + "epoch": 1.7192200557103066, + "grad_norm": 1.4139589071273804, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8898577094078064, + "num_tokens": 325035300.0, + "step": 9258 + }, + { + "epoch": 1.719405756731662, + "grad_norm": 1.3695296049118042, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8785273432731628, + "num_tokens": 325074481.0, + "step": 9259 + }, + { + "epoch": 1.7195914577530176, + "grad_norm": 1.5171455144882202, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8743731379508972, + "num_tokens": 325106249.0, + "step": 9260 + }, + { + "epoch": 1.7197771587743733, + "grad_norm": 1.4423072338104248, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8825628161430359, + "num_tokens": 325143743.0, + "step": 9261 + }, + { + "epoch": 1.7199628597957288, + "grad_norm": 1.5586614608764648, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8812978267669678, + "num_tokens": 325176235.0, + "step": 9262 + }, + { + "epoch": 1.7201485608170843, + "grad_norm": 1.44268798828125, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.892156720161438, + "num_tokens": 325209916.0, + "step": 9263 + }, + { + "epoch": 1.72033426183844, + "grad_norm": 1.4467194080352783, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8932551145553589, + "num_tokens": 325246096.0, + "step": 9264 + }, + { + "epoch": 1.7205199628597958, + "grad_norm": 1.4504270553588867, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8668365478515625, + "num_tokens": 325287323.0, + "step": 9265 + }, + { + "epoch": 1.7207056638811513, + "grad_norm": 1.4427002668380737, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8783880472183228, + "num_tokens": 325325144.0, + "step": 9266 + }, + { + "epoch": 1.7208913649025068, + "grad_norm": 1.5014028549194336, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8716478943824768, + "num_tokens": 325361672.0, + "step": 9267 + }, + { + "epoch": 1.7210770659238626, + "grad_norm": 1.715276837348938, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8929636478424072, + "num_tokens": 325385860.0, + "step": 9268 + }, + { + "epoch": 1.7212627669452183, + "grad_norm": 1.5768861770629883, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.8728477954864502, + "num_tokens": 325416321.0, + "step": 9269 + }, + { + "epoch": 1.7214484679665738, + "grad_norm": 1.4346526861190796, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8792722225189209, + "num_tokens": 325450787.0, + "step": 9270 + }, + { + "epoch": 1.7216341689879293, + "grad_norm": 1.4822384119033813, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8766764402389526, + "num_tokens": 325484876.0, + "step": 9271 + }, + { + "epoch": 1.721819870009285, + "grad_norm": 1.4267640113830566, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8759540319442749, + "num_tokens": 325524955.0, + "step": 9272 + }, + { + "epoch": 1.7220055710306408, + "grad_norm": 1.4788830280303955, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8836231231689453, + "num_tokens": 325559385.0, + "step": 9273 + }, + { + "epoch": 1.7221912720519963, + "grad_norm": 1.364314079284668, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8818932771682739, + "num_tokens": 325599787.0, + "step": 9274 + }, + { + "epoch": 1.7223769730733518, + "grad_norm": 1.671766996383667, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8825770020484924, + "num_tokens": 325633406.0, + "step": 9275 + }, + { + "epoch": 1.7225626740947075, + "grad_norm": 1.3574308156967163, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8865587115287781, + "num_tokens": 325671189.0, + "step": 9276 + }, + { + "epoch": 1.7227483751160633, + "grad_norm": 1.4569408893585205, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8856084942817688, + "num_tokens": 325706395.0, + "step": 9277 + }, + { + "epoch": 1.7229340761374188, + "grad_norm": 1.5008383989334106, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8829900026321411, + "num_tokens": 325741787.0, + "step": 9278 + }, + { + "epoch": 1.7231197771587743, + "grad_norm": 1.4804133176803589, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8907884359359741, + "num_tokens": 325778060.0, + "step": 9279 + }, + { + "epoch": 1.72330547818013, + "grad_norm": 1.4613628387451172, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8760901689529419, + "num_tokens": 325814816.0, + "step": 9280 + }, + { + "epoch": 1.7234911792014858, + "grad_norm": 1.424637794494629, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8716073036193848, + "num_tokens": 325855357.0, + "step": 9281 + }, + { + "epoch": 1.7236768802228413, + "grad_norm": 1.4964743852615356, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8639802932739258, + "num_tokens": 325893802.0, + "step": 9282 + }, + { + "epoch": 1.7238625812441968, + "grad_norm": 1.4670941829681396, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8803523778915405, + "num_tokens": 325928310.0, + "step": 9283 + }, + { + "epoch": 1.7240482822655525, + "grad_norm": 1.4095536470413208, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.8685389757156372, + "num_tokens": 325968547.0, + "step": 9284 + }, + { + "epoch": 1.724233983286908, + "grad_norm": 1.4763695001602173, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8755652904510498, + "num_tokens": 326005011.0, + "step": 9285 + }, + { + "epoch": 1.7244196843082635, + "grad_norm": 1.382331371307373, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8834021687507629, + "num_tokens": 326042341.0, + "step": 9286 + }, + { + "epoch": 1.7246053853296193, + "grad_norm": 1.4134243726730347, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8917331695556641, + "num_tokens": 326077188.0, + "step": 9287 + }, + { + "epoch": 1.724791086350975, + "grad_norm": 1.5690449476242065, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8769388794898987, + "num_tokens": 326108214.0, + "step": 9288 + }, + { + "epoch": 1.7249767873723305, + "grad_norm": 1.338325023651123, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8843488693237305, + "num_tokens": 326149567.0, + "step": 9289 + }, + { + "epoch": 1.725162488393686, + "grad_norm": 1.4793144464492798, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.862962543964386, + "num_tokens": 326188550.0, + "step": 9290 + }, + { + "epoch": 1.7253481894150418, + "grad_norm": 1.5326497554779053, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8805718421936035, + "num_tokens": 326221239.0, + "step": 9291 + }, + { + "epoch": 1.7255338904363975, + "grad_norm": 1.5501277446746826, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.8985276818275452, + "num_tokens": 326249302.0, + "step": 9292 + }, + { + "epoch": 1.725719591457753, + "grad_norm": 1.5406105518341064, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8905783891677856, + "num_tokens": 326280157.0, + "step": 9293 + }, + { + "epoch": 1.7259052924791085, + "grad_norm": 1.4735052585601807, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8887434005737305, + "num_tokens": 326313280.0, + "step": 9294 + }, + { + "epoch": 1.7260909935004642, + "grad_norm": 1.4978454113006592, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8700240850448608, + "num_tokens": 326353246.0, + "step": 9295 + }, + { + "epoch": 1.72627669452182, + "grad_norm": 1.6129796504974365, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.870507001876831, + "num_tokens": 326386785.0, + "step": 9296 + }, + { + "epoch": 1.7264623955431755, + "grad_norm": 1.4079607725143433, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8870825171470642, + "num_tokens": 326419151.0, + "step": 9297 + }, + { + "epoch": 1.726648096564531, + "grad_norm": 1.497596263885498, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8809713125228882, + "num_tokens": 326455138.0, + "step": 9298 + }, + { + "epoch": 1.7268337975858867, + "grad_norm": 1.473791241645813, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8771178126335144, + "num_tokens": 326494304.0, + "step": 9299 + }, + { + "epoch": 1.7270194986072425, + "grad_norm": 1.4571852684020996, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8805030584335327, + "num_tokens": 326531248.0, + "step": 9300 + }, + { + "epoch": 1.727205199628598, + "grad_norm": 1.5866762399673462, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8775771856307983, + "num_tokens": 326558618.0, + "step": 9301 + }, + { + "epoch": 1.7273909006499535, + "grad_norm": 1.5518840551376343, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8806279897689819, + "num_tokens": 326589134.0, + "step": 9302 + }, + { + "epoch": 1.7275766016713092, + "grad_norm": 1.386342167854309, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8817316293716431, + "num_tokens": 326626248.0, + "step": 9303 + }, + { + "epoch": 1.727762302692665, + "grad_norm": 1.4774419069290161, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8812986612319946, + "num_tokens": 326659955.0, + "step": 9304 + }, + { + "epoch": 1.7279480037140205, + "grad_norm": 1.5283350944519043, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8759761452674866, + "num_tokens": 326692626.0, + "step": 9305 + }, + { + "epoch": 1.728133704735376, + "grad_norm": 1.5777018070220947, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8802458047866821, + "num_tokens": 326722347.0, + "step": 9306 + }, + { + "epoch": 1.7283194057567317, + "grad_norm": 1.363656997680664, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8920872807502747, + "num_tokens": 326762231.0, + "step": 9307 + }, + { + "epoch": 1.7285051067780874, + "grad_norm": 1.4292813539505005, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8915042877197266, + "num_tokens": 326792687.0, + "step": 9308 + }, + { + "epoch": 1.728690807799443, + "grad_norm": 1.5081582069396973, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8918370604515076, + "num_tokens": 326829233.0, + "step": 9309 + }, + { + "epoch": 1.7288765088207985, + "grad_norm": 1.5949102640151978, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8786715865135193, + "num_tokens": 326860598.0, + "step": 9310 + }, + { + "epoch": 1.7290622098421542, + "grad_norm": 1.5364919900894165, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8708963394165039, + "num_tokens": 326894861.0, + "step": 9311 + }, + { + "epoch": 1.7292479108635097, + "grad_norm": 1.5049842596054077, + "learning_rate": 1e-06, + "loss": 0.4249, + "mean_token_accuracy": 0.8580890893936157, + "num_tokens": 326932804.0, + "step": 9312 + }, + { + "epoch": 1.7294336118848652, + "grad_norm": 1.4636509418487549, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8715031147003174, + "num_tokens": 326972472.0, + "step": 9313 + }, + { + "epoch": 1.729619312906221, + "grad_norm": 1.495318055152893, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8802715539932251, + "num_tokens": 327010079.0, + "step": 9314 + }, + { + "epoch": 1.7298050139275767, + "grad_norm": 1.579371452331543, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8790080547332764, + "num_tokens": 327043932.0, + "step": 9315 + }, + { + "epoch": 1.7299907149489322, + "grad_norm": 1.3941357135772705, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8780842423439026, + "num_tokens": 327084273.0, + "step": 9316 + }, + { + "epoch": 1.7301764159702877, + "grad_norm": 1.5901055335998535, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8890341520309448, + "num_tokens": 327112071.0, + "step": 9317 + }, + { + "epoch": 1.7303621169916434, + "grad_norm": 1.4956271648406982, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8786900043487549, + "num_tokens": 327144835.0, + "step": 9318 + }, + { + "epoch": 1.7305478180129992, + "grad_norm": 1.4834308624267578, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8769872188568115, + "num_tokens": 327180598.0, + "step": 9319 + }, + { + "epoch": 1.7307335190343547, + "grad_norm": 1.5740845203399658, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8752261400222778, + "num_tokens": 327212721.0, + "step": 9320 + }, + { + "epoch": 1.7309192200557102, + "grad_norm": 1.6349670886993408, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.8670311570167542, + "num_tokens": 327243731.0, + "step": 9321 + }, + { + "epoch": 1.731104921077066, + "grad_norm": 1.5140596628189087, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8809084892272949, + "num_tokens": 327276371.0, + "step": 9322 + }, + { + "epoch": 1.7312906220984217, + "grad_norm": 1.480149745941162, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8778412938117981, + "num_tokens": 327309586.0, + "step": 9323 + }, + { + "epoch": 1.7314763231197772, + "grad_norm": 1.6441502571105957, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8705376982688904, + "num_tokens": 327339752.0, + "step": 9324 + }, + { + "epoch": 1.7316620241411327, + "grad_norm": 1.3713074922561646, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8766241669654846, + "num_tokens": 327380052.0, + "step": 9325 + }, + { + "epoch": 1.7318477251624884, + "grad_norm": 1.5214115381240845, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8901028633117676, + "num_tokens": 327412889.0, + "step": 9326 + }, + { + "epoch": 1.7320334261838441, + "grad_norm": 1.6061882972717285, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.861531674861908, + "num_tokens": 327445164.0, + "step": 9327 + }, + { + "epoch": 1.7322191272051997, + "grad_norm": 1.532863736152649, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8825111985206604, + "num_tokens": 327479150.0, + "step": 9328 + }, + { + "epoch": 1.7324048282265552, + "grad_norm": 1.601087212562561, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8758355975151062, + "num_tokens": 327508337.0, + "step": 9329 + }, + { + "epoch": 1.732590529247911, + "grad_norm": 1.5796632766723633, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8953248858451843, + "num_tokens": 327540354.0, + "step": 9330 + }, + { + "epoch": 1.7327762302692666, + "grad_norm": 1.4456547498703003, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8845564126968384, + "num_tokens": 327575820.0, + "step": 9331 + }, + { + "epoch": 1.7329619312906221, + "grad_norm": 1.5563783645629883, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8683691620826721, + "num_tokens": 327609305.0, + "step": 9332 + }, + { + "epoch": 1.7331476323119777, + "grad_norm": 1.4223142862319946, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8962265253067017, + "num_tokens": 327643918.0, + "step": 9333 + }, + { + "epoch": 1.7333333333333334, + "grad_norm": 1.4071885347366333, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8853176832199097, + "num_tokens": 327679356.0, + "step": 9334 + }, + { + "epoch": 1.733519034354689, + "grad_norm": 1.5647019147872925, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8734453916549683, + "num_tokens": 327712332.0, + "step": 9335 + }, + { + "epoch": 1.7337047353760444, + "grad_norm": 1.4596149921417236, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8705549240112305, + "num_tokens": 327752305.0, + "step": 9336 + }, + { + "epoch": 1.7338904363974001, + "grad_norm": 1.4348115921020508, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8816186785697937, + "num_tokens": 327790737.0, + "step": 9337 + }, + { + "epoch": 1.7340761374187559, + "grad_norm": 1.4492430686950684, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.885145902633667, + "num_tokens": 327825446.0, + "step": 9338 + }, + { + "epoch": 1.7342618384401114, + "grad_norm": 1.459707260131836, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8680123686790466, + "num_tokens": 327865362.0, + "step": 9339 + }, + { + "epoch": 1.734447539461467, + "grad_norm": 1.4428807497024536, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.889374315738678, + "num_tokens": 327902292.0, + "step": 9340 + }, + { + "epoch": 1.7346332404828226, + "grad_norm": 1.3634999990463257, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.886489987373352, + "num_tokens": 327939707.0, + "step": 9341 + }, + { + "epoch": 1.7348189415041784, + "grad_norm": 1.482651948928833, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8765559792518616, + "num_tokens": 327974376.0, + "step": 9342 + }, + { + "epoch": 1.7350046425255339, + "grad_norm": 1.500563144683838, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8930890560150146, + "num_tokens": 328006455.0, + "step": 9343 + }, + { + "epoch": 1.7351903435468894, + "grad_norm": 1.4349894523620605, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8744685649871826, + "num_tokens": 328044470.0, + "step": 9344 + }, + { + "epoch": 1.7353760445682451, + "grad_norm": 1.5207895040512085, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8737972378730774, + "num_tokens": 328077053.0, + "step": 9345 + }, + { + "epoch": 1.7355617455896009, + "grad_norm": 1.501861572265625, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8884255886077881, + "num_tokens": 328108467.0, + "step": 9346 + }, + { + "epoch": 1.7357474466109564, + "grad_norm": 1.4785922765731812, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8758108019828796, + "num_tokens": 328145487.0, + "step": 9347 + }, + { + "epoch": 1.7359331476323119, + "grad_norm": 1.342721939086914, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8731729984283447, + "num_tokens": 328188696.0, + "step": 9348 + }, + { + "epoch": 1.7361188486536676, + "grad_norm": 1.4938246011734009, + "learning_rate": 1e-06, + "loss": 0.4259, + "mean_token_accuracy": 0.8550817966461182, + "num_tokens": 328226498.0, + "step": 9349 + }, + { + "epoch": 1.7363045496750233, + "grad_norm": 1.5375491380691528, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8850101232528687, + "num_tokens": 328259096.0, + "step": 9350 + }, + { + "epoch": 1.7364902506963789, + "grad_norm": 1.483258605003357, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8777925968170166, + "num_tokens": 328297752.0, + "step": 9351 + }, + { + "epoch": 1.7366759517177344, + "grad_norm": 1.4458231925964355, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8772444725036621, + "num_tokens": 328335705.0, + "step": 9352 + }, + { + "epoch": 1.73686165273909, + "grad_norm": 1.5831588506698608, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8719442486763, + "num_tokens": 328368841.0, + "step": 9353 + }, + { + "epoch": 1.7370473537604458, + "grad_norm": 1.6860190629959106, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.871482253074646, + "num_tokens": 328395568.0, + "step": 9354 + }, + { + "epoch": 1.7372330547818013, + "grad_norm": 1.4321670532226562, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8792768716812134, + "num_tokens": 328431690.0, + "step": 9355 + }, + { + "epoch": 1.7374187558031569, + "grad_norm": 1.4769673347473145, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8804435729980469, + "num_tokens": 328468530.0, + "step": 9356 + }, + { + "epoch": 1.7376044568245126, + "grad_norm": 1.3932158946990967, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.8776710033416748, + "num_tokens": 328510492.0, + "step": 9357 + }, + { + "epoch": 1.737790157845868, + "grad_norm": 1.7936384677886963, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8656638860702515, + "num_tokens": 328540471.0, + "step": 9358 + }, + { + "epoch": 1.7379758588672236, + "grad_norm": 1.406620979309082, + "learning_rate": 1e-06, + "loss": 0.3914, + "mean_token_accuracy": 0.8648349642753601, + "num_tokens": 328580374.0, + "step": 9359 + }, + { + "epoch": 1.7381615598885793, + "grad_norm": 1.3344472646713257, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.886864423751831, + "num_tokens": 328621283.0, + "step": 9360 + }, + { + "epoch": 1.738347260909935, + "grad_norm": 1.426645040512085, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8896574974060059, + "num_tokens": 328655668.0, + "step": 9361 + }, + { + "epoch": 1.7385329619312906, + "grad_norm": 1.3304692506790161, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.885805606842041, + "num_tokens": 328696357.0, + "step": 9362 + }, + { + "epoch": 1.738718662952646, + "grad_norm": 1.3929908275604248, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8746461868286133, + "num_tokens": 328737082.0, + "step": 9363 + }, + { + "epoch": 1.7389043639740018, + "grad_norm": 1.408086895942688, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8723299503326416, + "num_tokens": 328775831.0, + "step": 9364 + }, + { + "epoch": 1.7390900649953576, + "grad_norm": 1.3836328983306885, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8896297216415405, + "num_tokens": 328813478.0, + "step": 9365 + }, + { + "epoch": 1.739275766016713, + "grad_norm": 1.4686853885650635, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8732675313949585, + "num_tokens": 328850471.0, + "step": 9366 + }, + { + "epoch": 1.7394614670380686, + "grad_norm": 1.4865998029708862, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8823966979980469, + "num_tokens": 328884476.0, + "step": 9367 + }, + { + "epoch": 1.7396471680594243, + "grad_norm": 1.3780148029327393, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8848832845687866, + "num_tokens": 328920843.0, + "step": 9368 + }, + { + "epoch": 1.73983286908078, + "grad_norm": 1.506956934928894, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8729989528656006, + "num_tokens": 328958288.0, + "step": 9369 + }, + { + "epoch": 1.7400185701021356, + "grad_norm": 1.3769110441207886, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8808095455169678, + "num_tokens": 328997989.0, + "step": 9370 + }, + { + "epoch": 1.740204271123491, + "grad_norm": 1.481145977973938, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8750041723251343, + "num_tokens": 329033716.0, + "step": 9371 + }, + { + "epoch": 1.7403899721448468, + "grad_norm": 1.5694185495376587, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8786118030548096, + "num_tokens": 329065648.0, + "step": 9372 + }, + { + "epoch": 1.7405756731662025, + "grad_norm": 1.4679007530212402, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8836036920547485, + "num_tokens": 329098994.0, + "step": 9373 + }, + { + "epoch": 1.740761374187558, + "grad_norm": 1.4634569883346558, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8759656548500061, + "num_tokens": 329136491.0, + "step": 9374 + }, + { + "epoch": 1.7409470752089136, + "grad_norm": 1.4025641679763794, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8733576536178589, + "num_tokens": 329176833.0, + "step": 9375 + }, + { + "epoch": 1.7411327762302693, + "grad_norm": 1.6090567111968994, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.871513843536377, + "num_tokens": 329208807.0, + "step": 9376 + }, + { + "epoch": 1.741318477251625, + "grad_norm": 1.4724071025848389, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8697546720504761, + "num_tokens": 329246570.0, + "step": 9377 + }, + { + "epoch": 1.7415041782729805, + "grad_norm": 1.4899630546569824, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8750293254852295, + "num_tokens": 329286544.0, + "step": 9378 + }, + { + "epoch": 1.741689879294336, + "grad_norm": 1.601714015007019, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8863489627838135, + "num_tokens": 329317802.0, + "step": 9379 + }, + { + "epoch": 1.7418755803156918, + "grad_norm": 1.5155513286590576, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8783310651779175, + "num_tokens": 329350734.0, + "step": 9380 + }, + { + "epoch": 1.7420612813370475, + "grad_norm": 1.520936369895935, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8774890303611755, + "num_tokens": 329383619.0, + "step": 9381 + }, + { + "epoch": 1.7422469823584028, + "grad_norm": 1.46277916431427, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8657487630844116, + "num_tokens": 329422099.0, + "step": 9382 + }, + { + "epoch": 1.7424326833797585, + "grad_norm": 1.6366691589355469, + "learning_rate": 1e-06, + "loss": 0.4041, + "mean_token_accuracy": 0.8639167547225952, + "num_tokens": 329454180.0, + "step": 9383 + }, + { + "epoch": 1.7426183844011143, + "grad_norm": 1.4460278749465942, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8735345602035522, + "num_tokens": 329492726.0, + "step": 9384 + }, + { + "epoch": 1.7428040854224698, + "grad_norm": 1.5923781394958496, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8775520324707031, + "num_tokens": 329522082.0, + "step": 9385 + }, + { + "epoch": 1.7429897864438253, + "grad_norm": 1.4260388612747192, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8675824999809265, + "num_tokens": 329558668.0, + "step": 9386 + }, + { + "epoch": 1.743175487465181, + "grad_norm": 1.4501858949661255, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.884551465511322, + "num_tokens": 329594757.0, + "step": 9387 + }, + { + "epoch": 1.7433611884865368, + "grad_norm": 1.5476443767547607, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8639490604400635, + "num_tokens": 329628777.0, + "step": 9388 + }, + { + "epoch": 1.7435468895078923, + "grad_norm": 1.3157795667648315, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.871321439743042, + "num_tokens": 329674969.0, + "step": 9389 + }, + { + "epoch": 1.7437325905292478, + "grad_norm": 1.7726296186447144, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8694276809692383, + "num_tokens": 329700503.0, + "step": 9390 + }, + { + "epoch": 1.7439182915506035, + "grad_norm": 1.561362385749817, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8763606548309326, + "num_tokens": 329730385.0, + "step": 9391 + }, + { + "epoch": 1.7441039925719592, + "grad_norm": 1.4716999530792236, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8775738477706909, + "num_tokens": 329764369.0, + "step": 9392 + }, + { + "epoch": 1.7442896935933148, + "grad_norm": 1.6213905811309814, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8713594079017639, + "num_tokens": 329797551.0, + "step": 9393 + }, + { + "epoch": 1.7444753946146703, + "grad_norm": 1.5421783924102783, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8770429491996765, + "num_tokens": 329831147.0, + "step": 9394 + }, + { + "epoch": 1.744661095636026, + "grad_norm": 1.485251784324646, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8730722665786743, + "num_tokens": 329870056.0, + "step": 9395 + }, + { + "epoch": 1.7448467966573817, + "grad_norm": 1.5043566226959229, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.87736976146698, + "num_tokens": 329904295.0, + "step": 9396 + }, + { + "epoch": 1.7450324976787372, + "grad_norm": 1.5808292627334595, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8753608465194702, + "num_tokens": 329941083.0, + "step": 9397 + }, + { + "epoch": 1.7452181987000928, + "grad_norm": 1.5932869911193848, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8714048862457275, + "num_tokens": 329974267.0, + "step": 9398 + }, + { + "epoch": 1.7454038997214485, + "grad_norm": 1.4708051681518555, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8767998218536377, + "num_tokens": 330011243.0, + "step": 9399 + }, + { + "epoch": 1.7455896007428042, + "grad_norm": 1.4895014762878418, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8817625045776367, + "num_tokens": 330049624.0, + "step": 9400 + }, + { + "epoch": 1.7457753017641597, + "grad_norm": 1.5127453804016113, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8864389657974243, + "num_tokens": 330083968.0, + "step": 9401 + }, + { + "epoch": 1.7459610027855152, + "grad_norm": 1.5686964988708496, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8745770454406738, + "num_tokens": 330117495.0, + "step": 9402 + }, + { + "epoch": 1.746146703806871, + "grad_norm": 1.607538104057312, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8769514560699463, + "num_tokens": 330148532.0, + "step": 9403 + }, + { + "epoch": 1.7463324048282267, + "grad_norm": 1.5473753213882446, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8738908171653748, + "num_tokens": 330181192.0, + "step": 9404 + }, + { + "epoch": 1.7465181058495822, + "grad_norm": 1.414172887802124, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8827629089355469, + "num_tokens": 330216556.0, + "step": 9405 + }, + { + "epoch": 1.7467038068709377, + "grad_norm": 1.4055060148239136, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8881493806838989, + "num_tokens": 330252509.0, + "step": 9406 + }, + { + "epoch": 1.7468895078922935, + "grad_norm": 1.7178435325622559, + "learning_rate": 1e-06, + "loss": 0.4069, + "mean_token_accuracy": 0.8653359413146973, + "num_tokens": 330280843.0, + "step": 9407 + }, + { + "epoch": 1.747075208913649, + "grad_norm": 1.504550576210022, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8849141597747803, + "num_tokens": 330314945.0, + "step": 9408 + }, + { + "epoch": 1.7472609099350045, + "grad_norm": 1.4889723062515259, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8774576783180237, + "num_tokens": 330350265.0, + "step": 9409 + }, + { + "epoch": 1.7474466109563602, + "grad_norm": 1.5548309087753296, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8717051148414612, + "num_tokens": 330382956.0, + "step": 9410 + }, + { + "epoch": 1.747632311977716, + "grad_norm": 1.3942639827728271, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.8957173824310303, + "num_tokens": 330418843.0, + "step": 9411 + }, + { + "epoch": 1.7478180129990715, + "grad_norm": 1.762953519821167, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8686736822128296, + "num_tokens": 330443807.0, + "step": 9412 + }, + { + "epoch": 1.748003714020427, + "grad_norm": 1.619491696357727, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8732768297195435, + "num_tokens": 330477498.0, + "step": 9413 + }, + { + "epoch": 1.7481894150417827, + "grad_norm": 1.3894826173782349, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8927533626556396, + "num_tokens": 330513406.0, + "step": 9414 + }, + { + "epoch": 1.7483751160631384, + "grad_norm": 1.5509012937545776, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8819489479064941, + "num_tokens": 330547513.0, + "step": 9415 + }, + { + "epoch": 1.748560817084494, + "grad_norm": 1.3828752040863037, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8865640163421631, + "num_tokens": 330585924.0, + "step": 9416 + }, + { + "epoch": 1.7487465181058495, + "grad_norm": 1.4885280132293701, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8670608997344971, + "num_tokens": 330622520.0, + "step": 9417 + }, + { + "epoch": 1.7489322191272052, + "grad_norm": 1.670295000076294, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8721810579299927, + "num_tokens": 330651912.0, + "step": 9418 + }, + { + "epoch": 1.749117920148561, + "grad_norm": 1.5764232873916626, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8768468499183655, + "num_tokens": 330684503.0, + "step": 9419 + }, + { + "epoch": 1.7493036211699164, + "grad_norm": 1.4098094701766968, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8842846751213074, + "num_tokens": 330720286.0, + "step": 9420 + }, + { + "epoch": 1.749489322191272, + "grad_norm": 1.4233964681625366, + "learning_rate": 1e-06, + "loss": 0.3809, + "mean_token_accuracy": 0.8695687055587769, + "num_tokens": 330760145.0, + "step": 9421 + }, + { + "epoch": 1.7496750232126277, + "grad_norm": 1.609398365020752, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8723140358924866, + "num_tokens": 330790879.0, + "step": 9422 + }, + { + "epoch": 1.7498607242339834, + "grad_norm": 1.5765702724456787, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8767318725585938, + "num_tokens": 330824362.0, + "step": 9423 + }, + { + "epoch": 1.750046425255339, + "grad_norm": 1.4928956031799316, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8802129030227661, + "num_tokens": 330857446.0, + "step": 9424 + }, + { + "epoch": 1.7502321262766944, + "grad_norm": 1.436513900756836, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8694843053817749, + "num_tokens": 330898380.0, + "step": 9425 + }, + { + "epoch": 1.7504178272980502, + "grad_norm": 1.3452714681625366, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8736686706542969, + "num_tokens": 330941384.0, + "step": 9426 + }, + { + "epoch": 1.750603528319406, + "grad_norm": 1.4285880327224731, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8927879333496094, + "num_tokens": 330973939.0, + "step": 9427 + }, + { + "epoch": 1.7507892293407614, + "grad_norm": 1.5118443965911865, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8704975843429565, + "num_tokens": 331006140.0, + "step": 9428 + }, + { + "epoch": 1.750974930362117, + "grad_norm": 1.4104856252670288, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8868252635002136, + "num_tokens": 331044238.0, + "step": 9429 + }, + { + "epoch": 1.7511606313834727, + "grad_norm": 1.4189146757125854, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8878577947616577, + "num_tokens": 331081337.0, + "step": 9430 + }, + { + "epoch": 1.7513463324048282, + "grad_norm": 1.558678388595581, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8571147322654724, + "num_tokens": 331116772.0, + "step": 9431 + }, + { + "epoch": 1.7515320334261837, + "grad_norm": 1.3886570930480957, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8748981952667236, + "num_tokens": 331158032.0, + "step": 9432 + }, + { + "epoch": 1.7517177344475394, + "grad_norm": 1.4316567182540894, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8769559860229492, + "num_tokens": 331195236.0, + "step": 9433 + }, + { + "epoch": 1.7519034354688952, + "grad_norm": 1.4419759511947632, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8878281116485596, + "num_tokens": 331234256.0, + "step": 9434 + }, + { + "epoch": 1.7520891364902507, + "grad_norm": 1.405306339263916, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8826806545257568, + "num_tokens": 331273717.0, + "step": 9435 + }, + { + "epoch": 1.7522748375116062, + "grad_norm": 1.5018795728683472, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8781231641769409, + "num_tokens": 331308299.0, + "step": 9436 + }, + { + "epoch": 1.752460538532962, + "grad_norm": 1.400351643562317, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8974208831787109, + "num_tokens": 331343406.0, + "step": 9437 + }, + { + "epoch": 1.7526462395543176, + "grad_norm": 1.5133121013641357, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8855978846549988, + "num_tokens": 331374759.0, + "step": 9438 + }, + { + "epoch": 1.7528319405756732, + "grad_norm": 1.6598137617111206, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8790251016616821, + "num_tokens": 331401223.0, + "step": 9439 + }, + { + "epoch": 1.7530176415970287, + "grad_norm": 1.5967386960983276, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8816489577293396, + "num_tokens": 331434093.0, + "step": 9440 + }, + { + "epoch": 1.7532033426183844, + "grad_norm": 1.6733392477035522, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8738585710525513, + "num_tokens": 331463296.0, + "step": 9441 + }, + { + "epoch": 1.7533890436397401, + "grad_norm": 1.4724355936050415, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8888634443283081, + "num_tokens": 331499971.0, + "step": 9442 + }, + { + "epoch": 1.7535747446610956, + "grad_norm": 1.5302708148956299, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8727567791938782, + "num_tokens": 331533529.0, + "step": 9443 + }, + { + "epoch": 1.7537604456824512, + "grad_norm": 1.4846398830413818, + "learning_rate": 1e-06, + "loss": 0.421, + "mean_token_accuracy": 0.8587385416030884, + "num_tokens": 331572607.0, + "step": 9444 + }, + { + "epoch": 1.7539461467038069, + "grad_norm": 1.435542345046997, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8814634084701538, + "num_tokens": 331611410.0, + "step": 9445 + }, + { + "epoch": 1.7541318477251626, + "grad_norm": 1.4652224779129028, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8761260509490967, + "num_tokens": 331647585.0, + "step": 9446 + }, + { + "epoch": 1.7543175487465181, + "grad_norm": 1.4651589393615723, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8744176626205444, + "num_tokens": 331685695.0, + "step": 9447 + }, + { + "epoch": 1.7545032497678736, + "grad_norm": 1.597092866897583, + "learning_rate": 1e-06, + "loss": 0.3962, + "mean_token_accuracy": 0.8611665964126587, + "num_tokens": 331717880.0, + "step": 9448 + }, + { + "epoch": 1.7546889507892294, + "grad_norm": 1.6168400049209595, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8774803280830383, + "num_tokens": 331751340.0, + "step": 9449 + }, + { + "epoch": 1.754874651810585, + "grad_norm": 1.4256352186203003, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8743362426757812, + "num_tokens": 331789685.0, + "step": 9450 + }, + { + "epoch": 1.7550603528319406, + "grad_norm": 1.722670555114746, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8702952861785889, + "num_tokens": 331817243.0, + "step": 9451 + }, + { + "epoch": 1.7552460538532961, + "grad_norm": 1.3766591548919678, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8736526370048523, + "num_tokens": 331858645.0, + "step": 9452 + }, + { + "epoch": 1.7554317548746519, + "grad_norm": 1.4480128288269043, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.882708728313446, + "num_tokens": 331894300.0, + "step": 9453 + }, + { + "epoch": 1.7556174558960074, + "grad_norm": 1.7181650400161743, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.870072603225708, + "num_tokens": 331922481.0, + "step": 9454 + }, + { + "epoch": 1.7558031569173629, + "grad_norm": 1.3312636613845825, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8925955295562744, + "num_tokens": 331960266.0, + "step": 9455 + }, + { + "epoch": 1.7559888579387186, + "grad_norm": 1.4858993291854858, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8786572813987732, + "num_tokens": 331993013.0, + "step": 9456 + }, + { + "epoch": 1.7561745589600744, + "grad_norm": 1.5996592044830322, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8822965025901794, + "num_tokens": 332022828.0, + "step": 9457 + }, + { + "epoch": 1.7563602599814299, + "grad_norm": 1.5816797018051147, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.871367335319519, + "num_tokens": 332057695.0, + "step": 9458 + }, + { + "epoch": 1.7565459610027854, + "grad_norm": 1.2984660863876343, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8964370489120483, + "num_tokens": 332095053.0, + "step": 9459 + }, + { + "epoch": 1.756731662024141, + "grad_norm": 1.4927324056625366, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.876725435256958, + "num_tokens": 332134411.0, + "step": 9460 + }, + { + "epoch": 1.7569173630454968, + "grad_norm": 1.4493908882141113, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8715457320213318, + "num_tokens": 332172248.0, + "step": 9461 + }, + { + "epoch": 1.7571030640668523, + "grad_norm": 1.4762568473815918, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8834113478660583, + "num_tokens": 332205922.0, + "step": 9462 + }, + { + "epoch": 1.7572887650882079, + "grad_norm": 1.3718466758728027, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8866404294967651, + "num_tokens": 332242857.0, + "step": 9463 + }, + { + "epoch": 1.7574744661095636, + "grad_norm": 1.523510456085205, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8759366273880005, + "num_tokens": 332277137.0, + "step": 9464 + }, + { + "epoch": 1.7576601671309193, + "grad_norm": 1.4630720615386963, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8708820343017578, + "num_tokens": 332314962.0, + "step": 9465 + }, + { + "epoch": 1.7578458681522748, + "grad_norm": 1.476494550704956, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.877497673034668, + "num_tokens": 332349089.0, + "step": 9466 + }, + { + "epoch": 1.7580315691736303, + "grad_norm": 1.5224604606628418, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8803933262825012, + "num_tokens": 332384373.0, + "step": 9467 + }, + { + "epoch": 1.758217270194986, + "grad_norm": 1.3941431045532227, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.879958987236023, + "num_tokens": 332422480.0, + "step": 9468 + }, + { + "epoch": 1.7584029712163418, + "grad_norm": 1.5325708389282227, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8835092186927795, + "num_tokens": 332455582.0, + "step": 9469 + }, + { + "epoch": 1.7585886722376973, + "grad_norm": 1.436292290687561, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8757743835449219, + "num_tokens": 332494845.0, + "step": 9470 + }, + { + "epoch": 1.7587743732590528, + "grad_norm": 1.4194773435592651, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8855820894241333, + "num_tokens": 332534152.0, + "step": 9471 + }, + { + "epoch": 1.7589600742804086, + "grad_norm": 1.4692871570587158, + "learning_rate": 1e-06, + "loss": 0.3971, + "mean_token_accuracy": 0.8677147030830383, + "num_tokens": 332575756.0, + "step": 9472 + }, + { + "epoch": 1.7591457753017643, + "grad_norm": 1.5880308151245117, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8724292516708374, + "num_tokens": 332608592.0, + "step": 9473 + }, + { + "epoch": 1.7593314763231198, + "grad_norm": 1.5936932563781738, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8708900213241577, + "num_tokens": 332642028.0, + "step": 9474 + }, + { + "epoch": 1.7595171773444753, + "grad_norm": 1.5227922201156616, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.874113917350769, + "num_tokens": 332676929.0, + "step": 9475 + }, + { + "epoch": 1.759702878365831, + "grad_norm": 1.498376727104187, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8759454488754272, + "num_tokens": 332712230.0, + "step": 9476 + }, + { + "epoch": 1.7598885793871868, + "grad_norm": 1.5555543899536133, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8728046417236328, + "num_tokens": 332744921.0, + "step": 9477 + }, + { + "epoch": 1.7600742804085423, + "grad_norm": 1.4983645677566528, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8645159006118774, + "num_tokens": 332782214.0, + "step": 9478 + }, + { + "epoch": 1.7602599814298978, + "grad_norm": 1.813828468322754, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8874962329864502, + "num_tokens": 332807885.0, + "step": 9479 + }, + { + "epoch": 1.7604456824512535, + "grad_norm": 1.3516944646835327, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8777326345443726, + "num_tokens": 332848862.0, + "step": 9480 + }, + { + "epoch": 1.760631383472609, + "grad_norm": 1.517600655555725, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8717595338821411, + "num_tokens": 332882473.0, + "step": 9481 + }, + { + "epoch": 1.7608170844939646, + "grad_norm": 1.4286444187164307, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8829978108406067, + "num_tokens": 332918920.0, + "step": 9482 + }, + { + "epoch": 1.7610027855153203, + "grad_norm": 1.3893992900848389, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8802213668823242, + "num_tokens": 332954674.0, + "step": 9483 + }, + { + "epoch": 1.761188486536676, + "grad_norm": 1.5536309480667114, + "learning_rate": 1e-06, + "loss": 0.3995, + "mean_token_accuracy": 0.8648024201393127, + "num_tokens": 332992352.0, + "step": 9484 + }, + { + "epoch": 1.7613741875580315, + "grad_norm": 1.4106733798980713, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8831789493560791, + "num_tokens": 333028678.0, + "step": 9485 + }, + { + "epoch": 1.761559888579387, + "grad_norm": 1.4260892868041992, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8755828142166138, + "num_tokens": 333070686.0, + "step": 9486 + }, + { + "epoch": 1.7617455896007428, + "grad_norm": 1.4421298503875732, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8697742223739624, + "num_tokens": 333106453.0, + "step": 9487 + }, + { + "epoch": 1.7619312906220985, + "grad_norm": 1.5096380710601807, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8836402893066406, + "num_tokens": 333139656.0, + "step": 9488 + }, + { + "epoch": 1.762116991643454, + "grad_norm": 1.331357717514038, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8924520015716553, + "num_tokens": 333179507.0, + "step": 9489 + }, + { + "epoch": 1.7623026926648095, + "grad_norm": 1.3899657726287842, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.888725757598877, + "num_tokens": 333217881.0, + "step": 9490 + }, + { + "epoch": 1.7624883936861653, + "grad_norm": 1.419825792312622, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8761100769042969, + "num_tokens": 333254656.0, + "step": 9491 + }, + { + "epoch": 1.762674094707521, + "grad_norm": 1.480017900466919, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8713523149490356, + "num_tokens": 333290311.0, + "step": 9492 + }, + { + "epoch": 1.7628597957288765, + "grad_norm": 1.379866123199463, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8802525401115417, + "num_tokens": 333329150.0, + "step": 9493 + }, + { + "epoch": 1.763045496750232, + "grad_norm": 1.3906970024108887, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8930784463882446, + "num_tokens": 333370573.0, + "step": 9494 + }, + { + "epoch": 1.7632311977715878, + "grad_norm": 1.567305326461792, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8719019889831543, + "num_tokens": 333405746.0, + "step": 9495 + }, + { + "epoch": 1.7634168987929435, + "grad_norm": 1.479977011680603, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8831397294998169, + "num_tokens": 333443168.0, + "step": 9496 + }, + { + "epoch": 1.763602599814299, + "grad_norm": 1.4708787202835083, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8800773620605469, + "num_tokens": 333481414.0, + "step": 9497 + }, + { + "epoch": 1.7637883008356545, + "grad_norm": 1.4321222305297852, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8872596621513367, + "num_tokens": 333516459.0, + "step": 9498 + }, + { + "epoch": 1.7639740018570103, + "grad_norm": 1.5288344621658325, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8644266128540039, + "num_tokens": 333547765.0, + "step": 9499 + }, + { + "epoch": 1.764159702878366, + "grad_norm": 1.518728494644165, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8864091634750366, + "num_tokens": 333579455.0, + "step": 9500 + }, + { + "epoch": 1.7643454038997215, + "grad_norm": 1.5439859628677368, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8771936893463135, + "num_tokens": 333614752.0, + "step": 9501 + }, + { + "epoch": 1.764531104921077, + "grad_norm": 1.4244049787521362, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8653775453567505, + "num_tokens": 333652210.0, + "step": 9502 + }, + { + "epoch": 1.7647168059424327, + "grad_norm": 1.400075912475586, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8738237619400024, + "num_tokens": 333693709.0, + "step": 9503 + }, + { + "epoch": 1.7649025069637883, + "grad_norm": 1.411249041557312, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.893563985824585, + "num_tokens": 333730851.0, + "step": 9504 + }, + { + "epoch": 1.7650882079851438, + "grad_norm": 2.118680477142334, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8809610605239868, + "num_tokens": 333762792.0, + "step": 9505 + }, + { + "epoch": 1.7652739090064995, + "grad_norm": 1.3543990850448608, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.880244255065918, + "num_tokens": 333802444.0, + "step": 9506 + }, + { + "epoch": 1.7654596100278552, + "grad_norm": 1.437166452407837, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8724062442779541, + "num_tokens": 333838326.0, + "step": 9507 + }, + { + "epoch": 1.7656453110492107, + "grad_norm": 1.5375784635543823, + "learning_rate": 1e-06, + "loss": 0.402, + "mean_token_accuracy": 0.8648990988731384, + "num_tokens": 333874323.0, + "step": 9508 + }, + { + "epoch": 1.7658310120705663, + "grad_norm": 1.4314463138580322, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8846443891525269, + "num_tokens": 333908498.0, + "step": 9509 + }, + { + "epoch": 1.766016713091922, + "grad_norm": 1.4719139337539673, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8809641599655151, + "num_tokens": 333942475.0, + "step": 9510 + }, + { + "epoch": 1.7662024141132777, + "grad_norm": 1.5177830457687378, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8696253299713135, + "num_tokens": 333974768.0, + "step": 9511 + }, + { + "epoch": 1.7663881151346332, + "grad_norm": 1.5286097526550293, + "learning_rate": 1e-06, + "loss": 0.425, + "mean_token_accuracy": 0.8555039763450623, + "num_tokens": 334013029.0, + "step": 9512 + }, + { + "epoch": 1.7665738161559887, + "grad_norm": 1.7235691547393799, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8737918734550476, + "num_tokens": 334042684.0, + "step": 9513 + }, + { + "epoch": 1.7667595171773445, + "grad_norm": 1.5637376308441162, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8735479712486267, + "num_tokens": 334076870.0, + "step": 9514 + }, + { + "epoch": 1.7669452181987002, + "grad_norm": 1.364883303642273, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8721543550491333, + "num_tokens": 334115602.0, + "step": 9515 + }, + { + "epoch": 1.7671309192200557, + "grad_norm": 1.532891869544983, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8794869184494019, + "num_tokens": 334148368.0, + "step": 9516 + }, + { + "epoch": 1.7673166202414112, + "grad_norm": 1.504075527191162, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8765702247619629, + "num_tokens": 334182797.0, + "step": 9517 + }, + { + "epoch": 1.767502321262767, + "grad_norm": 1.4432048797607422, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8709259033203125, + "num_tokens": 334221982.0, + "step": 9518 + }, + { + "epoch": 1.7676880222841227, + "grad_norm": 1.4693565368652344, + "learning_rate": 1e-06, + "loss": 0.4024, + "mean_token_accuracy": 0.8640885353088379, + "num_tokens": 334263780.0, + "step": 9519 + }, + { + "epoch": 1.7678737233054782, + "grad_norm": 1.5503814220428467, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8914512395858765, + "num_tokens": 334294444.0, + "step": 9520 + }, + { + "epoch": 1.7680594243268337, + "grad_norm": 1.4756416082382202, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8751983046531677, + "num_tokens": 334330546.0, + "step": 9521 + }, + { + "epoch": 1.7682451253481895, + "grad_norm": 1.5116419792175293, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.871964693069458, + "num_tokens": 334367163.0, + "step": 9522 + }, + { + "epoch": 1.7684308263695452, + "grad_norm": 1.457940936088562, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8654732704162598, + "num_tokens": 334404178.0, + "step": 9523 + }, + { + "epoch": 1.7686165273909007, + "grad_norm": 1.451595664024353, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8729771375656128, + "num_tokens": 334443499.0, + "step": 9524 + }, + { + "epoch": 1.7688022284122562, + "grad_norm": 1.4507434368133545, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8843563795089722, + "num_tokens": 334479740.0, + "step": 9525 + }, + { + "epoch": 1.768987929433612, + "grad_norm": 1.5458996295928955, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.869470477104187, + "num_tokens": 334514302.0, + "step": 9526 + }, + { + "epoch": 1.7691736304549674, + "grad_norm": 1.4830114841461182, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8804483413696289, + "num_tokens": 334550736.0, + "step": 9527 + }, + { + "epoch": 1.769359331476323, + "grad_norm": 1.4484115839004517, + "learning_rate": 1e-06, + "loss": 0.4209, + "mean_token_accuracy": 0.8584945797920227, + "num_tokens": 334591069.0, + "step": 9528 + }, + { + "epoch": 1.7695450324976787, + "grad_norm": 1.4894423484802246, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8877930641174316, + "num_tokens": 334624934.0, + "step": 9529 + }, + { + "epoch": 1.7697307335190344, + "grad_norm": 1.3688623905181885, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8826422691345215, + "num_tokens": 334664757.0, + "step": 9530 + }, + { + "epoch": 1.76991643454039, + "grad_norm": 1.4947259426116943, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8769494295120239, + "num_tokens": 334699503.0, + "step": 9531 + }, + { + "epoch": 1.7701021355617454, + "grad_norm": 1.5633084774017334, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8719525933265686, + "num_tokens": 334732982.0, + "step": 9532 + }, + { + "epoch": 1.7702878365831012, + "grad_norm": 1.4194217920303345, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8875262141227722, + "num_tokens": 334767959.0, + "step": 9533 + }, + { + "epoch": 1.770473537604457, + "grad_norm": 1.5936875343322754, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8727362155914307, + "num_tokens": 334798675.0, + "step": 9534 + }, + { + "epoch": 1.7706592386258124, + "grad_norm": 1.587090015411377, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8859941959381104, + "num_tokens": 334827979.0, + "step": 9535 + }, + { + "epoch": 1.770844939647168, + "grad_norm": 1.410791277885437, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8848404884338379, + "num_tokens": 334863579.0, + "step": 9536 + }, + { + "epoch": 1.7710306406685237, + "grad_norm": 1.503042459487915, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8826303482055664, + "num_tokens": 334894439.0, + "step": 9537 + }, + { + "epoch": 1.7712163416898794, + "grad_norm": 1.431792974472046, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.875519871711731, + "num_tokens": 334933040.0, + "step": 9538 + }, + { + "epoch": 1.771402042711235, + "grad_norm": 1.3600579500198364, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8824580907821655, + "num_tokens": 334970331.0, + "step": 9539 + }, + { + "epoch": 1.7715877437325904, + "grad_norm": 1.654794692993164, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.879142165184021, + "num_tokens": 334998505.0, + "step": 9540 + }, + { + "epoch": 1.7717734447539462, + "grad_norm": 1.6461281776428223, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.874356746673584, + "num_tokens": 335029855.0, + "step": 9541 + }, + { + "epoch": 1.771959145775302, + "grad_norm": 1.3800607919692993, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8797758221626282, + "num_tokens": 335068953.0, + "step": 9542 + }, + { + "epoch": 1.7721448467966574, + "grad_norm": 1.3546335697174072, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8948571085929871, + "num_tokens": 335105308.0, + "step": 9543 + }, + { + "epoch": 1.772330547818013, + "grad_norm": 1.6279194355010986, + "learning_rate": 1e-06, + "loss": 0.4372, + "mean_token_accuracy": 0.8535488843917847, + "num_tokens": 335139904.0, + "step": 9544 + }, + { + "epoch": 1.7725162488393686, + "grad_norm": 1.420681118965149, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.873278796672821, + "num_tokens": 335178470.0, + "step": 9545 + }, + { + "epoch": 1.7727019498607244, + "grad_norm": 1.4381721019744873, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8823508024215698, + "num_tokens": 335215615.0, + "step": 9546 + }, + { + "epoch": 1.77288765088208, + "grad_norm": 1.508986473083496, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8843039870262146, + "num_tokens": 335250557.0, + "step": 9547 + }, + { + "epoch": 1.7730733519034354, + "grad_norm": 1.5668669939041138, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8729016780853271, + "num_tokens": 335282113.0, + "step": 9548 + }, + { + "epoch": 1.7732590529247911, + "grad_norm": 1.5483546257019043, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8733232021331787, + "num_tokens": 335315595.0, + "step": 9549 + }, + { + "epoch": 1.7734447539461469, + "grad_norm": 1.342029094696045, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8907365798950195, + "num_tokens": 335356875.0, + "step": 9550 + }, + { + "epoch": 1.7736304549675022, + "grad_norm": 1.6671621799468994, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8724977970123291, + "num_tokens": 335388303.0, + "step": 9551 + }, + { + "epoch": 1.773816155988858, + "grad_norm": 1.444156289100647, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8771518468856812, + "num_tokens": 335425859.0, + "step": 9552 + }, + { + "epoch": 1.7740018570102136, + "grad_norm": 1.3222084045410156, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8912954330444336, + "num_tokens": 335463131.0, + "step": 9553 + }, + { + "epoch": 1.7741875580315691, + "grad_norm": 1.4822168350219727, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8776055574417114, + "num_tokens": 335494045.0, + "step": 9554 + }, + { + "epoch": 1.7743732590529246, + "grad_norm": 1.3820077180862427, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8818860054016113, + "num_tokens": 335531364.0, + "step": 9555 + }, + { + "epoch": 1.7745589600742804, + "grad_norm": 1.4989876747131348, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8762507438659668, + "num_tokens": 335563559.0, + "step": 9556 + }, + { + "epoch": 1.7747446610956361, + "grad_norm": 1.5303535461425781, + "learning_rate": 1e-06, + "loss": 0.3751, + "mean_token_accuracy": 0.87436842918396, + "num_tokens": 335596739.0, + "step": 9557 + }, + { + "epoch": 1.7749303621169916, + "grad_norm": 1.5237658023834229, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8821302652359009, + "num_tokens": 335629731.0, + "step": 9558 + }, + { + "epoch": 1.7751160631383471, + "grad_norm": 1.3736628293991089, + "learning_rate": 1e-06, + "loss": 0.3742, + "mean_token_accuracy": 0.8733085989952087, + "num_tokens": 335669672.0, + "step": 9559 + }, + { + "epoch": 1.7753017641597029, + "grad_norm": 1.4262398481369019, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8777799606323242, + "num_tokens": 335705263.0, + "step": 9560 + }, + { + "epoch": 1.7754874651810586, + "grad_norm": 1.4970884323120117, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8873623609542847, + "num_tokens": 335739776.0, + "step": 9561 + }, + { + "epoch": 1.775673166202414, + "grad_norm": 1.585654854774475, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8829401731491089, + "num_tokens": 335770867.0, + "step": 9562 + }, + { + "epoch": 1.7758588672237696, + "grad_norm": 1.539563536643982, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8603417873382568, + "num_tokens": 335802803.0, + "step": 9563 + }, + { + "epoch": 1.7760445682451254, + "grad_norm": 1.5092566013336182, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8756407499313354, + "num_tokens": 335835875.0, + "step": 9564 + }, + { + "epoch": 1.776230269266481, + "grad_norm": 1.5366040468215942, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8752451539039612, + "num_tokens": 335869315.0, + "step": 9565 + }, + { + "epoch": 1.7764159702878366, + "grad_norm": 1.4094898700714111, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8853985071182251, + "num_tokens": 335905360.0, + "step": 9566 + }, + { + "epoch": 1.776601671309192, + "grad_norm": 1.9776477813720703, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8737646341323853, + "num_tokens": 335940449.0, + "step": 9567 + }, + { + "epoch": 1.7767873723305478, + "grad_norm": 1.321047067642212, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8848636150360107, + "num_tokens": 335981010.0, + "step": 9568 + }, + { + "epoch": 1.7769730733519036, + "grad_norm": 1.456976056098938, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8893064856529236, + "num_tokens": 336015855.0, + "step": 9569 + }, + { + "epoch": 1.777158774373259, + "grad_norm": 1.5179541110992432, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8761574029922485, + "num_tokens": 336051215.0, + "step": 9570 + }, + { + "epoch": 1.7773444753946146, + "grad_norm": 1.48856520652771, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8763019442558289, + "num_tokens": 336084874.0, + "step": 9571 + }, + { + "epoch": 1.7775301764159703, + "grad_norm": 1.487548589706421, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8756235241889954, + "num_tokens": 336119694.0, + "step": 9572 + }, + { + "epoch": 1.777715877437326, + "grad_norm": 1.4184786081314087, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8687976598739624, + "num_tokens": 336157991.0, + "step": 9573 + }, + { + "epoch": 1.7779015784586816, + "grad_norm": 1.4720299243927002, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8638885021209717, + "num_tokens": 336193182.0, + "step": 9574 + }, + { + "epoch": 1.778087279480037, + "grad_norm": 1.5115371942520142, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8872954845428467, + "num_tokens": 336223159.0, + "step": 9575 + }, + { + "epoch": 1.7782729805013928, + "grad_norm": 1.5629369020462036, + "learning_rate": 1e-06, + "loss": 0.3829, + "mean_token_accuracy": 0.8647558689117432, + "num_tokens": 336256628.0, + "step": 9576 + }, + { + "epoch": 1.7784586815227483, + "grad_norm": 1.5908676385879517, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.869758665561676, + "num_tokens": 336290962.0, + "step": 9577 + }, + { + "epoch": 1.7786443825441038, + "grad_norm": 1.5886421203613281, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8794248104095459, + "num_tokens": 336322342.0, + "step": 9578 + }, + { + "epoch": 1.7788300835654596, + "grad_norm": 1.3515424728393555, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.87799471616745, + "num_tokens": 336362760.0, + "step": 9579 + }, + { + "epoch": 1.7790157845868153, + "grad_norm": 1.4265947341918945, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.879758894443512, + "num_tokens": 336401617.0, + "step": 9580 + }, + { + "epoch": 1.7792014856081708, + "grad_norm": 1.4271690845489502, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8694748282432556, + "num_tokens": 336438299.0, + "step": 9581 + }, + { + "epoch": 1.7793871866295263, + "grad_norm": 1.4422414302825928, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8812447786331177, + "num_tokens": 336476399.0, + "step": 9582 + }, + { + "epoch": 1.779572887650882, + "grad_norm": 1.564560890197754, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8778678178787231, + "num_tokens": 336508509.0, + "step": 9583 + }, + { + "epoch": 1.7797585886722378, + "grad_norm": 1.4950908422470093, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8758887052536011, + "num_tokens": 336543429.0, + "step": 9584 + }, + { + "epoch": 1.7799442896935933, + "grad_norm": 1.4411240816116333, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8804104924201965, + "num_tokens": 336577096.0, + "step": 9585 + }, + { + "epoch": 1.7801299907149488, + "grad_norm": 1.3948187828063965, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8930076360702515, + "num_tokens": 336612974.0, + "step": 9586 + }, + { + "epoch": 1.7803156917363046, + "grad_norm": 1.3830236196517944, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.877447247505188, + "num_tokens": 336650839.0, + "step": 9587 + }, + { + "epoch": 1.7805013927576603, + "grad_norm": 1.5499118566513062, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.888564944267273, + "num_tokens": 336689916.0, + "step": 9588 + }, + { + "epoch": 1.7806870937790158, + "grad_norm": 1.4119023084640503, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8774569630622864, + "num_tokens": 336728417.0, + "step": 9589 + }, + { + "epoch": 1.7808727948003713, + "grad_norm": 1.5028961896896362, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.876956045627594, + "num_tokens": 336764018.0, + "step": 9590 + }, + { + "epoch": 1.781058495821727, + "grad_norm": 1.5433109998703003, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8743152618408203, + "num_tokens": 336797503.0, + "step": 9591 + }, + { + "epoch": 1.7812441968430828, + "grad_norm": 1.4457398653030396, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8754348754882812, + "num_tokens": 336834544.0, + "step": 9592 + }, + { + "epoch": 1.7814298978644383, + "grad_norm": 1.565483808517456, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8740129470825195, + "num_tokens": 336865579.0, + "step": 9593 + }, + { + "epoch": 1.7816155988857938, + "grad_norm": 1.4343883991241455, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.879361629486084, + "num_tokens": 336900859.0, + "step": 9594 + }, + { + "epoch": 1.7818012999071495, + "grad_norm": 1.4789469242095947, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8726513385772705, + "num_tokens": 336935556.0, + "step": 9595 + }, + { + "epoch": 1.7819870009285053, + "grad_norm": 1.4815278053283691, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8941731452941895, + "num_tokens": 336970166.0, + "step": 9596 + }, + { + "epoch": 1.7821727019498608, + "grad_norm": 1.6364961862564087, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8654834628105164, + "num_tokens": 337005314.0, + "step": 9597 + }, + { + "epoch": 1.7823584029712163, + "grad_norm": 1.6577982902526855, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8679634928703308, + "num_tokens": 337035381.0, + "step": 9598 + }, + { + "epoch": 1.782544103992572, + "grad_norm": 1.4442253112792969, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.884070873260498, + "num_tokens": 337073505.0, + "step": 9599 + }, + { + "epoch": 1.7827298050139275, + "grad_norm": 1.4867883920669556, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8724388480186462, + "num_tokens": 337109373.0, + "step": 9600 + }, + { + "epoch": 1.782915506035283, + "grad_norm": 1.544878602027893, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8780178427696228, + "num_tokens": 337145853.0, + "step": 9601 + }, + { + "epoch": 1.7831012070566388, + "grad_norm": 1.380784273147583, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8818086385726929, + "num_tokens": 337186279.0, + "step": 9602 + }, + { + "epoch": 1.7832869080779945, + "grad_norm": 1.513976812362671, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.876123309135437, + "num_tokens": 337220726.0, + "step": 9603 + }, + { + "epoch": 1.78347260909935, + "grad_norm": 1.5200978517532349, + "learning_rate": 1e-06, + "loss": 0.3698, + "mean_token_accuracy": 0.875734806060791, + "num_tokens": 337254823.0, + "step": 9604 + }, + { + "epoch": 1.7836583101207055, + "grad_norm": 1.7449408769607544, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8888552784919739, + "num_tokens": 337279941.0, + "step": 9605 + }, + { + "epoch": 1.7838440111420613, + "grad_norm": 1.5383089780807495, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8715883493423462, + "num_tokens": 337313528.0, + "step": 9606 + }, + { + "epoch": 1.784029712163417, + "grad_norm": 1.3597644567489624, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8802740573883057, + "num_tokens": 337353403.0, + "step": 9607 + }, + { + "epoch": 1.7842154131847725, + "grad_norm": 1.4508169889450073, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8948381543159485, + "num_tokens": 337389221.0, + "step": 9608 + }, + { + "epoch": 1.784401114206128, + "grad_norm": 1.5811866521835327, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.875206470489502, + "num_tokens": 337421294.0, + "step": 9609 + }, + { + "epoch": 1.7845868152274837, + "grad_norm": 1.3491857051849365, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8850021362304688, + "num_tokens": 337461542.0, + "step": 9610 + }, + { + "epoch": 1.7847725162488395, + "grad_norm": 1.6477223634719849, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.887143611907959, + "num_tokens": 337487433.0, + "step": 9611 + }, + { + "epoch": 1.784958217270195, + "grad_norm": 1.4672423601150513, + "learning_rate": 1e-06, + "loss": 0.3181, + "mean_token_accuracy": 0.8909776210784912, + "num_tokens": 337519932.0, + "step": 9612 + }, + { + "epoch": 1.7851439182915505, + "grad_norm": 1.50632643699646, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8864416480064392, + "num_tokens": 337551088.0, + "step": 9613 + }, + { + "epoch": 1.7853296193129062, + "grad_norm": 1.7190183401107788, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8710215091705322, + "num_tokens": 337580060.0, + "step": 9614 + }, + { + "epoch": 1.785515320334262, + "grad_norm": 1.4841077327728271, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.887150764465332, + "num_tokens": 337618242.0, + "step": 9615 + }, + { + "epoch": 1.7857010213556175, + "grad_norm": 1.5588951110839844, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8679361343383789, + "num_tokens": 337653342.0, + "step": 9616 + }, + { + "epoch": 1.785886722376973, + "grad_norm": 1.6366628408432007, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.8629752397537231, + "num_tokens": 337686367.0, + "step": 9617 + }, + { + "epoch": 1.7860724233983287, + "grad_norm": 1.5355207920074463, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8683596253395081, + "num_tokens": 337720926.0, + "step": 9618 + }, + { + "epoch": 1.7862581244196845, + "grad_norm": 1.434727430343628, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8849803805351257, + "num_tokens": 337755734.0, + "step": 9619 + }, + { + "epoch": 1.78644382544104, + "grad_norm": 1.5408570766448975, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8819079399108887, + "num_tokens": 337788584.0, + "step": 9620 + }, + { + "epoch": 1.7866295264623955, + "grad_norm": 1.58781099319458, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8749162554740906, + "num_tokens": 337821688.0, + "step": 9621 + }, + { + "epoch": 1.7868152274837512, + "grad_norm": 1.7733627557754517, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8785900473594666, + "num_tokens": 337846431.0, + "step": 9622 + }, + { + "epoch": 1.7870009285051067, + "grad_norm": 1.5325238704681396, + "learning_rate": 1e-06, + "loss": 0.3739, + "mean_token_accuracy": 0.8701949119567871, + "num_tokens": 337880021.0, + "step": 9623 + }, + { + "epoch": 1.7871866295264622, + "grad_norm": 1.5271307229995728, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8705848455429077, + "num_tokens": 337912590.0, + "step": 9624 + }, + { + "epoch": 1.787372330547818, + "grad_norm": 1.5146925449371338, + "learning_rate": 1e-06, + "loss": 0.4125, + "mean_token_accuracy": 0.8622750639915466, + "num_tokens": 337950567.0, + "step": 9625 + }, + { + "epoch": 1.7875580315691737, + "grad_norm": 1.298055648803711, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8881338834762573, + "num_tokens": 337992056.0, + "step": 9626 + }, + { + "epoch": 1.7877437325905292, + "grad_norm": 1.6726998090744019, + "learning_rate": 1e-06, + "loss": 0.401, + "mean_token_accuracy": 0.864249587059021, + "num_tokens": 338022543.0, + "step": 9627 + }, + { + "epoch": 1.7879294336118847, + "grad_norm": 1.4172558784484863, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8803805112838745, + "num_tokens": 338059144.0, + "step": 9628 + }, + { + "epoch": 1.7881151346332405, + "grad_norm": 1.4588115215301514, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8682592511177063, + "num_tokens": 338097429.0, + "step": 9629 + }, + { + "epoch": 1.7883008356545962, + "grad_norm": 1.41068434715271, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8907904624938965, + "num_tokens": 338138246.0, + "step": 9630 + }, + { + "epoch": 1.7884865366759517, + "grad_norm": 1.5171091556549072, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.866275429725647, + "num_tokens": 338175876.0, + "step": 9631 + }, + { + "epoch": 1.7886722376973072, + "grad_norm": 1.4866491556167603, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8824362754821777, + "num_tokens": 338210339.0, + "step": 9632 + }, + { + "epoch": 1.788857938718663, + "grad_norm": 1.586807370185852, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8790562152862549, + "num_tokens": 338242881.0, + "step": 9633 + }, + { + "epoch": 1.7890436397400187, + "grad_norm": 1.3528794050216675, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8796024322509766, + "num_tokens": 338284218.0, + "step": 9634 + }, + { + "epoch": 1.7892293407613742, + "grad_norm": 1.451638102531433, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8703776001930237, + "num_tokens": 338322912.0, + "step": 9635 + }, + { + "epoch": 1.7894150417827297, + "grad_norm": 1.4909700155258179, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8673155307769775, + "num_tokens": 338361490.0, + "step": 9636 + }, + { + "epoch": 1.7896007428040854, + "grad_norm": 1.6886029243469238, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8630719184875488, + "num_tokens": 338392452.0, + "step": 9637 + }, + { + "epoch": 1.7897864438254412, + "grad_norm": 1.4756239652633667, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8816381692886353, + "num_tokens": 338426991.0, + "step": 9638 + }, + { + "epoch": 1.7899721448467967, + "grad_norm": 1.5123498439788818, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8825056552886963, + "num_tokens": 338459360.0, + "step": 9639 + }, + { + "epoch": 1.7901578458681522, + "grad_norm": 1.3503799438476562, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8756252527236938, + "num_tokens": 338501170.0, + "step": 9640 + }, + { + "epoch": 1.790343546889508, + "grad_norm": 1.4475613832473755, + "learning_rate": 1e-06, + "loss": 0.3873, + "mean_token_accuracy": 0.8666083216667175, + "num_tokens": 338542794.0, + "step": 9641 + }, + { + "epoch": 1.7905292479108637, + "grad_norm": 1.4126240015029907, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8912575244903564, + "num_tokens": 338578813.0, + "step": 9642 + }, + { + "epoch": 1.7907149489322192, + "grad_norm": 1.4891432523727417, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8727632761001587, + "num_tokens": 338610967.0, + "step": 9643 + }, + { + "epoch": 1.7909006499535747, + "grad_norm": 1.3814425468444824, + "learning_rate": 1e-06, + "loss": 0.2752, + "mean_token_accuracy": 0.9037116765975952, + "num_tokens": 338643831.0, + "step": 9644 + }, + { + "epoch": 1.7910863509749304, + "grad_norm": 1.3705626726150513, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8894181847572327, + "num_tokens": 338679240.0, + "step": 9645 + }, + { + "epoch": 1.7912720519962861, + "grad_norm": 1.5786467790603638, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8810369968414307, + "num_tokens": 338709055.0, + "step": 9646 + }, + { + "epoch": 1.7914577530176417, + "grad_norm": 1.5042353868484497, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.882720947265625, + "num_tokens": 338742529.0, + "step": 9647 + }, + { + "epoch": 1.7916434540389972, + "grad_norm": 1.4824122190475464, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8913476467132568, + "num_tokens": 338775835.0, + "step": 9648 + }, + { + "epoch": 1.791829155060353, + "grad_norm": 1.4950034618377686, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.878205418586731, + "num_tokens": 338809400.0, + "step": 9649 + }, + { + "epoch": 1.7920148560817084, + "grad_norm": 1.4736840724945068, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8734967708587646, + "num_tokens": 338851532.0, + "step": 9650 + }, + { + "epoch": 1.792200557103064, + "grad_norm": 1.610504388809204, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8820174932479858, + "num_tokens": 338883062.0, + "step": 9651 + }, + { + "epoch": 1.7923862581244197, + "grad_norm": 1.5440703630447388, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8757287263870239, + "num_tokens": 338916982.0, + "step": 9652 + }, + { + "epoch": 1.7925719591457754, + "grad_norm": 1.6384447813034058, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.8688186407089233, + "num_tokens": 338950610.0, + "step": 9653 + }, + { + "epoch": 1.792757660167131, + "grad_norm": 1.7603367567062378, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8603387475013733, + "num_tokens": 338981185.0, + "step": 9654 + }, + { + "epoch": 1.7929433611884864, + "grad_norm": 1.5576506853103638, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8779394626617432, + "num_tokens": 339013599.0, + "step": 9655 + }, + { + "epoch": 1.7931290622098421, + "grad_norm": 1.3789974451065063, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8910033702850342, + "num_tokens": 339051298.0, + "step": 9656 + }, + { + "epoch": 1.7933147632311979, + "grad_norm": 1.4660325050354004, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8816421031951904, + "num_tokens": 339087636.0, + "step": 9657 + }, + { + "epoch": 1.7935004642525534, + "grad_norm": 1.5700405836105347, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8706619739532471, + "num_tokens": 339122097.0, + "step": 9658 + }, + { + "epoch": 1.793686165273909, + "grad_norm": 1.4040656089782715, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8860085010528564, + "num_tokens": 339157092.0, + "step": 9659 + }, + { + "epoch": 1.7938718662952646, + "grad_norm": 1.4598182439804077, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8858105540275574, + "num_tokens": 339192527.0, + "step": 9660 + }, + { + "epoch": 1.7940575673166204, + "grad_norm": 1.4949321746826172, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.880059540271759, + "num_tokens": 339228761.0, + "step": 9661 + }, + { + "epoch": 1.7942432683379759, + "grad_norm": 1.504841923713684, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8856117129325867, + "num_tokens": 339259999.0, + "step": 9662 + }, + { + "epoch": 1.7944289693593314, + "grad_norm": 1.3471674919128418, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.873484194278717, + "num_tokens": 339304347.0, + "step": 9663 + }, + { + "epoch": 1.7946146703806871, + "grad_norm": 1.4352233409881592, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8793761134147644, + "num_tokens": 339339596.0, + "step": 9664 + }, + { + "epoch": 1.7948003714020428, + "grad_norm": 1.6056208610534668, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8673070073127747, + "num_tokens": 339370165.0, + "step": 9665 + }, + { + "epoch": 1.7949860724233984, + "grad_norm": 1.4936388731002808, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8737326264381409, + "num_tokens": 339405531.0, + "step": 9666 + }, + { + "epoch": 1.7951717734447539, + "grad_norm": 1.465278148651123, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8785706162452698, + "num_tokens": 339442355.0, + "step": 9667 + }, + { + "epoch": 1.7953574744661096, + "grad_norm": 1.4624441862106323, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8708504438400269, + "num_tokens": 339477566.0, + "step": 9668 + }, + { + "epoch": 1.7955431754874653, + "grad_norm": 1.431213617324829, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8835736513137817, + "num_tokens": 339510311.0, + "step": 9669 + }, + { + "epoch": 1.7957288765088208, + "grad_norm": 1.4871058464050293, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8757332563400269, + "num_tokens": 339546536.0, + "step": 9670 + }, + { + "epoch": 1.7959145775301764, + "grad_norm": 1.3520559072494507, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.88392174243927, + "num_tokens": 339584881.0, + "step": 9671 + }, + { + "epoch": 1.796100278551532, + "grad_norm": 1.6815911531448364, + "learning_rate": 1e-06, + "loss": 0.3987, + "mean_token_accuracy": 0.8638290762901306, + "num_tokens": 339616239.0, + "step": 9672 + }, + { + "epoch": 1.7962859795728876, + "grad_norm": 1.4250048398971558, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8762787580490112, + "num_tokens": 339654279.0, + "step": 9673 + }, + { + "epoch": 1.7964716805942431, + "grad_norm": 1.4197418689727783, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8817275762557983, + "num_tokens": 339692150.0, + "step": 9674 + }, + { + "epoch": 1.7966573816155988, + "grad_norm": 1.3356914520263672, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8833779692649841, + "num_tokens": 339735065.0, + "step": 9675 + }, + { + "epoch": 1.7968430826369546, + "grad_norm": 1.4478013515472412, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8813048005104065, + "num_tokens": 339773660.0, + "step": 9676 + }, + { + "epoch": 1.79702878365831, + "grad_norm": 1.5026456117630005, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8839806318283081, + "num_tokens": 339808588.0, + "step": 9677 + }, + { + "epoch": 1.7972144846796656, + "grad_norm": 1.490017056465149, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.8705068826675415, + "num_tokens": 339843473.0, + "step": 9678 + }, + { + "epoch": 1.7974001857010213, + "grad_norm": 1.3216490745544434, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8846704959869385, + "num_tokens": 339883606.0, + "step": 9679 + }, + { + "epoch": 1.797585886722377, + "grad_norm": 1.4895195960998535, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8959658145904541, + "num_tokens": 339915281.0, + "step": 9680 + }, + { + "epoch": 1.7977715877437326, + "grad_norm": 1.496032476425171, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8836179375648499, + "num_tokens": 339950400.0, + "step": 9681 + }, + { + "epoch": 1.797957288765088, + "grad_norm": 1.5523053407669067, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8912748098373413, + "num_tokens": 339984158.0, + "step": 9682 + }, + { + "epoch": 1.7981429897864438, + "grad_norm": 1.378421664237976, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8837867975234985, + "num_tokens": 340022705.0, + "step": 9683 + }, + { + "epoch": 1.7983286908077996, + "grad_norm": 1.472977876663208, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8725141882896423, + "num_tokens": 340057517.0, + "step": 9684 + }, + { + "epoch": 1.798514391829155, + "grad_norm": 1.3984479904174805, + "learning_rate": 1e-06, + "loss": 0.3864, + "mean_token_accuracy": 0.870561420917511, + "num_tokens": 340100641.0, + "step": 9685 + }, + { + "epoch": 1.7987000928505106, + "grad_norm": 1.5475693941116333, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8858839273452759, + "num_tokens": 340132115.0, + "step": 9686 + }, + { + "epoch": 1.7988857938718663, + "grad_norm": 1.6647461652755737, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8763937950134277, + "num_tokens": 340158428.0, + "step": 9687 + }, + { + "epoch": 1.799071494893222, + "grad_norm": 1.597772240638733, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8688246011734009, + "num_tokens": 340193111.0, + "step": 9688 + }, + { + "epoch": 1.7992571959145776, + "grad_norm": 1.6141420602798462, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8824156522750854, + "num_tokens": 340223290.0, + "step": 9689 + }, + { + "epoch": 1.799442896935933, + "grad_norm": 1.600999116897583, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8830884099006653, + "num_tokens": 340254052.0, + "step": 9690 + }, + { + "epoch": 1.7996285979572888, + "grad_norm": 1.4129668474197388, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8823342323303223, + "num_tokens": 340288220.0, + "step": 9691 + }, + { + "epoch": 1.7998142989786445, + "grad_norm": 1.401289701461792, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8740788698196411, + "num_tokens": 340324570.0, + "step": 9692 + }, + { + "epoch": 1.8, + "grad_norm": 1.4095288515090942, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8807452321052551, + "num_tokens": 340360243.0, + "step": 9693 + }, + { + "epoch": 1.8001857010213556, + "grad_norm": 1.5507885217666626, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8768194913864136, + "num_tokens": 340391669.0, + "step": 9694 + }, + { + "epoch": 1.8003714020427113, + "grad_norm": 1.6272704601287842, + "learning_rate": 1e-06, + "loss": 0.4277, + "mean_token_accuracy": 0.855278730392456, + "num_tokens": 340425769.0, + "step": 9695 + }, + { + "epoch": 1.8005571030640668, + "grad_norm": 1.4282022714614868, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8862366676330566, + "num_tokens": 340458383.0, + "step": 9696 + }, + { + "epoch": 1.8007428040854223, + "grad_norm": 1.4051027297973633, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.889288067817688, + "num_tokens": 340498596.0, + "step": 9697 + }, + { + "epoch": 1.800928505106778, + "grad_norm": 1.4925442934036255, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8763025999069214, + "num_tokens": 340534451.0, + "step": 9698 + }, + { + "epoch": 1.8011142061281338, + "grad_norm": 1.6387723684310913, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8671717643737793, + "num_tokens": 340565788.0, + "step": 9699 + }, + { + "epoch": 1.8012999071494893, + "grad_norm": 1.4986844062805176, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8789995908737183, + "num_tokens": 340599004.0, + "step": 9700 + }, + { + "epoch": 1.8014856081708448, + "grad_norm": 1.4517996311187744, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8736283183097839, + "num_tokens": 340634017.0, + "step": 9701 + }, + { + "epoch": 1.8016713091922005, + "grad_norm": 1.4741764068603516, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8868972063064575, + "num_tokens": 340668298.0, + "step": 9702 + }, + { + "epoch": 1.8018570102135563, + "grad_norm": 1.505165457725525, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8809006810188293, + "num_tokens": 340704112.0, + "step": 9703 + }, + { + "epoch": 1.8020427112349118, + "grad_norm": 1.544061303138733, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8830742835998535, + "num_tokens": 340741457.0, + "step": 9704 + }, + { + "epoch": 1.8022284122562673, + "grad_norm": 1.572822093963623, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8927048444747925, + "num_tokens": 340770992.0, + "step": 9705 + }, + { + "epoch": 1.802414113277623, + "grad_norm": 1.4251171350479126, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8699915409088135, + "num_tokens": 340808982.0, + "step": 9706 + }, + { + "epoch": 1.8025998142989788, + "grad_norm": 1.7243657112121582, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.869436502456665, + "num_tokens": 340834950.0, + "step": 9707 + }, + { + "epoch": 1.8027855153203343, + "grad_norm": 1.688266634941101, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8735181093215942, + "num_tokens": 340867983.0, + "step": 9708 + }, + { + "epoch": 1.8029712163416898, + "grad_norm": 1.7361847162246704, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8937392234802246, + "num_tokens": 340897263.0, + "step": 9709 + }, + { + "epoch": 1.8031569173630455, + "grad_norm": 1.6845002174377441, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8752976059913635, + "num_tokens": 340927416.0, + "step": 9710 + }, + { + "epoch": 1.8033426183844012, + "grad_norm": 1.491685152053833, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8828696608543396, + "num_tokens": 340963128.0, + "step": 9711 + }, + { + "epoch": 1.8035283194057568, + "grad_norm": 1.4644384384155273, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8853684067726135, + "num_tokens": 340998375.0, + "step": 9712 + }, + { + "epoch": 1.8037140204271123, + "grad_norm": 1.4690021276474, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.88509202003479, + "num_tokens": 341030081.0, + "step": 9713 + }, + { + "epoch": 1.803899721448468, + "grad_norm": 1.4192723035812378, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8755648732185364, + "num_tokens": 341067789.0, + "step": 9714 + }, + { + "epoch": 1.8040854224698237, + "grad_norm": 1.67311692237854, + "learning_rate": 1e-06, + "loss": 0.4068, + "mean_token_accuracy": 0.8610424995422363, + "num_tokens": 341103393.0, + "step": 9715 + }, + { + "epoch": 1.8042711234911792, + "grad_norm": 1.7402634620666504, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8666512966156006, + "num_tokens": 341132229.0, + "step": 9716 + }, + { + "epoch": 1.8044568245125348, + "grad_norm": 1.523171067237854, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8797575831413269, + "num_tokens": 341168410.0, + "step": 9717 + }, + { + "epoch": 1.8046425255338905, + "grad_norm": 1.58690345287323, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8705213069915771, + "num_tokens": 341201616.0, + "step": 9718 + }, + { + "epoch": 1.8048282265552462, + "grad_norm": 1.467868447303772, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8826167583465576, + "num_tokens": 341239775.0, + "step": 9719 + }, + { + "epoch": 1.8050139275766015, + "grad_norm": 1.5564671754837036, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8691343069076538, + "num_tokens": 341276469.0, + "step": 9720 + }, + { + "epoch": 1.8051996285979572, + "grad_norm": 1.526078701019287, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8687098622322083, + "num_tokens": 341312254.0, + "step": 9721 + }, + { + "epoch": 1.805385329619313, + "grad_norm": 1.5122238397598267, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8780808448791504, + "num_tokens": 341344764.0, + "step": 9722 + }, + { + "epoch": 1.8055710306406685, + "grad_norm": 1.6338711977005005, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8763468861579895, + "num_tokens": 341373211.0, + "step": 9723 + }, + { + "epoch": 1.805756731662024, + "grad_norm": 1.4210387468338013, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8827341794967651, + "num_tokens": 341411623.0, + "step": 9724 + }, + { + "epoch": 1.8059424326833797, + "grad_norm": 1.4203850030899048, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8790543079376221, + "num_tokens": 341451132.0, + "step": 9725 + }, + { + "epoch": 1.8061281337047355, + "grad_norm": 1.5176634788513184, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8850382566452026, + "num_tokens": 341484638.0, + "step": 9726 + }, + { + "epoch": 1.806313834726091, + "grad_norm": 1.4377295970916748, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8861006498336792, + "num_tokens": 341519506.0, + "step": 9727 + }, + { + "epoch": 1.8064995357474465, + "grad_norm": 1.5658490657806396, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8799147009849548, + "num_tokens": 341551048.0, + "step": 9728 + }, + { + "epoch": 1.8066852367688022, + "grad_norm": 1.6406595706939697, + "learning_rate": 1e-06, + "loss": 0.3855, + "mean_token_accuracy": 0.8693063259124756, + "num_tokens": 341584751.0, + "step": 9729 + }, + { + "epoch": 1.806870937790158, + "grad_norm": 1.4547300338745117, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.877517580986023, + "num_tokens": 341621104.0, + "step": 9730 + }, + { + "epoch": 1.8070566388115135, + "grad_norm": 1.622651219367981, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8753519654273987, + "num_tokens": 341657576.0, + "step": 9731 + }, + { + "epoch": 1.807242339832869, + "grad_norm": 1.491495132446289, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8826849460601807, + "num_tokens": 341690169.0, + "step": 9732 + }, + { + "epoch": 1.8074280408542247, + "grad_norm": 1.4994001388549805, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8737140893936157, + "num_tokens": 341723670.0, + "step": 9733 + }, + { + "epoch": 1.8076137418755804, + "grad_norm": 1.437512993812561, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8766117095947266, + "num_tokens": 341758427.0, + "step": 9734 + }, + { + "epoch": 1.807799442896936, + "grad_norm": 1.444879174232483, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8735166788101196, + "num_tokens": 341799079.0, + "step": 9735 + }, + { + "epoch": 1.8079851439182915, + "grad_norm": 1.4665989875793457, + "learning_rate": 1e-06, + "loss": 0.3641, + "mean_token_accuracy": 0.8749473094940186, + "num_tokens": 341835263.0, + "step": 9736 + }, + { + "epoch": 1.8081708449396472, + "grad_norm": 1.5412591695785522, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8768076300621033, + "num_tokens": 341864056.0, + "step": 9737 + }, + { + "epoch": 1.808356545961003, + "grad_norm": 1.647049069404602, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8644952178001404, + "num_tokens": 341894765.0, + "step": 9738 + }, + { + "epoch": 1.8085422469823584, + "grad_norm": 1.3498079776763916, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8900186419487, + "num_tokens": 341930936.0, + "step": 9739 + }, + { + "epoch": 1.808727948003714, + "grad_norm": 1.5065685510635376, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8670031428337097, + "num_tokens": 341973007.0, + "step": 9740 + }, + { + "epoch": 1.8089136490250697, + "grad_norm": 1.4181753396987915, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8795359134674072, + "num_tokens": 342011728.0, + "step": 9741 + }, + { + "epoch": 1.8090993500464254, + "grad_norm": 1.435893177986145, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8813343048095703, + "num_tokens": 342048794.0, + "step": 9742 + }, + { + "epoch": 1.809285051067781, + "grad_norm": 1.500526785850525, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.87457275390625, + "num_tokens": 342088689.0, + "step": 9743 + }, + { + "epoch": 1.8094707520891364, + "grad_norm": 1.482037901878357, + "learning_rate": 1e-06, + "loss": 0.4128, + "mean_token_accuracy": 0.8567513227462769, + "num_tokens": 342126091.0, + "step": 9744 + }, + { + "epoch": 1.8096564531104922, + "grad_norm": 1.5235470533370972, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8838794231414795, + "num_tokens": 342157469.0, + "step": 9745 + }, + { + "epoch": 1.8098421541318477, + "grad_norm": 1.5624420642852783, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8818581700325012, + "num_tokens": 342188082.0, + "step": 9746 + }, + { + "epoch": 1.8100278551532032, + "grad_norm": 1.54691743850708, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.879214346408844, + "num_tokens": 342218947.0, + "step": 9747 + }, + { + "epoch": 1.810213556174559, + "grad_norm": 1.5018843412399292, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8908337354660034, + "num_tokens": 342256781.0, + "step": 9748 + }, + { + "epoch": 1.8103992571959147, + "grad_norm": 1.405580997467041, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8741264343261719, + "num_tokens": 342294403.0, + "step": 9749 + }, + { + "epoch": 1.8105849582172702, + "grad_norm": 1.6223125457763672, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8737571239471436, + "num_tokens": 342323831.0, + "step": 9750 + }, + { + "epoch": 1.8107706592386257, + "grad_norm": 1.4065924882888794, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8833370208740234, + "num_tokens": 342360660.0, + "step": 9751 + }, + { + "epoch": 1.8109563602599814, + "grad_norm": 1.4501538276672363, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8769126534461975, + "num_tokens": 342395515.0, + "step": 9752 + }, + { + "epoch": 1.8111420612813371, + "grad_norm": 1.385483980178833, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8792184591293335, + "num_tokens": 342430961.0, + "step": 9753 + }, + { + "epoch": 1.8113277623026927, + "grad_norm": 1.6338365077972412, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8858069181442261, + "num_tokens": 342460394.0, + "step": 9754 + }, + { + "epoch": 1.8115134633240482, + "grad_norm": 1.4838539361953735, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.890537679195404, + "num_tokens": 342493440.0, + "step": 9755 + }, + { + "epoch": 1.811699164345404, + "grad_norm": 1.3913114070892334, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8768558502197266, + "num_tokens": 342528810.0, + "step": 9756 + }, + { + "epoch": 1.8118848653667596, + "grad_norm": 1.4320406913757324, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8748776912689209, + "num_tokens": 342564907.0, + "step": 9757 + }, + { + "epoch": 1.8120705663881151, + "grad_norm": 1.4890421628952026, + "learning_rate": 1e-06, + "loss": 0.418, + "mean_token_accuracy": 0.8574411273002625, + "num_tokens": 342605268.0, + "step": 9758 + }, + { + "epoch": 1.8122562674094707, + "grad_norm": 1.558713436126709, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.866853654384613, + "num_tokens": 342642613.0, + "step": 9759 + }, + { + "epoch": 1.8124419684308264, + "grad_norm": 1.4677574634552002, + "learning_rate": 1e-06, + "loss": 0.384, + "mean_token_accuracy": 0.8681296110153198, + "num_tokens": 342679725.0, + "step": 9760 + }, + { + "epoch": 1.8126276694521821, + "grad_norm": 1.4913462400436401, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8708956837654114, + "num_tokens": 342715875.0, + "step": 9761 + }, + { + "epoch": 1.8128133704735376, + "grad_norm": 1.4953439235687256, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8650345802307129, + "num_tokens": 342753145.0, + "step": 9762 + }, + { + "epoch": 1.8129990714948931, + "grad_norm": 1.3975940942764282, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8767833709716797, + "num_tokens": 342787994.0, + "step": 9763 + }, + { + "epoch": 1.8131847725162489, + "grad_norm": 1.5496618747711182, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8758748769760132, + "num_tokens": 342820859.0, + "step": 9764 + }, + { + "epoch": 1.8133704735376046, + "grad_norm": 1.3743488788604736, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8817018270492554, + "num_tokens": 342857465.0, + "step": 9765 + }, + { + "epoch": 1.8135561745589601, + "grad_norm": 1.7062623500823975, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8859968185424805, + "num_tokens": 342887773.0, + "step": 9766 + }, + { + "epoch": 1.8137418755803156, + "grad_norm": 1.4923627376556396, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.87165766954422, + "num_tokens": 342923516.0, + "step": 9767 + }, + { + "epoch": 1.8139275766016714, + "grad_norm": 1.5061841011047363, + "learning_rate": 1e-06, + "loss": 0.3844, + "mean_token_accuracy": 0.8701572418212891, + "num_tokens": 342965228.0, + "step": 9768 + }, + { + "epoch": 1.8141132776230269, + "grad_norm": 1.520932674407959, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8720518350601196, + "num_tokens": 342999385.0, + "step": 9769 + }, + { + "epoch": 1.8142989786443824, + "grad_norm": 1.5264699459075928, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8738483190536499, + "num_tokens": 343030254.0, + "step": 9770 + }, + { + "epoch": 1.8144846796657381, + "grad_norm": 1.54618501663208, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8865498304367065, + "num_tokens": 343060789.0, + "step": 9771 + }, + { + "epoch": 1.8146703806870939, + "grad_norm": 1.8161637783050537, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8751152753829956, + "num_tokens": 343095650.0, + "step": 9772 + }, + { + "epoch": 1.8148560817084494, + "grad_norm": 1.4443328380584717, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.887876570224762, + "num_tokens": 343132089.0, + "step": 9773 + }, + { + "epoch": 1.8150417827298049, + "grad_norm": 1.479068398475647, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8621822595596313, + "num_tokens": 343170558.0, + "step": 9774 + }, + { + "epoch": 1.8152274837511606, + "grad_norm": 1.5203086137771606, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8916490077972412, + "num_tokens": 343201288.0, + "step": 9775 + }, + { + "epoch": 1.8154131847725163, + "grad_norm": 1.5746675729751587, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.88157719373703, + "num_tokens": 343236308.0, + "step": 9776 + }, + { + "epoch": 1.8155988857938719, + "grad_norm": 1.5911809206008911, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8812755346298218, + "num_tokens": 343268418.0, + "step": 9777 + }, + { + "epoch": 1.8157845868152274, + "grad_norm": 1.4873002767562866, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8796089887619019, + "num_tokens": 343301430.0, + "step": 9778 + }, + { + "epoch": 1.815970287836583, + "grad_norm": 1.5740013122558594, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.869907021522522, + "num_tokens": 343333589.0, + "step": 9779 + }, + { + "epoch": 1.8161559888579388, + "grad_norm": 1.4262940883636475, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8829606771469116, + "num_tokens": 343370246.0, + "step": 9780 + }, + { + "epoch": 1.8163416898792943, + "grad_norm": 1.4063574075698853, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8886115550994873, + "num_tokens": 343406728.0, + "step": 9781 + }, + { + "epoch": 1.8165273909006499, + "grad_norm": 1.3412598371505737, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8893110752105713, + "num_tokens": 343449893.0, + "step": 9782 + }, + { + "epoch": 1.8167130919220056, + "grad_norm": 1.47368323802948, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8891478776931763, + "num_tokens": 343483940.0, + "step": 9783 + }, + { + "epoch": 1.8168987929433613, + "grad_norm": 1.4605005979537964, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8727740049362183, + "num_tokens": 343519939.0, + "step": 9784 + }, + { + "epoch": 1.8170844939647168, + "grad_norm": 1.472678542137146, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.890066921710968, + "num_tokens": 343553959.0, + "step": 9785 + }, + { + "epoch": 1.8172701949860723, + "grad_norm": 1.58040452003479, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8682096004486084, + "num_tokens": 343586770.0, + "step": 9786 + }, + { + "epoch": 1.817455896007428, + "grad_norm": 1.4892125129699707, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8733108043670654, + "num_tokens": 343620058.0, + "step": 9787 + }, + { + "epoch": 1.8176415970287838, + "grad_norm": 1.5156766176223755, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8804571628570557, + "num_tokens": 343654906.0, + "step": 9788 + }, + { + "epoch": 1.8178272980501393, + "grad_norm": 1.7061783075332642, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8684546947479248, + "num_tokens": 343687053.0, + "step": 9789 + }, + { + "epoch": 1.8180129990714948, + "grad_norm": 1.5292757749557495, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8716967105865479, + "num_tokens": 343722786.0, + "step": 9790 + }, + { + "epoch": 1.8181987000928506, + "grad_norm": 1.455775260925293, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8779312372207642, + "num_tokens": 343758926.0, + "step": 9791 + }, + { + "epoch": 1.8183844011142063, + "grad_norm": 1.4642953872680664, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8846471309661865, + "num_tokens": 343789884.0, + "step": 9792 + }, + { + "epoch": 1.8185701021355616, + "grad_norm": 1.4011895656585693, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8811805248260498, + "num_tokens": 343827204.0, + "step": 9793 + }, + { + "epoch": 1.8187558031569173, + "grad_norm": 1.5132750272750854, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8856109380722046, + "num_tokens": 343854800.0, + "step": 9794 + }, + { + "epoch": 1.818941504178273, + "grad_norm": 1.5163623094558716, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8887695670127869, + "num_tokens": 343882736.0, + "step": 9795 + }, + { + "epoch": 1.8191272051996286, + "grad_norm": 1.385736346244812, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8745837807655334, + "num_tokens": 343923626.0, + "step": 9796 + }, + { + "epoch": 1.819312906220984, + "grad_norm": 1.493609070777893, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8807187080383301, + "num_tokens": 343958253.0, + "step": 9797 + }, + { + "epoch": 1.8194986072423398, + "grad_norm": 1.4657511711120605, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.873698353767395, + "num_tokens": 343991420.0, + "step": 9798 + }, + { + "epoch": 1.8196843082636955, + "grad_norm": 1.3857136964797974, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8876937031745911, + "num_tokens": 344029372.0, + "step": 9799 + }, + { + "epoch": 1.819870009285051, + "grad_norm": 1.4189388751983643, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8831033110618591, + "num_tokens": 344067721.0, + "step": 9800 + }, + { + "epoch": 1.8200557103064066, + "grad_norm": 1.3496050834655762, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8766042590141296, + "num_tokens": 344104925.0, + "step": 9801 + }, + { + "epoch": 1.8202414113277623, + "grad_norm": 1.48677659034729, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8696367144584656, + "num_tokens": 344141399.0, + "step": 9802 + }, + { + "epoch": 1.820427112349118, + "grad_norm": 1.3890905380249023, + "learning_rate": 1e-06, + "loss": 0.3815, + "mean_token_accuracy": 0.8709364533424377, + "num_tokens": 344181799.0, + "step": 9803 + }, + { + "epoch": 1.8206128133704735, + "grad_norm": 1.3353195190429688, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8779829740524292, + "num_tokens": 344225198.0, + "step": 9804 + }, + { + "epoch": 1.820798514391829, + "grad_norm": 1.377245306968689, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8798161745071411, + "num_tokens": 344266047.0, + "step": 9805 + }, + { + "epoch": 1.8209842154131848, + "grad_norm": 1.4870965480804443, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8822016716003418, + "num_tokens": 344302660.0, + "step": 9806 + }, + { + "epoch": 1.8211699164345405, + "grad_norm": 1.3967902660369873, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8778402805328369, + "num_tokens": 344339253.0, + "step": 9807 + }, + { + "epoch": 1.821355617455896, + "grad_norm": 1.461929440498352, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8721153736114502, + "num_tokens": 344376792.0, + "step": 9808 + }, + { + "epoch": 1.8215413184772515, + "grad_norm": 1.3835211992263794, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8866590857505798, + "num_tokens": 344412404.0, + "step": 9809 + }, + { + "epoch": 1.8217270194986073, + "grad_norm": 1.528024673461914, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8894496560096741, + "num_tokens": 344442180.0, + "step": 9810 + }, + { + "epoch": 1.821912720519963, + "grad_norm": 1.4066370725631714, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8836877346038818, + "num_tokens": 344476885.0, + "step": 9811 + }, + { + "epoch": 1.8220984215413185, + "grad_norm": 1.471688985824585, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8765807151794434, + "num_tokens": 344515946.0, + "step": 9812 + }, + { + "epoch": 1.822284122562674, + "grad_norm": 1.4479613304138184, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8791999816894531, + "num_tokens": 344552823.0, + "step": 9813 + }, + { + "epoch": 1.8224698235840298, + "grad_norm": 1.480523705482483, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8794443011283875, + "num_tokens": 344584787.0, + "step": 9814 + }, + { + "epoch": 1.8226555246053855, + "grad_norm": 1.439060926437378, + "learning_rate": 1e-06, + "loss": 0.2909, + "mean_token_accuracy": 0.894766092300415, + "num_tokens": 344615574.0, + "step": 9815 + }, + { + "epoch": 1.822841225626741, + "grad_norm": 1.5119543075561523, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8869633078575134, + "num_tokens": 344648015.0, + "step": 9816 + }, + { + "epoch": 1.8230269266480965, + "grad_norm": 1.4409453868865967, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8785752058029175, + "num_tokens": 344683758.0, + "step": 9817 + }, + { + "epoch": 1.8232126276694522, + "grad_norm": 1.5028533935546875, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8718714118003845, + "num_tokens": 344717833.0, + "step": 9818 + }, + { + "epoch": 1.8233983286908078, + "grad_norm": 1.5242054462432861, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8727912902832031, + "num_tokens": 344755287.0, + "step": 9819 + }, + { + "epoch": 1.8235840297121633, + "grad_norm": 1.5878320932388306, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8753291368484497, + "num_tokens": 344787413.0, + "step": 9820 + }, + { + "epoch": 1.823769730733519, + "grad_norm": 1.4024895429611206, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8757948279380798, + "num_tokens": 344827407.0, + "step": 9821 + }, + { + "epoch": 1.8239554317548747, + "grad_norm": 1.3585963249206543, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8918737769126892, + "num_tokens": 344864854.0, + "step": 9822 + }, + { + "epoch": 1.8241411327762302, + "grad_norm": 1.4477931261062622, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8850571513175964, + "num_tokens": 344902136.0, + "step": 9823 + }, + { + "epoch": 1.8243268337975858, + "grad_norm": 1.4743244647979736, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8866527080535889, + "num_tokens": 344936810.0, + "step": 9824 + }, + { + "epoch": 1.8245125348189415, + "grad_norm": 1.5351812839508057, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8770682215690613, + "num_tokens": 344966496.0, + "step": 9825 + }, + { + "epoch": 1.8246982358402972, + "grad_norm": 1.424666404724121, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8684563636779785, + "num_tokens": 345006781.0, + "step": 9826 + }, + { + "epoch": 1.8248839368616527, + "grad_norm": 1.6023281812667847, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8814563155174255, + "num_tokens": 345036144.0, + "step": 9827 + }, + { + "epoch": 1.8250696378830082, + "grad_norm": 1.576281189918518, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8895118236541748, + "num_tokens": 345065948.0, + "step": 9828 + }, + { + "epoch": 1.825255338904364, + "grad_norm": 1.4970780611038208, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.865827202796936, + "num_tokens": 345102071.0, + "step": 9829 + }, + { + "epoch": 1.8254410399257197, + "grad_norm": 1.4597363471984863, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8767971992492676, + "num_tokens": 345135318.0, + "step": 9830 + }, + { + "epoch": 1.8256267409470752, + "grad_norm": 1.4097645282745361, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8799242973327637, + "num_tokens": 345172197.0, + "step": 9831 + }, + { + "epoch": 1.8258124419684307, + "grad_norm": 1.4425995349884033, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8685234785079956, + "num_tokens": 345211509.0, + "step": 9832 + }, + { + "epoch": 1.8259981429897865, + "grad_norm": 1.2980512380599976, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8755119442939758, + "num_tokens": 345256347.0, + "step": 9833 + }, + { + "epoch": 1.8261838440111422, + "grad_norm": 1.6092878580093384, + "learning_rate": 1e-06, + "loss": 0.3886, + "mean_token_accuracy": 0.867703914642334, + "num_tokens": 345288154.0, + "step": 9834 + }, + { + "epoch": 1.8263695450324977, + "grad_norm": 1.5512359142303467, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8712933659553528, + "num_tokens": 345328433.0, + "step": 9835 + }, + { + "epoch": 1.8265552460538532, + "grad_norm": 1.7612563371658325, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8608628511428833, + "num_tokens": 345359950.0, + "step": 9836 + }, + { + "epoch": 1.826740947075209, + "grad_norm": 1.583086609840393, + "learning_rate": 1e-06, + "loss": 0.4032, + "mean_token_accuracy": 0.8634358644485474, + "num_tokens": 345392397.0, + "step": 9837 + }, + { + "epoch": 1.8269266480965647, + "grad_norm": 1.5183203220367432, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8829518556594849, + "num_tokens": 345424830.0, + "step": 9838 + }, + { + "epoch": 1.8271123491179202, + "grad_norm": 1.5057612657546997, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8889355063438416, + "num_tokens": 345461992.0, + "step": 9839 + }, + { + "epoch": 1.8272980501392757, + "grad_norm": 1.461918592453003, + "learning_rate": 1e-06, + "loss": 0.3952, + "mean_token_accuracy": 0.865970253944397, + "num_tokens": 345500177.0, + "step": 9840 + }, + { + "epoch": 1.8274837511606314, + "grad_norm": 1.4185596704483032, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8738028407096863, + "num_tokens": 345538367.0, + "step": 9841 + }, + { + "epoch": 1.827669452181987, + "grad_norm": 1.5782198905944824, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8620781302452087, + "num_tokens": 345575044.0, + "step": 9842 + }, + { + "epoch": 1.8278551532033425, + "grad_norm": 1.4124500751495361, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8787890672683716, + "num_tokens": 345611552.0, + "step": 9843 + }, + { + "epoch": 1.8280408542246982, + "grad_norm": 1.355387806892395, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8767956495285034, + "num_tokens": 345650018.0, + "step": 9844 + }, + { + "epoch": 1.828226555246054, + "grad_norm": 1.47592031955719, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8725094795227051, + "num_tokens": 345688371.0, + "step": 9845 + }, + { + "epoch": 1.8284122562674094, + "grad_norm": 1.5045074224472046, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8696682453155518, + "num_tokens": 345723209.0, + "step": 9846 + }, + { + "epoch": 1.828597957288765, + "grad_norm": 1.4187211990356445, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.864665150642395, + "num_tokens": 345762612.0, + "step": 9847 + }, + { + "epoch": 1.8287836583101207, + "grad_norm": 1.4741672277450562, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8646966814994812, + "num_tokens": 345798787.0, + "step": 9848 + }, + { + "epoch": 1.8289693593314764, + "grad_norm": 1.4918924570083618, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8839066624641418, + "num_tokens": 345835380.0, + "step": 9849 + }, + { + "epoch": 1.829155060352832, + "grad_norm": 1.461974024772644, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8785800933837891, + "num_tokens": 345872801.0, + "step": 9850 + }, + { + "epoch": 1.8293407613741874, + "grad_norm": 1.7489551305770874, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8714589476585388, + "num_tokens": 345900800.0, + "step": 9851 + }, + { + "epoch": 1.8295264623955432, + "grad_norm": 1.4947123527526855, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8798548579216003, + "num_tokens": 345933786.0, + "step": 9852 + }, + { + "epoch": 1.829712163416899, + "grad_norm": 1.5411579608917236, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8658818006515503, + "num_tokens": 345966335.0, + "step": 9853 + }, + { + "epoch": 1.8298978644382544, + "grad_norm": 1.481037974357605, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8720791339874268, + "num_tokens": 346004738.0, + "step": 9854 + }, + { + "epoch": 1.83008356545961, + "grad_norm": 1.4708514213562012, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8768233060836792, + "num_tokens": 346038156.0, + "step": 9855 + }, + { + "epoch": 1.8302692664809657, + "grad_norm": 1.446858286857605, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8653075098991394, + "num_tokens": 346074859.0, + "step": 9856 + }, + { + "epoch": 1.8304549675023214, + "grad_norm": 1.5523481369018555, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8946235179901123, + "num_tokens": 346105236.0, + "step": 9857 + }, + { + "epoch": 1.830640668523677, + "grad_norm": 1.3712681531906128, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8825911283493042, + "num_tokens": 346140047.0, + "step": 9858 + }, + { + "epoch": 1.8308263695450324, + "grad_norm": 1.383638858795166, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8902636170387268, + "num_tokens": 346173681.0, + "step": 9859 + }, + { + "epoch": 1.8310120705663882, + "grad_norm": 1.6462236642837524, + "learning_rate": 1e-06, + "loss": 0.4246, + "mean_token_accuracy": 0.8596661686897278, + "num_tokens": 346203112.0, + "step": 9860 + }, + { + "epoch": 1.8311977715877439, + "grad_norm": 1.4961577653884888, + "learning_rate": 1e-06, + "loss": 0.4033, + "mean_token_accuracy": 0.8600059151649475, + "num_tokens": 346238664.0, + "step": 9861 + }, + { + "epoch": 1.8313834726090994, + "grad_norm": 1.65534245967865, + "learning_rate": 1e-06, + "loss": 0.3911, + "mean_token_accuracy": 0.8693854808807373, + "num_tokens": 346265526.0, + "step": 9862 + }, + { + "epoch": 1.831569173630455, + "grad_norm": 1.4909532070159912, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8826191425323486, + "num_tokens": 346296709.0, + "step": 9863 + }, + { + "epoch": 1.8317548746518106, + "grad_norm": 1.408586025238037, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8845610618591309, + "num_tokens": 346333652.0, + "step": 9864 + }, + { + "epoch": 1.8319405756731661, + "grad_norm": 1.609248161315918, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8782489895820618, + "num_tokens": 346363516.0, + "step": 9865 + }, + { + "epoch": 1.8321262766945217, + "grad_norm": 1.39287531375885, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8852889537811279, + "num_tokens": 346401001.0, + "step": 9866 + }, + { + "epoch": 1.8323119777158774, + "grad_norm": 1.4815939664840698, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8790717720985413, + "num_tokens": 346435275.0, + "step": 9867 + }, + { + "epoch": 1.8324976787372331, + "grad_norm": 1.5114991664886475, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8734747767448425, + "num_tokens": 346469728.0, + "step": 9868 + }, + { + "epoch": 1.8326833797585886, + "grad_norm": 1.3449954986572266, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.8860741853713989, + "num_tokens": 346506034.0, + "step": 9869 + }, + { + "epoch": 1.8328690807799441, + "grad_norm": 1.4714323282241821, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8738393187522888, + "num_tokens": 346542595.0, + "step": 9870 + }, + { + "epoch": 1.8330547818012999, + "grad_norm": 1.6288710832595825, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.873685359954834, + "num_tokens": 346572596.0, + "step": 9871 + }, + { + "epoch": 1.8332404828226556, + "grad_norm": 1.6752064228057861, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8671869039535522, + "num_tokens": 346605460.0, + "step": 9872 + }, + { + "epoch": 1.8334261838440111, + "grad_norm": 1.4825892448425293, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8891699314117432, + "num_tokens": 346636825.0, + "step": 9873 + }, + { + "epoch": 1.8336118848653666, + "grad_norm": 1.5441347360610962, + "learning_rate": 1e-06, + "loss": 0.4138, + "mean_token_accuracy": 0.8575429916381836, + "num_tokens": 346675692.0, + "step": 9874 + }, + { + "epoch": 1.8337975858867224, + "grad_norm": 1.5222008228302002, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8777332305908203, + "num_tokens": 346710073.0, + "step": 9875 + }, + { + "epoch": 1.833983286908078, + "grad_norm": 1.4683773517608643, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.884605348110199, + "num_tokens": 346741221.0, + "step": 9876 + }, + { + "epoch": 1.8341689879294336, + "grad_norm": 1.7353113889694214, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8533244132995605, + "num_tokens": 346770397.0, + "step": 9877 + }, + { + "epoch": 1.8343546889507891, + "grad_norm": 1.39894700050354, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8793838620185852, + "num_tokens": 346810047.0, + "step": 9878 + }, + { + "epoch": 1.8345403899721449, + "grad_norm": 1.5331456661224365, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8829271793365479, + "num_tokens": 346846801.0, + "step": 9879 + }, + { + "epoch": 1.8347260909935006, + "grad_norm": 1.4857006072998047, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8798294067382812, + "num_tokens": 346884088.0, + "step": 9880 + }, + { + "epoch": 1.834911792014856, + "grad_norm": 1.5872993469238281, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8770302534103394, + "num_tokens": 346916119.0, + "step": 9881 + }, + { + "epoch": 1.8350974930362116, + "grad_norm": 1.5622256994247437, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8817689418792725, + "num_tokens": 346946313.0, + "step": 9882 + }, + { + "epoch": 1.8352831940575673, + "grad_norm": 1.4164341688156128, + "learning_rate": 1e-06, + "loss": 0.4153, + "mean_token_accuracy": 0.8561853170394897, + "num_tokens": 346989408.0, + "step": 9883 + }, + { + "epoch": 1.835468895078923, + "grad_norm": 1.4763270616531372, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8857429027557373, + "num_tokens": 347020993.0, + "step": 9884 + }, + { + "epoch": 1.8356545961002786, + "grad_norm": 1.716130018234253, + "learning_rate": 1e-06, + "loss": 0.4005, + "mean_token_accuracy": 0.8677061796188354, + "num_tokens": 347051979.0, + "step": 9885 + }, + { + "epoch": 1.835840297121634, + "grad_norm": 1.4474560022354126, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8750622272491455, + "num_tokens": 347088774.0, + "step": 9886 + }, + { + "epoch": 1.8360259981429898, + "grad_norm": 1.4425123929977417, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8777530193328857, + "num_tokens": 347127020.0, + "step": 9887 + }, + { + "epoch": 1.8362116991643456, + "grad_norm": 1.4586888551712036, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8796615600585938, + "num_tokens": 347163973.0, + "step": 9888 + }, + { + "epoch": 1.8363974001857009, + "grad_norm": 1.586899757385254, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8772488236427307, + "num_tokens": 347198376.0, + "step": 9889 + }, + { + "epoch": 1.8365831012070566, + "grad_norm": 1.4894671440124512, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8823050260543823, + "num_tokens": 347232694.0, + "step": 9890 + }, + { + "epoch": 1.8367688022284123, + "grad_norm": 1.5621647834777832, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.874479353427887, + "num_tokens": 347269613.0, + "step": 9891 + }, + { + "epoch": 1.8369545032497678, + "grad_norm": 1.4474681615829468, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8943442106246948, + "num_tokens": 347303030.0, + "step": 9892 + }, + { + "epoch": 1.8371402042711233, + "grad_norm": 1.5556061267852783, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8846258521080017, + "num_tokens": 347332145.0, + "step": 9893 + }, + { + "epoch": 1.837325905292479, + "grad_norm": 1.670199990272522, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8861863613128662, + "num_tokens": 347359604.0, + "step": 9894 + }, + { + "epoch": 1.8375116063138348, + "grad_norm": 1.5748131275177002, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8785120844841003, + "num_tokens": 347391986.0, + "step": 9895 + }, + { + "epoch": 1.8376973073351903, + "grad_norm": 1.4823319911956787, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8899006843566895, + "num_tokens": 347426460.0, + "step": 9896 + }, + { + "epoch": 1.8378830083565458, + "grad_norm": 1.7528390884399414, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8874028325080872, + "num_tokens": 347459156.0, + "step": 9897 + }, + { + "epoch": 1.8380687093779016, + "grad_norm": 1.617020606994629, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8773012757301331, + "num_tokens": 347492586.0, + "step": 9898 + }, + { + "epoch": 1.8382544103992573, + "grad_norm": 1.4619942903518677, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.879305362701416, + "num_tokens": 347532582.0, + "step": 9899 + }, + { + "epoch": 1.8384401114206128, + "grad_norm": 1.516048550605774, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8692725896835327, + "num_tokens": 347571418.0, + "step": 9900 + }, + { + "epoch": 1.8386258124419683, + "grad_norm": 1.3985621929168701, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8951441645622253, + "num_tokens": 347604880.0, + "step": 9901 + }, + { + "epoch": 1.838811513463324, + "grad_norm": 1.4124866724014282, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8845810890197754, + "num_tokens": 347641409.0, + "step": 9902 + }, + { + "epoch": 1.8389972144846798, + "grad_norm": 1.5414044857025146, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.87230384349823, + "num_tokens": 347671521.0, + "step": 9903 + }, + { + "epoch": 1.8391829155060353, + "grad_norm": 1.5194461345672607, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8775440454483032, + "num_tokens": 347704738.0, + "step": 9904 + }, + { + "epoch": 1.8393686165273908, + "grad_norm": 1.4979023933410645, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8726127743721008, + "num_tokens": 347740267.0, + "step": 9905 + }, + { + "epoch": 1.8395543175487465, + "grad_norm": 1.4032500982284546, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8766383528709412, + "num_tokens": 347780707.0, + "step": 9906 + }, + { + "epoch": 1.8397400185701023, + "grad_norm": 1.5717202425003052, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8704261183738708, + "num_tokens": 347816462.0, + "step": 9907 + }, + { + "epoch": 1.8399257195914578, + "grad_norm": 1.4993784427642822, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8882868885993958, + "num_tokens": 347848355.0, + "step": 9908 + }, + { + "epoch": 1.8401114206128133, + "grad_norm": 1.3648090362548828, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8749603629112244, + "num_tokens": 347885401.0, + "step": 9909 + }, + { + "epoch": 1.840297121634169, + "grad_norm": 1.5194138288497925, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8824481964111328, + "num_tokens": 347917614.0, + "step": 9910 + }, + { + "epoch": 1.8404828226555248, + "grad_norm": 1.5444563627243042, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8840153813362122, + "num_tokens": 347948446.0, + "step": 9911 + }, + { + "epoch": 1.8406685236768803, + "grad_norm": 1.5604921579360962, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8731688261032104, + "num_tokens": 347980443.0, + "step": 9912 + }, + { + "epoch": 1.8408542246982358, + "grad_norm": 1.4517558813095093, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8798198699951172, + "num_tokens": 348016215.0, + "step": 9913 + }, + { + "epoch": 1.8410399257195915, + "grad_norm": 1.3978999853134155, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8854344487190247, + "num_tokens": 348049837.0, + "step": 9914 + }, + { + "epoch": 1.841225626740947, + "grad_norm": 1.4508894681930542, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8858920931816101, + "num_tokens": 348087318.0, + "step": 9915 + }, + { + "epoch": 1.8414113277623025, + "grad_norm": 1.4337321519851685, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8885979652404785, + "num_tokens": 348120897.0, + "step": 9916 + }, + { + "epoch": 1.8415970287836583, + "grad_norm": 1.5351711511611938, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8712549209594727, + "num_tokens": 348152050.0, + "step": 9917 + }, + { + "epoch": 1.841782729805014, + "grad_norm": 1.4904917478561401, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8909640312194824, + "num_tokens": 348186468.0, + "step": 9918 + }, + { + "epoch": 1.8419684308263695, + "grad_norm": 1.6178791522979736, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8651829361915588, + "num_tokens": 348218169.0, + "step": 9919 + }, + { + "epoch": 1.842154131847725, + "grad_norm": 1.3822964429855347, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8773508071899414, + "num_tokens": 348257890.0, + "step": 9920 + }, + { + "epoch": 1.8423398328690808, + "grad_norm": 1.538068413734436, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8764268159866333, + "num_tokens": 348290169.0, + "step": 9921 + }, + { + "epoch": 1.8425255338904365, + "grad_norm": 1.3565360307693481, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8689010143280029, + "num_tokens": 348336822.0, + "step": 9922 + }, + { + "epoch": 1.842711234911792, + "grad_norm": 1.3417609930038452, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8778010606765747, + "num_tokens": 348378939.0, + "step": 9923 + }, + { + "epoch": 1.8428969359331475, + "grad_norm": 1.4406927824020386, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8820866942405701, + "num_tokens": 348412594.0, + "step": 9924 + }, + { + "epoch": 1.8430826369545033, + "grad_norm": 1.4207671880722046, + "learning_rate": 1e-06, + "loss": 0.3601, + "mean_token_accuracy": 0.8749051690101624, + "num_tokens": 348450809.0, + "step": 9925 + }, + { + "epoch": 1.843268337975859, + "grad_norm": 1.348730444908142, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8925014734268188, + "num_tokens": 348488975.0, + "step": 9926 + }, + { + "epoch": 1.8434540389972145, + "grad_norm": 1.3844821453094482, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.875377893447876, + "num_tokens": 348529261.0, + "step": 9927 + }, + { + "epoch": 1.84363974001857, + "grad_norm": 1.4471389055252075, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8703500032424927, + "num_tokens": 348564503.0, + "step": 9928 + }, + { + "epoch": 1.8438254410399257, + "grad_norm": 1.2717132568359375, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8884379863739014, + "num_tokens": 348605801.0, + "step": 9929 + }, + { + "epoch": 1.8440111420612815, + "grad_norm": 1.3809666633605957, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8863538503646851, + "num_tokens": 348644045.0, + "step": 9930 + }, + { + "epoch": 1.844196843082637, + "grad_norm": 1.5026469230651855, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8738583922386169, + "num_tokens": 348677157.0, + "step": 9931 + }, + { + "epoch": 1.8443825441039925, + "grad_norm": 1.470992088317871, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8806009888648987, + "num_tokens": 348712969.0, + "step": 9932 + }, + { + "epoch": 1.8445682451253482, + "grad_norm": 1.336861491203308, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8847122192382812, + "num_tokens": 348754585.0, + "step": 9933 + }, + { + "epoch": 1.844753946146704, + "grad_norm": 1.4922237396240234, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8799785375595093, + "num_tokens": 348790333.0, + "step": 9934 + }, + { + "epoch": 1.8449396471680595, + "grad_norm": 1.480596661567688, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.890641987323761, + "num_tokens": 348826101.0, + "step": 9935 + }, + { + "epoch": 1.845125348189415, + "grad_norm": 1.4547163248062134, + "learning_rate": 1e-06, + "loss": 0.3989, + "mean_token_accuracy": 0.864431619644165, + "num_tokens": 348868889.0, + "step": 9936 + }, + { + "epoch": 1.8453110492107707, + "grad_norm": 1.3005443811416626, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8828105926513672, + "num_tokens": 348913125.0, + "step": 9937 + }, + { + "epoch": 1.8454967502321262, + "grad_norm": 1.528576135635376, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8968557119369507, + "num_tokens": 348940949.0, + "step": 9938 + }, + { + "epoch": 1.8456824512534817, + "grad_norm": 1.3441635370254517, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8912410140037537, + "num_tokens": 348976194.0, + "step": 9939 + }, + { + "epoch": 1.8458681522748375, + "grad_norm": 1.5104190111160278, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8868688941001892, + "num_tokens": 349006748.0, + "step": 9940 + }, + { + "epoch": 1.8460538532961932, + "grad_norm": 1.6228514909744263, + "learning_rate": 1e-06, + "loss": 0.3991, + "mean_token_accuracy": 0.8614290952682495, + "num_tokens": 349035646.0, + "step": 9941 + }, + { + "epoch": 1.8462395543175487, + "grad_norm": 1.477252721786499, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8738908171653748, + "num_tokens": 349075156.0, + "step": 9942 + }, + { + "epoch": 1.8464252553389042, + "grad_norm": 1.507092833518982, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8768997192382812, + "num_tokens": 349115489.0, + "step": 9943 + }, + { + "epoch": 1.84661095636026, + "grad_norm": 1.4960548877716064, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8806859254837036, + "num_tokens": 349150590.0, + "step": 9944 + }, + { + "epoch": 1.8467966573816157, + "grad_norm": 1.4265509843826294, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8743480443954468, + "num_tokens": 349191873.0, + "step": 9945 + }, + { + "epoch": 1.8469823584029712, + "grad_norm": 1.507484793663025, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8883059024810791, + "num_tokens": 349222614.0, + "step": 9946 + }, + { + "epoch": 1.8471680594243267, + "grad_norm": 1.4666351079940796, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.882158100605011, + "num_tokens": 349262746.0, + "step": 9947 + }, + { + "epoch": 1.8473537604456824, + "grad_norm": 1.4970630407333374, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8808413147926331, + "num_tokens": 349298519.0, + "step": 9948 + }, + { + "epoch": 1.8475394614670382, + "grad_norm": 1.4960986375808716, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8685017228126526, + "num_tokens": 349336818.0, + "step": 9949 + }, + { + "epoch": 1.8477251624883937, + "grad_norm": 1.53229820728302, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8786695599555969, + "num_tokens": 349374140.0, + "step": 9950 + }, + { + "epoch": 1.8479108635097492, + "grad_norm": 1.48627507686615, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8813933730125427, + "num_tokens": 349409679.0, + "step": 9951 + }, + { + "epoch": 1.848096564531105, + "grad_norm": 1.4526677131652832, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8892451524734497, + "num_tokens": 349443836.0, + "step": 9952 + }, + { + "epoch": 1.8482822655524607, + "grad_norm": 1.495352864265442, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8756297826766968, + "num_tokens": 349479246.0, + "step": 9953 + }, + { + "epoch": 1.8484679665738162, + "grad_norm": 1.6077744960784912, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8832283020019531, + "num_tokens": 349506704.0, + "step": 9954 + }, + { + "epoch": 1.8486536675951717, + "grad_norm": 1.3536475896835327, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8803330063819885, + "num_tokens": 349547659.0, + "step": 9955 + }, + { + "epoch": 1.8488393686165274, + "grad_norm": 1.5342600345611572, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8858661651611328, + "num_tokens": 349578452.0, + "step": 9956 + }, + { + "epoch": 1.8490250696378832, + "grad_norm": 1.4152535200119019, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8848155736923218, + "num_tokens": 349617409.0, + "step": 9957 + }, + { + "epoch": 1.8492107706592387, + "grad_norm": 1.4399088621139526, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8735771775245667, + "num_tokens": 349656143.0, + "step": 9958 + }, + { + "epoch": 1.8493964716805942, + "grad_norm": 1.4623210430145264, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8854544758796692, + "num_tokens": 349688117.0, + "step": 9959 + }, + { + "epoch": 1.84958217270195, + "grad_norm": 1.6074320077896118, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8835819959640503, + "num_tokens": 349719799.0, + "step": 9960 + }, + { + "epoch": 1.8497678737233056, + "grad_norm": 1.515104055404663, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8775569200515747, + "num_tokens": 349755817.0, + "step": 9961 + }, + { + "epoch": 1.849953574744661, + "grad_norm": 1.4377142190933228, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8821262717247009, + "num_tokens": 349792645.0, + "step": 9962 + }, + { + "epoch": 1.8501392757660167, + "grad_norm": 1.5280152559280396, + "learning_rate": 1e-06, + "loss": 0.3661, + "mean_token_accuracy": 0.8705314993858337, + "num_tokens": 349828352.0, + "step": 9963 + }, + { + "epoch": 1.8503249767873724, + "grad_norm": 1.5470027923583984, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8806777000427246, + "num_tokens": 349864564.0, + "step": 9964 + }, + { + "epoch": 1.850510677808728, + "grad_norm": 1.4696402549743652, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8803149461746216, + "num_tokens": 349904473.0, + "step": 9965 + }, + { + "epoch": 1.8506963788300834, + "grad_norm": 1.3988386392593384, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8829139471054077, + "num_tokens": 349944913.0, + "step": 9966 + }, + { + "epoch": 1.8508820798514392, + "grad_norm": 1.5840191841125488, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8772480487823486, + "num_tokens": 349976349.0, + "step": 9967 + }, + { + "epoch": 1.8510677808727949, + "grad_norm": 1.4484208822250366, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8747856020927429, + "num_tokens": 350012978.0, + "step": 9968 + }, + { + "epoch": 1.8512534818941504, + "grad_norm": 1.3548040390014648, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8907061815261841, + "num_tokens": 350051196.0, + "step": 9969 + }, + { + "epoch": 1.851439182915506, + "grad_norm": 1.4572352170944214, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8877173662185669, + "num_tokens": 350084981.0, + "step": 9970 + }, + { + "epoch": 1.8516248839368616, + "grad_norm": 1.4278484582901, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8953091502189636, + "num_tokens": 350120133.0, + "step": 9971 + }, + { + "epoch": 1.8518105849582174, + "grad_norm": 1.432754635810852, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.868789792060852, + "num_tokens": 350159041.0, + "step": 9972 + }, + { + "epoch": 1.8519962859795729, + "grad_norm": 1.4626637697219849, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8779324293136597, + "num_tokens": 350194703.0, + "step": 9973 + }, + { + "epoch": 1.8521819870009284, + "grad_norm": 1.5342798233032227, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8827120065689087, + "num_tokens": 350225504.0, + "step": 9974 + }, + { + "epoch": 1.8523676880222841, + "grad_norm": 1.5227514505386353, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8723467588424683, + "num_tokens": 350259572.0, + "step": 9975 + }, + { + "epoch": 1.8525533890436399, + "grad_norm": 1.3991140127182007, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8883776068687439, + "num_tokens": 350297282.0, + "step": 9976 + }, + { + "epoch": 1.8527390900649954, + "grad_norm": 1.4244831800460815, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8965492248535156, + "num_tokens": 350329881.0, + "step": 9977 + }, + { + "epoch": 1.8529247910863509, + "grad_norm": 1.4147274494171143, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8726696968078613, + "num_tokens": 350366437.0, + "step": 9978 + }, + { + "epoch": 1.8531104921077066, + "grad_norm": 1.3904517889022827, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8816566467285156, + "num_tokens": 350404887.0, + "step": 9979 + }, + { + "epoch": 1.8532961931290624, + "grad_norm": 1.462617039680481, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8698021173477173, + "num_tokens": 350441666.0, + "step": 9980 + }, + { + "epoch": 1.8534818941504179, + "grad_norm": 1.5015106201171875, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8620558381080627, + "num_tokens": 350479681.0, + "step": 9981 + }, + { + "epoch": 1.8536675951717734, + "grad_norm": 1.6692945957183838, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8665812015533447, + "num_tokens": 350511193.0, + "step": 9982 + }, + { + "epoch": 1.853853296193129, + "grad_norm": 1.51924467086792, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8744127750396729, + "num_tokens": 350544415.0, + "step": 9983 + }, + { + "epoch": 1.8540389972144848, + "grad_norm": 1.5431700944900513, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8760668039321899, + "num_tokens": 350577829.0, + "step": 9984 + }, + { + "epoch": 1.8542246982358404, + "grad_norm": 1.2936253547668457, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8902174234390259, + "num_tokens": 350616624.0, + "step": 9985 + }, + { + "epoch": 1.8544103992571959, + "grad_norm": 1.4299248456954956, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.881528377532959, + "num_tokens": 350654815.0, + "step": 9986 + }, + { + "epoch": 1.8545961002785516, + "grad_norm": 1.4897031784057617, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8685928583145142, + "num_tokens": 350692443.0, + "step": 9987 + }, + { + "epoch": 1.854781801299907, + "grad_norm": 1.4691333770751953, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8718801736831665, + "num_tokens": 350727678.0, + "step": 9988 + }, + { + "epoch": 1.8549675023212626, + "grad_norm": 1.6236848831176758, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8763838410377502, + "num_tokens": 350757057.0, + "step": 9989 + }, + { + "epoch": 1.8551532033426184, + "grad_norm": 1.4254428148269653, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.879679799079895, + "num_tokens": 350793375.0, + "step": 9990 + }, + { + "epoch": 1.855338904363974, + "grad_norm": 1.3861552476882935, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8741108179092407, + "num_tokens": 350831357.0, + "step": 9991 + }, + { + "epoch": 1.8555246053853296, + "grad_norm": 1.5322985649108887, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8805922269821167, + "num_tokens": 350865626.0, + "step": 9992 + }, + { + "epoch": 1.855710306406685, + "grad_norm": 1.4725022315979004, + "learning_rate": 1e-06, + "loss": 0.3721, + "mean_token_accuracy": 0.8709955215454102, + "num_tokens": 350901231.0, + "step": 9993 + }, + { + "epoch": 1.8558960074280408, + "grad_norm": 1.5800625085830688, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8747228384017944, + "num_tokens": 350934344.0, + "step": 9994 + }, + { + "epoch": 1.8560817084493966, + "grad_norm": 1.4664865732192993, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8659789562225342, + "num_tokens": 350972149.0, + "step": 9995 + }, + { + "epoch": 1.856267409470752, + "grad_norm": 1.3638628721237183, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8905965089797974, + "num_tokens": 351012983.0, + "step": 9996 + }, + { + "epoch": 1.8564531104921076, + "grad_norm": 1.6712788343429565, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.881171464920044, + "num_tokens": 351038595.0, + "step": 9997 + }, + { + "epoch": 1.8566388115134633, + "grad_norm": 1.3603971004486084, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8894416093826294, + "num_tokens": 351074009.0, + "step": 9998 + }, + { + "epoch": 1.856824512534819, + "grad_norm": 1.42001473903656, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8857235908508301, + "num_tokens": 351111864.0, + "step": 9999 + }, + { + "epoch": 1.8570102135561746, + "grad_norm": 1.613539218902588, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8828973174095154, + "num_tokens": 351141115.0, + "step": 10000 + }, + { + "epoch": 1.85719591457753, + "grad_norm": 1.903971552848816, + "learning_rate": 1e-06, + "loss": 0.4543, + "mean_token_accuracy": 0.8524843454360962, + "num_tokens": 351169808.0, + "step": 10001 + }, + { + "epoch": 1.8573816155988858, + "grad_norm": 1.3872612714767456, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8846687078475952, + "num_tokens": 351205779.0, + "step": 10002 + }, + { + "epoch": 1.8575673166202415, + "grad_norm": 1.6840510368347168, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8652656078338623, + "num_tokens": 351239050.0, + "step": 10003 + }, + { + "epoch": 1.857753017641597, + "grad_norm": 2.441035509109497, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8756001591682434, + "num_tokens": 351275625.0, + "step": 10004 + }, + { + "epoch": 1.8579387186629526, + "grad_norm": 1.4271163940429688, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8881596326828003, + "num_tokens": 351311884.0, + "step": 10005 + }, + { + "epoch": 1.8581244196843083, + "grad_norm": 1.5869896411895752, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8731324672698975, + "num_tokens": 351346008.0, + "step": 10006 + }, + { + "epoch": 1.858310120705664, + "grad_norm": 1.477089285850525, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8824203014373779, + "num_tokens": 351381156.0, + "step": 10007 + }, + { + "epoch": 1.8584958217270195, + "grad_norm": 1.3528698682785034, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8908663988113403, + "num_tokens": 351419758.0, + "step": 10008 + }, + { + "epoch": 1.858681522748375, + "grad_norm": 1.636953353881836, + "learning_rate": 1e-06, + "loss": 0.4107, + "mean_token_accuracy": 0.8611089587211609, + "num_tokens": 351450857.0, + "step": 10009 + }, + { + "epoch": 1.8588672237697308, + "grad_norm": 1.4535800218582153, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.876529335975647, + "num_tokens": 351487735.0, + "step": 10010 + }, + { + "epoch": 1.8590529247910863, + "grad_norm": 1.440348744392395, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8863531351089478, + "num_tokens": 351527321.0, + "step": 10011 + }, + { + "epoch": 1.8592386258124418, + "grad_norm": 1.5843406915664673, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.8772794604301453, + "num_tokens": 351558932.0, + "step": 10012 + }, + { + "epoch": 1.8594243268337975, + "grad_norm": 1.5893170833587646, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8840726613998413, + "num_tokens": 351594063.0, + "step": 10013 + }, + { + "epoch": 1.8596100278551533, + "grad_norm": 1.4219508171081543, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.874396026134491, + "num_tokens": 351634107.0, + "step": 10014 + }, + { + "epoch": 1.8597957288765088, + "grad_norm": 1.5179592370986938, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8742257356643677, + "num_tokens": 351667676.0, + "step": 10015 + }, + { + "epoch": 1.8599814298978643, + "grad_norm": 1.5536621809005737, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.8583681583404541, + "num_tokens": 351703343.0, + "step": 10016 + }, + { + "epoch": 1.86016713091922, + "grad_norm": 1.50851571559906, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.882345974445343, + "num_tokens": 351734222.0, + "step": 10017 + }, + { + "epoch": 1.8603528319405758, + "grad_norm": 1.3925405740737915, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8689178228378296, + "num_tokens": 351774183.0, + "step": 10018 + }, + { + "epoch": 1.8605385329619313, + "grad_norm": 1.626694917678833, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8677729368209839, + "num_tokens": 351804736.0, + "step": 10019 + }, + { + "epoch": 1.8607242339832868, + "grad_norm": 1.4250640869140625, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8813506364822388, + "num_tokens": 351840708.0, + "step": 10020 + }, + { + "epoch": 1.8609099350046425, + "grad_norm": 1.600500226020813, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8883354663848877, + "num_tokens": 351872254.0, + "step": 10021 + }, + { + "epoch": 1.8610956360259983, + "grad_norm": 1.4543707370758057, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8926417827606201, + "num_tokens": 351905851.0, + "step": 10022 + }, + { + "epoch": 1.8612813370473538, + "grad_norm": 1.4550631046295166, + "learning_rate": 1e-06, + "loss": 0.3791, + "mean_token_accuracy": 0.868690013885498, + "num_tokens": 351941029.0, + "step": 10023 + }, + { + "epoch": 1.8614670380687093, + "grad_norm": 1.4444369077682495, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8679213523864746, + "num_tokens": 351983735.0, + "step": 10024 + }, + { + "epoch": 1.861652739090065, + "grad_norm": 1.6806604862213135, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8799982070922852, + "num_tokens": 352009426.0, + "step": 10025 + }, + { + "epoch": 1.8618384401114207, + "grad_norm": 1.6552664041519165, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8810492753982544, + "num_tokens": 352042182.0, + "step": 10026 + }, + { + "epoch": 1.8620241411327763, + "grad_norm": 1.5988540649414062, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.8580262064933777, + "num_tokens": 352076424.0, + "step": 10027 + }, + { + "epoch": 1.8622098421541318, + "grad_norm": 1.3894821405410767, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8790282607078552, + "num_tokens": 352115904.0, + "step": 10028 + }, + { + "epoch": 1.8623955431754875, + "grad_norm": 1.6329152584075928, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8856053352355957, + "num_tokens": 352144902.0, + "step": 10029 + }, + { + "epoch": 1.8625812441968432, + "grad_norm": 1.5983514785766602, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8783636093139648, + "num_tokens": 352176197.0, + "step": 10030 + }, + { + "epoch": 1.8627669452181987, + "grad_norm": 1.445473074913025, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8724783658981323, + "num_tokens": 352211911.0, + "step": 10031 + }, + { + "epoch": 1.8629526462395543, + "grad_norm": 1.5064600706100464, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8765538930892944, + "num_tokens": 352245287.0, + "step": 10032 + }, + { + "epoch": 1.86313834726091, + "grad_norm": 1.454919695854187, + "learning_rate": 1e-06, + "loss": 0.3923, + "mean_token_accuracy": 0.8667547106742859, + "num_tokens": 352284728.0, + "step": 10033 + }, + { + "epoch": 1.8633240482822655, + "grad_norm": 1.3822108507156372, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8776021003723145, + "num_tokens": 352325900.0, + "step": 10034 + }, + { + "epoch": 1.863509749303621, + "grad_norm": 1.5227985382080078, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.874000608921051, + "num_tokens": 352361202.0, + "step": 10035 + }, + { + "epoch": 1.8636954503249767, + "grad_norm": 1.500508427619934, + "learning_rate": 1e-06, + "loss": 0.3735, + "mean_token_accuracy": 0.8731095790863037, + "num_tokens": 352397815.0, + "step": 10036 + }, + { + "epoch": 1.8638811513463325, + "grad_norm": 1.538567066192627, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8690470457077026, + "num_tokens": 352434325.0, + "step": 10037 + }, + { + "epoch": 1.864066852367688, + "grad_norm": 1.3749982118606567, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8901766538619995, + "num_tokens": 352473802.0, + "step": 10038 + }, + { + "epoch": 1.8642525533890435, + "grad_norm": 1.3625397682189941, + "learning_rate": 1e-06, + "loss": 0.3698, + "mean_token_accuracy": 0.8747540712356567, + "num_tokens": 352514693.0, + "step": 10039 + }, + { + "epoch": 1.8644382544103992, + "grad_norm": 1.5971424579620361, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8796209096908569, + "num_tokens": 352545250.0, + "step": 10040 + }, + { + "epoch": 1.864623955431755, + "grad_norm": 1.5639867782592773, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8771620988845825, + "num_tokens": 352575033.0, + "step": 10041 + }, + { + "epoch": 1.8648096564531105, + "grad_norm": 1.4386496543884277, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8782122135162354, + "num_tokens": 352612248.0, + "step": 10042 + }, + { + "epoch": 1.864995357474466, + "grad_norm": 1.5185494422912598, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8834340572357178, + "num_tokens": 352646345.0, + "step": 10043 + }, + { + "epoch": 1.8651810584958217, + "grad_norm": 1.424972653388977, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8799267411231995, + "num_tokens": 352683800.0, + "step": 10044 + }, + { + "epoch": 1.8653667595171775, + "grad_norm": 1.4918372631072998, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.87751305103302, + "num_tokens": 352718298.0, + "step": 10045 + }, + { + "epoch": 1.865552460538533, + "grad_norm": 1.555650234222412, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8693192005157471, + "num_tokens": 352754486.0, + "step": 10046 + }, + { + "epoch": 1.8657381615598885, + "grad_norm": 1.527418851852417, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8787648677825928, + "num_tokens": 352789035.0, + "step": 10047 + }, + { + "epoch": 1.8659238625812442, + "grad_norm": 1.6814045906066895, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8717105388641357, + "num_tokens": 352820650.0, + "step": 10048 + }, + { + "epoch": 1.8661095636026, + "grad_norm": 1.506168246269226, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8843961954116821, + "num_tokens": 352853654.0, + "step": 10049 + }, + { + "epoch": 1.8662952646239555, + "grad_norm": 1.5683540105819702, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8822907209396362, + "num_tokens": 352888815.0, + "step": 10050 + }, + { + "epoch": 1.866480965645311, + "grad_norm": 1.4282010793685913, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8791376948356628, + "num_tokens": 352927042.0, + "step": 10051 + }, + { + "epoch": 1.8666666666666667, + "grad_norm": 1.5090500116348267, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8843145370483398, + "num_tokens": 352959511.0, + "step": 10052 + }, + { + "epoch": 1.8668523676880224, + "grad_norm": 1.4932255744934082, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8762164115905762, + "num_tokens": 352996217.0, + "step": 10053 + }, + { + "epoch": 1.867038068709378, + "grad_norm": 1.5656538009643555, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8915088176727295, + "num_tokens": 353031651.0, + "step": 10054 + }, + { + "epoch": 1.8672237697307335, + "grad_norm": 1.5636018514633179, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.882230818271637, + "num_tokens": 353063493.0, + "step": 10055 + }, + { + "epoch": 1.8674094707520892, + "grad_norm": 1.617763638496399, + "learning_rate": 1e-06, + "loss": 0.3918, + "mean_token_accuracy": 0.8705506920814514, + "num_tokens": 353092768.0, + "step": 10056 + }, + { + "epoch": 1.867595171773445, + "grad_norm": 1.5161378383636475, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8879519701004028, + "num_tokens": 353121828.0, + "step": 10057 + }, + { + "epoch": 1.8677808727948002, + "grad_norm": 1.3757888078689575, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8732157945632935, + "num_tokens": 353160119.0, + "step": 10058 + }, + { + "epoch": 1.867966573816156, + "grad_norm": 1.437291145324707, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8811008930206299, + "num_tokens": 353199693.0, + "step": 10059 + }, + { + "epoch": 1.8681522748375117, + "grad_norm": 1.601298451423645, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8709036707878113, + "num_tokens": 353231051.0, + "step": 10060 + }, + { + "epoch": 1.8683379758588672, + "grad_norm": 1.418503999710083, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8906910419464111, + "num_tokens": 353264228.0, + "step": 10061 + }, + { + "epoch": 1.8685236768802227, + "grad_norm": 1.3714596033096313, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8979305028915405, + "num_tokens": 353298899.0, + "step": 10062 + }, + { + "epoch": 1.8687093779015784, + "grad_norm": 1.6273901462554932, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8772718906402588, + "num_tokens": 353330610.0, + "step": 10063 + }, + { + "epoch": 1.8688950789229342, + "grad_norm": 1.5679105520248413, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8822072744369507, + "num_tokens": 353360286.0, + "step": 10064 + }, + { + "epoch": 1.8690807799442897, + "grad_norm": 1.5426039695739746, + "learning_rate": 1e-06, + "loss": 0.3927, + "mean_token_accuracy": 0.8677717447280884, + "num_tokens": 353394948.0, + "step": 10065 + }, + { + "epoch": 1.8692664809656452, + "grad_norm": 1.4968194961547852, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8707621693611145, + "num_tokens": 353435648.0, + "step": 10066 + }, + { + "epoch": 1.869452181987001, + "grad_norm": 1.5023159980773926, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8858105540275574, + "num_tokens": 353473250.0, + "step": 10067 + }, + { + "epoch": 1.8696378830083566, + "grad_norm": 1.42208993434906, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8869688510894775, + "num_tokens": 353511695.0, + "step": 10068 + }, + { + "epoch": 1.8698235840297122, + "grad_norm": 1.357892394065857, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8806872367858887, + "num_tokens": 353550456.0, + "step": 10069 + }, + { + "epoch": 1.8700092850510677, + "grad_norm": 1.3841854333877563, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8713629841804504, + "num_tokens": 353590057.0, + "step": 10070 + }, + { + "epoch": 1.8701949860724234, + "grad_norm": 1.626499056816101, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8771480321884155, + "num_tokens": 353619718.0, + "step": 10071 + }, + { + "epoch": 1.8703806870937791, + "grad_norm": 1.5965770483016968, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8936545848846436, + "num_tokens": 353657238.0, + "step": 10072 + }, + { + "epoch": 1.8705663881151346, + "grad_norm": 1.5487678050994873, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8757392764091492, + "num_tokens": 353691355.0, + "step": 10073 + }, + { + "epoch": 1.8707520891364902, + "grad_norm": 1.457199215888977, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8844120502471924, + "num_tokens": 353725550.0, + "step": 10074 + }, + { + "epoch": 1.870937790157846, + "grad_norm": 1.3798328638076782, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8858034610748291, + "num_tokens": 353764270.0, + "step": 10075 + }, + { + "epoch": 1.8711234911792016, + "grad_norm": 1.55367910861969, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8818141222000122, + "num_tokens": 353794712.0, + "step": 10076 + }, + { + "epoch": 1.8713091922005571, + "grad_norm": 1.7312010526657104, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8737185001373291, + "num_tokens": 353821948.0, + "step": 10077 + }, + { + "epoch": 1.8714948932219126, + "grad_norm": 1.709606647491455, + "learning_rate": 1e-06, + "loss": 0.4342, + "mean_token_accuracy": 0.8537615537643433, + "num_tokens": 353853391.0, + "step": 10078 + }, + { + "epoch": 1.8716805942432684, + "grad_norm": 1.4937156438827515, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8873708248138428, + "num_tokens": 353887703.0, + "step": 10079 + }, + { + "epoch": 1.8718662952646241, + "grad_norm": 1.5721940994262695, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8885456323623657, + "num_tokens": 353920319.0, + "step": 10080 + }, + { + "epoch": 1.8720519962859796, + "grad_norm": 1.5883458852767944, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8786554336547852, + "num_tokens": 353956368.0, + "step": 10081 + }, + { + "epoch": 1.8722376973073351, + "grad_norm": 1.602943778038025, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8836697936058044, + "num_tokens": 353982594.0, + "step": 10082 + }, + { + "epoch": 1.8724233983286909, + "grad_norm": 1.5245271921157837, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8909173607826233, + "num_tokens": 354013409.0, + "step": 10083 + }, + { + "epoch": 1.8726090993500464, + "grad_norm": 1.463756799697876, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8965301513671875, + "num_tokens": 354046820.0, + "step": 10084 + }, + { + "epoch": 1.872794800371402, + "grad_norm": 1.4390757083892822, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8703421950340271, + "num_tokens": 354083201.0, + "step": 10085 + }, + { + "epoch": 1.8729805013927576, + "grad_norm": 1.3463243246078491, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8742300868034363, + "num_tokens": 354125385.0, + "step": 10086 + }, + { + "epoch": 1.8731662024141134, + "grad_norm": 1.3916139602661133, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8778262734413147, + "num_tokens": 354162466.0, + "step": 10087 + }, + { + "epoch": 1.8733519034354689, + "grad_norm": 1.3665848970413208, + "learning_rate": 1e-06, + "loss": 0.4199, + "mean_token_accuracy": 0.8553242683410645, + "num_tokens": 354208599.0, + "step": 10088 + }, + { + "epoch": 1.8735376044568244, + "grad_norm": 1.4038735628128052, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8755642175674438, + "num_tokens": 354249922.0, + "step": 10089 + }, + { + "epoch": 1.8737233054781801, + "grad_norm": 1.4727479219436646, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8781024217605591, + "num_tokens": 354282464.0, + "step": 10090 + }, + { + "epoch": 1.8739090064995358, + "grad_norm": 1.3317891359329224, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8818572759628296, + "num_tokens": 354320464.0, + "step": 10091 + }, + { + "epoch": 1.8740947075208914, + "grad_norm": 1.4108494520187378, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8905618190765381, + "num_tokens": 354358047.0, + "step": 10092 + }, + { + "epoch": 1.8742804085422469, + "grad_norm": 1.7818585634231567, + "learning_rate": 1e-06, + "loss": 0.3972, + "mean_token_accuracy": 0.8618776798248291, + "num_tokens": 354383579.0, + "step": 10093 + }, + { + "epoch": 1.8744661095636026, + "grad_norm": 1.474886417388916, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8803222179412842, + "num_tokens": 354417376.0, + "step": 10094 + }, + { + "epoch": 1.8746518105849583, + "grad_norm": 1.4493645429611206, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.882820725440979, + "num_tokens": 354455606.0, + "step": 10095 + }, + { + "epoch": 1.8748375116063138, + "grad_norm": 1.3910340070724487, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8991619348526001, + "num_tokens": 354492134.0, + "step": 10096 + }, + { + "epoch": 1.8750232126276694, + "grad_norm": 1.4793518781661987, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8921367526054382, + "num_tokens": 354528024.0, + "step": 10097 + }, + { + "epoch": 1.875208913649025, + "grad_norm": 1.511146068572998, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8843352794647217, + "num_tokens": 354560344.0, + "step": 10098 + }, + { + "epoch": 1.8753946146703808, + "grad_norm": 1.3753232955932617, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8789459466934204, + "num_tokens": 354598897.0, + "step": 10099 + }, + { + "epoch": 1.8755803156917363, + "grad_norm": 1.4239919185638428, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8921966552734375, + "num_tokens": 354634573.0, + "step": 10100 + }, + { + "epoch": 1.8757660167130918, + "grad_norm": 1.4208765029907227, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8795489072799683, + "num_tokens": 354669191.0, + "step": 10101 + }, + { + "epoch": 1.8759517177344476, + "grad_norm": 1.383718490600586, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8862673044204712, + "num_tokens": 354703125.0, + "step": 10102 + }, + { + "epoch": 1.8761374187558033, + "grad_norm": 1.3637748956680298, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.891555666923523, + "num_tokens": 354745184.0, + "step": 10103 + }, + { + "epoch": 1.8763231197771588, + "grad_norm": 1.654647946357727, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8695849180221558, + "num_tokens": 354774058.0, + "step": 10104 + }, + { + "epoch": 1.8765088207985143, + "grad_norm": 1.5224413871765137, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8731191158294678, + "num_tokens": 354807227.0, + "step": 10105 + }, + { + "epoch": 1.87669452181987, + "grad_norm": 1.6910184621810913, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8668748140335083, + "num_tokens": 354837981.0, + "step": 10106 + }, + { + "epoch": 1.8768802228412256, + "grad_norm": 1.4871917963027954, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8786129355430603, + "num_tokens": 354873255.0, + "step": 10107 + }, + { + "epoch": 1.877065923862581, + "grad_norm": 1.558839201927185, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8831468820571899, + "num_tokens": 354905504.0, + "step": 10108 + }, + { + "epoch": 1.8772516248839368, + "grad_norm": 1.4486130475997925, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8912822604179382, + "num_tokens": 354942968.0, + "step": 10109 + }, + { + "epoch": 1.8774373259052926, + "grad_norm": 1.4192694425582886, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.886985182762146, + "num_tokens": 354980837.0, + "step": 10110 + }, + { + "epoch": 1.877623026926648, + "grad_norm": 1.6324763298034668, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8807275295257568, + "num_tokens": 355009922.0, + "step": 10111 + }, + { + "epoch": 1.8778087279480036, + "grad_norm": 1.4270093441009521, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8820944428443909, + "num_tokens": 355046537.0, + "step": 10112 + }, + { + "epoch": 1.8779944289693593, + "grad_norm": 1.5260329246520996, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8821589350700378, + "num_tokens": 355082441.0, + "step": 10113 + }, + { + "epoch": 1.878180129990715, + "grad_norm": 1.4743815660476685, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8768609762191772, + "num_tokens": 355115957.0, + "step": 10114 + }, + { + "epoch": 1.8783658310120706, + "grad_norm": 1.5439066886901855, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8790695667266846, + "num_tokens": 355146950.0, + "step": 10115 + }, + { + "epoch": 1.878551532033426, + "grad_norm": 1.3828375339508057, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8945868015289307, + "num_tokens": 355183330.0, + "step": 10116 + }, + { + "epoch": 1.8787372330547818, + "grad_norm": 1.4924588203430176, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8795366287231445, + "num_tokens": 355217336.0, + "step": 10117 + }, + { + "epoch": 1.8789229340761375, + "grad_norm": 1.512209177017212, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8867424726486206, + "num_tokens": 355249982.0, + "step": 10118 + }, + { + "epoch": 1.879108635097493, + "grad_norm": 1.5499056577682495, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8794733285903931, + "num_tokens": 355287875.0, + "step": 10119 + }, + { + "epoch": 1.8792943361188486, + "grad_norm": 1.4620227813720703, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8877394199371338, + "num_tokens": 355321333.0, + "step": 10120 + }, + { + "epoch": 1.8794800371402043, + "grad_norm": 1.4134440422058105, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8748735189437866, + "num_tokens": 355362053.0, + "step": 10121 + }, + { + "epoch": 1.87966573816156, + "grad_norm": 1.5745389461517334, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8755045533180237, + "num_tokens": 355393383.0, + "step": 10122 + }, + { + "epoch": 1.8798514391829155, + "grad_norm": 1.4186453819274902, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8887090682983398, + "num_tokens": 355430662.0, + "step": 10123 + }, + { + "epoch": 1.880037140204271, + "grad_norm": 1.3776265382766724, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8858667612075806, + "num_tokens": 355471448.0, + "step": 10124 + }, + { + "epoch": 1.8802228412256268, + "grad_norm": 1.4520705938339233, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8690611124038696, + "num_tokens": 355510103.0, + "step": 10125 + }, + { + "epoch": 1.8804085422469825, + "grad_norm": 1.4952698945999146, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8855661153793335, + "num_tokens": 355544616.0, + "step": 10126 + }, + { + "epoch": 1.880594243268338, + "grad_norm": 1.550007700920105, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8719817399978638, + "num_tokens": 355580784.0, + "step": 10127 + }, + { + "epoch": 1.8807799442896935, + "grad_norm": 1.4112730026245117, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8879110217094421, + "num_tokens": 355617443.0, + "step": 10128 + }, + { + "epoch": 1.8809656453110493, + "grad_norm": 1.580559492111206, + "learning_rate": 1e-06, + "loss": 0.4044, + "mean_token_accuracy": 0.8611228466033936, + "num_tokens": 355655810.0, + "step": 10129 + }, + { + "epoch": 1.881151346332405, + "grad_norm": 1.4489331245422363, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8860945105552673, + "num_tokens": 355693608.0, + "step": 10130 + }, + { + "epoch": 1.8813370473537603, + "grad_norm": 1.4099750518798828, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8796929717063904, + "num_tokens": 355735501.0, + "step": 10131 + }, + { + "epoch": 1.881522748375116, + "grad_norm": 1.5620077848434448, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8670721054077148, + "num_tokens": 355770771.0, + "step": 10132 + }, + { + "epoch": 1.8817084493964717, + "grad_norm": 1.3322134017944336, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8766205906867981, + "num_tokens": 355813050.0, + "step": 10133 + }, + { + "epoch": 1.8818941504178273, + "grad_norm": 1.4086294174194336, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8757883310317993, + "num_tokens": 355851079.0, + "step": 10134 + }, + { + "epoch": 1.8820798514391828, + "grad_norm": 1.3434687852859497, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8843456506729126, + "num_tokens": 355891875.0, + "step": 10135 + }, + { + "epoch": 1.8822655524605385, + "grad_norm": 1.5649752616882324, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8803311586380005, + "num_tokens": 355922918.0, + "step": 10136 + }, + { + "epoch": 1.8824512534818942, + "grad_norm": 1.5513447523117065, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8767980337142944, + "num_tokens": 355957075.0, + "step": 10137 + }, + { + "epoch": 1.8826369545032497, + "grad_norm": 1.4455924034118652, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8859411478042603, + "num_tokens": 355996757.0, + "step": 10138 + }, + { + "epoch": 1.8828226555246053, + "grad_norm": 1.5868736505508423, + "learning_rate": 1e-06, + "loss": 0.389, + "mean_token_accuracy": 0.8657093048095703, + "num_tokens": 356031819.0, + "step": 10139 + }, + { + "epoch": 1.883008356545961, + "grad_norm": 1.4394675493240356, + "learning_rate": 1e-06, + "loss": 0.2912, + "mean_token_accuracy": 0.8968695998191833, + "num_tokens": 356064760.0, + "step": 10140 + }, + { + "epoch": 1.8831940575673167, + "grad_norm": 1.4556859731674194, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8685862421989441, + "num_tokens": 356104514.0, + "step": 10141 + }, + { + "epoch": 1.8833797585886722, + "grad_norm": 1.5132397413253784, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8670288324356079, + "num_tokens": 356141247.0, + "step": 10142 + }, + { + "epoch": 1.8835654596100277, + "grad_norm": 1.459307074546814, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8853141069412231, + "num_tokens": 356176418.0, + "step": 10143 + }, + { + "epoch": 1.8837511606313835, + "grad_norm": 1.4469590187072754, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8889347314834595, + "num_tokens": 356207672.0, + "step": 10144 + }, + { + "epoch": 1.8839368616527392, + "grad_norm": 1.4291256666183472, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8905295133590698, + "num_tokens": 356242380.0, + "step": 10145 + }, + { + "epoch": 1.8841225626740947, + "grad_norm": 1.5267457962036133, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.882037341594696, + "num_tokens": 356276785.0, + "step": 10146 + }, + { + "epoch": 1.8843082636954502, + "grad_norm": 1.6151524782180786, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8684093952178955, + "num_tokens": 356305986.0, + "step": 10147 + }, + { + "epoch": 1.884493964716806, + "grad_norm": 1.5588570833206177, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8830668926239014, + "num_tokens": 356334594.0, + "step": 10148 + }, + { + "epoch": 1.8846796657381617, + "grad_norm": 1.4070565700531006, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8757421374320984, + "num_tokens": 356377812.0, + "step": 10149 + }, + { + "epoch": 1.8848653667595172, + "grad_norm": 1.5764589309692383, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8762028217315674, + "num_tokens": 356414311.0, + "step": 10150 + }, + { + "epoch": 1.8850510677808727, + "grad_norm": 1.5617055892944336, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8860495090484619, + "num_tokens": 356444968.0, + "step": 10151 + }, + { + "epoch": 1.8852367688022285, + "grad_norm": 1.6104391813278198, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8764519691467285, + "num_tokens": 356478008.0, + "step": 10152 + }, + { + "epoch": 1.8854224698235842, + "grad_norm": 1.5926539897918701, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8796139359474182, + "num_tokens": 356508481.0, + "step": 10153 + }, + { + "epoch": 1.8856081708449397, + "grad_norm": 1.4891191720962524, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8824535608291626, + "num_tokens": 356545865.0, + "step": 10154 + }, + { + "epoch": 1.8857938718662952, + "grad_norm": 1.4325166940689087, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8910978436470032, + "num_tokens": 356582858.0, + "step": 10155 + }, + { + "epoch": 1.885979572887651, + "grad_norm": 1.37608003616333, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8857885599136353, + "num_tokens": 356621455.0, + "step": 10156 + }, + { + "epoch": 1.8861652739090065, + "grad_norm": 1.3974759578704834, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8853217363357544, + "num_tokens": 356660579.0, + "step": 10157 + }, + { + "epoch": 1.886350974930362, + "grad_norm": 1.5142837762832642, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8779391646385193, + "num_tokens": 356694402.0, + "step": 10158 + }, + { + "epoch": 1.8865366759517177, + "grad_norm": 1.4847012758255005, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8775805234909058, + "num_tokens": 356734878.0, + "step": 10159 + }, + { + "epoch": 1.8867223769730734, + "grad_norm": 1.514927625656128, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8936728239059448, + "num_tokens": 356769653.0, + "step": 10160 + }, + { + "epoch": 1.886908077994429, + "grad_norm": 1.690525770187378, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8761394023895264, + "num_tokens": 356800568.0, + "step": 10161 + }, + { + "epoch": 1.8870937790157845, + "grad_norm": 1.3427678346633911, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8869860172271729, + "num_tokens": 356838177.0, + "step": 10162 + }, + { + "epoch": 1.8872794800371402, + "grad_norm": 1.483932375907898, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.880851686000824, + "num_tokens": 356874087.0, + "step": 10163 + }, + { + "epoch": 1.887465181058496, + "grad_norm": 1.3990296125411987, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8793420195579529, + "num_tokens": 356912182.0, + "step": 10164 + }, + { + "epoch": 1.8876508820798514, + "grad_norm": 1.631726622581482, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8789221048355103, + "num_tokens": 356942425.0, + "step": 10165 + }, + { + "epoch": 1.887836583101207, + "grad_norm": 1.321426510810852, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8745225071907043, + "num_tokens": 356990076.0, + "step": 10166 + }, + { + "epoch": 1.8880222841225627, + "grad_norm": 1.6254851818084717, + "learning_rate": 1e-06, + "loss": 0.4198, + "mean_token_accuracy": 0.8598792552947998, + "num_tokens": 357023553.0, + "step": 10167 + }, + { + "epoch": 1.8882079851439184, + "grad_norm": 1.3378159999847412, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8864067196846008, + "num_tokens": 357061846.0, + "step": 10168 + }, + { + "epoch": 1.888393686165274, + "grad_norm": 1.5667359828948975, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8740353584289551, + "num_tokens": 357094530.0, + "step": 10169 + }, + { + "epoch": 1.8885793871866294, + "grad_norm": 1.5202330350875854, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8753422498703003, + "num_tokens": 357130295.0, + "step": 10170 + }, + { + "epoch": 1.8887650882079852, + "grad_norm": 1.4404630661010742, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8770444393157959, + "num_tokens": 357167067.0, + "step": 10171 + }, + { + "epoch": 1.888950789229341, + "grad_norm": 1.4678984880447388, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8761325478553772, + "num_tokens": 357200300.0, + "step": 10172 + }, + { + "epoch": 1.8891364902506964, + "grad_norm": 1.4354223012924194, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8661398887634277, + "num_tokens": 357239338.0, + "step": 10173 + }, + { + "epoch": 1.889322191272052, + "grad_norm": 1.5334088802337646, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8752732872962952, + "num_tokens": 357274078.0, + "step": 10174 + }, + { + "epoch": 1.8895078922934077, + "grad_norm": 1.416873574256897, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8731874227523804, + "num_tokens": 357313953.0, + "step": 10175 + }, + { + "epoch": 1.8896935933147634, + "grad_norm": 1.488900899887085, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8875885009765625, + "num_tokens": 357346934.0, + "step": 10176 + }, + { + "epoch": 1.889879294336119, + "grad_norm": 1.5486923456192017, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8778600692749023, + "num_tokens": 357379175.0, + "step": 10177 + }, + { + "epoch": 1.8900649953574744, + "grad_norm": 1.4826289415359497, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8701945543289185, + "num_tokens": 357418348.0, + "step": 10178 + }, + { + "epoch": 1.8902506963788301, + "grad_norm": 1.616636872291565, + "learning_rate": 1e-06, + "loss": 0.4505, + "mean_token_accuracy": 0.8529222011566162, + "num_tokens": 357458122.0, + "step": 10179 + }, + { + "epoch": 1.8904363974001857, + "grad_norm": 1.7591607570648193, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8797325491905212, + "num_tokens": 357487577.0, + "step": 10180 + }, + { + "epoch": 1.8906220984215412, + "grad_norm": 1.4161471128463745, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8735227584838867, + "num_tokens": 357528542.0, + "step": 10181 + }, + { + "epoch": 1.890807799442897, + "grad_norm": 1.5587760210037231, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8643782138824463, + "num_tokens": 357565038.0, + "step": 10182 + }, + { + "epoch": 1.8909935004642526, + "grad_norm": 1.5234568119049072, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8726486563682556, + "num_tokens": 357598886.0, + "step": 10183 + }, + { + "epoch": 1.8911792014856081, + "grad_norm": 1.6343404054641724, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8821488618850708, + "num_tokens": 357630804.0, + "step": 10184 + }, + { + "epoch": 1.8913649025069637, + "grad_norm": 1.4674301147460938, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8751119375228882, + "num_tokens": 357666616.0, + "step": 10185 + }, + { + "epoch": 1.8915506035283194, + "grad_norm": 1.5714181661605835, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8835958242416382, + "num_tokens": 357691607.0, + "step": 10186 + }, + { + "epoch": 1.8917363045496751, + "grad_norm": 1.5748960971832275, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8840975165367126, + "num_tokens": 357723333.0, + "step": 10187 + }, + { + "epoch": 1.8919220055710306, + "grad_norm": 1.579006552696228, + "learning_rate": 1e-06, + "loss": 0.3908, + "mean_token_accuracy": 0.8620372414588928, + "num_tokens": 357757882.0, + "step": 10188 + }, + { + "epoch": 1.8921077065923861, + "grad_norm": 1.507799506187439, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8837665319442749, + "num_tokens": 357789310.0, + "step": 10189 + }, + { + "epoch": 1.8922934076137419, + "grad_norm": 1.5604872703552246, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8754284381866455, + "num_tokens": 357819874.0, + "step": 10190 + }, + { + "epoch": 1.8924791086350976, + "grad_norm": 1.4353854656219482, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8842533230781555, + "num_tokens": 357853884.0, + "step": 10191 + }, + { + "epoch": 1.8926648096564531, + "grad_norm": 1.4965970516204834, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8708357810974121, + "num_tokens": 357888820.0, + "step": 10192 + }, + { + "epoch": 1.8928505106778086, + "grad_norm": 1.517321228981018, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8805059194564819, + "num_tokens": 357924461.0, + "step": 10193 + }, + { + "epoch": 1.8930362116991644, + "grad_norm": 1.5248091220855713, + "learning_rate": 1e-06, + "loss": 0.383, + "mean_token_accuracy": 0.8702455163002014, + "num_tokens": 357958438.0, + "step": 10194 + }, + { + "epoch": 1.89322191272052, + "grad_norm": 1.503233790397644, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8827731609344482, + "num_tokens": 357989461.0, + "step": 10195 + }, + { + "epoch": 1.8934076137418756, + "grad_norm": 1.548359990119934, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8826370239257812, + "num_tokens": 358024245.0, + "step": 10196 + }, + { + "epoch": 1.8935933147632311, + "grad_norm": 1.560876488685608, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.866448700428009, + "num_tokens": 358059728.0, + "step": 10197 + }, + { + "epoch": 1.8937790157845868, + "grad_norm": 1.5235048532485962, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.889718234539032, + "num_tokens": 358091698.0, + "step": 10198 + }, + { + "epoch": 1.8939647168059426, + "grad_norm": 1.5089539289474487, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.879438579082489, + "num_tokens": 358125068.0, + "step": 10199 + }, + { + "epoch": 1.894150417827298, + "grad_norm": 1.3780676126480103, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8943204283714294, + "num_tokens": 358162138.0, + "step": 10200 + }, + { + "epoch": 1.8943361188486536, + "grad_norm": 1.510728359222412, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8677631616592407, + "num_tokens": 358201010.0, + "step": 10201 + }, + { + "epoch": 1.8945218198700093, + "grad_norm": 1.6559760570526123, + "learning_rate": 1e-06, + "loss": 0.3716, + "mean_token_accuracy": 0.8765161633491516, + "num_tokens": 358230271.0, + "step": 10202 + }, + { + "epoch": 1.8947075208913648, + "grad_norm": 1.7848296165466309, + "learning_rate": 1e-06, + "loss": 0.3983, + "mean_token_accuracy": 0.8617461919784546, + "num_tokens": 358262669.0, + "step": 10203 + }, + { + "epoch": 1.8948932219127204, + "grad_norm": 1.4619255065917969, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8780519962310791, + "num_tokens": 358301122.0, + "step": 10204 + }, + { + "epoch": 1.895078922934076, + "grad_norm": 1.417346477508545, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8830057382583618, + "num_tokens": 358338488.0, + "step": 10205 + }, + { + "epoch": 1.8952646239554318, + "grad_norm": 1.5204508304595947, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8913518190383911, + "num_tokens": 358368146.0, + "step": 10206 + }, + { + "epoch": 1.8954503249767873, + "grad_norm": 1.376792550086975, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8814494609832764, + "num_tokens": 358406848.0, + "step": 10207 + }, + { + "epoch": 1.8956360259981428, + "grad_norm": 1.555914282798767, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8776156902313232, + "num_tokens": 358438655.0, + "step": 10208 + }, + { + "epoch": 1.8958217270194986, + "grad_norm": 1.3833636045455933, + "learning_rate": 1e-06, + "loss": 0.292, + "mean_token_accuracy": 0.8969686031341553, + "num_tokens": 358474954.0, + "step": 10209 + }, + { + "epoch": 1.8960074280408543, + "grad_norm": 1.5002782344818115, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8769498467445374, + "num_tokens": 358510256.0, + "step": 10210 + }, + { + "epoch": 1.8961931290622098, + "grad_norm": 1.422637939453125, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8859286308288574, + "num_tokens": 358545317.0, + "step": 10211 + }, + { + "epoch": 1.8963788300835653, + "grad_norm": 1.353248119354248, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8862802982330322, + "num_tokens": 358583751.0, + "step": 10212 + }, + { + "epoch": 1.896564531104921, + "grad_norm": 1.5071086883544922, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.87261962890625, + "num_tokens": 358617790.0, + "step": 10213 + }, + { + "epoch": 1.8967502321262768, + "grad_norm": 1.3701117038726807, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.890099048614502, + "num_tokens": 358654061.0, + "step": 10214 + }, + { + "epoch": 1.8969359331476323, + "grad_norm": 1.4593329429626465, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8779771327972412, + "num_tokens": 358688387.0, + "step": 10215 + }, + { + "epoch": 1.8971216341689878, + "grad_norm": 1.4345682859420776, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8821666240692139, + "num_tokens": 358727072.0, + "step": 10216 + }, + { + "epoch": 1.8973073351903436, + "grad_norm": 1.6290203332901, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8672356009483337, + "num_tokens": 358756016.0, + "step": 10217 + }, + { + "epoch": 1.8974930362116993, + "grad_norm": 1.463753581047058, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8873886466026306, + "num_tokens": 358787807.0, + "step": 10218 + }, + { + "epoch": 1.8976787372330548, + "grad_norm": 1.534497618675232, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8893665671348572, + "num_tokens": 358822276.0, + "step": 10219 + }, + { + "epoch": 1.8978644382544103, + "grad_norm": 1.4114958047866821, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8860061168670654, + "num_tokens": 358857465.0, + "step": 10220 + }, + { + "epoch": 1.898050139275766, + "grad_norm": 1.4442110061645508, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8820455074310303, + "num_tokens": 358895182.0, + "step": 10221 + }, + { + "epoch": 1.8982358402971218, + "grad_norm": 1.5709823369979858, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8827313780784607, + "num_tokens": 358927120.0, + "step": 10222 + }, + { + "epoch": 1.8984215413184773, + "grad_norm": 1.4421296119689941, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8673204183578491, + "num_tokens": 358967353.0, + "step": 10223 + }, + { + "epoch": 1.8986072423398328, + "grad_norm": 1.4616479873657227, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.877981424331665, + "num_tokens": 359007710.0, + "step": 10224 + }, + { + "epoch": 1.8987929433611885, + "grad_norm": 1.5336041450500488, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8714350461959839, + "num_tokens": 359039398.0, + "step": 10225 + }, + { + "epoch": 1.8989786443825443, + "grad_norm": 1.5311567783355713, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8805790543556213, + "num_tokens": 359071227.0, + "step": 10226 + }, + { + "epoch": 1.8991643454038996, + "grad_norm": 1.4599039554595947, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8827826976776123, + "num_tokens": 359106175.0, + "step": 10227 + }, + { + "epoch": 1.8993500464252553, + "grad_norm": 1.4713150262832642, + "learning_rate": 1e-06, + "loss": 0.419, + "mean_token_accuracy": 0.8645316362380981, + "num_tokens": 359147418.0, + "step": 10228 + }, + { + "epoch": 1.899535747446611, + "grad_norm": 1.384419560432434, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.87751305103302, + "num_tokens": 359189496.0, + "step": 10229 + }, + { + "epoch": 1.8997214484679665, + "grad_norm": 1.3799152374267578, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.883150577545166, + "num_tokens": 359229751.0, + "step": 10230 + }, + { + "epoch": 1.899907149489322, + "grad_norm": 1.4684611558914185, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8923687934875488, + "num_tokens": 359261852.0, + "step": 10231 + }, + { + "epoch": 1.9000928505106778, + "grad_norm": 1.555762529373169, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8811134099960327, + "num_tokens": 359293994.0, + "step": 10232 + }, + { + "epoch": 1.9002785515320335, + "grad_norm": 1.6641888618469238, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.871135950088501, + "num_tokens": 359322812.0, + "step": 10233 + }, + { + "epoch": 1.900464252553389, + "grad_norm": 1.4668824672698975, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8792451620101929, + "num_tokens": 359357664.0, + "step": 10234 + }, + { + "epoch": 1.9006499535747445, + "grad_norm": 2.6548588275909424, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8796807527542114, + "num_tokens": 359391748.0, + "step": 10235 + }, + { + "epoch": 1.9008356545961003, + "grad_norm": 1.4711151123046875, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8717489838600159, + "num_tokens": 359429924.0, + "step": 10236 + }, + { + "epoch": 1.901021355617456, + "grad_norm": 1.3852112293243408, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8933274745941162, + "num_tokens": 359469496.0, + "step": 10237 + }, + { + "epoch": 1.9012070566388115, + "grad_norm": 1.495985984802246, + "learning_rate": 1e-06, + "loss": 0.3784, + "mean_token_accuracy": 0.8722400665283203, + "num_tokens": 359503786.0, + "step": 10238 + }, + { + "epoch": 1.901392757660167, + "grad_norm": 1.314659595489502, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8799054622650146, + "num_tokens": 359549079.0, + "step": 10239 + }, + { + "epoch": 1.9015784586815228, + "grad_norm": 1.4325531721115112, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8757468461990356, + "num_tokens": 359587801.0, + "step": 10240 + }, + { + "epoch": 1.9017641597028785, + "grad_norm": 1.5944582223892212, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8800954222679138, + "num_tokens": 359620417.0, + "step": 10241 + }, + { + "epoch": 1.901949860724234, + "grad_norm": 1.4006584882736206, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8846017718315125, + "num_tokens": 359661231.0, + "step": 10242 + }, + { + "epoch": 1.9021355617455895, + "grad_norm": 1.5168524980545044, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8686672449111938, + "num_tokens": 359697258.0, + "step": 10243 + }, + { + "epoch": 1.9023212627669452, + "grad_norm": 1.407926082611084, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8793998956680298, + "num_tokens": 359735224.0, + "step": 10244 + }, + { + "epoch": 1.902506963788301, + "grad_norm": 1.571977972984314, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8707292675971985, + "num_tokens": 359769181.0, + "step": 10245 + }, + { + "epoch": 1.9026926648096565, + "grad_norm": 1.429465889930725, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8815754652023315, + "num_tokens": 359803185.0, + "step": 10246 + }, + { + "epoch": 1.902878365831012, + "grad_norm": 1.437234878540039, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8853769302368164, + "num_tokens": 359838715.0, + "step": 10247 + }, + { + "epoch": 1.9030640668523677, + "grad_norm": 1.413488745689392, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8887392282485962, + "num_tokens": 359872841.0, + "step": 10248 + }, + { + "epoch": 1.9032497678737235, + "grad_norm": 1.4164576530456543, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8604322671890259, + "num_tokens": 359915893.0, + "step": 10249 + }, + { + "epoch": 1.903435468895079, + "grad_norm": 1.4233115911483765, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8885415196418762, + "num_tokens": 359951729.0, + "step": 10250 + }, + { + "epoch": 1.9036211699164345, + "grad_norm": 1.4570361375808716, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8797600269317627, + "num_tokens": 359984623.0, + "step": 10251 + }, + { + "epoch": 1.9038068709377902, + "grad_norm": 1.4663870334625244, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8874796628952026, + "num_tokens": 360017988.0, + "step": 10252 + }, + { + "epoch": 1.9039925719591457, + "grad_norm": 1.5062096118927002, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8780394196510315, + "num_tokens": 360051751.0, + "step": 10253 + }, + { + "epoch": 1.9041782729805012, + "grad_norm": 1.481917142868042, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8897923827171326, + "num_tokens": 360083437.0, + "step": 10254 + }, + { + "epoch": 1.904363974001857, + "grad_norm": 1.3733289241790771, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8760111927986145, + "num_tokens": 360122308.0, + "step": 10255 + }, + { + "epoch": 1.9045496750232127, + "grad_norm": 1.3547865152359009, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.875768780708313, + "num_tokens": 360163315.0, + "step": 10256 + }, + { + "epoch": 1.9047353760445682, + "grad_norm": 1.4104233980178833, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.880577564239502, + "num_tokens": 360201558.0, + "step": 10257 + }, + { + "epoch": 1.9049210770659237, + "grad_norm": 1.445272445678711, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8841946721076965, + "num_tokens": 360234812.0, + "step": 10258 + }, + { + "epoch": 1.9051067780872795, + "grad_norm": 1.5212178230285645, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8840121030807495, + "num_tokens": 360268274.0, + "step": 10259 + }, + { + "epoch": 1.9052924791086352, + "grad_norm": 1.564035415649414, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.867967963218689, + "num_tokens": 360301544.0, + "step": 10260 + }, + { + "epoch": 1.9054781801299907, + "grad_norm": 1.4235328435897827, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8883510231971741, + "num_tokens": 360334472.0, + "step": 10261 + }, + { + "epoch": 1.9056638811513462, + "grad_norm": 1.3876858949661255, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8800917267799377, + "num_tokens": 360372338.0, + "step": 10262 + }, + { + "epoch": 1.905849582172702, + "grad_norm": 1.357133388519287, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8884385824203491, + "num_tokens": 360408737.0, + "step": 10263 + }, + { + "epoch": 1.9060352831940577, + "grad_norm": 1.4558064937591553, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.8784346580505371, + "num_tokens": 360445869.0, + "step": 10264 + }, + { + "epoch": 1.9062209842154132, + "grad_norm": 1.6713145971298218, + "learning_rate": 1e-06, + "loss": 0.3764, + "mean_token_accuracy": 0.8755402565002441, + "num_tokens": 360475613.0, + "step": 10265 + }, + { + "epoch": 1.9064066852367687, + "grad_norm": 1.419674038887024, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8727664947509766, + "num_tokens": 360512887.0, + "step": 10266 + }, + { + "epoch": 1.9065923862581244, + "grad_norm": 1.4670416116714478, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8788912892341614, + "num_tokens": 360553474.0, + "step": 10267 + }, + { + "epoch": 1.9067780872794802, + "grad_norm": 1.5618054866790771, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.878719687461853, + "num_tokens": 360584935.0, + "step": 10268 + }, + { + "epoch": 1.9069637883008357, + "grad_norm": 1.523593544960022, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8808819055557251, + "num_tokens": 360614100.0, + "step": 10269 + }, + { + "epoch": 1.9071494893221912, + "grad_norm": 1.4490375518798828, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8717746138572693, + "num_tokens": 360652296.0, + "step": 10270 + }, + { + "epoch": 1.907335190343547, + "grad_norm": 1.441971778869629, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8765487670898438, + "num_tokens": 360691102.0, + "step": 10271 + }, + { + "epoch": 1.9075208913649027, + "grad_norm": 1.4777436256408691, + "learning_rate": 1e-06, + "loss": 0.397, + "mean_token_accuracy": 0.8635998964309692, + "num_tokens": 360731361.0, + "step": 10272 + }, + { + "epoch": 1.9077065923862582, + "grad_norm": 1.4853527545928955, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8883081674575806, + "num_tokens": 360762903.0, + "step": 10273 + }, + { + "epoch": 1.9078922934076137, + "grad_norm": 1.5750702619552612, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8732261657714844, + "num_tokens": 360794724.0, + "step": 10274 + }, + { + "epoch": 1.9080779944289694, + "grad_norm": 1.3842111825942993, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8777909278869629, + "num_tokens": 360835424.0, + "step": 10275 + }, + { + "epoch": 1.908263695450325, + "grad_norm": 1.5007202625274658, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8777445554733276, + "num_tokens": 360871855.0, + "step": 10276 + }, + { + "epoch": 1.9084493964716804, + "grad_norm": 1.508440375328064, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8785998821258545, + "num_tokens": 360905347.0, + "step": 10277 + }, + { + "epoch": 1.9086350974930362, + "grad_norm": 1.5271772146224976, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8715071082115173, + "num_tokens": 360941284.0, + "step": 10278 + }, + { + "epoch": 1.908820798514392, + "grad_norm": 1.7136316299438477, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.879305362701416, + "num_tokens": 360967623.0, + "step": 10279 + }, + { + "epoch": 1.9090064995357474, + "grad_norm": 1.5434279441833496, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8814527988433838, + "num_tokens": 361003054.0, + "step": 10280 + }, + { + "epoch": 1.909192200557103, + "grad_norm": 1.4728132486343384, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8815206289291382, + "num_tokens": 361039396.0, + "step": 10281 + }, + { + "epoch": 1.9093779015784587, + "grad_norm": 1.5621285438537598, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8736023306846619, + "num_tokens": 361074041.0, + "step": 10282 + }, + { + "epoch": 1.9095636025998144, + "grad_norm": 1.4443970918655396, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8898864388465881, + "num_tokens": 361105565.0, + "step": 10283 + }, + { + "epoch": 1.90974930362117, + "grad_norm": 1.390452265739441, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8878796100616455, + "num_tokens": 361141058.0, + "step": 10284 + }, + { + "epoch": 1.9099350046425254, + "grad_norm": 1.4533473253250122, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.872626543045044, + "num_tokens": 361176495.0, + "step": 10285 + }, + { + "epoch": 1.9101207056638811, + "grad_norm": 1.4049543142318726, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8862596750259399, + "num_tokens": 361209248.0, + "step": 10286 + }, + { + "epoch": 1.9103064066852369, + "grad_norm": 1.485805630683899, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8818321228027344, + "num_tokens": 361243707.0, + "step": 10287 + }, + { + "epoch": 1.9104921077065924, + "grad_norm": 1.6296627521514893, + "learning_rate": 1e-06, + "loss": 0.4142, + "mean_token_accuracy": 0.8597291707992554, + "num_tokens": 361273606.0, + "step": 10288 + }, + { + "epoch": 1.910677808727948, + "grad_norm": 1.4959220886230469, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.877257227897644, + "num_tokens": 361308440.0, + "step": 10289 + }, + { + "epoch": 1.9108635097493036, + "grad_norm": 1.419871211051941, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8816056847572327, + "num_tokens": 361344265.0, + "step": 10290 + }, + { + "epoch": 1.9110492107706594, + "grad_norm": 1.483917236328125, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.881029486656189, + "num_tokens": 361379387.0, + "step": 10291 + }, + { + "epoch": 1.9112349117920149, + "grad_norm": 1.4744524955749512, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8885306119918823, + "num_tokens": 361412562.0, + "step": 10292 + }, + { + "epoch": 1.9114206128133704, + "grad_norm": 1.4539848566055298, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8779513239860535, + "num_tokens": 361446918.0, + "step": 10293 + }, + { + "epoch": 1.9116063138347261, + "grad_norm": 1.4819601774215698, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8810397982597351, + "num_tokens": 361482093.0, + "step": 10294 + }, + { + "epoch": 1.9117920148560819, + "grad_norm": 1.3935426473617554, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8862152695655823, + "num_tokens": 361521060.0, + "step": 10295 + }, + { + "epoch": 1.9119777158774374, + "grad_norm": 1.5209190845489502, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8872445821762085, + "num_tokens": 361553560.0, + "step": 10296 + }, + { + "epoch": 1.9121634168987929, + "grad_norm": 1.4627315998077393, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8848353028297424, + "num_tokens": 361587893.0, + "step": 10297 + }, + { + "epoch": 1.9123491179201486, + "grad_norm": 1.4528180360794067, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8894673585891724, + "num_tokens": 361622749.0, + "step": 10298 + }, + { + "epoch": 1.9125348189415043, + "grad_norm": 1.4979795217514038, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8767372369766235, + "num_tokens": 361659117.0, + "step": 10299 + }, + { + "epoch": 1.9127205199628596, + "grad_norm": 1.5794103145599365, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8784469962120056, + "num_tokens": 361690582.0, + "step": 10300 + }, + { + "epoch": 1.9129062209842154, + "grad_norm": 1.7067439556121826, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8903701305389404, + "num_tokens": 361717175.0, + "step": 10301 + }, + { + "epoch": 1.913091922005571, + "grad_norm": 1.3820427656173706, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.8667755126953125, + "num_tokens": 361759039.0, + "step": 10302 + }, + { + "epoch": 1.9132776230269266, + "grad_norm": 1.5903152227401733, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.864355206489563, + "num_tokens": 361790867.0, + "step": 10303 + }, + { + "epoch": 1.9134633240482821, + "grad_norm": 1.5140249729156494, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8831019401550293, + "num_tokens": 361825354.0, + "step": 10304 + }, + { + "epoch": 1.9136490250696379, + "grad_norm": 1.3989462852478027, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8784152269363403, + "num_tokens": 361863417.0, + "step": 10305 + }, + { + "epoch": 1.9138347260909936, + "grad_norm": 1.4556471109390259, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8762879371643066, + "num_tokens": 361897945.0, + "step": 10306 + }, + { + "epoch": 1.914020427112349, + "grad_norm": 1.381649374961853, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8821255564689636, + "num_tokens": 361937683.0, + "step": 10307 + }, + { + "epoch": 1.9142061281337046, + "grad_norm": 1.515786051750183, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8797276616096497, + "num_tokens": 361970427.0, + "step": 10308 + }, + { + "epoch": 1.9143918291550603, + "grad_norm": 1.5550616979599, + "learning_rate": 1e-06, + "loss": 0.3678, + "mean_token_accuracy": 0.8702907562255859, + "num_tokens": 362005578.0, + "step": 10309 + }, + { + "epoch": 1.914577530176416, + "grad_norm": 1.650913953781128, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8800450563430786, + "num_tokens": 362037109.0, + "step": 10310 + }, + { + "epoch": 1.9147632311977716, + "grad_norm": 1.4734472036361694, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.88154137134552, + "num_tokens": 362073671.0, + "step": 10311 + }, + { + "epoch": 1.914948932219127, + "grad_norm": 1.4818085432052612, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.8646335601806641, + "num_tokens": 362111923.0, + "step": 10312 + }, + { + "epoch": 1.9151346332404828, + "grad_norm": 1.4758143424987793, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8836377263069153, + "num_tokens": 362149580.0, + "step": 10313 + }, + { + "epoch": 1.9153203342618386, + "grad_norm": 1.4712142944335938, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8835430145263672, + "num_tokens": 362182999.0, + "step": 10314 + }, + { + "epoch": 1.915506035283194, + "grad_norm": 1.4995616674423218, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8884256482124329, + "num_tokens": 362216791.0, + "step": 10315 + }, + { + "epoch": 1.9156917363045496, + "grad_norm": 1.4747059345245361, + "learning_rate": 1e-06, + "loss": 0.3777, + "mean_token_accuracy": 0.8699308633804321, + "num_tokens": 362255104.0, + "step": 10316 + }, + { + "epoch": 1.9158774373259053, + "grad_norm": 1.4439619779586792, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8900114297866821, + "num_tokens": 362288895.0, + "step": 10317 + }, + { + "epoch": 1.916063138347261, + "grad_norm": 1.4976580142974854, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8761693239212036, + "num_tokens": 362326033.0, + "step": 10318 + }, + { + "epoch": 1.9162488393686166, + "grad_norm": 1.4401063919067383, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8783556818962097, + "num_tokens": 362362836.0, + "step": 10319 + }, + { + "epoch": 1.916434540389972, + "grad_norm": 1.2872679233551025, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8797980546951294, + "num_tokens": 362406693.0, + "step": 10320 + }, + { + "epoch": 1.9166202414113278, + "grad_norm": 1.5064339637756348, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8881254196166992, + "num_tokens": 362437497.0, + "step": 10321 + }, + { + "epoch": 1.9168059424326835, + "grad_norm": 1.5329325199127197, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8719730377197266, + "num_tokens": 362469061.0, + "step": 10322 + }, + { + "epoch": 1.916991643454039, + "grad_norm": 1.5191423892974854, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8875576853752136, + "num_tokens": 362501874.0, + "step": 10323 + }, + { + "epoch": 1.9171773444753946, + "grad_norm": 1.5140695571899414, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8792463541030884, + "num_tokens": 362537948.0, + "step": 10324 + }, + { + "epoch": 1.9173630454967503, + "grad_norm": 1.4950467348098755, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8856104612350464, + "num_tokens": 362571682.0, + "step": 10325 + }, + { + "epoch": 1.9175487465181058, + "grad_norm": 1.322222352027893, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8822404146194458, + "num_tokens": 362612450.0, + "step": 10326 + }, + { + "epoch": 1.9177344475394613, + "grad_norm": 1.408987045288086, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8717573881149292, + "num_tokens": 362649548.0, + "step": 10327 + }, + { + "epoch": 1.917920148560817, + "grad_norm": 1.5855698585510254, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8737930059432983, + "num_tokens": 362678400.0, + "step": 10328 + }, + { + "epoch": 1.9181058495821728, + "grad_norm": 1.4420533180236816, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8751978278160095, + "num_tokens": 362718313.0, + "step": 10329 + }, + { + "epoch": 1.9182915506035283, + "grad_norm": 1.3978643417358398, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8841425776481628, + "num_tokens": 362756864.0, + "step": 10330 + }, + { + "epoch": 1.9184772516248838, + "grad_norm": 1.4224497079849243, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8724322319030762, + "num_tokens": 362797000.0, + "step": 10331 + }, + { + "epoch": 1.9186629526462395, + "grad_norm": 1.469377040863037, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8770596981048584, + "num_tokens": 362830602.0, + "step": 10332 + }, + { + "epoch": 1.9188486536675953, + "grad_norm": 1.400086522102356, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8788092732429504, + "num_tokens": 362866978.0, + "step": 10333 + }, + { + "epoch": 1.9190343546889508, + "grad_norm": 1.5383098125457764, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8598651885986328, + "num_tokens": 362903638.0, + "step": 10334 + }, + { + "epoch": 1.9192200557103063, + "grad_norm": 1.6402320861816406, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8925672769546509, + "num_tokens": 362927654.0, + "step": 10335 + }, + { + "epoch": 1.919405756731662, + "grad_norm": 1.6251120567321777, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8804023265838623, + "num_tokens": 362955748.0, + "step": 10336 + }, + { + "epoch": 1.9195914577530178, + "grad_norm": 1.3595860004425049, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8732184171676636, + "num_tokens": 362998343.0, + "step": 10337 + }, + { + "epoch": 1.9197771587743733, + "grad_norm": 1.3471574783325195, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8863439559936523, + "num_tokens": 363037723.0, + "step": 10338 + }, + { + "epoch": 1.9199628597957288, + "grad_norm": 1.2820402383804321, + "learning_rate": 1e-06, + "loss": 0.2996, + "mean_token_accuracy": 0.8964115381240845, + "num_tokens": 363078524.0, + "step": 10339 + }, + { + "epoch": 1.9201485608170845, + "grad_norm": 1.4872294664382935, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8718881607055664, + "num_tokens": 363114380.0, + "step": 10340 + }, + { + "epoch": 1.9203342618384402, + "grad_norm": 1.3695440292358398, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8848656415939331, + "num_tokens": 363151674.0, + "step": 10341 + }, + { + "epoch": 1.9205199628597958, + "grad_norm": 1.4462051391601562, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8800340890884399, + "num_tokens": 363190114.0, + "step": 10342 + }, + { + "epoch": 1.9207056638811513, + "grad_norm": 1.3437260389328003, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8868358135223389, + "num_tokens": 363229483.0, + "step": 10343 + }, + { + "epoch": 1.920891364902507, + "grad_norm": 1.4601587057113647, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8815112113952637, + "num_tokens": 363265666.0, + "step": 10344 + }, + { + "epoch": 1.9210770659238627, + "grad_norm": 1.5678669214248657, + "learning_rate": 1e-06, + "loss": 0.422, + "mean_token_accuracy": 0.8541007041931152, + "num_tokens": 363302106.0, + "step": 10345 + }, + { + "epoch": 1.9212627669452182, + "grad_norm": 1.5532108545303345, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8771194815635681, + "num_tokens": 363334698.0, + "step": 10346 + }, + { + "epoch": 1.9214484679665738, + "grad_norm": 1.4972290992736816, + "learning_rate": 1e-06, + "loss": 0.3969, + "mean_token_accuracy": 0.8628807067871094, + "num_tokens": 363371001.0, + "step": 10347 + }, + { + "epoch": 1.9216341689879295, + "grad_norm": 1.520427942276001, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.9003421664237976, + "num_tokens": 363398918.0, + "step": 10348 + }, + { + "epoch": 1.921819870009285, + "grad_norm": 1.3785711526870728, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.881998598575592, + "num_tokens": 363435693.0, + "step": 10349 + }, + { + "epoch": 1.9220055710306405, + "grad_norm": 1.5229028463363647, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8754386901855469, + "num_tokens": 363470060.0, + "step": 10350 + }, + { + "epoch": 1.9221912720519962, + "grad_norm": 1.4010365009307861, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8793854713439941, + "num_tokens": 363509354.0, + "step": 10351 + }, + { + "epoch": 1.922376973073352, + "grad_norm": 1.542720913887024, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8814126253128052, + "num_tokens": 363542986.0, + "step": 10352 + }, + { + "epoch": 1.9225626740947075, + "grad_norm": 1.4307204484939575, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8920681476593018, + "num_tokens": 363579008.0, + "step": 10353 + }, + { + "epoch": 1.922748375116063, + "grad_norm": 1.4863678216934204, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.87974613904953, + "num_tokens": 363613559.0, + "step": 10354 + }, + { + "epoch": 1.9229340761374187, + "grad_norm": 1.5952131748199463, + "learning_rate": 1e-06, + "loss": 0.4153, + "mean_token_accuracy": 0.860170304775238, + "num_tokens": 363647191.0, + "step": 10355 + }, + { + "epoch": 1.9231197771587745, + "grad_norm": 1.4890931844711304, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8828412890434265, + "num_tokens": 363680950.0, + "step": 10356 + }, + { + "epoch": 1.92330547818013, + "grad_norm": 1.4512838125228882, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8769535422325134, + "num_tokens": 363719215.0, + "step": 10357 + }, + { + "epoch": 1.9234911792014855, + "grad_norm": 1.6130188703536987, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8983333110809326, + "num_tokens": 363745299.0, + "step": 10358 + }, + { + "epoch": 1.9236768802228412, + "grad_norm": 1.540607213973999, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8758513927459717, + "num_tokens": 363779737.0, + "step": 10359 + }, + { + "epoch": 1.923862581244197, + "grad_norm": 1.421500563621521, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8668314218521118, + "num_tokens": 363816782.0, + "step": 10360 + }, + { + "epoch": 1.9240482822655525, + "grad_norm": 1.3996254205703735, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8739455342292786, + "num_tokens": 363861935.0, + "step": 10361 + }, + { + "epoch": 1.924233983286908, + "grad_norm": 1.6092119216918945, + "learning_rate": 1e-06, + "loss": 0.4134, + "mean_token_accuracy": 0.8602246642112732, + "num_tokens": 363895589.0, + "step": 10362 + }, + { + "epoch": 1.9244196843082637, + "grad_norm": 1.3673841953277588, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8912690281867981, + "num_tokens": 363933400.0, + "step": 10363 + }, + { + "epoch": 1.9246053853296194, + "grad_norm": 1.3176506757736206, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8632873892784119, + "num_tokens": 363979208.0, + "step": 10364 + }, + { + "epoch": 1.924791086350975, + "grad_norm": 1.4688879251480103, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8762235641479492, + "num_tokens": 364014634.0, + "step": 10365 + }, + { + "epoch": 1.9249767873723305, + "grad_norm": 1.5612479448318481, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8678430318832397, + "num_tokens": 364050556.0, + "step": 10366 + }, + { + "epoch": 1.9251624883936862, + "grad_norm": 1.441741704940796, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8833861351013184, + "num_tokens": 364088746.0, + "step": 10367 + }, + { + "epoch": 1.925348189415042, + "grad_norm": 1.5825691223144531, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8789930939674377, + "num_tokens": 364119303.0, + "step": 10368 + }, + { + "epoch": 1.9255338904363974, + "grad_norm": 1.4634456634521484, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8660920858383179, + "num_tokens": 364156812.0, + "step": 10369 + }, + { + "epoch": 1.925719591457753, + "grad_norm": 1.4837332963943481, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8770585656166077, + "num_tokens": 364199046.0, + "step": 10370 + }, + { + "epoch": 1.9259052924791087, + "grad_norm": 1.5916048288345337, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8807400465011597, + "num_tokens": 364229683.0, + "step": 10371 + }, + { + "epoch": 1.9260909935004642, + "grad_norm": 1.479608178138733, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8709476590156555, + "num_tokens": 364268705.0, + "step": 10372 + }, + { + "epoch": 1.9262766945218197, + "grad_norm": 1.4551135301589966, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8820770978927612, + "num_tokens": 364304589.0, + "step": 10373 + }, + { + "epoch": 1.9264623955431754, + "grad_norm": 1.5195139646530151, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8759894967079163, + "num_tokens": 364337238.0, + "step": 10374 + }, + { + "epoch": 1.9266480965645312, + "grad_norm": 1.5607331991195679, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8825822472572327, + "num_tokens": 364371886.0, + "step": 10375 + }, + { + "epoch": 1.9268337975858867, + "grad_norm": 1.4651525020599365, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8749620914459229, + "num_tokens": 364408636.0, + "step": 10376 + }, + { + "epoch": 1.9270194986072422, + "grad_norm": 1.5723706483840942, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8629934787750244, + "num_tokens": 364441425.0, + "step": 10377 + }, + { + "epoch": 1.927205199628598, + "grad_norm": 1.5969138145446777, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8829558491706848, + "num_tokens": 364468485.0, + "step": 10378 + }, + { + "epoch": 1.9273909006499537, + "grad_norm": 1.527929663658142, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8897802829742432, + "num_tokens": 364502798.0, + "step": 10379 + }, + { + "epoch": 1.9275766016713092, + "grad_norm": 1.446848750114441, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8753791451454163, + "num_tokens": 364541613.0, + "step": 10380 + }, + { + "epoch": 1.9277623026926647, + "grad_norm": 1.5048835277557373, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8743901252746582, + "num_tokens": 364576100.0, + "step": 10381 + }, + { + "epoch": 1.9279480037140204, + "grad_norm": 1.7052134275436401, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8754768371582031, + "num_tokens": 364604690.0, + "step": 10382 + }, + { + "epoch": 1.9281337047353762, + "grad_norm": 1.439171314239502, + "learning_rate": 1e-06, + "loss": 0.3916, + "mean_token_accuracy": 0.8630200028419495, + "num_tokens": 364645632.0, + "step": 10383 + }, + { + "epoch": 1.9283194057567317, + "grad_norm": 1.4187101125717163, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8790691494941711, + "num_tokens": 364682511.0, + "step": 10384 + }, + { + "epoch": 1.9285051067780872, + "grad_norm": 1.5389541387557983, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8759878873825073, + "num_tokens": 364715989.0, + "step": 10385 + }, + { + "epoch": 1.928690807799443, + "grad_norm": 1.4939219951629639, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8796433806419373, + "num_tokens": 364751178.0, + "step": 10386 + }, + { + "epoch": 1.9288765088207986, + "grad_norm": 1.5112124681472778, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8745689392089844, + "num_tokens": 364785765.0, + "step": 10387 + }, + { + "epoch": 1.9290622098421542, + "grad_norm": 1.4596470594406128, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8917301893234253, + "num_tokens": 364817020.0, + "step": 10388 + }, + { + "epoch": 1.9292479108635097, + "grad_norm": 1.5373053550720215, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.878242552280426, + "num_tokens": 364851448.0, + "step": 10389 + }, + { + "epoch": 1.9294336118848654, + "grad_norm": 1.6167445182800293, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8673926591873169, + "num_tokens": 364884192.0, + "step": 10390 + }, + { + "epoch": 1.9296193129062211, + "grad_norm": 1.5357476472854614, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8872203826904297, + "num_tokens": 364913482.0, + "step": 10391 + }, + { + "epoch": 1.9298050139275766, + "grad_norm": 1.3411588668823242, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8776944875717163, + "num_tokens": 364952516.0, + "step": 10392 + }, + { + "epoch": 1.9299907149489322, + "grad_norm": 1.4051321744918823, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8723627328872681, + "num_tokens": 364990272.0, + "step": 10393 + }, + { + "epoch": 1.9301764159702879, + "grad_norm": 1.5323505401611328, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8687673807144165, + "num_tokens": 365027667.0, + "step": 10394 + }, + { + "epoch": 1.9303621169916436, + "grad_norm": 1.3723713159561157, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8900488615036011, + "num_tokens": 365067236.0, + "step": 10395 + }, + { + "epoch": 1.930547818012999, + "grad_norm": 1.438718557357788, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8890135884284973, + "num_tokens": 365100094.0, + "step": 10396 + }, + { + "epoch": 1.9307335190343546, + "grad_norm": 1.568070650100708, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8837875127792358, + "num_tokens": 365131540.0, + "step": 10397 + }, + { + "epoch": 1.9309192200557104, + "grad_norm": 1.4342927932739258, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8838411569595337, + "num_tokens": 365168956.0, + "step": 10398 + }, + { + "epoch": 1.9311049210770659, + "grad_norm": 1.3628941774368286, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8846832513809204, + "num_tokens": 365206841.0, + "step": 10399 + }, + { + "epoch": 1.9312906220984214, + "grad_norm": 1.3936212062835693, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8765175342559814, + "num_tokens": 365248028.0, + "step": 10400 + }, + { + "epoch": 1.9314763231197771, + "grad_norm": 1.630143404006958, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8896253705024719, + "num_tokens": 365277223.0, + "step": 10401 + }, + { + "epoch": 1.9316620241411329, + "grad_norm": 1.4413752555847168, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8810802698135376, + "num_tokens": 365313560.0, + "step": 10402 + }, + { + "epoch": 1.9318477251624884, + "grad_norm": 1.3304115533828735, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.889072060585022, + "num_tokens": 365353533.0, + "step": 10403 + }, + { + "epoch": 1.9320334261838439, + "grad_norm": 1.498477816581726, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.874748945236206, + "num_tokens": 365388613.0, + "step": 10404 + }, + { + "epoch": 1.9322191272051996, + "grad_norm": 1.5441266298294067, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8820639848709106, + "num_tokens": 365423917.0, + "step": 10405 + }, + { + "epoch": 1.9324048282265553, + "grad_norm": 1.4187322854995728, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8768506050109863, + "num_tokens": 365459932.0, + "step": 10406 + }, + { + "epoch": 1.9325905292479109, + "grad_norm": 1.6817340850830078, + "learning_rate": 1e-06, + "loss": 0.3979, + "mean_token_accuracy": 0.8658145666122437, + "num_tokens": 365489892.0, + "step": 10407 + }, + { + "epoch": 1.9327762302692664, + "grad_norm": 1.4276610612869263, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8765550851821899, + "num_tokens": 365528629.0, + "step": 10408 + }, + { + "epoch": 1.932961931290622, + "grad_norm": 1.5377267599105835, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8806527853012085, + "num_tokens": 365559141.0, + "step": 10409 + }, + { + "epoch": 1.9331476323119778, + "grad_norm": 1.4058457612991333, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8727881908416748, + "num_tokens": 365596671.0, + "step": 10410 + }, + { + "epoch": 1.9333333333333333, + "grad_norm": 1.4503990411758423, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8821784257888794, + "num_tokens": 365634624.0, + "step": 10411 + }, + { + "epoch": 1.9335190343546889, + "grad_norm": 1.4167712926864624, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8862718343734741, + "num_tokens": 365671186.0, + "step": 10412 + }, + { + "epoch": 1.9337047353760446, + "grad_norm": 1.4147446155548096, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8796239495277405, + "num_tokens": 365705613.0, + "step": 10413 + }, + { + "epoch": 1.9338904363974003, + "grad_norm": 1.588230013847351, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8830273747444153, + "num_tokens": 365734837.0, + "step": 10414 + }, + { + "epoch": 1.9340761374187558, + "grad_norm": 1.5943645238876343, + "learning_rate": 1e-06, + "loss": 0.4107, + "mean_token_accuracy": 0.8614895939826965, + "num_tokens": 365772003.0, + "step": 10415 + }, + { + "epoch": 1.9342618384401113, + "grad_norm": 1.3779163360595703, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8777077198028564, + "num_tokens": 365813691.0, + "step": 10416 + }, + { + "epoch": 1.934447539461467, + "grad_norm": 1.6144431829452515, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8806707859039307, + "num_tokens": 365841604.0, + "step": 10417 + }, + { + "epoch": 1.9346332404828228, + "grad_norm": 1.3690211772918701, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8831318616867065, + "num_tokens": 365880893.0, + "step": 10418 + }, + { + "epoch": 1.9348189415041783, + "grad_norm": 1.5147883892059326, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8727200031280518, + "num_tokens": 365914974.0, + "step": 10419 + }, + { + "epoch": 1.9350046425255338, + "grad_norm": 1.5473620891571045, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8821282386779785, + "num_tokens": 365946200.0, + "step": 10420 + }, + { + "epoch": 1.9351903435468896, + "grad_norm": 1.4665617942810059, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8759243488311768, + "num_tokens": 365984778.0, + "step": 10421 + }, + { + "epoch": 1.935376044568245, + "grad_norm": 1.5594851970672607, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8822594285011292, + "num_tokens": 366016735.0, + "step": 10422 + }, + { + "epoch": 1.9355617455896006, + "grad_norm": 1.5151162147521973, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8787911534309387, + "num_tokens": 366053667.0, + "step": 10423 + }, + { + "epoch": 1.9357474466109563, + "grad_norm": 1.5454763174057007, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.889844536781311, + "num_tokens": 366083085.0, + "step": 10424 + }, + { + "epoch": 1.935933147632312, + "grad_norm": 1.4561561346054077, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.878413200378418, + "num_tokens": 366117643.0, + "step": 10425 + }, + { + "epoch": 1.9361188486536676, + "grad_norm": 1.7562426328659058, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8711971044540405, + "num_tokens": 366145608.0, + "step": 10426 + }, + { + "epoch": 1.936304549675023, + "grad_norm": 1.5247657299041748, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8730902671813965, + "num_tokens": 366179887.0, + "step": 10427 + }, + { + "epoch": 1.9364902506963788, + "grad_norm": 1.6458120346069336, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8712470531463623, + "num_tokens": 366211065.0, + "step": 10428 + }, + { + "epoch": 1.9366759517177345, + "grad_norm": 1.5448062419891357, + "learning_rate": 1e-06, + "loss": 0.3082, + "mean_token_accuracy": 0.8907727003097534, + "num_tokens": 366241530.0, + "step": 10429 + }, + { + "epoch": 1.93686165273909, + "grad_norm": 1.6549313068389893, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.870471715927124, + "num_tokens": 366271816.0, + "step": 10430 + }, + { + "epoch": 1.9370473537604456, + "grad_norm": 1.6115657091140747, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8727880716323853, + "num_tokens": 366306763.0, + "step": 10431 + }, + { + "epoch": 1.9372330547818013, + "grad_norm": 1.5149965286254883, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8805672526359558, + "num_tokens": 366341691.0, + "step": 10432 + }, + { + "epoch": 1.937418755803157, + "grad_norm": 1.4457393884658813, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8753724694252014, + "num_tokens": 366376085.0, + "step": 10433 + }, + { + "epoch": 1.9376044568245125, + "grad_norm": 1.474204182624817, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8739413619041443, + "num_tokens": 366411881.0, + "step": 10434 + }, + { + "epoch": 1.937790157845868, + "grad_norm": 1.4349149465560913, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8717774152755737, + "num_tokens": 366448960.0, + "step": 10435 + }, + { + "epoch": 1.9379758588672238, + "grad_norm": 1.64727783203125, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.886785626411438, + "num_tokens": 366477519.0, + "step": 10436 + }, + { + "epoch": 1.9381615598885795, + "grad_norm": 1.5288593769073486, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8743472099304199, + "num_tokens": 366509263.0, + "step": 10437 + }, + { + "epoch": 1.938347260909935, + "grad_norm": 1.434583306312561, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8866760730743408, + "num_tokens": 366542229.0, + "step": 10438 + }, + { + "epoch": 1.9385329619312905, + "grad_norm": 1.4460917711257935, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.87836092710495, + "num_tokens": 366581335.0, + "step": 10439 + }, + { + "epoch": 1.9387186629526463, + "grad_norm": 1.5222172737121582, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8717124462127686, + "num_tokens": 366615459.0, + "step": 10440 + }, + { + "epoch": 1.938904363974002, + "grad_norm": 1.3863662481307983, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8819667100906372, + "num_tokens": 366653789.0, + "step": 10441 + }, + { + "epoch": 1.9390900649953575, + "grad_norm": 1.557685136795044, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.890136182308197, + "num_tokens": 366685313.0, + "step": 10442 + }, + { + "epoch": 1.939275766016713, + "grad_norm": 1.678979516029358, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.879184365272522, + "num_tokens": 366714790.0, + "step": 10443 + }, + { + "epoch": 1.9394614670380688, + "grad_norm": 1.6976391077041626, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8810067176818848, + "num_tokens": 366740639.0, + "step": 10444 + }, + { + "epoch": 1.9396471680594243, + "grad_norm": 1.419313669204712, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8785401582717896, + "num_tokens": 366775991.0, + "step": 10445 + }, + { + "epoch": 1.9398328690807798, + "grad_norm": 1.3451260328292847, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8898016214370728, + "num_tokens": 366812292.0, + "step": 10446 + }, + { + "epoch": 1.9400185701021355, + "grad_norm": 1.6116682291030884, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8914715051651001, + "num_tokens": 366840081.0, + "step": 10447 + }, + { + "epoch": 1.9402042711234913, + "grad_norm": 1.445889949798584, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8826165199279785, + "num_tokens": 366874991.0, + "step": 10448 + }, + { + "epoch": 1.9403899721448468, + "grad_norm": 1.4643789529800415, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8832142353057861, + "num_tokens": 366909273.0, + "step": 10449 + }, + { + "epoch": 1.9405756731662023, + "grad_norm": 1.4221880435943604, + "learning_rate": 1e-06, + "loss": 0.4025, + "mean_token_accuracy": 0.8646149635314941, + "num_tokens": 366953553.0, + "step": 10450 + }, + { + "epoch": 1.940761374187558, + "grad_norm": 1.421341896057129, + "learning_rate": 1e-06, + "loss": 0.4085, + "mean_token_accuracy": 0.8617073893547058, + "num_tokens": 366994282.0, + "step": 10451 + }, + { + "epoch": 1.9409470752089137, + "grad_norm": 1.6284286975860596, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8758434057235718, + "num_tokens": 367024752.0, + "step": 10452 + }, + { + "epoch": 1.9411327762302693, + "grad_norm": 1.4681169986724854, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8807593584060669, + "num_tokens": 367061815.0, + "step": 10453 + }, + { + "epoch": 1.9413184772516248, + "grad_norm": 1.6010494232177734, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.867316484451294, + "num_tokens": 367094815.0, + "step": 10454 + }, + { + "epoch": 1.9415041782729805, + "grad_norm": 1.4517104625701904, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8817986249923706, + "num_tokens": 367130275.0, + "step": 10455 + }, + { + "epoch": 1.9416898792943362, + "grad_norm": 1.3459501266479492, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8812156915664673, + "num_tokens": 367173602.0, + "step": 10456 + }, + { + "epoch": 1.9418755803156917, + "grad_norm": 1.394016981124878, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8767459392547607, + "num_tokens": 367214399.0, + "step": 10457 + }, + { + "epoch": 1.9420612813370473, + "grad_norm": 1.4880996942520142, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8814220428466797, + "num_tokens": 367250576.0, + "step": 10458 + }, + { + "epoch": 1.942246982358403, + "grad_norm": 1.5999526977539062, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8697653412818909, + "num_tokens": 367280841.0, + "step": 10459 + }, + { + "epoch": 1.9424326833797587, + "grad_norm": 1.3967984914779663, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8849428296089172, + "num_tokens": 367317252.0, + "step": 10460 + }, + { + "epoch": 1.9426183844011142, + "grad_norm": 1.5127942562103271, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8842290639877319, + "num_tokens": 367350555.0, + "step": 10461 + }, + { + "epoch": 1.9428040854224697, + "grad_norm": 1.5141322612762451, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8661671876907349, + "num_tokens": 367383201.0, + "step": 10462 + }, + { + "epoch": 1.9429897864438255, + "grad_norm": 1.5836186408996582, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8802129030227661, + "num_tokens": 367412409.0, + "step": 10463 + }, + { + "epoch": 1.9431754874651812, + "grad_norm": 1.4383004903793335, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8835313320159912, + "num_tokens": 367444911.0, + "step": 10464 + }, + { + "epoch": 1.9433611884865367, + "grad_norm": 1.4798482656478882, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8906970024108887, + "num_tokens": 367479267.0, + "step": 10465 + }, + { + "epoch": 1.9435468895078922, + "grad_norm": 1.7278333902359009, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8826647400856018, + "num_tokens": 367505125.0, + "step": 10466 + }, + { + "epoch": 1.943732590529248, + "grad_norm": 1.4750639200210571, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8871757984161377, + "num_tokens": 367539978.0, + "step": 10467 + }, + { + "epoch": 1.9439182915506037, + "grad_norm": 1.4696391820907593, + "learning_rate": 1e-06, + "loss": 0.4061, + "mean_token_accuracy": 0.865284264087677, + "num_tokens": 367582905.0, + "step": 10468 + }, + { + "epoch": 1.944103992571959, + "grad_norm": 1.4161732196807861, + "learning_rate": 1e-06, + "loss": 0.2932, + "mean_token_accuracy": 0.8970427513122559, + "num_tokens": 367616603.0, + "step": 10469 + }, + { + "epoch": 1.9442896935933147, + "grad_norm": 1.4292479753494263, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8887110948562622, + "num_tokens": 367655373.0, + "step": 10470 + }, + { + "epoch": 1.9444753946146704, + "grad_norm": 1.453342318534851, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8758384585380554, + "num_tokens": 367689342.0, + "step": 10471 + }, + { + "epoch": 1.944661095636026, + "grad_norm": 1.4499735832214355, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8843342065811157, + "num_tokens": 367727011.0, + "step": 10472 + }, + { + "epoch": 1.9448467966573815, + "grad_norm": 1.4650083780288696, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8832225203514099, + "num_tokens": 367759504.0, + "step": 10473 + }, + { + "epoch": 1.9450324976787372, + "grad_norm": 1.6010483503341675, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.877424955368042, + "num_tokens": 367790945.0, + "step": 10474 + }, + { + "epoch": 1.945218198700093, + "grad_norm": 1.447875738143921, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8646148443222046, + "num_tokens": 367833902.0, + "step": 10475 + }, + { + "epoch": 1.9454038997214484, + "grad_norm": 1.422773838043213, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8731831312179565, + "num_tokens": 367874859.0, + "step": 10476 + }, + { + "epoch": 1.945589600742804, + "grad_norm": 1.4173225164413452, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8836886882781982, + "num_tokens": 367911704.0, + "step": 10477 + }, + { + "epoch": 1.9457753017641597, + "grad_norm": 1.3901302814483643, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8892182111740112, + "num_tokens": 367949079.0, + "step": 10478 + }, + { + "epoch": 1.9459610027855154, + "grad_norm": 1.5833666324615479, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8855686187744141, + "num_tokens": 367979911.0, + "step": 10479 + }, + { + "epoch": 1.946146703806871, + "grad_norm": 1.4474124908447266, + "learning_rate": 1e-06, + "loss": 0.3996, + "mean_token_accuracy": 0.8670583367347717, + "num_tokens": 368021375.0, + "step": 10480 + }, + { + "epoch": 1.9463324048282264, + "grad_norm": 1.5630828142166138, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8755078911781311, + "num_tokens": 368056111.0, + "step": 10481 + }, + { + "epoch": 1.9465181058495822, + "grad_norm": 1.667886734008789, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8744006752967834, + "num_tokens": 368083312.0, + "step": 10482 + }, + { + "epoch": 1.946703806870938, + "grad_norm": 1.386502742767334, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8753842711448669, + "num_tokens": 368122113.0, + "step": 10483 + }, + { + "epoch": 1.9468895078922934, + "grad_norm": 1.4480276107788086, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8716391324996948, + "num_tokens": 368165141.0, + "step": 10484 + }, + { + "epoch": 1.947075208913649, + "grad_norm": 1.4240752458572388, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8749029636383057, + "num_tokens": 368202619.0, + "step": 10485 + }, + { + "epoch": 1.9472609099350047, + "grad_norm": 1.6155308485031128, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8697270154953003, + "num_tokens": 368236312.0, + "step": 10486 + }, + { + "epoch": 1.9474466109563604, + "grad_norm": 1.509408950805664, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8724489808082581, + "num_tokens": 368270956.0, + "step": 10487 + }, + { + "epoch": 1.947632311977716, + "grad_norm": 1.5314184427261353, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8772450685501099, + "num_tokens": 368304881.0, + "step": 10488 + }, + { + "epoch": 1.9478180129990714, + "grad_norm": 1.4733167886734009, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8811648488044739, + "num_tokens": 368339908.0, + "step": 10489 + }, + { + "epoch": 1.9480037140204272, + "grad_norm": 1.448451280593872, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8867011070251465, + "num_tokens": 368374517.0, + "step": 10490 + }, + { + "epoch": 1.948189415041783, + "grad_norm": 1.55381441116333, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8837484121322632, + "num_tokens": 368406918.0, + "step": 10491 + }, + { + "epoch": 1.9483751160631384, + "grad_norm": 1.4101790189743042, + "learning_rate": 1e-06, + "loss": 0.3879, + "mean_token_accuracy": 0.8690418601036072, + "num_tokens": 368448744.0, + "step": 10492 + }, + { + "epoch": 1.948560817084494, + "grad_norm": 1.4943963289260864, + "learning_rate": 1e-06, + "loss": 0.3909, + "mean_token_accuracy": 0.8682993054389954, + "num_tokens": 368484058.0, + "step": 10493 + }, + { + "epoch": 1.9487465181058496, + "grad_norm": 1.6610081195831299, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8697380423545837, + "num_tokens": 368512114.0, + "step": 10494 + }, + { + "epoch": 1.9489322191272052, + "grad_norm": 1.6752541065216064, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8847976922988892, + "num_tokens": 368539677.0, + "step": 10495 + }, + { + "epoch": 1.9491179201485607, + "grad_norm": 1.4370607137680054, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8904991149902344, + "num_tokens": 368573324.0, + "step": 10496 + }, + { + "epoch": 1.9493036211699164, + "grad_norm": 1.4365503787994385, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8723003268241882, + "num_tokens": 368609868.0, + "step": 10497 + }, + { + "epoch": 1.9494893221912721, + "grad_norm": 1.5077099800109863, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8639350533485413, + "num_tokens": 368644396.0, + "step": 10498 + }, + { + "epoch": 1.9496750232126276, + "grad_norm": 1.4534083604812622, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8640737533569336, + "num_tokens": 368682480.0, + "step": 10499 + }, + { + "epoch": 1.9498607242339832, + "grad_norm": 1.5359923839569092, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8699870705604553, + "num_tokens": 368714283.0, + "step": 10500 + }, + { + "epoch": 1.9500464252553389, + "grad_norm": 1.5423938035964966, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8823035955429077, + "num_tokens": 368746265.0, + "step": 10501 + }, + { + "epoch": 1.9502321262766946, + "grad_norm": 1.66069757938385, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8676456212997437, + "num_tokens": 368776360.0, + "step": 10502 + }, + { + "epoch": 1.9504178272980501, + "grad_norm": 1.4787743091583252, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8816722631454468, + "num_tokens": 368811223.0, + "step": 10503 + }, + { + "epoch": 1.9506035283194056, + "grad_norm": 1.6634316444396973, + "learning_rate": 1e-06, + "loss": 0.4256, + "mean_token_accuracy": 0.8568642139434814, + "num_tokens": 368842919.0, + "step": 10504 + }, + { + "epoch": 1.9507892293407614, + "grad_norm": 1.447019100189209, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.882262110710144, + "num_tokens": 368877136.0, + "step": 10505 + }, + { + "epoch": 1.950974930362117, + "grad_norm": 1.5080301761627197, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8880090117454529, + "num_tokens": 368909778.0, + "step": 10506 + }, + { + "epoch": 1.9511606313834726, + "grad_norm": 1.3345882892608643, + "learning_rate": 1e-06, + "loss": 0.2964, + "mean_token_accuracy": 0.8954945802688599, + "num_tokens": 368943564.0, + "step": 10507 + }, + { + "epoch": 1.9513463324048281, + "grad_norm": 1.3774265050888062, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8694398999214172, + "num_tokens": 368982952.0, + "step": 10508 + }, + { + "epoch": 1.9515320334261839, + "grad_norm": 1.4525072574615479, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.878059983253479, + "num_tokens": 369018777.0, + "step": 10509 + }, + { + "epoch": 1.9517177344475396, + "grad_norm": 1.5641365051269531, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8890727162361145, + "num_tokens": 369052050.0, + "step": 10510 + }, + { + "epoch": 1.951903435468895, + "grad_norm": 1.3518588542938232, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8708650469779968, + "num_tokens": 369095798.0, + "step": 10511 + }, + { + "epoch": 1.9520891364902506, + "grad_norm": 1.4735957384109497, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8778064250946045, + "num_tokens": 369131893.0, + "step": 10512 + }, + { + "epoch": 1.9522748375116064, + "grad_norm": 1.5863384008407593, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8652543425559998, + "num_tokens": 369167590.0, + "step": 10513 + }, + { + "epoch": 1.952460538532962, + "grad_norm": 1.573550820350647, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8764333724975586, + "num_tokens": 369202987.0, + "step": 10514 + }, + { + "epoch": 1.9526462395543176, + "grad_norm": 1.3981478214263916, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8807023763656616, + "num_tokens": 369241065.0, + "step": 10515 + }, + { + "epoch": 1.952831940575673, + "grad_norm": 1.6487787961959839, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.878976583480835, + "num_tokens": 369269163.0, + "step": 10516 + }, + { + "epoch": 1.9530176415970288, + "grad_norm": 1.5739789009094238, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8900129795074463, + "num_tokens": 369299224.0, + "step": 10517 + }, + { + "epoch": 1.9532033426183844, + "grad_norm": 1.4775800704956055, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.873751699924469, + "num_tokens": 369343337.0, + "step": 10518 + }, + { + "epoch": 1.9533890436397399, + "grad_norm": 1.449442744255066, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8635944128036499, + "num_tokens": 369383513.0, + "step": 10519 + }, + { + "epoch": 1.9535747446610956, + "grad_norm": 1.3337280750274658, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.87945556640625, + "num_tokens": 369423413.0, + "step": 10520 + }, + { + "epoch": 1.9537604456824513, + "grad_norm": 1.5736740827560425, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8645657300949097, + "num_tokens": 369456067.0, + "step": 10521 + }, + { + "epoch": 1.9539461467038068, + "grad_norm": 1.6000902652740479, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8758615851402283, + "num_tokens": 369490188.0, + "step": 10522 + }, + { + "epoch": 1.9541318477251624, + "grad_norm": 1.669302225112915, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8878526091575623, + "num_tokens": 369515539.0, + "step": 10523 + }, + { + "epoch": 1.954317548746518, + "grad_norm": 1.3701368570327759, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8808826208114624, + "num_tokens": 369554809.0, + "step": 10524 + }, + { + "epoch": 1.9545032497678738, + "grad_norm": 1.46781325340271, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8747504949569702, + "num_tokens": 369591517.0, + "step": 10525 + }, + { + "epoch": 1.9546889507892293, + "grad_norm": 1.4338730573654175, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8755243420600891, + "num_tokens": 369626865.0, + "step": 10526 + }, + { + "epoch": 1.9548746518105848, + "grad_norm": 1.460500955581665, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8752270936965942, + "num_tokens": 369662305.0, + "step": 10527 + }, + { + "epoch": 1.9550603528319406, + "grad_norm": 1.6528609991073608, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.8720159530639648, + "num_tokens": 369697799.0, + "step": 10528 + }, + { + "epoch": 1.9552460538532963, + "grad_norm": 1.5733203887939453, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.8719737529754639, + "num_tokens": 369734671.0, + "step": 10529 + }, + { + "epoch": 1.9554317548746518, + "grad_norm": 1.469521164894104, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8751577734947205, + "num_tokens": 369774735.0, + "step": 10530 + }, + { + "epoch": 1.9556174558960073, + "grad_norm": 1.4104193449020386, + "learning_rate": 1e-06, + "loss": 0.2798, + "mean_token_accuracy": 0.8989558219909668, + "num_tokens": 369806834.0, + "step": 10531 + }, + { + "epoch": 1.955803156917363, + "grad_norm": 1.4505209922790527, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8722985982894897, + "num_tokens": 369845091.0, + "step": 10532 + }, + { + "epoch": 1.9559888579387188, + "grad_norm": 1.5815184116363525, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8765065670013428, + "num_tokens": 369876033.0, + "step": 10533 + }, + { + "epoch": 1.9561745589600743, + "grad_norm": 1.5665864944458008, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8917031288146973, + "num_tokens": 369909221.0, + "step": 10534 + }, + { + "epoch": 1.9563602599814298, + "grad_norm": 1.4051116704940796, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8858082294464111, + "num_tokens": 369948446.0, + "step": 10535 + }, + { + "epoch": 1.9565459610027855, + "grad_norm": 1.572307825088501, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8783608675003052, + "num_tokens": 369982948.0, + "step": 10536 + }, + { + "epoch": 1.9567316620241413, + "grad_norm": 1.4215574264526367, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8698880672454834, + "num_tokens": 370022165.0, + "step": 10537 + }, + { + "epoch": 1.9569173630454968, + "grad_norm": 1.5201499462127686, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8826769590377808, + "num_tokens": 370056761.0, + "step": 10538 + }, + { + "epoch": 1.9571030640668523, + "grad_norm": 1.4976812601089478, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8883181214332581, + "num_tokens": 370090891.0, + "step": 10539 + }, + { + "epoch": 1.957288765088208, + "grad_norm": 1.4248589277267456, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8828684091567993, + "num_tokens": 370127816.0, + "step": 10540 + }, + { + "epoch": 1.9574744661095635, + "grad_norm": 1.6569807529449463, + "learning_rate": 1e-06, + "loss": 0.3789, + "mean_token_accuracy": 0.8687966465950012, + "num_tokens": 370164464.0, + "step": 10541 + }, + { + "epoch": 1.957660167130919, + "grad_norm": 1.5179795026779175, + "learning_rate": 1e-06, + "loss": 0.3945, + "mean_token_accuracy": 0.8624954223632812, + "num_tokens": 370204856.0, + "step": 10542 + }, + { + "epoch": 1.9578458681522748, + "grad_norm": 1.4346261024475098, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8855950236320496, + "num_tokens": 370240446.0, + "step": 10543 + }, + { + "epoch": 1.9580315691736305, + "grad_norm": 1.58798348903656, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8717653751373291, + "num_tokens": 370277631.0, + "step": 10544 + }, + { + "epoch": 1.958217270194986, + "grad_norm": 1.540840983390808, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.870629072189331, + "num_tokens": 370311476.0, + "step": 10545 + }, + { + "epoch": 1.9584029712163415, + "grad_norm": 1.6022770404815674, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.872639000415802, + "num_tokens": 370343916.0, + "step": 10546 + }, + { + "epoch": 1.9585886722376973, + "grad_norm": 1.3439773321151733, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8803694248199463, + "num_tokens": 370384053.0, + "step": 10547 + }, + { + "epoch": 1.958774373259053, + "grad_norm": 1.5292696952819824, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8886229991912842, + "num_tokens": 370412293.0, + "step": 10548 + }, + { + "epoch": 1.9589600742804085, + "grad_norm": 1.5031042098999023, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8731105923652649, + "num_tokens": 370447473.0, + "step": 10549 + }, + { + "epoch": 1.959145775301764, + "grad_norm": 1.4785337448120117, + "learning_rate": 1e-06, + "loss": 0.3744, + "mean_token_accuracy": 0.8696197271347046, + "num_tokens": 370479996.0, + "step": 10550 + }, + { + "epoch": 1.9593314763231198, + "grad_norm": 1.5918596982955933, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8734678030014038, + "num_tokens": 370512278.0, + "step": 10551 + }, + { + "epoch": 1.9595171773444755, + "grad_norm": 1.4302842617034912, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8752611875534058, + "num_tokens": 370553370.0, + "step": 10552 + }, + { + "epoch": 1.959702878365831, + "grad_norm": 1.4758634567260742, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8652597665786743, + "num_tokens": 370591818.0, + "step": 10553 + }, + { + "epoch": 1.9598885793871865, + "grad_norm": 1.4380760192871094, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8806436061859131, + "num_tokens": 370628316.0, + "step": 10554 + }, + { + "epoch": 1.9600742804085423, + "grad_norm": 1.5104187726974487, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8765460252761841, + "num_tokens": 370660555.0, + "step": 10555 + }, + { + "epoch": 1.960259981429898, + "grad_norm": 1.4747579097747803, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8768796920776367, + "num_tokens": 370696462.0, + "step": 10556 + }, + { + "epoch": 1.9604456824512535, + "grad_norm": 1.5101876258850098, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8757894039154053, + "num_tokens": 370732553.0, + "step": 10557 + }, + { + "epoch": 1.960631383472609, + "grad_norm": 1.3716782331466675, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8886733055114746, + "num_tokens": 370768627.0, + "step": 10558 + }, + { + "epoch": 1.9608170844939647, + "grad_norm": 1.599124550819397, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8807591199874878, + "num_tokens": 370798464.0, + "step": 10559 + }, + { + "epoch": 1.9610027855153205, + "grad_norm": 1.5802361965179443, + "learning_rate": 1e-06, + "loss": 0.4233, + "mean_token_accuracy": 0.8584250211715698, + "num_tokens": 370833739.0, + "step": 10560 + }, + { + "epoch": 1.961188486536676, + "grad_norm": 1.6127022504806519, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8778014183044434, + "num_tokens": 370864094.0, + "step": 10561 + }, + { + "epoch": 1.9613741875580315, + "grad_norm": 1.7045133113861084, + "learning_rate": 1e-06, + "loss": 0.4336, + "mean_token_accuracy": 0.8591047525405884, + "num_tokens": 370899560.0, + "step": 10562 + }, + { + "epoch": 1.9615598885793872, + "grad_norm": 1.4540412425994873, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8859338760375977, + "num_tokens": 370934866.0, + "step": 10563 + }, + { + "epoch": 1.961745589600743, + "grad_norm": 1.4671686887741089, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8890297412872314, + "num_tokens": 370972504.0, + "step": 10564 + }, + { + "epoch": 1.9619312906220983, + "grad_norm": 1.5172144174575806, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8908835053443909, + "num_tokens": 371007362.0, + "step": 10565 + }, + { + "epoch": 1.962116991643454, + "grad_norm": 1.6299498081207275, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8906377553939819, + "num_tokens": 371033650.0, + "step": 10566 + }, + { + "epoch": 1.9623026926648097, + "grad_norm": 1.2872929573059082, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8843939304351807, + "num_tokens": 371075283.0, + "step": 10567 + }, + { + "epoch": 1.9624883936861652, + "grad_norm": 1.4095547199249268, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8711487054824829, + "num_tokens": 371113332.0, + "step": 10568 + }, + { + "epoch": 1.9626740947075207, + "grad_norm": 1.5145167112350464, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8859701752662659, + "num_tokens": 371142941.0, + "step": 10569 + }, + { + "epoch": 1.9628597957288765, + "grad_norm": 1.484338402748108, + "learning_rate": 1e-06, + "loss": 0.4039, + "mean_token_accuracy": 0.8661577701568604, + "num_tokens": 371177729.0, + "step": 10570 + }, + { + "epoch": 1.9630454967502322, + "grad_norm": 1.4751131534576416, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8842192888259888, + "num_tokens": 371209596.0, + "step": 10571 + }, + { + "epoch": 1.9632311977715877, + "grad_norm": 1.5485345125198364, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8666020631790161, + "num_tokens": 371242653.0, + "step": 10572 + }, + { + "epoch": 1.9634168987929432, + "grad_norm": 1.677384614944458, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8713344931602478, + "num_tokens": 371272695.0, + "step": 10573 + }, + { + "epoch": 1.963602599814299, + "grad_norm": 1.4907231330871582, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8887646794319153, + "num_tokens": 371307671.0, + "step": 10574 + }, + { + "epoch": 1.9637883008356547, + "grad_norm": 1.4495612382888794, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.8747317790985107, + "num_tokens": 371348946.0, + "step": 10575 + }, + { + "epoch": 1.9639740018570102, + "grad_norm": 1.506889820098877, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8717398643493652, + "num_tokens": 371383784.0, + "step": 10576 + }, + { + "epoch": 1.9641597028783657, + "grad_norm": 1.3915529251098633, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8899483680725098, + "num_tokens": 371418936.0, + "step": 10577 + }, + { + "epoch": 1.9643454038997215, + "grad_norm": 1.432287573814392, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8834609389305115, + "num_tokens": 371456874.0, + "step": 10578 + }, + { + "epoch": 1.9645311049210772, + "grad_norm": 1.6326993703842163, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8816566467285156, + "num_tokens": 371493382.0, + "step": 10579 + }, + { + "epoch": 1.9647168059424327, + "grad_norm": 1.6368157863616943, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.866983950138092, + "num_tokens": 371524448.0, + "step": 10580 + }, + { + "epoch": 1.9649025069637882, + "grad_norm": 1.3117536306381226, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.887545108795166, + "num_tokens": 371562749.0, + "step": 10581 + }, + { + "epoch": 1.965088207985144, + "grad_norm": 1.8628705739974976, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.870867133140564, + "num_tokens": 371587764.0, + "step": 10582 + }, + { + "epoch": 1.9652739090064997, + "grad_norm": 1.6383148431777954, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8829132318496704, + "num_tokens": 371617175.0, + "step": 10583 + }, + { + "epoch": 1.9654596100278552, + "grad_norm": 1.4229472875595093, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8656755089759827, + "num_tokens": 371659947.0, + "step": 10584 + }, + { + "epoch": 1.9656453110492107, + "grad_norm": 1.6303660869598389, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.872062623500824, + "num_tokens": 371692389.0, + "step": 10585 + }, + { + "epoch": 1.9658310120705664, + "grad_norm": 1.529966115951538, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8728504180908203, + "num_tokens": 371725034.0, + "step": 10586 + }, + { + "epoch": 1.9660167130919222, + "grad_norm": 1.3733192682266235, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.891331672668457, + "num_tokens": 371760773.0, + "step": 10587 + }, + { + "epoch": 1.9662024141132777, + "grad_norm": 1.4684624671936035, + "learning_rate": 1e-06, + "loss": 0.4055, + "mean_token_accuracy": 0.86400306224823, + "num_tokens": 371799114.0, + "step": 10588 + }, + { + "epoch": 1.9663881151346332, + "grad_norm": 1.5299220085144043, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8711746335029602, + "num_tokens": 371832305.0, + "step": 10589 + }, + { + "epoch": 1.966573816155989, + "grad_norm": 1.64337158203125, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8768329620361328, + "num_tokens": 371862858.0, + "step": 10590 + }, + { + "epoch": 1.9667595171773444, + "grad_norm": 1.4480390548706055, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8845860362052917, + "num_tokens": 371898322.0, + "step": 10591 + }, + { + "epoch": 1.9669452181987, + "grad_norm": 1.3903083801269531, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8623663187026978, + "num_tokens": 371940107.0, + "step": 10592 + }, + { + "epoch": 1.9671309192200557, + "grad_norm": 1.4173158407211304, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8939043283462524, + "num_tokens": 371975149.0, + "step": 10593 + }, + { + "epoch": 1.9673166202414114, + "grad_norm": 1.4816707372665405, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8732572793960571, + "num_tokens": 372011101.0, + "step": 10594 + }, + { + "epoch": 1.967502321262767, + "grad_norm": 1.511784553527832, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8659429550170898, + "num_tokens": 372047697.0, + "step": 10595 + }, + { + "epoch": 1.9676880222841224, + "grad_norm": 1.6879372596740723, + "learning_rate": 1e-06, + "loss": 0.3838, + "mean_token_accuracy": 0.8759375810623169, + "num_tokens": 372081731.0, + "step": 10596 + }, + { + "epoch": 1.9678737233054782, + "grad_norm": 1.4535056352615356, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8832137584686279, + "num_tokens": 372114991.0, + "step": 10597 + }, + { + "epoch": 1.968059424326834, + "grad_norm": 1.301591396331787, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8921430110931396, + "num_tokens": 372155644.0, + "step": 10598 + }, + { + "epoch": 1.9682451253481894, + "grad_norm": 1.5007448196411133, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8808125853538513, + "num_tokens": 372186078.0, + "step": 10599 + }, + { + "epoch": 1.968430826369545, + "grad_norm": 1.470309853553772, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8917301893234253, + "num_tokens": 372217557.0, + "step": 10600 + }, + { + "epoch": 1.9686165273909007, + "grad_norm": 1.4729485511779785, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8747602701187134, + "num_tokens": 372254140.0, + "step": 10601 + }, + { + "epoch": 1.9688022284122564, + "grad_norm": 1.449264407157898, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8765612244606018, + "num_tokens": 372289430.0, + "step": 10602 + }, + { + "epoch": 1.968987929433612, + "grad_norm": 1.5671268701553345, + "learning_rate": 1e-06, + "loss": 0.4447, + "mean_token_accuracy": 0.849309504032135, + "num_tokens": 372323370.0, + "step": 10603 + }, + { + "epoch": 1.9691736304549674, + "grad_norm": 1.588789939880371, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8745120167732239, + "num_tokens": 372352418.0, + "step": 10604 + }, + { + "epoch": 1.9693593314763231, + "grad_norm": 1.4898953437805176, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.874862790107727, + "num_tokens": 372385421.0, + "step": 10605 + }, + { + "epoch": 1.9695450324976789, + "grad_norm": 1.4428871870040894, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8887218236923218, + "num_tokens": 372418485.0, + "step": 10606 + }, + { + "epoch": 1.9697307335190344, + "grad_norm": 1.3319891691207886, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8871111273765564, + "num_tokens": 372458970.0, + "step": 10607 + }, + { + "epoch": 1.96991643454039, + "grad_norm": 1.633319616317749, + "learning_rate": 1e-06, + "loss": 0.4229, + "mean_token_accuracy": 0.8581719398498535, + "num_tokens": 372491595.0, + "step": 10608 + }, + { + "epoch": 1.9701021355617456, + "grad_norm": 1.3372663259506226, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8903412818908691, + "num_tokens": 372531700.0, + "step": 10609 + }, + { + "epoch": 1.9702878365831014, + "grad_norm": 1.476779580116272, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8776149153709412, + "num_tokens": 372568793.0, + "step": 10610 + }, + { + "epoch": 1.9704735376044569, + "grad_norm": 1.6222087144851685, + "learning_rate": 1e-06, + "loss": 0.4183, + "mean_token_accuracy": 0.8597879409790039, + "num_tokens": 372602196.0, + "step": 10611 + }, + { + "epoch": 1.9706592386258124, + "grad_norm": 1.4689151048660278, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8882490396499634, + "num_tokens": 372633843.0, + "step": 10612 + }, + { + "epoch": 1.9708449396471681, + "grad_norm": 1.5310404300689697, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8719732165336609, + "num_tokens": 372672807.0, + "step": 10613 + }, + { + "epoch": 1.9710306406685236, + "grad_norm": 1.7550630569458008, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8902480602264404, + "num_tokens": 372704031.0, + "step": 10614 + }, + { + "epoch": 1.9712163416898791, + "grad_norm": 1.3538658618927002, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8829809427261353, + "num_tokens": 372746619.0, + "step": 10615 + }, + { + "epoch": 1.9714020427112349, + "grad_norm": 1.494044303894043, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8715529441833496, + "num_tokens": 372782383.0, + "step": 10616 + }, + { + "epoch": 1.9715877437325906, + "grad_norm": 1.4991230964660645, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8725711107254028, + "num_tokens": 372815335.0, + "step": 10617 + }, + { + "epoch": 1.9717734447539461, + "grad_norm": 1.4271957874298096, + "learning_rate": 1e-06, + "loss": 0.3929, + "mean_token_accuracy": 0.8710493445396423, + "num_tokens": 372854550.0, + "step": 10618 + }, + { + "epoch": 1.9719591457753016, + "grad_norm": 1.5576896667480469, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8880667686462402, + "num_tokens": 372882833.0, + "step": 10619 + }, + { + "epoch": 1.9721448467966574, + "grad_norm": 1.6198904514312744, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.885931134223938, + "num_tokens": 372910856.0, + "step": 10620 + }, + { + "epoch": 1.972330547818013, + "grad_norm": 1.390291690826416, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8773374557495117, + "num_tokens": 372950178.0, + "step": 10621 + }, + { + "epoch": 1.9725162488393686, + "grad_norm": 1.545655608177185, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8762974739074707, + "num_tokens": 372980791.0, + "step": 10622 + }, + { + "epoch": 1.9727019498607241, + "grad_norm": 1.423251748085022, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8839249610900879, + "num_tokens": 373018338.0, + "step": 10623 + }, + { + "epoch": 1.9728876508820798, + "grad_norm": 1.3815337419509888, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8882955312728882, + "num_tokens": 373056237.0, + "step": 10624 + }, + { + "epoch": 1.9730733519034356, + "grad_norm": 1.4155099391937256, + "learning_rate": 1e-06, + "loss": 0.2887, + "mean_token_accuracy": 0.896294355392456, + "num_tokens": 373087440.0, + "step": 10625 + }, + { + "epoch": 1.973259052924791, + "grad_norm": 1.4220318794250488, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8898577690124512, + "num_tokens": 373121053.0, + "step": 10626 + }, + { + "epoch": 1.9734447539461466, + "grad_norm": 1.358633279800415, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8649550676345825, + "num_tokens": 373167483.0, + "step": 10627 + }, + { + "epoch": 1.9736304549675023, + "grad_norm": 1.5479646921157837, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8711027503013611, + "num_tokens": 373204578.0, + "step": 10628 + }, + { + "epoch": 1.973816155988858, + "grad_norm": 1.496979832649231, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8757815361022949, + "num_tokens": 373240283.0, + "step": 10629 + }, + { + "epoch": 1.9740018570102136, + "grad_norm": 1.8138794898986816, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.875220537185669, + "num_tokens": 373266444.0, + "step": 10630 + }, + { + "epoch": 1.974187558031569, + "grad_norm": 1.5266070365905762, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.865851879119873, + "num_tokens": 373302832.0, + "step": 10631 + }, + { + "epoch": 1.9743732590529248, + "grad_norm": 1.565361738204956, + "learning_rate": 1e-06, + "loss": 0.4226, + "mean_token_accuracy": 0.8644397258758545, + "num_tokens": 373337714.0, + "step": 10632 + }, + { + "epoch": 1.9745589600742806, + "grad_norm": 1.6045420169830322, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8648892641067505, + "num_tokens": 373372043.0, + "step": 10633 + }, + { + "epoch": 1.974744661095636, + "grad_norm": 1.4944208860397339, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8760093450546265, + "num_tokens": 373409289.0, + "step": 10634 + }, + { + "epoch": 1.9749303621169916, + "grad_norm": 1.5108251571655273, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8780182600021362, + "num_tokens": 373442988.0, + "step": 10635 + }, + { + "epoch": 1.9751160631383473, + "grad_norm": 1.5445621013641357, + "learning_rate": 1e-06, + "loss": 0.3881, + "mean_token_accuracy": 0.8684914112091064, + "num_tokens": 373478373.0, + "step": 10636 + }, + { + "epoch": 1.975301764159703, + "grad_norm": 1.4410444498062134, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8789006471633911, + "num_tokens": 373516034.0, + "step": 10637 + }, + { + "epoch": 1.9754874651810583, + "grad_norm": 1.4370205402374268, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8894714713096619, + "num_tokens": 373552726.0, + "step": 10638 + }, + { + "epoch": 1.975673166202414, + "grad_norm": 1.4760645627975464, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8704562187194824, + "num_tokens": 373587581.0, + "step": 10639 + }, + { + "epoch": 1.9758588672237698, + "grad_norm": 1.4091432094573975, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8952258229255676, + "num_tokens": 373619848.0, + "step": 10640 + }, + { + "epoch": 1.9760445682451253, + "grad_norm": 1.5545010566711426, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8733419179916382, + "num_tokens": 373649160.0, + "step": 10641 + }, + { + "epoch": 1.9762302692664808, + "grad_norm": 1.585790991783142, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.881216824054718, + "num_tokens": 373679516.0, + "step": 10642 + }, + { + "epoch": 1.9764159702878366, + "grad_norm": 1.3952791690826416, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8814963102340698, + "num_tokens": 373720433.0, + "step": 10643 + }, + { + "epoch": 1.9766016713091923, + "grad_norm": 1.507411241531372, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8815852403640747, + "num_tokens": 373756465.0, + "step": 10644 + }, + { + "epoch": 1.9767873723305478, + "grad_norm": 1.5075764656066895, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8938703536987305, + "num_tokens": 373791139.0, + "step": 10645 + }, + { + "epoch": 1.9769730733519033, + "grad_norm": 1.395105004310608, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8736448884010315, + "num_tokens": 373831859.0, + "step": 10646 + }, + { + "epoch": 1.977158774373259, + "grad_norm": 1.4507054090499878, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8854036927223206, + "num_tokens": 373866389.0, + "step": 10647 + }, + { + "epoch": 1.9773444753946148, + "grad_norm": 1.609206199645996, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8809860944747925, + "num_tokens": 373897505.0, + "step": 10648 + }, + { + "epoch": 1.9775301764159703, + "grad_norm": 1.3696074485778809, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.890339732170105, + "num_tokens": 373934598.0, + "step": 10649 + }, + { + "epoch": 1.9777158774373258, + "grad_norm": 1.5638782978057861, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8731933236122131, + "num_tokens": 373968182.0, + "step": 10650 + }, + { + "epoch": 1.9779015784586815, + "grad_norm": 1.4654743671417236, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.871238112449646, + "num_tokens": 374003157.0, + "step": 10651 + }, + { + "epoch": 1.9780872794800373, + "grad_norm": 1.432658314704895, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8672304749488831, + "num_tokens": 374046304.0, + "step": 10652 + }, + { + "epoch": 1.9782729805013928, + "grad_norm": 1.5140236616134644, + "learning_rate": 1e-06, + "loss": 0.392, + "mean_token_accuracy": 0.8687800765037537, + "num_tokens": 374081604.0, + "step": 10653 + }, + { + "epoch": 1.9784586815227483, + "grad_norm": 1.4922749996185303, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8796354532241821, + "num_tokens": 374116815.0, + "step": 10654 + }, + { + "epoch": 1.978644382544104, + "grad_norm": 1.5358989238739014, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.864133894443512, + "num_tokens": 374150523.0, + "step": 10655 + }, + { + "epoch": 1.9788300835654598, + "grad_norm": 1.3608710765838623, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8768285512924194, + "num_tokens": 374193530.0, + "step": 10656 + }, + { + "epoch": 1.9790157845868153, + "grad_norm": 1.5097397565841675, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8863514065742493, + "num_tokens": 374229310.0, + "step": 10657 + }, + { + "epoch": 1.9792014856081708, + "grad_norm": 1.420630931854248, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8820672631263733, + "num_tokens": 374267710.0, + "step": 10658 + }, + { + "epoch": 1.9793871866295265, + "grad_norm": 1.4051889181137085, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8720927238464355, + "num_tokens": 374305398.0, + "step": 10659 + }, + { + "epoch": 1.9795728876508822, + "grad_norm": 1.522030234336853, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8700984120368958, + "num_tokens": 374341324.0, + "step": 10660 + }, + { + "epoch": 1.9797585886722378, + "grad_norm": 1.373948335647583, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8871831893920898, + "num_tokens": 374378445.0, + "step": 10661 + }, + { + "epoch": 1.9799442896935933, + "grad_norm": 1.5045517683029175, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.893608570098877, + "num_tokens": 374414287.0, + "step": 10662 + }, + { + "epoch": 1.980129990714949, + "grad_norm": 1.4775402545928955, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8739058375358582, + "num_tokens": 374449137.0, + "step": 10663 + }, + { + "epoch": 1.9803156917363045, + "grad_norm": 1.519575595855713, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8779624104499817, + "num_tokens": 374480516.0, + "step": 10664 + }, + { + "epoch": 1.98050139275766, + "grad_norm": 1.517945647239685, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8864286541938782, + "num_tokens": 374515551.0, + "step": 10665 + }, + { + "epoch": 1.9806870937790158, + "grad_norm": 1.5054779052734375, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8824353218078613, + "num_tokens": 374550058.0, + "step": 10666 + }, + { + "epoch": 1.9808727948003715, + "grad_norm": 1.5034327507019043, + "learning_rate": 1e-06, + "loss": 0.4132, + "mean_token_accuracy": 0.8605828285217285, + "num_tokens": 374588697.0, + "step": 10667 + }, + { + "epoch": 1.981058495821727, + "grad_norm": 1.4275271892547607, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8817393779754639, + "num_tokens": 374624325.0, + "step": 10668 + }, + { + "epoch": 1.9812441968430825, + "grad_norm": 1.507590413093567, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8782361745834351, + "num_tokens": 374660913.0, + "step": 10669 + }, + { + "epoch": 1.9814298978644382, + "grad_norm": 1.6750258207321167, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8702819347381592, + "num_tokens": 374691457.0, + "step": 10670 + }, + { + "epoch": 1.981615598885794, + "grad_norm": 1.3525185585021973, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8853715658187866, + "num_tokens": 374731002.0, + "step": 10671 + }, + { + "epoch": 1.9818012999071495, + "grad_norm": 1.4760547876358032, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8868868947029114, + "num_tokens": 374765696.0, + "step": 10672 + }, + { + "epoch": 1.981987000928505, + "grad_norm": 1.4876948595046997, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8722389936447144, + "num_tokens": 374801086.0, + "step": 10673 + }, + { + "epoch": 1.9821727019498607, + "grad_norm": 1.4659470319747925, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8894658088684082, + "num_tokens": 374834052.0, + "step": 10674 + }, + { + "epoch": 1.9823584029712165, + "grad_norm": 1.4729374647140503, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8698527216911316, + "num_tokens": 374871683.0, + "step": 10675 + }, + { + "epoch": 1.982544103992572, + "grad_norm": 1.3744479417800903, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8973867893218994, + "num_tokens": 374907501.0, + "step": 10676 + }, + { + "epoch": 1.9827298050139275, + "grad_norm": 1.6860675811767578, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.877836287021637, + "num_tokens": 374934009.0, + "step": 10677 + }, + { + "epoch": 1.9829155060352832, + "grad_norm": 1.3923025131225586, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.870074987411499, + "num_tokens": 374975391.0, + "step": 10678 + }, + { + "epoch": 1.983101207056639, + "grad_norm": 1.5281907320022583, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8778867721557617, + "num_tokens": 375008568.0, + "step": 10679 + }, + { + "epoch": 1.9832869080779945, + "grad_norm": 1.453789472579956, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8782854080200195, + "num_tokens": 375042203.0, + "step": 10680 + }, + { + "epoch": 1.98347260909935, + "grad_norm": 1.575681447982788, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8726447224617004, + "num_tokens": 375073563.0, + "step": 10681 + }, + { + "epoch": 1.9836583101207057, + "grad_norm": 1.4333419799804688, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.887871265411377, + "num_tokens": 375108784.0, + "step": 10682 + }, + { + "epoch": 1.9838440111420614, + "grad_norm": 1.6341917514801025, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8760517835617065, + "num_tokens": 375137352.0, + "step": 10683 + }, + { + "epoch": 1.984029712163417, + "grad_norm": 1.4143404960632324, + "learning_rate": 1e-06, + "loss": 0.3771, + "mean_token_accuracy": 0.8703869581222534, + "num_tokens": 375178498.0, + "step": 10684 + }, + { + "epoch": 1.9842154131847725, + "grad_norm": 1.3969171047210693, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8840471506118774, + "num_tokens": 375216701.0, + "step": 10685 + }, + { + "epoch": 1.9844011142061282, + "grad_norm": 1.4488627910614014, + "learning_rate": 1e-06, + "loss": 0.3637, + "mean_token_accuracy": 0.8754355907440186, + "num_tokens": 375251702.0, + "step": 10686 + }, + { + "epoch": 1.9845868152274837, + "grad_norm": 1.5630205869674683, + "learning_rate": 1e-06, + "loss": 0.3665, + "mean_token_accuracy": 0.8764451742172241, + "num_tokens": 375283037.0, + "step": 10687 + }, + { + "epoch": 1.9847725162488392, + "grad_norm": 1.4692089557647705, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8896728157997131, + "num_tokens": 375319397.0, + "step": 10688 + }, + { + "epoch": 1.984958217270195, + "grad_norm": 1.4832061529159546, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8777592182159424, + "num_tokens": 375356509.0, + "step": 10689 + }, + { + "epoch": 1.9851439182915507, + "grad_norm": 1.3288190364837646, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8776345252990723, + "num_tokens": 375401818.0, + "step": 10690 + }, + { + "epoch": 1.9853296193129062, + "grad_norm": 1.397233247756958, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8838062882423401, + "num_tokens": 375440624.0, + "step": 10691 + }, + { + "epoch": 1.9855153203342617, + "grad_norm": 1.488592267036438, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8813824653625488, + "num_tokens": 375476285.0, + "step": 10692 + }, + { + "epoch": 1.9857010213556174, + "grad_norm": 1.640318751335144, + "learning_rate": 1e-06, + "loss": 0.3874, + "mean_token_accuracy": 0.8725765347480774, + "num_tokens": 375506057.0, + "step": 10693 + }, + { + "epoch": 1.9858867223769732, + "grad_norm": 1.3653926849365234, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8670388460159302, + "num_tokens": 375548627.0, + "step": 10694 + }, + { + "epoch": 1.9860724233983287, + "grad_norm": 1.3806511163711548, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8929049968719482, + "num_tokens": 375585323.0, + "step": 10695 + }, + { + "epoch": 1.9862581244196842, + "grad_norm": 1.461492896080017, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8696421980857849, + "num_tokens": 375625056.0, + "step": 10696 + }, + { + "epoch": 1.98644382544104, + "grad_norm": 1.4572309255599976, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8769640326499939, + "num_tokens": 375659302.0, + "step": 10697 + }, + { + "epoch": 1.9866295264623957, + "grad_norm": 1.302543044090271, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8950382471084595, + "num_tokens": 375699554.0, + "step": 10698 + }, + { + "epoch": 1.9868152274837512, + "grad_norm": 1.409212350845337, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8739067316055298, + "num_tokens": 375738556.0, + "step": 10699 + }, + { + "epoch": 1.9870009285051067, + "grad_norm": 1.4782369136810303, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8897485136985779, + "num_tokens": 375772173.0, + "step": 10700 + }, + { + "epoch": 1.9871866295264624, + "grad_norm": 1.573319673538208, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.883987545967102, + "num_tokens": 375810499.0, + "step": 10701 + }, + { + "epoch": 1.9873723305478181, + "grad_norm": 1.4792418479919434, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.878794252872467, + "num_tokens": 375847216.0, + "step": 10702 + }, + { + "epoch": 1.9875580315691737, + "grad_norm": 1.5615417957305908, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8712003231048584, + "num_tokens": 375881277.0, + "step": 10703 + }, + { + "epoch": 1.9877437325905292, + "grad_norm": 1.4697353839874268, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8895167112350464, + "num_tokens": 375916159.0, + "step": 10704 + }, + { + "epoch": 1.987929433611885, + "grad_norm": 1.4760822057724, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8782885074615479, + "num_tokens": 375951705.0, + "step": 10705 + }, + { + "epoch": 1.9881151346332406, + "grad_norm": 1.5012050867080688, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.885566771030426, + "num_tokens": 375986326.0, + "step": 10706 + }, + { + "epoch": 1.9883008356545961, + "grad_norm": 1.44100022315979, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8892223834991455, + "num_tokens": 376024765.0, + "step": 10707 + }, + { + "epoch": 1.9884865366759517, + "grad_norm": 1.6514132022857666, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8857654333114624, + "num_tokens": 376053226.0, + "step": 10708 + }, + { + "epoch": 1.9886722376973074, + "grad_norm": 1.4288703203201294, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.889557421207428, + "num_tokens": 376089767.0, + "step": 10709 + }, + { + "epoch": 1.988857938718663, + "grad_norm": 1.529897689819336, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8776756525039673, + "num_tokens": 376122766.0, + "step": 10710 + }, + { + "epoch": 1.9890436397400184, + "grad_norm": 1.476406455039978, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8813215494155884, + "num_tokens": 376156948.0, + "step": 10711 + }, + { + "epoch": 1.9892293407613741, + "grad_norm": 1.6209636926651, + "learning_rate": 1e-06, + "loss": 0.4019, + "mean_token_accuracy": 0.8620725274085999, + "num_tokens": 376187972.0, + "step": 10712 + }, + { + "epoch": 1.9894150417827299, + "grad_norm": 1.5450998544692993, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8745011687278748, + "num_tokens": 376221002.0, + "step": 10713 + }, + { + "epoch": 1.9896007428040854, + "grad_norm": 1.3727754354476929, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8880313634872437, + "num_tokens": 376258340.0, + "step": 10714 + }, + { + "epoch": 1.989786443825441, + "grad_norm": 1.3788552284240723, + "learning_rate": 1e-06, + "loss": 0.3912, + "mean_token_accuracy": 0.866115927696228, + "num_tokens": 376301462.0, + "step": 10715 + }, + { + "epoch": 1.9899721448467966, + "grad_norm": 1.5573606491088867, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.8916090726852417, + "num_tokens": 376334321.0, + "step": 10716 + }, + { + "epoch": 1.9901578458681524, + "grad_norm": 1.4773484468460083, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.877243161201477, + "num_tokens": 376371003.0, + "step": 10717 + }, + { + "epoch": 1.9903435468895079, + "grad_norm": 1.4023104906082153, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8841184377670288, + "num_tokens": 376407325.0, + "step": 10718 + }, + { + "epoch": 1.9905292479108634, + "grad_norm": 1.4745302200317383, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8822020292282104, + "num_tokens": 376444015.0, + "step": 10719 + }, + { + "epoch": 1.9907149489322191, + "grad_norm": 1.476750373840332, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8797314167022705, + "num_tokens": 376480199.0, + "step": 10720 + }, + { + "epoch": 1.9909006499535749, + "grad_norm": 1.6304939985275269, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8726145029067993, + "num_tokens": 376514451.0, + "step": 10721 + }, + { + "epoch": 1.9910863509749304, + "grad_norm": 1.5474761724472046, + "learning_rate": 1e-06, + "loss": 0.3912, + "mean_token_accuracy": 0.8632495403289795, + "num_tokens": 376546484.0, + "step": 10722 + }, + { + "epoch": 1.9912720519962859, + "grad_norm": 1.4253056049346924, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8955002427101135, + "num_tokens": 376579464.0, + "step": 10723 + }, + { + "epoch": 1.9914577530176416, + "grad_norm": 1.5052286386489868, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.868328332901001, + "num_tokens": 376616822.0, + "step": 10724 + }, + { + "epoch": 1.9916434540389973, + "grad_norm": 1.4812082052230835, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8822648525238037, + "num_tokens": 376652102.0, + "step": 10725 + }, + { + "epoch": 1.9918291550603529, + "grad_norm": 1.4833639860153198, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8833757638931274, + "num_tokens": 376684709.0, + "step": 10726 + }, + { + "epoch": 1.9920148560817084, + "grad_norm": 1.6144421100616455, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8828675150871277, + "num_tokens": 376712281.0, + "step": 10727 + }, + { + "epoch": 1.992200557103064, + "grad_norm": 1.5861172676086426, + "learning_rate": 1e-06, + "loss": 0.3526, + "mean_token_accuracy": 0.8787559866905212, + "num_tokens": 376743910.0, + "step": 10728 + }, + { + "epoch": 1.9923862581244198, + "grad_norm": 1.3175536394119263, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8978039026260376, + "num_tokens": 376782015.0, + "step": 10729 + }, + { + "epoch": 1.9925719591457753, + "grad_norm": 1.3884438276290894, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8729889392852783, + "num_tokens": 376825036.0, + "step": 10730 + }, + { + "epoch": 1.9927576601671309, + "grad_norm": 1.5433433055877686, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8818299770355225, + "num_tokens": 376855446.0, + "step": 10731 + }, + { + "epoch": 1.9929433611884866, + "grad_norm": 1.4710047245025635, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8853458166122437, + "num_tokens": 376886132.0, + "step": 10732 + }, + { + "epoch": 1.9931290622098423, + "grad_norm": 1.3593555688858032, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8766554594039917, + "num_tokens": 376927402.0, + "step": 10733 + }, + { + "epoch": 1.9933147632311976, + "grad_norm": 1.5115783214569092, + "learning_rate": 1e-06, + "loss": 0.4082, + "mean_token_accuracy": 0.8622550368309021, + "num_tokens": 376963856.0, + "step": 10734 + }, + { + "epoch": 1.9935004642525533, + "grad_norm": 1.391640543937683, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8807529211044312, + "num_tokens": 377000014.0, + "step": 10735 + }, + { + "epoch": 1.993686165273909, + "grad_norm": 1.5884695053100586, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.8653028607368469, + "num_tokens": 377032975.0, + "step": 10736 + }, + { + "epoch": 1.9938718662952646, + "grad_norm": 1.6733297109603882, + "learning_rate": 1e-06, + "loss": 0.4109, + "mean_token_accuracy": 0.8658930063247681, + "num_tokens": 377064023.0, + "step": 10737 + }, + { + "epoch": 1.99405756731662, + "grad_norm": 1.4743943214416504, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8806817531585693, + "num_tokens": 377096623.0, + "step": 10738 + }, + { + "epoch": 1.9942432683379758, + "grad_norm": 1.5033584833145142, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8716509342193604, + "num_tokens": 377131075.0, + "step": 10739 + }, + { + "epoch": 1.9944289693593316, + "grad_norm": 1.4612919092178345, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8700295090675354, + "num_tokens": 377168456.0, + "step": 10740 + }, + { + "epoch": 1.994614670380687, + "grad_norm": 1.3985145092010498, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8860743641853333, + "num_tokens": 377201246.0, + "step": 10741 + }, + { + "epoch": 1.9948003714020426, + "grad_norm": 1.4706876277923584, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8799206018447876, + "num_tokens": 377240144.0, + "step": 10742 + }, + { + "epoch": 1.9949860724233983, + "grad_norm": 1.397261619567871, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8800708055496216, + "num_tokens": 377276868.0, + "step": 10743 + }, + { + "epoch": 1.995171773444754, + "grad_norm": 1.5334007740020752, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8687870502471924, + "num_tokens": 377313264.0, + "step": 10744 + }, + { + "epoch": 1.9953574744661096, + "grad_norm": 1.3954252004623413, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.876665472984314, + "num_tokens": 377350776.0, + "step": 10745 + }, + { + "epoch": 1.995543175487465, + "grad_norm": 1.4851856231689453, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8772599101066589, + "num_tokens": 377389547.0, + "step": 10746 + }, + { + "epoch": 1.9957288765088208, + "grad_norm": 1.4273927211761475, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.885231077671051, + "num_tokens": 377425976.0, + "step": 10747 + }, + { + "epoch": 1.9959145775301765, + "grad_norm": 1.419633388519287, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8951036930084229, + "num_tokens": 377459389.0, + "step": 10748 + }, + { + "epoch": 1.996100278551532, + "grad_norm": 1.3308906555175781, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8897771835327148, + "num_tokens": 377498931.0, + "step": 10749 + }, + { + "epoch": 1.9962859795728876, + "grad_norm": 1.3346810340881348, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8769893646240234, + "num_tokens": 377542002.0, + "step": 10750 + }, + { + "epoch": 1.9964716805942433, + "grad_norm": 1.5807788372039795, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.877974271774292, + "num_tokens": 377576052.0, + "step": 10751 + }, + { + "epoch": 1.996657381615599, + "grad_norm": 1.4539695978164673, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8873293399810791, + "num_tokens": 377610572.0, + "step": 10752 + }, + { + "epoch": 1.9968430826369545, + "grad_norm": 1.5126416683197021, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8777966499328613, + "num_tokens": 377648368.0, + "step": 10753 + }, + { + "epoch": 1.99702878365831, + "grad_norm": 1.8341277837753296, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8750392198562622, + "num_tokens": 377671901.0, + "step": 10754 + }, + { + "epoch": 1.9972144846796658, + "grad_norm": 1.4673066139221191, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8857275247573853, + "num_tokens": 377702261.0, + "step": 10755 + }, + { + "epoch": 1.9974001857010215, + "grad_norm": 1.5341297388076782, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8868738412857056, + "num_tokens": 377734053.0, + "step": 10756 + }, + { + "epoch": 1.997585886722377, + "grad_norm": 1.4623581171035767, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8900433778762817, + "num_tokens": 377765910.0, + "step": 10757 + }, + { + "epoch": 1.9977715877437325, + "grad_norm": 1.4076837301254272, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8858787417411804, + "num_tokens": 377801661.0, + "step": 10758 + }, + { + "epoch": 1.9979572887650883, + "grad_norm": 1.5056846141815186, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8837850689888, + "num_tokens": 377833253.0, + "step": 10759 + }, + { + "epoch": 1.9981429897864438, + "grad_norm": 1.5335471630096436, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8614603281021118, + "num_tokens": 377869883.0, + "step": 10760 + }, + { + "epoch": 1.9983286908077993, + "grad_norm": 1.606916069984436, + "learning_rate": 1e-06, + "loss": 0.3758, + "mean_token_accuracy": 0.8713672757148743, + "num_tokens": 377902878.0, + "step": 10761 + }, + { + "epoch": 1.998514391829155, + "grad_norm": 1.430823802947998, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8688192963600159, + "num_tokens": 377941003.0, + "step": 10762 + }, + { + "epoch": 1.9987000928505108, + "grad_norm": 1.5078860521316528, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8767833709716797, + "num_tokens": 377973618.0, + "step": 10763 + }, + { + "epoch": 1.9988857938718663, + "grad_norm": 1.387211799621582, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8791534304618835, + "num_tokens": 378011837.0, + "step": 10764 + }, + { + "epoch": 1.9990714948932218, + "grad_norm": 1.376235842704773, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8787863850593567, + "num_tokens": 378049039.0, + "step": 10765 + }, + { + "epoch": 1.9992571959145775, + "grad_norm": 1.3906368017196655, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8795962333679199, + "num_tokens": 378089261.0, + "step": 10766 + }, + { + "epoch": 1.9994428969359332, + "grad_norm": 1.417059302330017, + "learning_rate": 1e-06, + "loss": 0.3773, + "mean_token_accuracy": 0.8679862022399902, + "num_tokens": 378131571.0, + "step": 10767 + }, + { + "epoch": 1.9996285979572888, + "grad_norm": 1.4769755601882935, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8723353147506714, + "num_tokens": 378168362.0, + "step": 10768 + }, + { + "epoch": 1.9998142989786443, + "grad_norm": 1.473858118057251, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8857890963554382, + "num_tokens": 378202156.0, + "step": 10769 + }, + { + "epoch": 2.0, + "grad_norm": 1.4676363468170166, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8688924312591553, + "num_tokens": 378238909.0, + "step": 10770 + }, + { + "epoch": 2.0001857010213557, + "grad_norm": 1.4894965887069702, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8892278075218201, + "num_tokens": 378276912.0, + "step": 10771 + }, + { + "epoch": 2.000371402042711, + "grad_norm": 1.3933545351028442, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8932313919067383, + "num_tokens": 378313212.0, + "step": 10772 + }, + { + "epoch": 2.0005571030640668, + "grad_norm": 1.4065524339675903, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8913867473602295, + "num_tokens": 378350776.0, + "step": 10773 + }, + { + "epoch": 2.0007428040854225, + "grad_norm": 1.6162707805633545, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8782854080200195, + "num_tokens": 378380037.0, + "step": 10774 + }, + { + "epoch": 2.000928505106778, + "grad_norm": 1.4364678859710693, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8782004117965698, + "num_tokens": 378418206.0, + "step": 10775 + }, + { + "epoch": 2.0011142061281335, + "grad_norm": 1.356304407119751, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8909703493118286, + "num_tokens": 378457545.0, + "step": 10776 + }, + { + "epoch": 2.0012999071494892, + "grad_norm": 1.398741364479065, + "learning_rate": 1e-06, + "loss": 0.2793, + "mean_token_accuracy": 0.8994650840759277, + "num_tokens": 378493157.0, + "step": 10777 + }, + { + "epoch": 2.001485608170845, + "grad_norm": 1.5182629823684692, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8935950398445129, + "num_tokens": 378526880.0, + "step": 10778 + }, + { + "epoch": 2.0016713091922007, + "grad_norm": 1.606223464012146, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8708179593086243, + "num_tokens": 378562851.0, + "step": 10779 + }, + { + "epoch": 2.001857010213556, + "grad_norm": 1.560755968093872, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8863807916641235, + "num_tokens": 378596727.0, + "step": 10780 + }, + { + "epoch": 2.0020427112349117, + "grad_norm": 1.571855068206787, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8867785930633545, + "num_tokens": 378630426.0, + "step": 10781 + }, + { + "epoch": 2.0022284122562675, + "grad_norm": 1.5779279470443726, + "learning_rate": 1e-06, + "loss": 0.2648, + "mean_token_accuracy": 0.9038788676261902, + "num_tokens": 378660774.0, + "step": 10782 + }, + { + "epoch": 2.002414113277623, + "grad_norm": 1.7234505414962769, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.889878511428833, + "num_tokens": 378689063.0, + "step": 10783 + }, + { + "epoch": 2.0025998142989785, + "grad_norm": 1.5804800987243652, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8916759490966797, + "num_tokens": 378722801.0, + "step": 10784 + }, + { + "epoch": 2.002785515320334, + "grad_norm": 1.5723187923431396, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8859899640083313, + "num_tokens": 378760836.0, + "step": 10785 + }, + { + "epoch": 2.00297121634169, + "grad_norm": 1.579805612564087, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8830969333648682, + "num_tokens": 378793864.0, + "step": 10786 + }, + { + "epoch": 2.0031569173630457, + "grad_norm": 1.5092041492462158, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8895636796951294, + "num_tokens": 378828577.0, + "step": 10787 + }, + { + "epoch": 2.003342618384401, + "grad_norm": 1.6175745725631714, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8845454454421997, + "num_tokens": 378859032.0, + "step": 10788 + }, + { + "epoch": 2.0035283194057567, + "grad_norm": 1.341167688369751, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8912858366966248, + "num_tokens": 378902553.0, + "step": 10789 + }, + { + "epoch": 2.0037140204271124, + "grad_norm": 1.503484845161438, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8819716572761536, + "num_tokens": 378941704.0, + "step": 10790 + }, + { + "epoch": 2.003899721448468, + "grad_norm": 1.5594276189804077, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8665109872817993, + "num_tokens": 378980882.0, + "step": 10791 + }, + { + "epoch": 2.0040854224698235, + "grad_norm": 1.3350962400436401, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8968181610107422, + "num_tokens": 379020006.0, + "step": 10792 + }, + { + "epoch": 2.004271123491179, + "grad_norm": 1.5611460208892822, + "learning_rate": 1e-06, + "loss": 0.2766, + "mean_token_accuracy": 0.9019099473953247, + "num_tokens": 379048961.0, + "step": 10793 + }, + { + "epoch": 2.004456824512535, + "grad_norm": 1.4629027843475342, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8840799927711487, + "num_tokens": 379089361.0, + "step": 10794 + }, + { + "epoch": 2.00464252553389, + "grad_norm": 1.5139487981796265, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8881829380989075, + "num_tokens": 379121815.0, + "step": 10795 + }, + { + "epoch": 2.004828226555246, + "grad_norm": 1.480171799659729, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8839632868766785, + "num_tokens": 379157463.0, + "step": 10796 + }, + { + "epoch": 2.0050139275766017, + "grad_norm": 1.5638580322265625, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.886635422706604, + "num_tokens": 379191729.0, + "step": 10797 + }, + { + "epoch": 2.0051996285979574, + "grad_norm": 1.597779631614685, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8798786401748657, + "num_tokens": 379225239.0, + "step": 10798 + }, + { + "epoch": 2.0053853296193127, + "grad_norm": 1.499456524848938, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8922189474105835, + "num_tokens": 379258351.0, + "step": 10799 + }, + { + "epoch": 2.0055710306406684, + "grad_norm": 1.4596574306488037, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8815740346908569, + "num_tokens": 379298047.0, + "step": 10800 + }, + { + "epoch": 2.005756731662024, + "grad_norm": 1.4286515712738037, + "learning_rate": 1e-06, + "loss": 0.2769, + "mean_token_accuracy": 0.9022356867790222, + "num_tokens": 379333684.0, + "step": 10801 + }, + { + "epoch": 2.00594243268338, + "grad_norm": 1.6496450901031494, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8771862387657166, + "num_tokens": 379365596.0, + "step": 10802 + }, + { + "epoch": 2.006128133704735, + "grad_norm": 1.5171939134597778, + "learning_rate": 1e-06, + "loss": 0.2877, + "mean_token_accuracy": 0.8969978094100952, + "num_tokens": 379398279.0, + "step": 10803 + }, + { + "epoch": 2.006313834726091, + "grad_norm": 1.4338963031768799, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8822987675666809, + "num_tokens": 379437895.0, + "step": 10804 + }, + { + "epoch": 2.0064995357474467, + "grad_norm": 1.4890456199645996, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8765058517456055, + "num_tokens": 379476020.0, + "step": 10805 + }, + { + "epoch": 2.0066852367688024, + "grad_norm": 1.3390763998031616, + "learning_rate": 1e-06, + "loss": 0.2722, + "mean_token_accuracy": 0.9017021059989929, + "num_tokens": 379515736.0, + "step": 10806 + }, + { + "epoch": 2.0068709377901577, + "grad_norm": 1.4155598878860474, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8821949362754822, + "num_tokens": 379556055.0, + "step": 10807 + }, + { + "epoch": 2.0070566388115134, + "grad_norm": 1.5317344665527344, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8812415599822998, + "num_tokens": 379589187.0, + "step": 10808 + }, + { + "epoch": 2.007242339832869, + "grad_norm": 1.4807997941970825, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8809231519699097, + "num_tokens": 379627947.0, + "step": 10809 + }, + { + "epoch": 2.007428040854225, + "grad_norm": 1.468921422958374, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.9003388285636902, + "num_tokens": 379665354.0, + "step": 10810 + }, + { + "epoch": 2.00761374187558, + "grad_norm": 1.4705265760421753, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8950562477111816, + "num_tokens": 379702883.0, + "step": 10811 + }, + { + "epoch": 2.007799442896936, + "grad_norm": 1.4691064357757568, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8829845190048218, + "num_tokens": 379743924.0, + "step": 10812 + }, + { + "epoch": 2.0079851439182916, + "grad_norm": 1.6454520225524902, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8893052339553833, + "num_tokens": 379776746.0, + "step": 10813 + }, + { + "epoch": 2.0081708449396474, + "grad_norm": 1.5018929243087769, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8904786109924316, + "num_tokens": 379813223.0, + "step": 10814 + }, + { + "epoch": 2.0083565459610027, + "grad_norm": 1.7306106090545654, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8862682580947876, + "num_tokens": 379841720.0, + "step": 10815 + }, + { + "epoch": 2.0085422469823584, + "grad_norm": 1.5545817613601685, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8922984004020691, + "num_tokens": 379875163.0, + "step": 10816 + }, + { + "epoch": 2.008727948003714, + "grad_norm": 1.5182279348373413, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.887263298034668, + "num_tokens": 379911987.0, + "step": 10817 + }, + { + "epoch": 2.00891364902507, + "grad_norm": 1.5428459644317627, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8892655968666077, + "num_tokens": 379948671.0, + "step": 10818 + }, + { + "epoch": 2.009099350046425, + "grad_norm": 1.8808165788650513, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8780585527420044, + "num_tokens": 379974529.0, + "step": 10819 + }, + { + "epoch": 2.009285051067781, + "grad_norm": 1.6292554140090942, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.880906343460083, + "num_tokens": 380007783.0, + "step": 10820 + }, + { + "epoch": 2.0094707520891366, + "grad_norm": 1.5229170322418213, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8949188590049744, + "num_tokens": 380044074.0, + "step": 10821 + }, + { + "epoch": 2.009656453110492, + "grad_norm": 1.7902588844299316, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8792546987533569, + "num_tokens": 380070234.0, + "step": 10822 + }, + { + "epoch": 2.0098421541318476, + "grad_norm": 1.5710248947143555, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.877840518951416, + "num_tokens": 380107171.0, + "step": 10823 + }, + { + "epoch": 2.0100278551532034, + "grad_norm": 1.3826013803482056, + "learning_rate": 1e-06, + "loss": 0.2966, + "mean_token_accuracy": 0.8956927061080933, + "num_tokens": 380151007.0, + "step": 10824 + }, + { + "epoch": 2.010213556174559, + "grad_norm": 1.637595772743225, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8861862421035767, + "num_tokens": 380182110.0, + "step": 10825 + }, + { + "epoch": 2.0103992571959144, + "grad_norm": 1.5303783416748047, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8978553414344788, + "num_tokens": 380216729.0, + "step": 10826 + }, + { + "epoch": 2.01058495821727, + "grad_norm": 1.5467150211334229, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.886093258857727, + "num_tokens": 380251488.0, + "step": 10827 + }, + { + "epoch": 2.010770659238626, + "grad_norm": 1.6405363082885742, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8704233169555664, + "num_tokens": 380282880.0, + "step": 10828 + }, + { + "epoch": 2.0109563602599816, + "grad_norm": 1.6699374914169312, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8854455947875977, + "num_tokens": 380313565.0, + "step": 10829 + }, + { + "epoch": 2.011142061281337, + "grad_norm": 1.451024055480957, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8931326866149902, + "num_tokens": 380351048.0, + "step": 10830 + }, + { + "epoch": 2.0113277623026926, + "grad_norm": 1.5668362379074097, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8966503143310547, + "num_tokens": 380388398.0, + "step": 10831 + }, + { + "epoch": 2.0115134633240483, + "grad_norm": 1.4213147163391113, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8897556662559509, + "num_tokens": 380428021.0, + "step": 10832 + }, + { + "epoch": 2.011699164345404, + "grad_norm": 1.4483082294464111, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8831351399421692, + "num_tokens": 380469812.0, + "step": 10833 + }, + { + "epoch": 2.0118848653667594, + "grad_norm": 1.6402548551559448, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8915948867797852, + "num_tokens": 380502575.0, + "step": 10834 + }, + { + "epoch": 2.012070566388115, + "grad_norm": 1.5539964437484741, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8881593346595764, + "num_tokens": 380539083.0, + "step": 10835 + }, + { + "epoch": 2.012256267409471, + "grad_norm": 1.4423633813858032, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8919792175292969, + "num_tokens": 380578255.0, + "step": 10836 + }, + { + "epoch": 2.0124419684308266, + "grad_norm": 1.6207562685012817, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.887017548084259, + "num_tokens": 380610946.0, + "step": 10837 + }, + { + "epoch": 2.012627669452182, + "grad_norm": 1.5454572439193726, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8698613047599792, + "num_tokens": 380649554.0, + "step": 10838 + }, + { + "epoch": 2.0128133704735376, + "grad_norm": 1.6043305397033691, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8813308477401733, + "num_tokens": 380680690.0, + "step": 10839 + }, + { + "epoch": 2.0129990714948933, + "grad_norm": 1.4228663444519043, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8832006454467773, + "num_tokens": 380722683.0, + "step": 10840 + }, + { + "epoch": 2.013184772516249, + "grad_norm": 1.5259175300598145, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8790839910507202, + "num_tokens": 380759923.0, + "step": 10841 + }, + { + "epoch": 2.0133704735376043, + "grad_norm": 1.5783236026763916, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8893639445304871, + "num_tokens": 380792362.0, + "step": 10842 + }, + { + "epoch": 2.01355617455896, + "grad_norm": 1.4675551652908325, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8887529969215393, + "num_tokens": 380828130.0, + "step": 10843 + }, + { + "epoch": 2.013741875580316, + "grad_norm": 1.5890768766403198, + "learning_rate": 1e-06, + "loss": 0.3067, + "mean_token_accuracy": 0.8915866613388062, + "num_tokens": 380858439.0, + "step": 10844 + }, + { + "epoch": 2.013927576601671, + "grad_norm": 1.4788395166397095, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8858385682106018, + "num_tokens": 380894816.0, + "step": 10845 + }, + { + "epoch": 2.014113277623027, + "grad_norm": 1.5836105346679688, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.890182375907898, + "num_tokens": 380927968.0, + "step": 10846 + }, + { + "epoch": 2.0142989786443826, + "grad_norm": 1.5085476636886597, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8808615207672119, + "num_tokens": 380967394.0, + "step": 10847 + }, + { + "epoch": 2.0144846796657383, + "grad_norm": 1.5619522333145142, + "learning_rate": 1e-06, + "loss": 0.2863, + "mean_token_accuracy": 0.8993066549301147, + "num_tokens": 380997885.0, + "step": 10848 + }, + { + "epoch": 2.0146703806870936, + "grad_norm": 1.5248534679412842, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.884833574295044, + "num_tokens": 381035182.0, + "step": 10849 + }, + { + "epoch": 2.0148560817084493, + "grad_norm": 1.4605227708816528, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.8946653604507446, + "num_tokens": 381073260.0, + "step": 10850 + }, + { + "epoch": 2.015041782729805, + "grad_norm": 1.534440279006958, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8862160444259644, + "num_tokens": 381106020.0, + "step": 10851 + }, + { + "epoch": 2.015227483751161, + "grad_norm": 1.744023084640503, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8714439272880554, + "num_tokens": 381135466.0, + "step": 10852 + }, + { + "epoch": 2.015413184772516, + "grad_norm": 1.6311719417572021, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.877354621887207, + "num_tokens": 381169074.0, + "step": 10853 + }, + { + "epoch": 2.015598885793872, + "grad_norm": 1.4856147766113281, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.897874116897583, + "num_tokens": 381205141.0, + "step": 10854 + }, + { + "epoch": 2.0157845868152275, + "grad_norm": 1.5972974300384521, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8941311836242676, + "num_tokens": 381236298.0, + "step": 10855 + }, + { + "epoch": 2.0159702878365833, + "grad_norm": 1.5757522583007812, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8835961818695068, + "num_tokens": 381273329.0, + "step": 10856 + }, + { + "epoch": 2.0161559888579386, + "grad_norm": 1.5005309581756592, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8879420757293701, + "num_tokens": 381312420.0, + "step": 10857 + }, + { + "epoch": 2.0163416898792943, + "grad_norm": 1.589887022972107, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.877434492111206, + "num_tokens": 381346479.0, + "step": 10858 + }, + { + "epoch": 2.01652739090065, + "grad_norm": 1.3753029108047485, + "learning_rate": 1e-06, + "loss": 0.2793, + "mean_token_accuracy": 0.9025518894195557, + "num_tokens": 381383121.0, + "step": 10859 + }, + { + "epoch": 2.0167130919220058, + "grad_norm": 1.5098259449005127, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8866652250289917, + "num_tokens": 381420516.0, + "step": 10860 + }, + { + "epoch": 2.016898792943361, + "grad_norm": 1.7279797792434692, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8956155776977539, + "num_tokens": 381450590.0, + "step": 10861 + }, + { + "epoch": 2.017084493964717, + "grad_norm": 1.630918264389038, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8779124021530151, + "num_tokens": 381486551.0, + "step": 10862 + }, + { + "epoch": 2.0172701949860725, + "grad_norm": 1.6360085010528564, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8808969259262085, + "num_tokens": 381520372.0, + "step": 10863 + }, + { + "epoch": 2.0174558960074283, + "grad_norm": 1.788382649421692, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8674520254135132, + "num_tokens": 381554016.0, + "step": 10864 + }, + { + "epoch": 2.0176415970287835, + "grad_norm": 1.545330286026001, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8786172270774841, + "num_tokens": 381589650.0, + "step": 10865 + }, + { + "epoch": 2.0178272980501393, + "grad_norm": 1.4707595109939575, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8892836570739746, + "num_tokens": 381629324.0, + "step": 10866 + }, + { + "epoch": 2.018012999071495, + "grad_norm": 1.5231345891952515, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8903087377548218, + "num_tokens": 381666845.0, + "step": 10867 + }, + { + "epoch": 2.0181987000928503, + "grad_norm": 1.6176223754882812, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8775680661201477, + "num_tokens": 381699252.0, + "step": 10868 + }, + { + "epoch": 2.018384401114206, + "grad_norm": 1.5506519079208374, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8798116445541382, + "num_tokens": 381736869.0, + "step": 10869 + }, + { + "epoch": 2.0185701021355618, + "grad_norm": 1.4687968492507935, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8847430348396301, + "num_tokens": 381775345.0, + "step": 10870 + }, + { + "epoch": 2.0187558031569175, + "grad_norm": 1.5106418132781982, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8833143711090088, + "num_tokens": 381816254.0, + "step": 10871 + }, + { + "epoch": 2.018941504178273, + "grad_norm": 1.5625252723693848, + "learning_rate": 1e-06, + "loss": 0.2793, + "mean_token_accuracy": 0.8987325429916382, + "num_tokens": 381847846.0, + "step": 10872 + }, + { + "epoch": 2.0191272051996285, + "grad_norm": 1.487365961074829, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8713258504867554, + "num_tokens": 381886496.0, + "step": 10873 + }, + { + "epoch": 2.0193129062209842, + "grad_norm": 1.6168408393859863, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8792541027069092, + "num_tokens": 381921357.0, + "step": 10874 + }, + { + "epoch": 2.01949860724234, + "grad_norm": 1.5980298519134521, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8909907341003418, + "num_tokens": 381956660.0, + "step": 10875 + }, + { + "epoch": 2.0196843082636953, + "grad_norm": 1.5793442726135254, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8810348510742188, + "num_tokens": 381989805.0, + "step": 10876 + }, + { + "epoch": 2.019870009285051, + "grad_norm": 1.569149374961853, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.884262204170227, + "num_tokens": 382026553.0, + "step": 10877 + }, + { + "epoch": 2.0200557103064067, + "grad_norm": 1.5575963258743286, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8911380171775818, + "num_tokens": 382058026.0, + "step": 10878 + }, + { + "epoch": 2.0202414113277625, + "grad_norm": 1.5056816339492798, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8957719206809998, + "num_tokens": 382091696.0, + "step": 10879 + }, + { + "epoch": 2.0204271123491178, + "grad_norm": 1.6821428537368774, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8906053304672241, + "num_tokens": 382122991.0, + "step": 10880 + }, + { + "epoch": 2.0206128133704735, + "grad_norm": 1.5830285549163818, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.885287880897522, + "num_tokens": 382156716.0, + "step": 10881 + }, + { + "epoch": 2.0207985143918292, + "grad_norm": 1.5702486038208008, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8891342878341675, + "num_tokens": 382189386.0, + "step": 10882 + }, + { + "epoch": 2.020984215413185, + "grad_norm": 1.5497043132781982, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8975048065185547, + "num_tokens": 382222100.0, + "step": 10883 + }, + { + "epoch": 2.0211699164345402, + "grad_norm": 1.6720764636993408, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8763330578804016, + "num_tokens": 382251909.0, + "step": 10884 + }, + { + "epoch": 2.021355617455896, + "grad_norm": 1.6930736303329468, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8873623609542847, + "num_tokens": 382282146.0, + "step": 10885 + }, + { + "epoch": 2.0215413184772517, + "grad_norm": 1.4856324195861816, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.882011890411377, + "num_tokens": 382324429.0, + "step": 10886 + }, + { + "epoch": 2.0217270194986074, + "grad_norm": 1.6052882671356201, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8895827531814575, + "num_tokens": 382358838.0, + "step": 10887 + }, + { + "epoch": 2.0219127205199627, + "grad_norm": 1.5105736255645752, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8731928467750549, + "num_tokens": 382397895.0, + "step": 10888 + }, + { + "epoch": 2.0220984215413185, + "grad_norm": 1.692707896232605, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8924709558486938, + "num_tokens": 382426750.0, + "step": 10889 + }, + { + "epoch": 2.022284122562674, + "grad_norm": 1.3849611282348633, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8979208469390869, + "num_tokens": 382465713.0, + "step": 10890 + }, + { + "epoch": 2.0224698235840295, + "grad_norm": 1.7020633220672607, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.884479820728302, + "num_tokens": 382495946.0, + "step": 10891 + }, + { + "epoch": 2.0226555246053852, + "grad_norm": 1.477876901626587, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.9001427888870239, + "num_tokens": 382529436.0, + "step": 10892 + }, + { + "epoch": 2.022841225626741, + "grad_norm": 1.4613966941833496, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8948404788970947, + "num_tokens": 382563912.0, + "step": 10893 + }, + { + "epoch": 2.0230269266480967, + "grad_norm": 1.5545330047607422, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8761561512947083, + "num_tokens": 382600140.0, + "step": 10894 + }, + { + "epoch": 2.023212627669452, + "grad_norm": 1.5245141983032227, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.891221821308136, + "num_tokens": 382637510.0, + "step": 10895 + }, + { + "epoch": 2.0233983286908077, + "grad_norm": 1.5957882404327393, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8826838135719299, + "num_tokens": 382671882.0, + "step": 10896 + }, + { + "epoch": 2.0235840297121634, + "grad_norm": 1.700777530670166, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8920869827270508, + "num_tokens": 382703787.0, + "step": 10897 + }, + { + "epoch": 2.023769730733519, + "grad_norm": 1.7809419631958008, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8730595707893372, + "num_tokens": 382733939.0, + "step": 10898 + }, + { + "epoch": 2.0239554317548745, + "grad_norm": 1.6408945322036743, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8870162963867188, + "num_tokens": 382768440.0, + "step": 10899 + }, + { + "epoch": 2.02414113277623, + "grad_norm": 1.797465205192566, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8889881372451782, + "num_tokens": 382799337.0, + "step": 10900 + }, + { + "epoch": 2.024326833797586, + "grad_norm": 1.5070538520812988, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8884543180465698, + "num_tokens": 382835304.0, + "step": 10901 + }, + { + "epoch": 2.0245125348189417, + "grad_norm": 1.656411051750183, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8981985449790955, + "num_tokens": 382861867.0, + "step": 10902 + }, + { + "epoch": 2.024698235840297, + "grad_norm": 1.6454085111618042, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8847782611846924, + "num_tokens": 382894715.0, + "step": 10903 + }, + { + "epoch": 2.0248839368616527, + "grad_norm": 1.505952000617981, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8867233991622925, + "num_tokens": 382929523.0, + "step": 10904 + }, + { + "epoch": 2.0250696378830084, + "grad_norm": 1.635652780532837, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.893737256526947, + "num_tokens": 382960766.0, + "step": 10905 + }, + { + "epoch": 2.025255338904364, + "grad_norm": 1.64242422580719, + "learning_rate": 1e-06, + "loss": 0.4103, + "mean_token_accuracy": 0.8604335784912109, + "num_tokens": 382996594.0, + "step": 10906 + }, + { + "epoch": 2.0254410399257194, + "grad_norm": 1.4955333471298218, + "learning_rate": 1e-06, + "loss": 0.3803, + "mean_token_accuracy": 0.8717496991157532, + "num_tokens": 383038035.0, + "step": 10907 + }, + { + "epoch": 2.025626740947075, + "grad_norm": 1.4129960536956787, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8861618041992188, + "num_tokens": 383079868.0, + "step": 10908 + }, + { + "epoch": 2.025812441968431, + "grad_norm": 1.59121835231781, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8888299465179443, + "num_tokens": 383110051.0, + "step": 10909 + }, + { + "epoch": 2.0259981429897866, + "grad_norm": 1.755950927734375, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.88735032081604, + "num_tokens": 383136311.0, + "step": 10910 + }, + { + "epoch": 2.026183844011142, + "grad_norm": 1.4418572187423706, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8843268156051636, + "num_tokens": 383175837.0, + "step": 10911 + }, + { + "epoch": 2.0263695450324977, + "grad_norm": 1.5141756534576416, + "learning_rate": 1e-06, + "loss": 0.2756, + "mean_token_accuracy": 0.9014674425125122, + "num_tokens": 383210728.0, + "step": 10912 + }, + { + "epoch": 2.0265552460538534, + "grad_norm": 1.4322184324264526, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8919879198074341, + "num_tokens": 383248862.0, + "step": 10913 + }, + { + "epoch": 2.026740947075209, + "grad_norm": 1.5853430032730103, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8817612528800964, + "num_tokens": 383282989.0, + "step": 10914 + }, + { + "epoch": 2.0269266480965644, + "grad_norm": 1.494850993156433, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8794676661491394, + "num_tokens": 383320084.0, + "step": 10915 + }, + { + "epoch": 2.02711234911792, + "grad_norm": 1.6051180362701416, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8890595436096191, + "num_tokens": 383349094.0, + "step": 10916 + }, + { + "epoch": 2.027298050139276, + "grad_norm": 1.5478318929672241, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8828413486480713, + "num_tokens": 383383815.0, + "step": 10917 + }, + { + "epoch": 2.027483751160631, + "grad_norm": 1.6673928499221802, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8800641298294067, + "num_tokens": 383416634.0, + "step": 10918 + }, + { + "epoch": 2.027669452181987, + "grad_norm": 1.5325313806533813, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8880923986434937, + "num_tokens": 383454186.0, + "step": 10919 + }, + { + "epoch": 2.0278551532033426, + "grad_norm": 1.4960689544677734, + "learning_rate": 1e-06, + "loss": 0.2807, + "mean_token_accuracy": 0.8985142111778259, + "num_tokens": 383487695.0, + "step": 10920 + }, + { + "epoch": 2.0280408542246984, + "grad_norm": 1.4407775402069092, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.891994297504425, + "num_tokens": 383528194.0, + "step": 10921 + }, + { + "epoch": 2.0282265552460537, + "grad_norm": 1.592448353767395, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8806341290473938, + "num_tokens": 383562337.0, + "step": 10922 + }, + { + "epoch": 2.0284122562674094, + "grad_norm": 1.5303730964660645, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8925037384033203, + "num_tokens": 383596627.0, + "step": 10923 + }, + { + "epoch": 2.028597957288765, + "grad_norm": 1.6567738056182861, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8824707865715027, + "num_tokens": 383628768.0, + "step": 10924 + }, + { + "epoch": 2.028783658310121, + "grad_norm": 1.5130956172943115, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8854540586471558, + "num_tokens": 383665672.0, + "step": 10925 + }, + { + "epoch": 2.028969359331476, + "grad_norm": 1.4439665079116821, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8975234031677246, + "num_tokens": 383701854.0, + "step": 10926 + }, + { + "epoch": 2.029155060352832, + "grad_norm": 1.6387274265289307, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8696529269218445, + "num_tokens": 383734364.0, + "step": 10927 + }, + { + "epoch": 2.0293407613741876, + "grad_norm": 1.7593828439712524, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8844928741455078, + "num_tokens": 383760601.0, + "step": 10928 + }, + { + "epoch": 2.0295264623955434, + "grad_norm": 1.693583369255066, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8873926401138306, + "num_tokens": 383794185.0, + "step": 10929 + }, + { + "epoch": 2.0297121634168986, + "grad_norm": 1.5019922256469727, + "learning_rate": 1e-06, + "loss": 0.2707, + "mean_token_accuracy": 0.9038659334182739, + "num_tokens": 383831766.0, + "step": 10930 + }, + { + "epoch": 2.0298978644382544, + "grad_norm": 1.5679289102554321, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8846222758293152, + "num_tokens": 383869621.0, + "step": 10931 + }, + { + "epoch": 2.03008356545961, + "grad_norm": 1.6748785972595215, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8733856678009033, + "num_tokens": 383903862.0, + "step": 10932 + }, + { + "epoch": 2.030269266480966, + "grad_norm": 1.6251779794692993, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.883664071559906, + "num_tokens": 383939175.0, + "step": 10933 + }, + { + "epoch": 2.030454967502321, + "grad_norm": 1.565937876701355, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8779749870300293, + "num_tokens": 383976860.0, + "step": 10934 + }, + { + "epoch": 2.030640668523677, + "grad_norm": 1.4287134408950806, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8948585987091064, + "num_tokens": 384015794.0, + "step": 10935 + }, + { + "epoch": 2.0308263695450326, + "grad_norm": 1.556146264076233, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8713718056678772, + "num_tokens": 384053077.0, + "step": 10936 + }, + { + "epoch": 2.0310120705663883, + "grad_norm": 1.5922430753707886, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8886041045188904, + "num_tokens": 384088612.0, + "step": 10937 + }, + { + "epoch": 2.0311977715877436, + "grad_norm": 1.5263493061065674, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8902477025985718, + "num_tokens": 384124322.0, + "step": 10938 + }, + { + "epoch": 2.0313834726090993, + "grad_norm": 1.5119752883911133, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8839820623397827, + "num_tokens": 384159432.0, + "step": 10939 + }, + { + "epoch": 2.031569173630455, + "grad_norm": 1.6102174520492554, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.887956976890564, + "num_tokens": 384193654.0, + "step": 10940 + }, + { + "epoch": 2.0317548746518104, + "grad_norm": 1.3990827798843384, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8971991539001465, + "num_tokens": 384232223.0, + "step": 10941 + }, + { + "epoch": 2.031940575673166, + "grad_norm": 1.5866841077804565, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8981108069419861, + "num_tokens": 384269877.0, + "step": 10942 + }, + { + "epoch": 2.032126276694522, + "grad_norm": 1.6386240720748901, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8748988509178162, + "num_tokens": 384301064.0, + "step": 10943 + }, + { + "epoch": 2.0323119777158776, + "grad_norm": 1.5744413137435913, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8851449489593506, + "num_tokens": 384335382.0, + "step": 10944 + }, + { + "epoch": 2.032497678737233, + "grad_norm": 1.59678053855896, + "learning_rate": 1e-06, + "loss": 0.2538, + "mean_token_accuracy": 0.9100906252861023, + "num_tokens": 384364264.0, + "step": 10945 + }, + { + "epoch": 2.0326833797585886, + "grad_norm": 1.504859209060669, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8834197521209717, + "num_tokens": 384403006.0, + "step": 10946 + }, + { + "epoch": 2.0328690807799443, + "grad_norm": 1.524973750114441, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8857993483543396, + "num_tokens": 384441819.0, + "step": 10947 + }, + { + "epoch": 2.0330547818013, + "grad_norm": 1.5668327808380127, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8952679634094238, + "num_tokens": 384475541.0, + "step": 10948 + }, + { + "epoch": 2.0332404828226553, + "grad_norm": 1.4372076988220215, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8878134489059448, + "num_tokens": 384516410.0, + "step": 10949 + }, + { + "epoch": 2.033426183844011, + "grad_norm": 1.437078833580017, + "learning_rate": 1e-06, + "loss": 0.2659, + "mean_token_accuracy": 0.9077666401863098, + "num_tokens": 384551138.0, + "step": 10950 + }, + { + "epoch": 2.033611884865367, + "grad_norm": 1.5209473371505737, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8914654850959778, + "num_tokens": 384593604.0, + "step": 10951 + }, + { + "epoch": 2.0337975858867225, + "grad_norm": 2.719261407852173, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8936187028884888, + "num_tokens": 384629988.0, + "step": 10952 + }, + { + "epoch": 2.033983286908078, + "grad_norm": 1.5100451707839966, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8955913782119751, + "num_tokens": 384662909.0, + "step": 10953 + }, + { + "epoch": 2.0341689879294336, + "grad_norm": 1.681408166885376, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8798995018005371, + "num_tokens": 384695700.0, + "step": 10954 + }, + { + "epoch": 2.0343546889507893, + "grad_norm": 1.616613507270813, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8677963018417358, + "num_tokens": 384733453.0, + "step": 10955 + }, + { + "epoch": 2.034540389972145, + "grad_norm": 1.715506672859192, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8763039708137512, + "num_tokens": 384763271.0, + "step": 10956 + }, + { + "epoch": 2.0347260909935003, + "grad_norm": 1.4017744064331055, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8885000348091125, + "num_tokens": 384804115.0, + "step": 10957 + }, + { + "epoch": 2.034911792014856, + "grad_norm": 1.55242121219635, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8818879723548889, + "num_tokens": 384840034.0, + "step": 10958 + }, + { + "epoch": 2.035097493036212, + "grad_norm": 1.5129525661468506, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.873583197593689, + "num_tokens": 384880922.0, + "step": 10959 + }, + { + "epoch": 2.0352831940575675, + "grad_norm": 1.525273084640503, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.884607195854187, + "num_tokens": 384920422.0, + "step": 10960 + }, + { + "epoch": 2.035468895078923, + "grad_norm": 1.6849509477615356, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8791540861129761, + "num_tokens": 384954398.0, + "step": 10961 + }, + { + "epoch": 2.0356545961002785, + "grad_norm": 1.6329326629638672, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8924935460090637, + "num_tokens": 384984825.0, + "step": 10962 + }, + { + "epoch": 2.0358402971216343, + "grad_norm": 1.519437551498413, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8860171437263489, + "num_tokens": 385021510.0, + "step": 10963 + }, + { + "epoch": 2.0360259981429896, + "grad_norm": 1.5399647951126099, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8983680009841919, + "num_tokens": 385053987.0, + "step": 10964 + }, + { + "epoch": 2.0362116991643453, + "grad_norm": 1.5050822496414185, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8774423599243164, + "num_tokens": 385089255.0, + "step": 10965 + }, + { + "epoch": 2.036397400185701, + "grad_norm": 1.570314884185791, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.888392984867096, + "num_tokens": 385125425.0, + "step": 10966 + }, + { + "epoch": 2.0365831012070568, + "grad_norm": 1.663075566291809, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.876401960849762, + "num_tokens": 385157670.0, + "step": 10967 + }, + { + "epoch": 2.036768802228412, + "grad_norm": 1.573643445968628, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.884221613407135, + "num_tokens": 385195691.0, + "step": 10968 + }, + { + "epoch": 2.036954503249768, + "grad_norm": 1.5420314073562622, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.9011714458465576, + "num_tokens": 385230243.0, + "step": 10969 + }, + { + "epoch": 2.0371402042711235, + "grad_norm": 1.703697919845581, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.9000318050384521, + "num_tokens": 385256542.0, + "step": 10970 + }, + { + "epoch": 2.0373259052924793, + "grad_norm": 1.5368767976760864, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.884732723236084, + "num_tokens": 385290844.0, + "step": 10971 + }, + { + "epoch": 2.0375116063138345, + "grad_norm": 1.5708502531051636, + "learning_rate": 1e-06, + "loss": 0.2795, + "mean_token_accuracy": 0.904011607170105, + "num_tokens": 385321436.0, + "step": 10972 + }, + { + "epoch": 2.0376973073351903, + "grad_norm": 1.4386569261550903, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.8980138301849365, + "num_tokens": 385359914.0, + "step": 10973 + }, + { + "epoch": 2.037883008356546, + "grad_norm": 1.7127134799957275, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.888514518737793, + "num_tokens": 385391366.0, + "step": 10974 + }, + { + "epoch": 2.0380687093779017, + "grad_norm": 1.4661349058151245, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8815750479698181, + "num_tokens": 385429603.0, + "step": 10975 + }, + { + "epoch": 2.038254410399257, + "grad_norm": 1.9048981666564941, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8825833797454834, + "num_tokens": 385452860.0, + "step": 10976 + }, + { + "epoch": 2.0384401114206128, + "grad_norm": 1.6595088243484497, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8799090385437012, + "num_tokens": 385491134.0, + "step": 10977 + }, + { + "epoch": 2.0386258124419685, + "grad_norm": 1.5838810205459595, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8920594453811646, + "num_tokens": 385524668.0, + "step": 10978 + }, + { + "epoch": 2.0388115134633242, + "grad_norm": 1.7427053451538086, + "learning_rate": 1e-06, + "loss": 0.3851, + "mean_token_accuracy": 0.8704974055290222, + "num_tokens": 385559403.0, + "step": 10979 + }, + { + "epoch": 2.0389972144846795, + "grad_norm": 1.5209639072418213, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8776867985725403, + "num_tokens": 385596350.0, + "step": 10980 + }, + { + "epoch": 2.0391829155060353, + "grad_norm": 1.5790166854858398, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8892371654510498, + "num_tokens": 385627547.0, + "step": 10981 + }, + { + "epoch": 2.039368616527391, + "grad_norm": 1.4737625122070312, + "learning_rate": 1e-06, + "loss": 0.2741, + "mean_token_accuracy": 0.9023045301437378, + "num_tokens": 385661034.0, + "step": 10982 + }, + { + "epoch": 2.0395543175487467, + "grad_norm": 1.598167896270752, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8959653377532959, + "num_tokens": 385691169.0, + "step": 10983 + }, + { + "epoch": 2.039740018570102, + "grad_norm": 1.514204978942871, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8930578827857971, + "num_tokens": 385725549.0, + "step": 10984 + }, + { + "epoch": 2.0399257195914577, + "grad_norm": 1.4460418224334717, + "learning_rate": 1e-06, + "loss": 0.2759, + "mean_token_accuracy": 0.9022521376609802, + "num_tokens": 385762667.0, + "step": 10985 + }, + { + "epoch": 2.0401114206128135, + "grad_norm": 1.5683690309524536, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8866684436798096, + "num_tokens": 385798835.0, + "step": 10986 + }, + { + "epoch": 2.040297121634169, + "grad_norm": 1.4845656156539917, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.8989466428756714, + "num_tokens": 385833010.0, + "step": 10987 + }, + { + "epoch": 2.0404828226555245, + "grad_norm": 1.6481505632400513, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8930586576461792, + "num_tokens": 385864034.0, + "step": 10988 + }, + { + "epoch": 2.0406685236768802, + "grad_norm": 1.6507067680358887, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8741879463195801, + "num_tokens": 385899277.0, + "step": 10989 + }, + { + "epoch": 2.040854224698236, + "grad_norm": 1.6121063232421875, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8686489462852478, + "num_tokens": 385934229.0, + "step": 10990 + }, + { + "epoch": 2.0410399257195913, + "grad_norm": 1.9988330602645874, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8838754296302795, + "num_tokens": 385967697.0, + "step": 10991 + }, + { + "epoch": 2.041225626740947, + "grad_norm": 1.494202971458435, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.8955218195915222, + "num_tokens": 386003911.0, + "step": 10992 + }, + { + "epoch": 2.0414113277623027, + "grad_norm": 1.6081136465072632, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8901485204696655, + "num_tokens": 386037720.0, + "step": 10993 + }, + { + "epoch": 2.0415970287836585, + "grad_norm": 1.7995445728302002, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.866571843624115, + "num_tokens": 386067188.0, + "step": 10994 + }, + { + "epoch": 2.0417827298050137, + "grad_norm": 1.6721484661102295, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8794055581092834, + "num_tokens": 386099337.0, + "step": 10995 + }, + { + "epoch": 2.0419684308263695, + "grad_norm": 1.716754674911499, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8853135704994202, + "num_tokens": 386129665.0, + "step": 10996 + }, + { + "epoch": 2.042154131847725, + "grad_norm": 1.646254062652588, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8697035312652588, + "num_tokens": 386163890.0, + "step": 10997 + }, + { + "epoch": 2.042339832869081, + "grad_norm": 1.6448732614517212, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.885076642036438, + "num_tokens": 386196722.0, + "step": 10998 + }, + { + "epoch": 2.0425255338904362, + "grad_norm": 1.3763140439987183, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8924185633659363, + "num_tokens": 386239201.0, + "step": 10999 + }, + { + "epoch": 2.042711234911792, + "grad_norm": 1.5151381492614746, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.9008246660232544, + "num_tokens": 386272730.0, + "step": 11000 + }, + { + "epoch": 2.0428969359331477, + "grad_norm": 1.7283095121383667, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8703899383544922, + "num_tokens": 386306652.0, + "step": 11001 + }, + { + "epoch": 2.0430826369545034, + "grad_norm": 1.4607280492782593, + "learning_rate": 1e-06, + "loss": 0.2725, + "mean_token_accuracy": 0.9035759568214417, + "num_tokens": 386344980.0, + "step": 11002 + }, + { + "epoch": 2.0432683379758587, + "grad_norm": 1.6165251731872559, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8880617618560791, + "num_tokens": 386379720.0, + "step": 11003 + }, + { + "epoch": 2.0434540389972145, + "grad_norm": 1.679414987564087, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8898468017578125, + "num_tokens": 386408813.0, + "step": 11004 + }, + { + "epoch": 2.04363974001857, + "grad_norm": 1.5820001363754272, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8839508891105652, + "num_tokens": 386441808.0, + "step": 11005 + }, + { + "epoch": 2.043825441039926, + "grad_norm": 1.4990438222885132, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8747255206108093, + "num_tokens": 386480443.0, + "step": 11006 + }, + { + "epoch": 2.044011142061281, + "grad_norm": 1.6570278406143188, + "learning_rate": 1e-06, + "loss": 0.4171, + "mean_token_accuracy": 0.8575859069824219, + "num_tokens": 386517061.0, + "step": 11007 + }, + { + "epoch": 2.044196843082637, + "grad_norm": 1.5901992321014404, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8859199285507202, + "num_tokens": 386550099.0, + "step": 11008 + }, + { + "epoch": 2.0443825441039927, + "grad_norm": 1.641082763671875, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8858226537704468, + "num_tokens": 386583124.0, + "step": 11009 + }, + { + "epoch": 2.0445682451253484, + "grad_norm": 1.3991998434066772, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8943582773208618, + "num_tokens": 386621218.0, + "step": 11010 + }, + { + "epoch": 2.0447539461467037, + "grad_norm": 1.5431917905807495, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8906375169754028, + "num_tokens": 386656131.0, + "step": 11011 + }, + { + "epoch": 2.0449396471680594, + "grad_norm": 1.593873143196106, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8752657175064087, + "num_tokens": 386690555.0, + "step": 11012 + }, + { + "epoch": 2.045125348189415, + "grad_norm": 1.5378116369247437, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8959587812423706, + "num_tokens": 386721476.0, + "step": 11013 + }, + { + "epoch": 2.0453110492107704, + "grad_norm": 1.6731871366500854, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8826179504394531, + "num_tokens": 386751885.0, + "step": 11014 + }, + { + "epoch": 2.045496750232126, + "grad_norm": 1.5912617444992065, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8880391120910645, + "num_tokens": 386785874.0, + "step": 11015 + }, + { + "epoch": 2.045682451253482, + "grad_norm": 1.5302428007125854, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8747996091842651, + "num_tokens": 386821372.0, + "step": 11016 + }, + { + "epoch": 2.0458681522748376, + "grad_norm": 1.638904094696045, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8778672218322754, + "num_tokens": 386855003.0, + "step": 11017 + }, + { + "epoch": 2.046053853296193, + "grad_norm": 1.5940605401992798, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8885469436645508, + "num_tokens": 386888865.0, + "step": 11018 + }, + { + "epoch": 2.0462395543175487, + "grad_norm": 1.5607160329818726, + "learning_rate": 1e-06, + "loss": 0.2947, + "mean_token_accuracy": 0.8967082500457764, + "num_tokens": 386921532.0, + "step": 11019 + }, + { + "epoch": 2.0464252553389044, + "grad_norm": 1.5243827104568481, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8868600130081177, + "num_tokens": 386956727.0, + "step": 11020 + }, + { + "epoch": 2.04661095636026, + "grad_norm": 1.5503472089767456, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8814425468444824, + "num_tokens": 386991440.0, + "step": 11021 + }, + { + "epoch": 2.0467966573816154, + "grad_norm": 1.5227539539337158, + "learning_rate": 1e-06, + "loss": 0.3023, + "mean_token_accuracy": 0.8912468552589417, + "num_tokens": 387028263.0, + "step": 11022 + }, + { + "epoch": 2.046982358402971, + "grad_norm": 1.59433114528656, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8855962157249451, + "num_tokens": 387065014.0, + "step": 11023 + }, + { + "epoch": 2.047168059424327, + "grad_norm": 1.6326837539672852, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8884415626525879, + "num_tokens": 387099792.0, + "step": 11024 + }, + { + "epoch": 2.0473537604456826, + "grad_norm": 1.6213680505752563, + "learning_rate": 1e-06, + "loss": 0.3602, + "mean_token_accuracy": 0.873335599899292, + "num_tokens": 387135177.0, + "step": 11025 + }, + { + "epoch": 2.047539461467038, + "grad_norm": 1.8198471069335938, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8900453448295593, + "num_tokens": 387167229.0, + "step": 11026 + }, + { + "epoch": 2.0477251624883936, + "grad_norm": 1.5163394212722778, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8767071962356567, + "num_tokens": 387205800.0, + "step": 11027 + }, + { + "epoch": 2.0479108635097494, + "grad_norm": 1.6115856170654297, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8863024711608887, + "num_tokens": 387236516.0, + "step": 11028 + }, + { + "epoch": 2.048096564531105, + "grad_norm": 1.483786702156067, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8860867023468018, + "num_tokens": 387275472.0, + "step": 11029 + }, + { + "epoch": 2.0482822655524604, + "grad_norm": 1.6340469121932983, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8810197114944458, + "num_tokens": 387307951.0, + "step": 11030 + }, + { + "epoch": 2.048467966573816, + "grad_norm": 1.5077414512634277, + "learning_rate": 1e-06, + "loss": 0.2678, + "mean_token_accuracy": 0.9051641821861267, + "num_tokens": 387340282.0, + "step": 11031 + }, + { + "epoch": 2.048653667595172, + "grad_norm": 1.5017549991607666, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8823766708374023, + "num_tokens": 387377975.0, + "step": 11032 + }, + { + "epoch": 2.0488393686165276, + "grad_norm": 1.5987612009048462, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.874174952507019, + "num_tokens": 387415214.0, + "step": 11033 + }, + { + "epoch": 2.049025069637883, + "grad_norm": 1.5169258117675781, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8815362453460693, + "num_tokens": 387454098.0, + "step": 11034 + }, + { + "epoch": 2.0492107706592386, + "grad_norm": 1.511409878730774, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8738030791282654, + "num_tokens": 387492512.0, + "step": 11035 + }, + { + "epoch": 2.0493964716805944, + "grad_norm": 1.4337832927703857, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8915770053863525, + "num_tokens": 387529991.0, + "step": 11036 + }, + { + "epoch": 2.0495821727019496, + "grad_norm": 1.5535029172897339, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8893075585365295, + "num_tokens": 387563584.0, + "step": 11037 + }, + { + "epoch": 2.0497678737233054, + "grad_norm": 1.5898735523223877, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8786407709121704, + "num_tokens": 387598825.0, + "step": 11038 + }, + { + "epoch": 2.049953574744661, + "grad_norm": 1.5045416355133057, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8825216889381409, + "num_tokens": 387637699.0, + "step": 11039 + }, + { + "epoch": 2.050139275766017, + "grad_norm": 1.591210126876831, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8773807287216187, + "num_tokens": 387671714.0, + "step": 11040 + }, + { + "epoch": 2.050324976787372, + "grad_norm": 1.5521999597549438, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8901220560073853, + "num_tokens": 387709055.0, + "step": 11041 + }, + { + "epoch": 2.050510677808728, + "grad_norm": 1.6560819149017334, + "learning_rate": 1e-06, + "loss": 0.2829, + "mean_token_accuracy": 0.902043342590332, + "num_tokens": 387737155.0, + "step": 11042 + }, + { + "epoch": 2.0506963788300836, + "grad_norm": 1.5531855821609497, + "learning_rate": 1e-06, + "loss": 0.3896, + "mean_token_accuracy": 0.8645520210266113, + "num_tokens": 387777509.0, + "step": 11043 + }, + { + "epoch": 2.0508820798514393, + "grad_norm": 1.5431548357009888, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8824134469032288, + "num_tokens": 387820958.0, + "step": 11044 + }, + { + "epoch": 2.0510677808727946, + "grad_norm": 1.7087523937225342, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8833884596824646, + "num_tokens": 387851996.0, + "step": 11045 + }, + { + "epoch": 2.0512534818941504, + "grad_norm": 1.4670414924621582, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8884100914001465, + "num_tokens": 387892919.0, + "step": 11046 + }, + { + "epoch": 2.051439182915506, + "grad_norm": 1.501143217086792, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8924945592880249, + "num_tokens": 387926779.0, + "step": 11047 + }, + { + "epoch": 2.051624883936862, + "grad_norm": 1.423555612564087, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.8988370895385742, + "num_tokens": 387964428.0, + "step": 11048 + }, + { + "epoch": 2.051810584958217, + "grad_norm": 1.6092827320098877, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8806885480880737, + "num_tokens": 387995774.0, + "step": 11049 + }, + { + "epoch": 2.051996285979573, + "grad_norm": 1.482820987701416, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8828747272491455, + "num_tokens": 388032537.0, + "step": 11050 + }, + { + "epoch": 2.0521819870009286, + "grad_norm": 1.4516416788101196, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8806322813034058, + "num_tokens": 388072032.0, + "step": 11051 + }, + { + "epoch": 2.0523676880222843, + "grad_norm": 1.5791319608688354, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8938195109367371, + "num_tokens": 388105988.0, + "step": 11052 + }, + { + "epoch": 2.0525533890436396, + "grad_norm": 1.602616786956787, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8939394950866699, + "num_tokens": 388139302.0, + "step": 11053 + }, + { + "epoch": 2.0527390900649953, + "grad_norm": 1.4859486818313599, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8944347500801086, + "num_tokens": 388174825.0, + "step": 11054 + }, + { + "epoch": 2.052924791086351, + "grad_norm": 1.5893681049346924, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8896777629852295, + "num_tokens": 388206489.0, + "step": 11055 + }, + { + "epoch": 2.053110492107707, + "grad_norm": 1.5003408193588257, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8802372217178345, + "num_tokens": 388248193.0, + "step": 11056 + }, + { + "epoch": 2.053296193129062, + "grad_norm": 1.5141632556915283, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8878214359283447, + "num_tokens": 388283140.0, + "step": 11057 + }, + { + "epoch": 2.053481894150418, + "grad_norm": 1.474145770072937, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8808718323707581, + "num_tokens": 388322941.0, + "step": 11058 + }, + { + "epoch": 2.0536675951717736, + "grad_norm": 1.5792512893676758, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8875540494918823, + "num_tokens": 388355789.0, + "step": 11059 + }, + { + "epoch": 2.053853296193129, + "grad_norm": 1.5386735200881958, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8891298770904541, + "num_tokens": 388391131.0, + "step": 11060 + }, + { + "epoch": 2.0540389972144846, + "grad_norm": 1.6333098411560059, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8809220790863037, + "num_tokens": 388423919.0, + "step": 11061 + }, + { + "epoch": 2.0542246982358403, + "grad_norm": 1.4353662729263306, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8929048776626587, + "num_tokens": 388463825.0, + "step": 11062 + }, + { + "epoch": 2.054410399257196, + "grad_norm": 1.5087672472000122, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8912404775619507, + "num_tokens": 388500847.0, + "step": 11063 + }, + { + "epoch": 2.0545961002785513, + "grad_norm": 1.6306339502334595, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.892264187335968, + "num_tokens": 388531994.0, + "step": 11064 + }, + { + "epoch": 2.054781801299907, + "grad_norm": 1.6164960861206055, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8792344331741333, + "num_tokens": 388568457.0, + "step": 11065 + }, + { + "epoch": 2.054967502321263, + "grad_norm": 1.6018621921539307, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8878618478775024, + "num_tokens": 388603506.0, + "step": 11066 + }, + { + "epoch": 2.0551532033426185, + "grad_norm": 1.46689772605896, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.8992090225219727, + "num_tokens": 388638736.0, + "step": 11067 + }, + { + "epoch": 2.055338904363974, + "grad_norm": 1.5236282348632812, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8871037364006042, + "num_tokens": 388675609.0, + "step": 11068 + }, + { + "epoch": 2.0555246053853296, + "grad_norm": 1.5350053310394287, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8939732909202576, + "num_tokens": 388708633.0, + "step": 11069 + }, + { + "epoch": 2.0557103064066853, + "grad_norm": 1.5671273469924927, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8913875818252563, + "num_tokens": 388741465.0, + "step": 11070 + }, + { + "epoch": 2.055896007428041, + "grad_norm": 1.69670569896698, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8733111619949341, + "num_tokens": 388774172.0, + "step": 11071 + }, + { + "epoch": 2.0560817084493963, + "grad_norm": 1.62786066532135, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8825926780700684, + "num_tokens": 388804993.0, + "step": 11072 + }, + { + "epoch": 2.056267409470752, + "grad_norm": 1.5695548057556152, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8949205875396729, + "num_tokens": 388837109.0, + "step": 11073 + }, + { + "epoch": 2.0564531104921078, + "grad_norm": 1.6015591621398926, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.8973702192306519, + "num_tokens": 388870688.0, + "step": 11074 + }, + { + "epoch": 2.0566388115134635, + "grad_norm": 1.6180477142333984, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8865774869918823, + "num_tokens": 388902553.0, + "step": 11075 + }, + { + "epoch": 2.056824512534819, + "grad_norm": 1.3859866857528687, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8912575244903564, + "num_tokens": 388945319.0, + "step": 11076 + }, + { + "epoch": 2.0570102135561745, + "grad_norm": 1.4612643718719482, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8904550671577454, + "num_tokens": 388982362.0, + "step": 11077 + }, + { + "epoch": 2.0571959145775303, + "grad_norm": 1.5828522443771362, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8951998949050903, + "num_tokens": 389014501.0, + "step": 11078 + }, + { + "epoch": 2.057381615598886, + "grad_norm": 1.525193691253662, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8919426202774048, + "num_tokens": 389045886.0, + "step": 11079 + }, + { + "epoch": 2.0575673166202413, + "grad_norm": 1.437203288078308, + "learning_rate": 1e-06, + "loss": 0.2747, + "mean_token_accuracy": 0.9063316583633423, + "num_tokens": 389078306.0, + "step": 11080 + }, + { + "epoch": 2.057753017641597, + "grad_norm": 1.7266368865966797, + "learning_rate": 1e-06, + "loss": 0.3782, + "mean_token_accuracy": 0.8684182167053223, + "num_tokens": 389107858.0, + "step": 11081 + }, + { + "epoch": 2.0579387186629527, + "grad_norm": 1.416930913925171, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8957779407501221, + "num_tokens": 389145284.0, + "step": 11082 + }, + { + "epoch": 2.0581244196843085, + "grad_norm": 1.5130436420440674, + "learning_rate": 1e-06, + "loss": 0.2803, + "mean_token_accuracy": 0.899634599685669, + "num_tokens": 389176244.0, + "step": 11083 + }, + { + "epoch": 2.0583101207056638, + "grad_norm": 1.489863395690918, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8892691135406494, + "num_tokens": 389213276.0, + "step": 11084 + }, + { + "epoch": 2.0584958217270195, + "grad_norm": 1.5580391883850098, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8828641772270203, + "num_tokens": 389246629.0, + "step": 11085 + }, + { + "epoch": 2.0586815227483752, + "grad_norm": 1.5330467224121094, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8914957642555237, + "num_tokens": 389281277.0, + "step": 11086 + }, + { + "epoch": 2.0588672237697305, + "grad_norm": 1.5586122274398804, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8917341232299805, + "num_tokens": 389315049.0, + "step": 11087 + }, + { + "epoch": 2.0590529247910863, + "grad_norm": 1.517279863357544, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8912638425827026, + "num_tokens": 389350897.0, + "step": 11088 + }, + { + "epoch": 2.059238625812442, + "grad_norm": 1.4444077014923096, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8853212594985962, + "num_tokens": 389393503.0, + "step": 11089 + }, + { + "epoch": 2.0594243268337977, + "grad_norm": 1.6801577806472778, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8832277059555054, + "num_tokens": 389425746.0, + "step": 11090 + }, + { + "epoch": 2.059610027855153, + "grad_norm": 1.5106613636016846, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.872046172618866, + "num_tokens": 389467270.0, + "step": 11091 + }, + { + "epoch": 2.0597957288765087, + "grad_norm": 1.5351357460021973, + "learning_rate": 1e-06, + "loss": 0.2702, + "mean_token_accuracy": 0.9022732973098755, + "num_tokens": 389502242.0, + "step": 11092 + }, + { + "epoch": 2.0599814298978645, + "grad_norm": 1.4826123714447021, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8875013589859009, + "num_tokens": 389541592.0, + "step": 11093 + }, + { + "epoch": 2.06016713091922, + "grad_norm": 1.4469388723373413, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8957867622375488, + "num_tokens": 389579578.0, + "step": 11094 + }, + { + "epoch": 2.0603528319405755, + "grad_norm": 1.5867254734039307, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8930857181549072, + "num_tokens": 389613400.0, + "step": 11095 + }, + { + "epoch": 2.0605385329619312, + "grad_norm": 1.4951280355453491, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8875996470451355, + "num_tokens": 389651555.0, + "step": 11096 + }, + { + "epoch": 2.060724233983287, + "grad_norm": 1.4572328329086304, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8997104167938232, + "num_tokens": 389688875.0, + "step": 11097 + }, + { + "epoch": 2.0609099350046427, + "grad_norm": 1.6148250102996826, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8946625590324402, + "num_tokens": 389720076.0, + "step": 11098 + }, + { + "epoch": 2.061095636025998, + "grad_norm": 1.58696448802948, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8814923763275146, + "num_tokens": 389754347.0, + "step": 11099 + }, + { + "epoch": 2.0612813370473537, + "grad_norm": 1.4433022737503052, + "learning_rate": 1e-06, + "loss": 0.2765, + "mean_token_accuracy": 0.9031490683555603, + "num_tokens": 389788079.0, + "step": 11100 + }, + { + "epoch": 2.0614670380687095, + "grad_norm": 1.5146801471710205, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8888448476791382, + "num_tokens": 389825018.0, + "step": 11101 + }, + { + "epoch": 2.061652739090065, + "grad_norm": 1.6136647462844849, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8887691497802734, + "num_tokens": 389859420.0, + "step": 11102 + }, + { + "epoch": 2.0618384401114205, + "grad_norm": 1.493728518486023, + "learning_rate": 1e-06, + "loss": 0.3084, + "mean_token_accuracy": 0.8915702104568481, + "num_tokens": 389895911.0, + "step": 11103 + }, + { + "epoch": 2.062024141132776, + "grad_norm": 1.523529291152954, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8822152018547058, + "num_tokens": 389935074.0, + "step": 11104 + }, + { + "epoch": 2.062209842154132, + "grad_norm": 1.3971960544586182, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8869998455047607, + "num_tokens": 389978242.0, + "step": 11105 + }, + { + "epoch": 2.0623955431754877, + "grad_norm": 1.5457744598388672, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8857081532478333, + "num_tokens": 390013180.0, + "step": 11106 + }, + { + "epoch": 2.062581244196843, + "grad_norm": 1.5433465242385864, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8867661952972412, + "num_tokens": 390048706.0, + "step": 11107 + }, + { + "epoch": 2.0627669452181987, + "grad_norm": 1.4440480470657349, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8782558441162109, + "num_tokens": 390090635.0, + "step": 11108 + }, + { + "epoch": 2.0629526462395544, + "grad_norm": 1.671927571296692, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8871826529502869, + "num_tokens": 390120399.0, + "step": 11109 + }, + { + "epoch": 2.0631383472609097, + "grad_norm": 1.6988054513931274, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8948291540145874, + "num_tokens": 390148756.0, + "step": 11110 + }, + { + "epoch": 2.0633240482822655, + "grad_norm": 1.6372159719467163, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.9019114971160889, + "num_tokens": 390177256.0, + "step": 11111 + }, + { + "epoch": 2.063509749303621, + "grad_norm": 1.532109022140503, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8739030361175537, + "num_tokens": 390218004.0, + "step": 11112 + }, + { + "epoch": 2.063695450324977, + "grad_norm": 1.6082829236984253, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8887519836425781, + "num_tokens": 390250623.0, + "step": 11113 + }, + { + "epoch": 2.063881151346332, + "grad_norm": 1.6353298425674438, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8760186433792114, + "num_tokens": 390284234.0, + "step": 11114 + }, + { + "epoch": 2.064066852367688, + "grad_norm": 1.6640080213546753, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8798884153366089, + "num_tokens": 390316052.0, + "step": 11115 + }, + { + "epoch": 2.0642525533890437, + "grad_norm": 1.5119843482971191, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8871132135391235, + "num_tokens": 390353633.0, + "step": 11116 + }, + { + "epoch": 2.0644382544103994, + "grad_norm": 1.510332703590393, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.890565037727356, + "num_tokens": 390390838.0, + "step": 11117 + }, + { + "epoch": 2.0646239554317547, + "grad_norm": 1.4436782598495483, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8905380964279175, + "num_tokens": 390430538.0, + "step": 11118 + }, + { + "epoch": 2.0648096564531104, + "grad_norm": 1.7151437997817993, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8768502473831177, + "num_tokens": 390460029.0, + "step": 11119 + }, + { + "epoch": 2.064995357474466, + "grad_norm": 1.5272125005722046, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8867422342300415, + "num_tokens": 390495474.0, + "step": 11120 + }, + { + "epoch": 2.065181058495822, + "grad_norm": 1.476971983909607, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8806670308113098, + "num_tokens": 390535147.0, + "step": 11121 + }, + { + "epoch": 2.065366759517177, + "grad_norm": 1.6570886373519897, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8764535188674927, + "num_tokens": 390567602.0, + "step": 11122 + }, + { + "epoch": 2.065552460538533, + "grad_norm": 1.4329639673233032, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8962984085083008, + "num_tokens": 390603484.0, + "step": 11123 + }, + { + "epoch": 2.0657381615598887, + "grad_norm": 1.6616463661193848, + "learning_rate": 1e-06, + "loss": 0.2717, + "mean_token_accuracy": 0.9011590480804443, + "num_tokens": 390630795.0, + "step": 11124 + }, + { + "epoch": 2.0659238625812444, + "grad_norm": 1.6187372207641602, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8909175395965576, + "num_tokens": 390664192.0, + "step": 11125 + }, + { + "epoch": 2.0661095636025997, + "grad_norm": 1.563692331314087, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8862290382385254, + "num_tokens": 390697571.0, + "step": 11126 + }, + { + "epoch": 2.0662952646239554, + "grad_norm": 1.4281636476516724, + "learning_rate": 1e-06, + "loss": 0.2774, + "mean_token_accuracy": 0.9007188081741333, + "num_tokens": 390734433.0, + "step": 11127 + }, + { + "epoch": 2.066480965645311, + "grad_norm": 1.4664771556854248, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.892224907875061, + "num_tokens": 390769908.0, + "step": 11128 + }, + { + "epoch": 2.066666666666667, + "grad_norm": 1.7135870456695557, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8956773281097412, + "num_tokens": 390796557.0, + "step": 11129 + }, + { + "epoch": 2.066852367688022, + "grad_norm": 1.6190184354782104, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8825340270996094, + "num_tokens": 390829918.0, + "step": 11130 + }, + { + "epoch": 2.067038068709378, + "grad_norm": 1.5899851322174072, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8855626583099365, + "num_tokens": 390866038.0, + "step": 11131 + }, + { + "epoch": 2.0672237697307336, + "grad_norm": 1.459640383720398, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8958798050880432, + "num_tokens": 390900872.0, + "step": 11132 + }, + { + "epoch": 2.0674094707520894, + "grad_norm": 1.4400683641433716, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8836511969566345, + "num_tokens": 390940194.0, + "step": 11133 + }, + { + "epoch": 2.0675951717734447, + "grad_norm": 1.5316745042800903, + "learning_rate": 1e-06, + "loss": 0.2732, + "mean_token_accuracy": 0.9028317928314209, + "num_tokens": 390972120.0, + "step": 11134 + }, + { + "epoch": 2.0677808727948004, + "grad_norm": 1.5148121118545532, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8938642740249634, + "num_tokens": 391004804.0, + "step": 11135 + }, + { + "epoch": 2.067966573816156, + "grad_norm": 1.644466757774353, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8857805728912354, + "num_tokens": 391038932.0, + "step": 11136 + }, + { + "epoch": 2.0681522748375114, + "grad_norm": 1.6514519453048706, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8902492523193359, + "num_tokens": 391068873.0, + "step": 11137 + }, + { + "epoch": 2.068337975858867, + "grad_norm": 1.6617507934570312, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8822617530822754, + "num_tokens": 391098442.0, + "step": 11138 + }, + { + "epoch": 2.068523676880223, + "grad_norm": 1.6553255319595337, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8864498138427734, + "num_tokens": 391129439.0, + "step": 11139 + }, + { + "epoch": 2.0687093779015786, + "grad_norm": 1.5624014139175415, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8803905248641968, + "num_tokens": 391164118.0, + "step": 11140 + }, + { + "epoch": 2.068895078922934, + "grad_norm": 1.5287234783172607, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8894098997116089, + "num_tokens": 391199526.0, + "step": 11141 + }, + { + "epoch": 2.0690807799442896, + "grad_norm": 1.6789727210998535, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8873507976531982, + "num_tokens": 391233997.0, + "step": 11142 + }, + { + "epoch": 2.0692664809656454, + "grad_norm": 1.640042781829834, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8837355971336365, + "num_tokens": 391268249.0, + "step": 11143 + }, + { + "epoch": 2.069452181987001, + "grad_norm": 1.5587555170059204, + "learning_rate": 1e-06, + "loss": 0.2875, + "mean_token_accuracy": 0.8977614045143127, + "num_tokens": 391301049.0, + "step": 11144 + }, + { + "epoch": 2.0696378830083564, + "grad_norm": 1.6983742713928223, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8812592625617981, + "num_tokens": 391331005.0, + "step": 11145 + }, + { + "epoch": 2.069823584029712, + "grad_norm": 1.5828375816345215, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8885990381240845, + "num_tokens": 391363103.0, + "step": 11146 + }, + { + "epoch": 2.070009285051068, + "grad_norm": 1.5902553796768188, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8886069655418396, + "num_tokens": 391395915.0, + "step": 11147 + }, + { + "epoch": 2.0701949860724236, + "grad_norm": 1.6224033832550049, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8808193802833557, + "num_tokens": 391427388.0, + "step": 11148 + }, + { + "epoch": 2.070380687093779, + "grad_norm": 1.5850889682769775, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8907442092895508, + "num_tokens": 391458521.0, + "step": 11149 + }, + { + "epoch": 2.0705663881151346, + "grad_norm": 1.43691086769104, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8966575264930725, + "num_tokens": 391492087.0, + "step": 11150 + }, + { + "epoch": 2.0707520891364903, + "grad_norm": 1.5896919965744019, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8887669444084167, + "num_tokens": 391524782.0, + "step": 11151 + }, + { + "epoch": 2.070937790157846, + "grad_norm": 1.4737486839294434, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8852975368499756, + "num_tokens": 391559799.0, + "step": 11152 + }, + { + "epoch": 2.0711234911792014, + "grad_norm": 1.6247646808624268, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8948073983192444, + "num_tokens": 391590881.0, + "step": 11153 + }, + { + "epoch": 2.071309192200557, + "grad_norm": 1.6828079223632812, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8833305239677429, + "num_tokens": 391626417.0, + "step": 11154 + }, + { + "epoch": 2.071494893221913, + "grad_norm": 1.5213285684585571, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8916659951210022, + "num_tokens": 391663045.0, + "step": 11155 + }, + { + "epoch": 2.0716805942432686, + "grad_norm": 1.6404012441635132, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8841561079025269, + "num_tokens": 391696087.0, + "step": 11156 + }, + { + "epoch": 2.071866295264624, + "grad_norm": 1.6416646242141724, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8891648650169373, + "num_tokens": 391728040.0, + "step": 11157 + }, + { + "epoch": 2.0720519962859796, + "grad_norm": 1.5739766359329224, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8841747045516968, + "num_tokens": 391764862.0, + "step": 11158 + }, + { + "epoch": 2.0722376973073353, + "grad_norm": 1.5223376750946045, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8810999989509583, + "num_tokens": 391804675.0, + "step": 11159 + }, + { + "epoch": 2.0724233983286906, + "grad_norm": 1.6734020709991455, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8841910362243652, + "num_tokens": 391832615.0, + "step": 11160 + }, + { + "epoch": 2.0726090993500463, + "grad_norm": 1.5967457294464111, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8857495784759521, + "num_tokens": 391867032.0, + "step": 11161 + }, + { + "epoch": 2.072794800371402, + "grad_norm": 1.5932812690734863, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8818701505661011, + "num_tokens": 391903690.0, + "step": 11162 + }, + { + "epoch": 2.072980501392758, + "grad_norm": 1.5677471160888672, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.883426308631897, + "num_tokens": 391939845.0, + "step": 11163 + }, + { + "epoch": 2.073166202414113, + "grad_norm": 1.6823270320892334, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8772635459899902, + "num_tokens": 391973583.0, + "step": 11164 + }, + { + "epoch": 2.073351903435469, + "grad_norm": 1.4690064191818237, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8839849233627319, + "num_tokens": 392015070.0, + "step": 11165 + }, + { + "epoch": 2.0735376044568246, + "grad_norm": 1.5321770906448364, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8868894577026367, + "num_tokens": 392056095.0, + "step": 11166 + }, + { + "epoch": 2.0737233054781803, + "grad_norm": 1.548171043395996, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.8930149078369141, + "num_tokens": 392090223.0, + "step": 11167 + }, + { + "epoch": 2.0739090064995356, + "grad_norm": 1.5987930297851562, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8934996128082275, + "num_tokens": 392122584.0, + "step": 11168 + }, + { + "epoch": 2.0740947075208913, + "grad_norm": 1.6172349452972412, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8756923675537109, + "num_tokens": 392156411.0, + "step": 11169 + }, + { + "epoch": 2.074280408542247, + "grad_norm": 1.4782071113586426, + "learning_rate": 1e-06, + "loss": 0.2858, + "mean_token_accuracy": 0.896053671836853, + "num_tokens": 392190955.0, + "step": 11170 + }, + { + "epoch": 2.074466109563603, + "grad_norm": 1.5435914993286133, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8891297578811646, + "num_tokens": 392225615.0, + "step": 11171 + }, + { + "epoch": 2.074651810584958, + "grad_norm": 1.5916999578475952, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8849906921386719, + "num_tokens": 392259475.0, + "step": 11172 + }, + { + "epoch": 2.074837511606314, + "grad_norm": 1.4440040588378906, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8918005228042603, + "num_tokens": 392299597.0, + "step": 11173 + }, + { + "epoch": 2.0750232126276695, + "grad_norm": 1.600759744644165, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8841161727905273, + "num_tokens": 392335010.0, + "step": 11174 + }, + { + "epoch": 2.0752089136490253, + "grad_norm": 1.634539008140564, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8778281211853027, + "num_tokens": 392368616.0, + "step": 11175 + }, + { + "epoch": 2.0753946146703806, + "grad_norm": 1.4255895614624023, + "learning_rate": 1e-06, + "loss": 0.2851, + "mean_token_accuracy": 0.8963403701782227, + "num_tokens": 392406509.0, + "step": 11176 + }, + { + "epoch": 2.0755803156917363, + "grad_norm": 1.6392531394958496, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8779619932174683, + "num_tokens": 392441068.0, + "step": 11177 + }, + { + "epoch": 2.075766016713092, + "grad_norm": 1.514399528503418, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8947433233261108, + "num_tokens": 392479130.0, + "step": 11178 + }, + { + "epoch": 2.0759517177344478, + "grad_norm": 1.4378217458724976, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8959159255027771, + "num_tokens": 392517057.0, + "step": 11179 + }, + { + "epoch": 2.076137418755803, + "grad_norm": 1.5132263898849487, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8912655115127563, + "num_tokens": 392551604.0, + "step": 11180 + }, + { + "epoch": 2.0763231197771588, + "grad_norm": 1.629373550415039, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8822235465049744, + "num_tokens": 392587136.0, + "step": 11181 + }, + { + "epoch": 2.0765088207985145, + "grad_norm": 1.6325300931930542, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.882259726524353, + "num_tokens": 392621792.0, + "step": 11182 + }, + { + "epoch": 2.07669452181987, + "grad_norm": 1.6078778505325317, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8871618509292603, + "num_tokens": 392654394.0, + "step": 11183 + }, + { + "epoch": 2.0768802228412255, + "grad_norm": 1.6267889738082886, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.875099241733551, + "num_tokens": 392690046.0, + "step": 11184 + }, + { + "epoch": 2.0770659238625813, + "grad_norm": 1.6696887016296387, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8800346851348877, + "num_tokens": 392723608.0, + "step": 11185 + }, + { + "epoch": 2.077251624883937, + "grad_norm": 1.47701895236969, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8882904052734375, + "num_tokens": 392764864.0, + "step": 11186 + }, + { + "epoch": 2.0774373259052923, + "grad_norm": 1.695835828781128, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8835580348968506, + "num_tokens": 392799572.0, + "step": 11187 + }, + { + "epoch": 2.077623026926648, + "grad_norm": 1.5267709493637085, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8909651041030884, + "num_tokens": 392834152.0, + "step": 11188 + }, + { + "epoch": 2.0778087279480038, + "grad_norm": 1.5708162784576416, + "learning_rate": 1e-06, + "loss": 0.2693, + "mean_token_accuracy": 0.9030312299728394, + "num_tokens": 392863067.0, + "step": 11189 + }, + { + "epoch": 2.0779944289693595, + "grad_norm": 1.4862629175186157, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8836005926132202, + "num_tokens": 392901508.0, + "step": 11190 + }, + { + "epoch": 2.0781801299907148, + "grad_norm": 1.5429093837738037, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8938789963722229, + "num_tokens": 392932267.0, + "step": 11191 + }, + { + "epoch": 2.0783658310120705, + "grad_norm": 1.5801490545272827, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8766928911209106, + "num_tokens": 392966343.0, + "step": 11192 + }, + { + "epoch": 2.0785515320334262, + "grad_norm": 1.4418331384658813, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8770444393157959, + "num_tokens": 393004882.0, + "step": 11193 + }, + { + "epoch": 2.078737233054782, + "grad_norm": 1.5710515975952148, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8778612613677979, + "num_tokens": 393045865.0, + "step": 11194 + }, + { + "epoch": 2.0789229340761373, + "grad_norm": 1.606008768081665, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8881442546844482, + "num_tokens": 393078852.0, + "step": 11195 + }, + { + "epoch": 2.079108635097493, + "grad_norm": 1.4461705684661865, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8854145407676697, + "num_tokens": 393119855.0, + "step": 11196 + }, + { + "epoch": 2.0792943361188487, + "grad_norm": 1.4991499185562134, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.883663535118103, + "num_tokens": 393156765.0, + "step": 11197 + }, + { + "epoch": 2.0794800371402045, + "grad_norm": 1.4965484142303467, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8984260559082031, + "num_tokens": 393193257.0, + "step": 11198 + }, + { + "epoch": 2.0796657381615598, + "grad_norm": 1.5800141096115112, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8797010183334351, + "num_tokens": 393228761.0, + "step": 11199 + }, + { + "epoch": 2.0798514391829155, + "grad_norm": 1.556652307510376, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8831665515899658, + "num_tokens": 393264430.0, + "step": 11200 + }, + { + "epoch": 2.080037140204271, + "grad_norm": 1.6359046697616577, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8699272871017456, + "num_tokens": 393301646.0, + "step": 11201 + }, + { + "epoch": 2.080222841225627, + "grad_norm": 1.56398606300354, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8815864324569702, + "num_tokens": 393341869.0, + "step": 11202 + }, + { + "epoch": 2.0804085422469822, + "grad_norm": 1.5105990171432495, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.890997052192688, + "num_tokens": 393380206.0, + "step": 11203 + }, + { + "epoch": 2.080594243268338, + "grad_norm": 1.709905743598938, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.8945848345756531, + "num_tokens": 393406932.0, + "step": 11204 + }, + { + "epoch": 2.0807799442896937, + "grad_norm": 1.44977605342865, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8795686960220337, + "num_tokens": 393446016.0, + "step": 11205 + }, + { + "epoch": 2.080965645311049, + "grad_norm": 1.5678519010543823, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8889938592910767, + "num_tokens": 393477221.0, + "step": 11206 + }, + { + "epoch": 2.0811513463324047, + "grad_norm": 1.6034923791885376, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8856828212738037, + "num_tokens": 393508166.0, + "step": 11207 + }, + { + "epoch": 2.0813370473537605, + "grad_norm": 1.649457335472107, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8860427737236023, + "num_tokens": 393539269.0, + "step": 11208 + }, + { + "epoch": 2.081522748375116, + "grad_norm": 1.5975079536437988, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8934062719345093, + "num_tokens": 393569038.0, + "step": 11209 + }, + { + "epoch": 2.0817084493964715, + "grad_norm": 1.6136832237243652, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8798380494117737, + "num_tokens": 393603376.0, + "step": 11210 + }, + { + "epoch": 2.081894150417827, + "grad_norm": 1.7103559970855713, + "learning_rate": 1e-06, + "loss": 0.4003, + "mean_token_accuracy": 0.8669726848602295, + "num_tokens": 393636781.0, + "step": 11211 + }, + { + "epoch": 2.082079851439183, + "grad_norm": 1.4827091693878174, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8977624177932739, + "num_tokens": 393673297.0, + "step": 11212 + }, + { + "epoch": 2.0822655524605387, + "grad_norm": 1.5073230266571045, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8902872800827026, + "num_tokens": 393708692.0, + "step": 11213 + }, + { + "epoch": 2.082451253481894, + "grad_norm": 1.4686847925186157, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8865513801574707, + "num_tokens": 393749382.0, + "step": 11214 + }, + { + "epoch": 2.0826369545032497, + "grad_norm": 1.417169451713562, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8908053636550903, + "num_tokens": 393787048.0, + "step": 11215 + }, + { + "epoch": 2.0828226555246054, + "grad_norm": 1.5323824882507324, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8632053732872009, + "num_tokens": 393828059.0, + "step": 11216 + }, + { + "epoch": 2.083008356545961, + "grad_norm": 1.5738580226898193, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8921627998352051, + "num_tokens": 393862997.0, + "step": 11217 + }, + { + "epoch": 2.0831940575673165, + "grad_norm": 1.4189906120300293, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8935351371765137, + "num_tokens": 393900633.0, + "step": 11218 + }, + { + "epoch": 2.083379758588672, + "grad_norm": 1.7126076221466064, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8864872455596924, + "num_tokens": 393928826.0, + "step": 11219 + }, + { + "epoch": 2.083565459610028, + "grad_norm": 1.6873230934143066, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8600948452949524, + "num_tokens": 393963656.0, + "step": 11220 + }, + { + "epoch": 2.0837511606313837, + "grad_norm": 1.5511500835418701, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8846105337142944, + "num_tokens": 393996599.0, + "step": 11221 + }, + { + "epoch": 2.083936861652739, + "grad_norm": 1.6755307912826538, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8805526494979858, + "num_tokens": 394028937.0, + "step": 11222 + }, + { + "epoch": 2.0841225626740947, + "grad_norm": 1.614611268043518, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8892396092414856, + "num_tokens": 394065522.0, + "step": 11223 + }, + { + "epoch": 2.0843082636954504, + "grad_norm": 1.603410243988037, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8853176236152649, + "num_tokens": 394097868.0, + "step": 11224 + }, + { + "epoch": 2.084493964716806, + "grad_norm": 1.639966368675232, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8835512399673462, + "num_tokens": 394132119.0, + "step": 11225 + }, + { + "epoch": 2.0846796657381614, + "grad_norm": 1.5060533285140991, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8973665237426758, + "num_tokens": 394168675.0, + "step": 11226 + }, + { + "epoch": 2.084865366759517, + "grad_norm": 1.4074807167053223, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.88176429271698, + "num_tokens": 394210826.0, + "step": 11227 + }, + { + "epoch": 2.085051067780873, + "grad_norm": 1.6600532531738281, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8761855363845825, + "num_tokens": 394241766.0, + "step": 11228 + }, + { + "epoch": 2.085236768802228, + "grad_norm": 1.5476880073547363, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8940744400024414, + "num_tokens": 394275108.0, + "step": 11229 + }, + { + "epoch": 2.085422469823584, + "grad_norm": 1.5169926881790161, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8919357061386108, + "num_tokens": 394312370.0, + "step": 11230 + }, + { + "epoch": 2.0856081708449397, + "grad_norm": 1.5359654426574707, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8922523260116577, + "num_tokens": 394347618.0, + "step": 11231 + }, + { + "epoch": 2.0857938718662954, + "grad_norm": 1.6096667051315308, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8905163407325745, + "num_tokens": 394375984.0, + "step": 11232 + }, + { + "epoch": 2.0859795728876507, + "grad_norm": 1.4634836912155151, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8939025402069092, + "num_tokens": 394411161.0, + "step": 11233 + }, + { + "epoch": 2.0861652739090064, + "grad_norm": 1.518552541732788, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8835544586181641, + "num_tokens": 394449238.0, + "step": 11234 + }, + { + "epoch": 2.086350974930362, + "grad_norm": 1.666317105293274, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8952299952507019, + "num_tokens": 394478067.0, + "step": 11235 + }, + { + "epoch": 2.086536675951718, + "grad_norm": 1.5094289779663086, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8898967504501343, + "num_tokens": 394510809.0, + "step": 11236 + }, + { + "epoch": 2.086722376973073, + "grad_norm": 1.5736680030822754, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8913259506225586, + "num_tokens": 394545399.0, + "step": 11237 + }, + { + "epoch": 2.086908077994429, + "grad_norm": 1.4405932426452637, + "learning_rate": 1e-06, + "loss": 0.2984, + "mean_token_accuracy": 0.8936570286750793, + "num_tokens": 394582962.0, + "step": 11238 + }, + { + "epoch": 2.0870937790157846, + "grad_norm": 1.5355316400527954, + "learning_rate": 1e-06, + "loss": 0.2878, + "mean_token_accuracy": 0.8952377438545227, + "num_tokens": 394615392.0, + "step": 11239 + }, + { + "epoch": 2.0872794800371404, + "grad_norm": 1.7076520919799805, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8898788690567017, + "num_tokens": 394646259.0, + "step": 11240 + }, + { + "epoch": 2.0874651810584957, + "grad_norm": 1.5724073648452759, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.891575276851654, + "num_tokens": 394681658.0, + "step": 11241 + }, + { + "epoch": 2.0876508820798514, + "grad_norm": 1.545134425163269, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8907560110092163, + "num_tokens": 394718184.0, + "step": 11242 + }, + { + "epoch": 2.087836583101207, + "grad_norm": 1.5928974151611328, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8776422142982483, + "num_tokens": 394755254.0, + "step": 11243 + }, + { + "epoch": 2.088022284122563, + "grad_norm": 1.6232460737228394, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8699051737785339, + "num_tokens": 394792755.0, + "step": 11244 + }, + { + "epoch": 2.088207985143918, + "grad_norm": 1.5453425645828247, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8914324045181274, + "num_tokens": 394823682.0, + "step": 11245 + }, + { + "epoch": 2.088393686165274, + "grad_norm": 1.6267497539520264, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8826178312301636, + "num_tokens": 394856603.0, + "step": 11246 + }, + { + "epoch": 2.0885793871866296, + "grad_norm": 1.513949990272522, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8863214254379272, + "num_tokens": 394897628.0, + "step": 11247 + }, + { + "epoch": 2.0887650882079853, + "grad_norm": 1.7326860427856445, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8750247955322266, + "num_tokens": 394925009.0, + "step": 11248 + }, + { + "epoch": 2.0889507892293406, + "grad_norm": 1.6297316551208496, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8863441348075867, + "num_tokens": 394959259.0, + "step": 11249 + }, + { + "epoch": 2.0891364902506964, + "grad_norm": 1.7324110269546509, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8930831551551819, + "num_tokens": 394990923.0, + "step": 11250 + }, + { + "epoch": 2.089322191272052, + "grad_norm": 1.5651566982269287, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8941619396209717, + "num_tokens": 395026034.0, + "step": 11251 + }, + { + "epoch": 2.0895078922934074, + "grad_norm": 1.811869740486145, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8708817958831787, + "num_tokens": 395054405.0, + "step": 11252 + }, + { + "epoch": 2.089693593314763, + "grad_norm": 1.4605176448822021, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8895578384399414, + "num_tokens": 395090999.0, + "step": 11253 + }, + { + "epoch": 2.089879294336119, + "grad_norm": 1.4509283304214478, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8887096643447876, + "num_tokens": 395130249.0, + "step": 11254 + }, + { + "epoch": 2.0900649953574746, + "grad_norm": 1.5079752206802368, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8955317735671997, + "num_tokens": 395166289.0, + "step": 11255 + }, + { + "epoch": 2.09025069637883, + "grad_norm": 1.6917742490768433, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8849034905433655, + "num_tokens": 395196809.0, + "step": 11256 + }, + { + "epoch": 2.0904363974001856, + "grad_norm": 1.5641460418701172, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8798285722732544, + "num_tokens": 395234391.0, + "step": 11257 + }, + { + "epoch": 2.0906220984215413, + "grad_norm": 1.5986409187316895, + "learning_rate": 1e-06, + "loss": 0.2947, + "mean_token_accuracy": 0.8972654342651367, + "num_tokens": 395263645.0, + "step": 11258 + }, + { + "epoch": 2.090807799442897, + "grad_norm": 1.6784464120864868, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8768473863601685, + "num_tokens": 395298316.0, + "step": 11259 + }, + { + "epoch": 2.0909935004642524, + "grad_norm": 1.4724217653274536, + "learning_rate": 1e-06, + "loss": 0.292, + "mean_token_accuracy": 0.8961325883865356, + "num_tokens": 395335870.0, + "step": 11260 + }, + { + "epoch": 2.091179201485608, + "grad_norm": 1.5553853511810303, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8844073414802551, + "num_tokens": 395376542.0, + "step": 11261 + }, + { + "epoch": 2.091364902506964, + "grad_norm": 1.9523959159851074, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8928074240684509, + "num_tokens": 395407260.0, + "step": 11262 + }, + { + "epoch": 2.0915506035283196, + "grad_norm": 1.6246001720428467, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8848353624343872, + "num_tokens": 395438463.0, + "step": 11263 + }, + { + "epoch": 2.091736304549675, + "grad_norm": 1.5594955682754517, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8716615438461304, + "num_tokens": 395475990.0, + "step": 11264 + }, + { + "epoch": 2.0919220055710306, + "grad_norm": 1.6849322319030762, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8899285793304443, + "num_tokens": 395503376.0, + "step": 11265 + }, + { + "epoch": 2.0921077065923863, + "grad_norm": 1.6068662405014038, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8936108350753784, + "num_tokens": 395534385.0, + "step": 11266 + }, + { + "epoch": 2.092293407613742, + "grad_norm": 1.428237795829773, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8864609599113464, + "num_tokens": 395577850.0, + "step": 11267 + }, + { + "epoch": 2.0924791086350973, + "grad_norm": 1.7188422679901123, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.893182098865509, + "num_tokens": 395604429.0, + "step": 11268 + }, + { + "epoch": 2.092664809656453, + "grad_norm": 1.579450249671936, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8858456611633301, + "num_tokens": 395638351.0, + "step": 11269 + }, + { + "epoch": 2.092850510677809, + "grad_norm": 1.6641535758972168, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8811559677124023, + "num_tokens": 395673934.0, + "step": 11270 + }, + { + "epoch": 2.0930362116991645, + "grad_norm": 1.5203512907028198, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8911409378051758, + "num_tokens": 395711768.0, + "step": 11271 + }, + { + "epoch": 2.09322191272052, + "grad_norm": 1.3980355262756348, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.8954201340675354, + "num_tokens": 395751544.0, + "step": 11272 + }, + { + "epoch": 2.0934076137418756, + "grad_norm": 1.5919983386993408, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8805797696113586, + "num_tokens": 395786153.0, + "step": 11273 + }, + { + "epoch": 2.0935933147632313, + "grad_norm": 1.586613655090332, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.88946533203125, + "num_tokens": 395821839.0, + "step": 11274 + }, + { + "epoch": 2.093779015784587, + "grad_norm": 1.474643349647522, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8867906332015991, + "num_tokens": 395860723.0, + "step": 11275 + }, + { + "epoch": 2.0939647168059423, + "grad_norm": 1.8492637872695923, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8858778476715088, + "num_tokens": 395887615.0, + "step": 11276 + }, + { + "epoch": 2.094150417827298, + "grad_norm": 1.7208203077316284, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8701372146606445, + "num_tokens": 395918905.0, + "step": 11277 + }, + { + "epoch": 2.094336118848654, + "grad_norm": 1.5336412191390991, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.873406708240509, + "num_tokens": 395957408.0, + "step": 11278 + }, + { + "epoch": 2.094521819870009, + "grad_norm": 1.734397053718567, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8883373737335205, + "num_tokens": 395985014.0, + "step": 11279 + }, + { + "epoch": 2.094707520891365, + "grad_norm": 1.5168856382369995, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8869913220405579, + "num_tokens": 396022384.0, + "step": 11280 + }, + { + "epoch": 2.0948932219127205, + "grad_norm": 1.59219229221344, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8829588294029236, + "num_tokens": 396053953.0, + "step": 11281 + }, + { + "epoch": 2.0950789229340763, + "grad_norm": 2.035576820373535, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.869721531867981, + "num_tokens": 396089846.0, + "step": 11282 + }, + { + "epoch": 2.0952646239554316, + "grad_norm": 1.534534215927124, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8835210204124451, + "num_tokens": 396124806.0, + "step": 11283 + }, + { + "epoch": 2.0954503249767873, + "grad_norm": 1.605177402496338, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8856015801429749, + "num_tokens": 396158792.0, + "step": 11284 + }, + { + "epoch": 2.095636025998143, + "grad_norm": 1.5851157903671265, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8892019987106323, + "num_tokens": 396191193.0, + "step": 11285 + }, + { + "epoch": 2.0958217270194988, + "grad_norm": 1.6022529602050781, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8902715444564819, + "num_tokens": 396222789.0, + "step": 11286 + }, + { + "epoch": 2.096007428040854, + "grad_norm": 1.4384130239486694, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.8985382318496704, + "num_tokens": 396259724.0, + "step": 11287 + }, + { + "epoch": 2.09619312906221, + "grad_norm": 1.4979984760284424, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.886594831943512, + "num_tokens": 396300439.0, + "step": 11288 + }, + { + "epoch": 2.0963788300835655, + "grad_norm": 1.5809465646743774, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8743512630462646, + "num_tokens": 396335744.0, + "step": 11289 + }, + { + "epoch": 2.0965645311049212, + "grad_norm": 1.5259146690368652, + "learning_rate": 1e-06, + "loss": 0.2696, + "mean_token_accuracy": 0.9043622612953186, + "num_tokens": 396368258.0, + "step": 11290 + }, + { + "epoch": 2.0967502321262765, + "grad_norm": 1.8238131999969482, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8803098797798157, + "num_tokens": 396401580.0, + "step": 11291 + }, + { + "epoch": 2.0969359331476323, + "grad_norm": 1.7433841228485107, + "learning_rate": 1e-06, + "loss": 0.3999, + "mean_token_accuracy": 0.8614715337753296, + "num_tokens": 396435462.0, + "step": 11292 + }, + { + "epoch": 2.097121634168988, + "grad_norm": 1.5127619504928589, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8908988237380981, + "num_tokens": 396470852.0, + "step": 11293 + }, + { + "epoch": 2.0973073351903437, + "grad_norm": 1.572949767112732, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8888967037200928, + "num_tokens": 396504656.0, + "step": 11294 + }, + { + "epoch": 2.097493036211699, + "grad_norm": 1.4386727809906006, + "learning_rate": 1e-06, + "loss": 0.2684, + "mean_token_accuracy": 0.9052520990371704, + "num_tokens": 396541145.0, + "step": 11295 + }, + { + "epoch": 2.0976787372330548, + "grad_norm": 1.4946436882019043, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.897274911403656, + "num_tokens": 396577151.0, + "step": 11296 + }, + { + "epoch": 2.0978644382544105, + "grad_norm": 1.7788567543029785, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.879472017288208, + "num_tokens": 396607818.0, + "step": 11297 + }, + { + "epoch": 2.0980501392757662, + "grad_norm": 1.8579939603805542, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8808356523513794, + "num_tokens": 396638390.0, + "step": 11298 + }, + { + "epoch": 2.0982358402971215, + "grad_norm": 1.800583839416504, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8856458067893982, + "num_tokens": 396669014.0, + "step": 11299 + }, + { + "epoch": 2.0984215413184772, + "grad_norm": 1.6923061609268188, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8848966956138611, + "num_tokens": 396703760.0, + "step": 11300 + }, + { + "epoch": 2.098607242339833, + "grad_norm": 1.5754258632659912, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8971080780029297, + "num_tokens": 396737123.0, + "step": 11301 + }, + { + "epoch": 2.0987929433611887, + "grad_norm": 1.6851590871810913, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8837864398956299, + "num_tokens": 396765879.0, + "step": 11302 + }, + { + "epoch": 2.098978644382544, + "grad_norm": 1.5226784944534302, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8888648748397827, + "num_tokens": 396802617.0, + "step": 11303 + }, + { + "epoch": 2.0991643454038997, + "grad_norm": 1.7461105585098267, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8765861988067627, + "num_tokens": 396835178.0, + "step": 11304 + }, + { + "epoch": 2.0993500464252555, + "grad_norm": 1.6202372312545776, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.882774829864502, + "num_tokens": 396869404.0, + "step": 11305 + }, + { + "epoch": 2.0995357474466108, + "grad_norm": 1.7878873348236084, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8843942880630493, + "num_tokens": 396899425.0, + "step": 11306 + }, + { + "epoch": 2.0997214484679665, + "grad_norm": 1.5647844076156616, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.897007405757904, + "num_tokens": 396932207.0, + "step": 11307 + }, + { + "epoch": 2.0999071494893222, + "grad_norm": 1.4284214973449707, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8870531320571899, + "num_tokens": 396974915.0, + "step": 11308 + }, + { + "epoch": 2.100092850510678, + "grad_norm": 1.5227974653244019, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8790677785873413, + "num_tokens": 397012817.0, + "step": 11309 + }, + { + "epoch": 2.1002785515320332, + "grad_norm": 1.6224684715270996, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8822501301765442, + "num_tokens": 397046085.0, + "step": 11310 + }, + { + "epoch": 2.100464252553389, + "grad_norm": 1.5574743747711182, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8886303901672363, + "num_tokens": 397079655.0, + "step": 11311 + }, + { + "epoch": 2.1006499535747447, + "grad_norm": 1.499914526939392, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8774375915527344, + "num_tokens": 397119785.0, + "step": 11312 + }, + { + "epoch": 2.1008356545961004, + "grad_norm": 1.460023283958435, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8866455554962158, + "num_tokens": 397160980.0, + "step": 11313 + }, + { + "epoch": 2.1010213556174557, + "grad_norm": 1.626390814781189, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8926424980163574, + "num_tokens": 397190851.0, + "step": 11314 + }, + { + "epoch": 2.1012070566388115, + "grad_norm": 1.504406452178955, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.892114520072937, + "num_tokens": 397227042.0, + "step": 11315 + }, + { + "epoch": 2.101392757660167, + "grad_norm": 1.6404125690460205, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8975889682769775, + "num_tokens": 397256820.0, + "step": 11316 + }, + { + "epoch": 2.101578458681523, + "grad_norm": 1.6034446954727173, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8940951824188232, + "num_tokens": 397288964.0, + "step": 11317 + }, + { + "epoch": 2.101764159702878, + "grad_norm": 1.551830530166626, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8732166886329651, + "num_tokens": 397326185.0, + "step": 11318 + }, + { + "epoch": 2.101949860724234, + "grad_norm": 1.5303903818130493, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.882079005241394, + "num_tokens": 397363805.0, + "step": 11319 + }, + { + "epoch": 2.1021355617455897, + "grad_norm": 1.4816360473632812, + "learning_rate": 1e-06, + "loss": 0.297, + "mean_token_accuracy": 0.8912530541419983, + "num_tokens": 397400619.0, + "step": 11320 + }, + { + "epoch": 2.1023212627669454, + "grad_norm": 1.5706911087036133, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.886897623538971, + "num_tokens": 397435808.0, + "step": 11321 + }, + { + "epoch": 2.1025069637883007, + "grad_norm": 1.5291740894317627, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8810490965843201, + "num_tokens": 397474501.0, + "step": 11322 + }, + { + "epoch": 2.1026926648096564, + "grad_norm": 1.6731715202331543, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8830710053443909, + "num_tokens": 397504119.0, + "step": 11323 + }, + { + "epoch": 2.102878365831012, + "grad_norm": 1.5574675798416138, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8880795240402222, + "num_tokens": 397542969.0, + "step": 11324 + }, + { + "epoch": 2.103064066852368, + "grad_norm": 1.6440836191177368, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8735738396644592, + "num_tokens": 397576573.0, + "step": 11325 + }, + { + "epoch": 2.103249767873723, + "grad_norm": 1.526102066040039, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8895636796951294, + "num_tokens": 397611789.0, + "step": 11326 + }, + { + "epoch": 2.103435468895079, + "grad_norm": 1.531511902809143, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8895071744918823, + "num_tokens": 397648044.0, + "step": 11327 + }, + { + "epoch": 2.1036211699164347, + "grad_norm": 1.5850012302398682, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8906254768371582, + "num_tokens": 397680132.0, + "step": 11328 + }, + { + "epoch": 2.10380687093779, + "grad_norm": 1.6018515825271606, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8835273385047913, + "num_tokens": 397713514.0, + "step": 11329 + }, + { + "epoch": 2.1039925719591457, + "grad_norm": 1.6079659461975098, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.89200758934021, + "num_tokens": 397743579.0, + "step": 11330 + }, + { + "epoch": 2.1041782729805014, + "grad_norm": 1.3855717182159424, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.8962728381156921, + "num_tokens": 397782431.0, + "step": 11331 + }, + { + "epoch": 2.104363974001857, + "grad_norm": 1.5755094289779663, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8728618621826172, + "num_tokens": 397820837.0, + "step": 11332 + }, + { + "epoch": 2.1045496750232124, + "grad_norm": 1.6149734258651733, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8961127400398254, + "num_tokens": 397849145.0, + "step": 11333 + }, + { + "epoch": 2.104735376044568, + "grad_norm": 1.5092084407806396, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8879891633987427, + "num_tokens": 397886517.0, + "step": 11334 + }, + { + "epoch": 2.104921077065924, + "grad_norm": 1.521681547164917, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8906129002571106, + "num_tokens": 397920697.0, + "step": 11335 + }, + { + "epoch": 2.1051067780872796, + "grad_norm": 1.4963734149932861, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8702259659767151, + "num_tokens": 397961886.0, + "step": 11336 + }, + { + "epoch": 2.105292479108635, + "grad_norm": 1.5564032793045044, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.883682906627655, + "num_tokens": 397998550.0, + "step": 11337 + }, + { + "epoch": 2.1054781801299907, + "grad_norm": 1.5579874515533447, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8920360803604126, + "num_tokens": 398029491.0, + "step": 11338 + }, + { + "epoch": 2.1056638811513464, + "grad_norm": 1.6025817394256592, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8846921920776367, + "num_tokens": 398067202.0, + "step": 11339 + }, + { + "epoch": 2.105849582172702, + "grad_norm": 1.5715357065200806, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8973767161369324, + "num_tokens": 398099613.0, + "step": 11340 + }, + { + "epoch": 2.1060352831940574, + "grad_norm": 1.6057977676391602, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8920843601226807, + "num_tokens": 398132439.0, + "step": 11341 + }, + { + "epoch": 2.106220984215413, + "grad_norm": 1.5180939435958862, + "learning_rate": 1e-06, + "loss": 0.2717, + "mean_token_accuracy": 0.9034346342086792, + "num_tokens": 398165205.0, + "step": 11342 + }, + { + "epoch": 2.106406685236769, + "grad_norm": 1.6690196990966797, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8946347832679749, + "num_tokens": 398196538.0, + "step": 11343 + }, + { + "epoch": 2.1065923862581246, + "grad_norm": 1.6694049835205078, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8839311599731445, + "num_tokens": 398229112.0, + "step": 11344 + }, + { + "epoch": 2.10677808727948, + "grad_norm": 1.5584794282913208, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8985906839370728, + "num_tokens": 398259745.0, + "step": 11345 + }, + { + "epoch": 2.1069637883008356, + "grad_norm": 1.395076036453247, + "learning_rate": 1e-06, + "loss": 0.2792, + "mean_token_accuracy": 0.9014107584953308, + "num_tokens": 398299364.0, + "step": 11346 + }, + { + "epoch": 2.1071494893221914, + "grad_norm": 1.612066626548767, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8895997405052185, + "num_tokens": 398333739.0, + "step": 11347 + }, + { + "epoch": 2.107335190343547, + "grad_norm": 1.4854656457901, + "learning_rate": 1e-06, + "loss": 0.2781, + "mean_token_accuracy": 0.8969573974609375, + "num_tokens": 398370013.0, + "step": 11348 + }, + { + "epoch": 2.1075208913649024, + "grad_norm": 1.609588384628296, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8845235109329224, + "num_tokens": 398403718.0, + "step": 11349 + }, + { + "epoch": 2.107706592386258, + "grad_norm": 1.6096749305725098, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8869615793228149, + "num_tokens": 398434514.0, + "step": 11350 + }, + { + "epoch": 2.107892293407614, + "grad_norm": 1.692578673362732, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8883160352706909, + "num_tokens": 398466532.0, + "step": 11351 + }, + { + "epoch": 2.108077994428969, + "grad_norm": 1.6196742057800293, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8814854621887207, + "num_tokens": 398502373.0, + "step": 11352 + }, + { + "epoch": 2.108263695450325, + "grad_norm": 1.6535282135009766, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8787661194801331, + "num_tokens": 398534742.0, + "step": 11353 + }, + { + "epoch": 2.1084493964716806, + "grad_norm": 1.5247308015823364, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8906573057174683, + "num_tokens": 398569777.0, + "step": 11354 + }, + { + "epoch": 2.1086350974930363, + "grad_norm": 1.5620516538619995, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8817808032035828, + "num_tokens": 398607684.0, + "step": 11355 + }, + { + "epoch": 2.1088207985143916, + "grad_norm": 1.6205315589904785, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8833705186843872, + "num_tokens": 398640858.0, + "step": 11356 + }, + { + "epoch": 2.1090064995357474, + "grad_norm": 1.5190426111221313, + "learning_rate": 1e-06, + "loss": 0.3016, + "mean_token_accuracy": 0.8935337662696838, + "num_tokens": 398675732.0, + "step": 11357 + }, + { + "epoch": 2.109192200557103, + "grad_norm": 1.5682768821716309, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8854325413703918, + "num_tokens": 398708247.0, + "step": 11358 + }, + { + "epoch": 2.109377901578459, + "grad_norm": 1.502038598060608, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8892303109169006, + "num_tokens": 398745577.0, + "step": 11359 + }, + { + "epoch": 2.109563602599814, + "grad_norm": 1.5208759307861328, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8877682685852051, + "num_tokens": 398781936.0, + "step": 11360 + }, + { + "epoch": 2.10974930362117, + "grad_norm": 1.7053780555725098, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8816969394683838, + "num_tokens": 398813334.0, + "step": 11361 + }, + { + "epoch": 2.1099350046425256, + "grad_norm": 1.7102956771850586, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8838635683059692, + "num_tokens": 398845332.0, + "step": 11362 + }, + { + "epoch": 2.1101207056638813, + "grad_norm": 1.6367624998092651, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8885406255722046, + "num_tokens": 398875841.0, + "step": 11363 + }, + { + "epoch": 2.1103064066852366, + "grad_norm": 1.5205390453338623, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8961008787155151, + "num_tokens": 398908716.0, + "step": 11364 + }, + { + "epoch": 2.1104921077065923, + "grad_norm": 1.5684484243392944, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8875259160995483, + "num_tokens": 398943534.0, + "step": 11365 + }, + { + "epoch": 2.110677808727948, + "grad_norm": 1.5409703254699707, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8818219900131226, + "num_tokens": 398978981.0, + "step": 11366 + }, + { + "epoch": 2.110863509749304, + "grad_norm": 1.492472529411316, + "learning_rate": 1e-06, + "loss": 0.2546, + "mean_token_accuracy": 0.9070299863815308, + "num_tokens": 399011705.0, + "step": 11367 + }, + { + "epoch": 2.111049210770659, + "grad_norm": 1.403721809387207, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8971868753433228, + "num_tokens": 399051064.0, + "step": 11368 + }, + { + "epoch": 2.111234911792015, + "grad_norm": 1.5749566555023193, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8835918307304382, + "num_tokens": 399084586.0, + "step": 11369 + }, + { + "epoch": 2.1114206128133706, + "grad_norm": 1.417171835899353, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8823829293251038, + "num_tokens": 399126073.0, + "step": 11370 + }, + { + "epoch": 2.1116063138347263, + "grad_norm": 1.5587362051010132, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8874582648277283, + "num_tokens": 399160974.0, + "step": 11371 + }, + { + "epoch": 2.1117920148560816, + "grad_norm": 1.5357977151870728, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8853440284729004, + "num_tokens": 399197775.0, + "step": 11372 + }, + { + "epoch": 2.1119777158774373, + "grad_norm": 1.6934881210327148, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8810245990753174, + "num_tokens": 399230286.0, + "step": 11373 + }, + { + "epoch": 2.112163416898793, + "grad_norm": 1.608860969543457, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8723998069763184, + "num_tokens": 399265121.0, + "step": 11374 + }, + { + "epoch": 2.1123491179201483, + "grad_norm": 1.514072299003601, + "learning_rate": 1e-06, + "loss": 0.2977, + "mean_token_accuracy": 0.8935233950614929, + "num_tokens": 399304080.0, + "step": 11375 + }, + { + "epoch": 2.112534818941504, + "grad_norm": 1.4846526384353638, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8905764222145081, + "num_tokens": 399344025.0, + "step": 11376 + }, + { + "epoch": 2.11272051996286, + "grad_norm": 1.7334942817687988, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8864704966545105, + "num_tokens": 399371680.0, + "step": 11377 + }, + { + "epoch": 2.1129062209842155, + "grad_norm": 1.6451669931411743, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8837499618530273, + "num_tokens": 399405552.0, + "step": 11378 + }, + { + "epoch": 2.113091922005571, + "grad_norm": 1.598963737487793, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8869258165359497, + "num_tokens": 399439142.0, + "step": 11379 + }, + { + "epoch": 2.1132776230269266, + "grad_norm": 1.4723747968673706, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8752010464668274, + "num_tokens": 399482345.0, + "step": 11380 + }, + { + "epoch": 2.1134633240482823, + "grad_norm": 1.5438357591629028, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8774468898773193, + "num_tokens": 399521745.0, + "step": 11381 + }, + { + "epoch": 2.113649025069638, + "grad_norm": 1.5581742525100708, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8915148973464966, + "num_tokens": 399555044.0, + "step": 11382 + }, + { + "epoch": 2.1138347260909933, + "grad_norm": 1.607313871383667, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8953067064285278, + "num_tokens": 399583856.0, + "step": 11383 + }, + { + "epoch": 2.114020427112349, + "grad_norm": 1.6079213619232178, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8752741813659668, + "num_tokens": 399618227.0, + "step": 11384 + }, + { + "epoch": 2.114206128133705, + "grad_norm": 1.6423488855361938, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8814955949783325, + "num_tokens": 399653488.0, + "step": 11385 + }, + { + "epoch": 2.1143918291550605, + "grad_norm": 1.4754968881607056, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.890161395072937, + "num_tokens": 399691497.0, + "step": 11386 + }, + { + "epoch": 2.114577530176416, + "grad_norm": 1.4143253564834595, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.894819974899292, + "num_tokens": 399729177.0, + "step": 11387 + }, + { + "epoch": 2.1147632311977715, + "grad_norm": 1.4978221654891968, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8797639012336731, + "num_tokens": 399768574.0, + "step": 11388 + }, + { + "epoch": 2.1149489322191273, + "grad_norm": 1.6710760593414307, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8750337362289429, + "num_tokens": 399803693.0, + "step": 11389 + }, + { + "epoch": 2.115134633240483, + "grad_norm": 1.531178593635559, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8997514247894287, + "num_tokens": 399836209.0, + "step": 11390 + }, + { + "epoch": 2.1153203342618383, + "grad_norm": 1.8063602447509766, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.868035614490509, + "num_tokens": 399870030.0, + "step": 11391 + }, + { + "epoch": 2.115506035283194, + "grad_norm": 1.6493916511535645, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8948684930801392, + "num_tokens": 399898341.0, + "step": 11392 + }, + { + "epoch": 2.1156917363045498, + "grad_norm": 1.6254814863204956, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.885650634765625, + "num_tokens": 399931682.0, + "step": 11393 + }, + { + "epoch": 2.1158774373259055, + "grad_norm": 1.4382357597351074, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8906508088111877, + "num_tokens": 399970379.0, + "step": 11394 + }, + { + "epoch": 2.116063138347261, + "grad_norm": 1.3905351161956787, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.8944076299667358, + "num_tokens": 400009724.0, + "step": 11395 + }, + { + "epoch": 2.1162488393686165, + "grad_norm": 1.4904589653015137, + "learning_rate": 1e-06, + "loss": 0.2676, + "mean_token_accuracy": 0.9059464931488037, + "num_tokens": 400042640.0, + "step": 11396 + }, + { + "epoch": 2.1164345403899723, + "grad_norm": 1.584601879119873, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8814589381217957, + "num_tokens": 400077048.0, + "step": 11397 + }, + { + "epoch": 2.1166202414113275, + "grad_norm": 1.5080726146697998, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8940696716308594, + "num_tokens": 400109319.0, + "step": 11398 + }, + { + "epoch": 2.1168059424326833, + "grad_norm": 1.5525983572006226, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8793687224388123, + "num_tokens": 400145739.0, + "step": 11399 + }, + { + "epoch": 2.116991643454039, + "grad_norm": 1.4702354669570923, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8953187465667725, + "num_tokens": 400180647.0, + "step": 11400 + }, + { + "epoch": 2.1171773444753947, + "grad_norm": 1.4885146617889404, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8863836526870728, + "num_tokens": 400219491.0, + "step": 11401 + }, + { + "epoch": 2.11736304549675, + "grad_norm": 1.497029423713684, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8952478170394897, + "num_tokens": 400252723.0, + "step": 11402 + }, + { + "epoch": 2.1175487465181058, + "grad_norm": 1.4920140504837036, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8906808495521545, + "num_tokens": 400291466.0, + "step": 11403 + }, + { + "epoch": 2.1177344475394615, + "grad_norm": 1.6191052198410034, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.86916184425354, + "num_tokens": 400328350.0, + "step": 11404 + }, + { + "epoch": 2.1179201485608172, + "grad_norm": 1.61723792552948, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8889727592468262, + "num_tokens": 400360884.0, + "step": 11405 + }, + { + "epoch": 2.1181058495821725, + "grad_norm": 1.7468657493591309, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8783137202262878, + "num_tokens": 400389385.0, + "step": 11406 + }, + { + "epoch": 2.1182915506035283, + "grad_norm": 1.5450525283813477, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8852002024650574, + "num_tokens": 400423460.0, + "step": 11407 + }, + { + "epoch": 2.118477251624884, + "grad_norm": 1.3938356637954712, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.896687388420105, + "num_tokens": 400463580.0, + "step": 11408 + }, + { + "epoch": 2.1186629526462397, + "grad_norm": 1.4427896738052368, + "learning_rate": 1e-06, + "loss": 0.2821, + "mean_token_accuracy": 0.9002437591552734, + "num_tokens": 400498294.0, + "step": 11409 + }, + { + "epoch": 2.118848653667595, + "grad_norm": 1.547082543373108, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8866257071495056, + "num_tokens": 400534890.0, + "step": 11410 + }, + { + "epoch": 2.1190343546889507, + "grad_norm": 1.9259483814239502, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8730149865150452, + "num_tokens": 400562675.0, + "step": 11411 + }, + { + "epoch": 2.1192200557103065, + "grad_norm": 1.533572793006897, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8860338926315308, + "num_tokens": 400598871.0, + "step": 11412 + }, + { + "epoch": 2.119405756731662, + "grad_norm": 1.4925669431686401, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8869180679321289, + "num_tokens": 400637083.0, + "step": 11413 + }, + { + "epoch": 2.1195914577530175, + "grad_norm": 1.6522181034088135, + "learning_rate": 1e-06, + "loss": 0.3513, + "mean_token_accuracy": 0.8776283264160156, + "num_tokens": 400670616.0, + "step": 11414 + }, + { + "epoch": 2.1197771587743732, + "grad_norm": 1.6001653671264648, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.89708411693573, + "num_tokens": 400701383.0, + "step": 11415 + }, + { + "epoch": 2.119962859795729, + "grad_norm": 1.6756218671798706, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8632490634918213, + "num_tokens": 400735078.0, + "step": 11416 + }, + { + "epoch": 2.1201485608170847, + "grad_norm": 1.4759466648101807, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8812578916549683, + "num_tokens": 400772042.0, + "step": 11417 + }, + { + "epoch": 2.12033426183844, + "grad_norm": 1.7105324268341064, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8794857263565063, + "num_tokens": 400801764.0, + "step": 11418 + }, + { + "epoch": 2.1205199628597957, + "grad_norm": 1.6139034032821655, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8872445821762085, + "num_tokens": 400834364.0, + "step": 11419 + }, + { + "epoch": 2.1207056638811514, + "grad_norm": 1.4914666414260864, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.8701586723327637, + "num_tokens": 400877818.0, + "step": 11420 + }, + { + "epoch": 2.1208913649025067, + "grad_norm": 1.448911190032959, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8941917419433594, + "num_tokens": 400916412.0, + "step": 11421 + }, + { + "epoch": 2.1210770659238625, + "grad_norm": 1.6626874208450317, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8861154317855835, + "num_tokens": 400949266.0, + "step": 11422 + }, + { + "epoch": 2.121262766945218, + "grad_norm": 1.5857430696487427, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8853299617767334, + "num_tokens": 400981962.0, + "step": 11423 + }, + { + "epoch": 2.121448467966574, + "grad_norm": 1.7534860372543335, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8873268365859985, + "num_tokens": 401009088.0, + "step": 11424 + }, + { + "epoch": 2.1216341689879292, + "grad_norm": 1.4982892274856567, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.880622386932373, + "num_tokens": 401049096.0, + "step": 11425 + }, + { + "epoch": 2.121819870009285, + "grad_norm": 1.608364462852478, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8887060880661011, + "num_tokens": 401084838.0, + "step": 11426 + }, + { + "epoch": 2.1220055710306407, + "grad_norm": 1.6713722944259644, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8758256435394287, + "num_tokens": 401120603.0, + "step": 11427 + }, + { + "epoch": 2.1221912720519964, + "grad_norm": 1.4964263439178467, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8955546021461487, + "num_tokens": 401159483.0, + "step": 11428 + }, + { + "epoch": 2.1223769730733517, + "grad_norm": 1.6079695224761963, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8765080571174622, + "num_tokens": 401196157.0, + "step": 11429 + }, + { + "epoch": 2.1225626740947074, + "grad_norm": 1.6741498708724976, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8783544301986694, + "num_tokens": 401227776.0, + "step": 11430 + }, + { + "epoch": 2.122748375116063, + "grad_norm": 1.4809304475784302, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8848814368247986, + "num_tokens": 401266252.0, + "step": 11431 + }, + { + "epoch": 2.122934076137419, + "grad_norm": 1.5280190706253052, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8796553611755371, + "num_tokens": 401304106.0, + "step": 11432 + }, + { + "epoch": 2.123119777158774, + "grad_norm": 1.6136417388916016, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8767604231834412, + "num_tokens": 401340743.0, + "step": 11433 + }, + { + "epoch": 2.12330547818013, + "grad_norm": 1.5172709226608276, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.879550576210022, + "num_tokens": 401379091.0, + "step": 11434 + }, + { + "epoch": 2.1234911792014857, + "grad_norm": 1.4662593603134155, + "learning_rate": 1e-06, + "loss": 0.2742, + "mean_token_accuracy": 0.9024382829666138, + "num_tokens": 401415240.0, + "step": 11435 + }, + { + "epoch": 2.1236768802228414, + "grad_norm": 1.6883463859558105, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.8921236991882324, + "num_tokens": 401443468.0, + "step": 11436 + }, + { + "epoch": 2.1238625812441967, + "grad_norm": 1.6956058740615845, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.875059187412262, + "num_tokens": 401475151.0, + "step": 11437 + }, + { + "epoch": 2.1240482822655524, + "grad_norm": 1.4620401859283447, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8873237371444702, + "num_tokens": 401512491.0, + "step": 11438 + }, + { + "epoch": 2.124233983286908, + "grad_norm": 1.53303062915802, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.891867995262146, + "num_tokens": 401548595.0, + "step": 11439 + }, + { + "epoch": 2.124419684308264, + "grad_norm": 1.673980951309204, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8725289106369019, + "num_tokens": 401582659.0, + "step": 11440 + }, + { + "epoch": 2.124605385329619, + "grad_norm": 1.625731348991394, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8836286067962646, + "num_tokens": 401616023.0, + "step": 11441 + }, + { + "epoch": 2.124791086350975, + "grad_norm": 1.5906370878219604, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8679600954055786, + "num_tokens": 401652862.0, + "step": 11442 + }, + { + "epoch": 2.1249767873723306, + "grad_norm": 1.4940704107284546, + "learning_rate": 1e-06, + "loss": 0.2943, + "mean_token_accuracy": 0.894632875919342, + "num_tokens": 401689219.0, + "step": 11443 + }, + { + "epoch": 2.125162488393686, + "grad_norm": 1.7151857614517212, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8914373517036438, + "num_tokens": 401720316.0, + "step": 11444 + }, + { + "epoch": 2.1253481894150417, + "grad_norm": 1.5651588439941406, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8903403282165527, + "num_tokens": 401753403.0, + "step": 11445 + }, + { + "epoch": 2.1255338904363974, + "grad_norm": 1.7056635618209839, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8791252374649048, + "num_tokens": 401785242.0, + "step": 11446 + }, + { + "epoch": 2.125719591457753, + "grad_norm": 1.608744502067566, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8897081613540649, + "num_tokens": 401818093.0, + "step": 11447 + }, + { + "epoch": 2.125905292479109, + "grad_norm": 1.5662394762039185, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8889740705490112, + "num_tokens": 401855424.0, + "step": 11448 + }, + { + "epoch": 2.126090993500464, + "grad_norm": 1.748504400253296, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8931092023849487, + "num_tokens": 401887279.0, + "step": 11449 + }, + { + "epoch": 2.12627669452182, + "grad_norm": 1.6095575094223022, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8859006762504578, + "num_tokens": 401923663.0, + "step": 11450 + }, + { + "epoch": 2.1264623955431756, + "grad_norm": 1.5924456119537354, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8926673531532288, + "num_tokens": 401953488.0, + "step": 11451 + }, + { + "epoch": 2.126648096564531, + "grad_norm": 1.5545003414154053, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.889257550239563, + "num_tokens": 401987707.0, + "step": 11452 + }, + { + "epoch": 2.1268337975858866, + "grad_norm": 1.7290246486663818, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8719912767410278, + "num_tokens": 402023663.0, + "step": 11453 + }, + { + "epoch": 2.1270194986072424, + "grad_norm": 1.7582613229751587, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8883440494537354, + "num_tokens": 402056059.0, + "step": 11454 + }, + { + "epoch": 2.127205199628598, + "grad_norm": 1.7173528671264648, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8793891072273254, + "num_tokens": 402086276.0, + "step": 11455 + }, + { + "epoch": 2.1273909006499534, + "grad_norm": 1.5668162107467651, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.870996356010437, + "num_tokens": 402125278.0, + "step": 11456 + }, + { + "epoch": 2.127576601671309, + "grad_norm": 1.5346674919128418, + "learning_rate": 1e-06, + "loss": 0.276, + "mean_token_accuracy": 0.9009170532226562, + "num_tokens": 402157416.0, + "step": 11457 + }, + { + "epoch": 2.127762302692665, + "grad_norm": 1.4969685077667236, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8841953277587891, + "num_tokens": 402195600.0, + "step": 11458 + }, + { + "epoch": 2.1279480037140206, + "grad_norm": 1.519521951675415, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.895391583442688, + "num_tokens": 402231760.0, + "step": 11459 + }, + { + "epoch": 2.128133704735376, + "grad_norm": 1.578476071357727, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.885673463344574, + "num_tokens": 402262317.0, + "step": 11460 + }, + { + "epoch": 2.1283194057567316, + "grad_norm": 1.5378762483596802, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8871315717697144, + "num_tokens": 402302339.0, + "step": 11461 + }, + { + "epoch": 2.1285051067780874, + "grad_norm": 1.720800518989563, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8893651962280273, + "num_tokens": 402331534.0, + "step": 11462 + }, + { + "epoch": 2.128690807799443, + "grad_norm": 1.508121371269226, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8781605958938599, + "num_tokens": 402376695.0, + "step": 11463 + }, + { + "epoch": 2.1288765088207984, + "grad_norm": 1.5652931928634644, + "learning_rate": 1e-06, + "loss": 0.3643, + "mean_token_accuracy": 0.8735548257827759, + "num_tokens": 402413221.0, + "step": 11464 + }, + { + "epoch": 2.129062209842154, + "grad_norm": 1.6242454051971436, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8976906538009644, + "num_tokens": 402442861.0, + "step": 11465 + }, + { + "epoch": 2.12924791086351, + "grad_norm": 1.403394103050232, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8893333077430725, + "num_tokens": 402480766.0, + "step": 11466 + }, + { + "epoch": 2.1294336118848656, + "grad_norm": 1.6369000673294067, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8861200213432312, + "num_tokens": 402513148.0, + "step": 11467 + }, + { + "epoch": 2.129619312906221, + "grad_norm": 1.5578763484954834, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8894150257110596, + "num_tokens": 402546167.0, + "step": 11468 + }, + { + "epoch": 2.1298050139275766, + "grad_norm": 1.4741066694259644, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8865278959274292, + "num_tokens": 402583345.0, + "step": 11469 + }, + { + "epoch": 2.1299907149489323, + "grad_norm": 1.4684951305389404, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8865483999252319, + "num_tokens": 402620364.0, + "step": 11470 + }, + { + "epoch": 2.130176415970288, + "grad_norm": 1.5270273685455322, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8888370990753174, + "num_tokens": 402656391.0, + "step": 11471 + }, + { + "epoch": 2.1303621169916434, + "grad_norm": 1.5528090000152588, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.8965755701065063, + "num_tokens": 402688086.0, + "step": 11472 + }, + { + "epoch": 2.130547818012999, + "grad_norm": 1.5518759489059448, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8875312805175781, + "num_tokens": 402724374.0, + "step": 11473 + }, + { + "epoch": 2.130733519034355, + "grad_norm": 1.7444101572036743, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.877292275428772, + "num_tokens": 402752944.0, + "step": 11474 + }, + { + "epoch": 2.13091922005571, + "grad_norm": 1.5191330909729004, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8934739828109741, + "num_tokens": 402786534.0, + "step": 11475 + }, + { + "epoch": 2.131104921077066, + "grad_norm": 1.4801589250564575, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8851553797721863, + "num_tokens": 402826163.0, + "step": 11476 + }, + { + "epoch": 2.1312906220984216, + "grad_norm": 1.5591248273849487, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8849096894264221, + "num_tokens": 402861573.0, + "step": 11477 + }, + { + "epoch": 2.1314763231197773, + "grad_norm": 1.6770533323287964, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8767346143722534, + "num_tokens": 402893525.0, + "step": 11478 + }, + { + "epoch": 2.1316620241411326, + "grad_norm": 1.5415011644363403, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8805158734321594, + "num_tokens": 402929099.0, + "step": 11479 + }, + { + "epoch": 2.1318477251624883, + "grad_norm": 1.523959994316101, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8789310455322266, + "num_tokens": 402966070.0, + "step": 11480 + }, + { + "epoch": 2.132033426183844, + "grad_norm": 1.4243884086608887, + "learning_rate": 1e-06, + "loss": 0.372, + "mean_token_accuracy": 0.8705254197120667, + "num_tokens": 403008555.0, + "step": 11481 + }, + { + "epoch": 2.1322191272052, + "grad_norm": 1.338997483253479, + "learning_rate": 1e-06, + "loss": 0.2804, + "mean_token_accuracy": 0.8968218564987183, + "num_tokens": 403053452.0, + "step": 11482 + }, + { + "epoch": 2.132404828226555, + "grad_norm": 1.6612493991851807, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8865740299224854, + "num_tokens": 403083681.0, + "step": 11483 + }, + { + "epoch": 2.132590529247911, + "grad_norm": 1.444332480430603, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.899519145488739, + "num_tokens": 403119542.0, + "step": 11484 + }, + { + "epoch": 2.1327762302692665, + "grad_norm": 1.576462745666504, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8822629451751709, + "num_tokens": 403154835.0, + "step": 11485 + }, + { + "epoch": 2.1329619312906223, + "grad_norm": 1.7498633861541748, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8888617753982544, + "num_tokens": 403182020.0, + "step": 11486 + }, + { + "epoch": 2.1331476323119776, + "grad_norm": 1.6037629842758179, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8852245807647705, + "num_tokens": 403215721.0, + "step": 11487 + }, + { + "epoch": 2.1333333333333333, + "grad_norm": 1.834386944770813, + "learning_rate": 1e-06, + "loss": 0.2718, + "mean_token_accuracy": 0.8985857963562012, + "num_tokens": 403239611.0, + "step": 11488 + }, + { + "epoch": 2.133519034354689, + "grad_norm": 1.427713394165039, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8893709182739258, + "num_tokens": 403282329.0, + "step": 11489 + }, + { + "epoch": 2.1337047353760448, + "grad_norm": 1.4783539772033691, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8814613223075867, + "num_tokens": 403325532.0, + "step": 11490 + }, + { + "epoch": 2.1338904363974, + "grad_norm": 1.5155174732208252, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8784034252166748, + "num_tokens": 403371067.0, + "step": 11491 + }, + { + "epoch": 2.134076137418756, + "grad_norm": 1.7106842994689941, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8730179071426392, + "num_tokens": 403407859.0, + "step": 11492 + }, + { + "epoch": 2.1342618384401115, + "grad_norm": 1.5580239295959473, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8957637548446655, + "num_tokens": 403445756.0, + "step": 11493 + }, + { + "epoch": 2.1344475394614673, + "grad_norm": 1.6830579042434692, + "learning_rate": 1e-06, + "loss": 0.4225, + "mean_token_accuracy": 0.863155722618103, + "num_tokens": 403482750.0, + "step": 11494 + }, + { + "epoch": 2.1346332404828225, + "grad_norm": 1.622788667678833, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8976195454597473, + "num_tokens": 403514892.0, + "step": 11495 + }, + { + "epoch": 2.1348189415041783, + "grad_norm": 1.692848563194275, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8802478313446045, + "num_tokens": 403545734.0, + "step": 11496 + }, + { + "epoch": 2.135004642525534, + "grad_norm": 1.6550953388214111, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8876948356628418, + "num_tokens": 403579160.0, + "step": 11497 + }, + { + "epoch": 2.1351903435468893, + "grad_norm": 1.435133457183838, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8928115367889404, + "num_tokens": 403614620.0, + "step": 11498 + }, + { + "epoch": 2.135376044568245, + "grad_norm": 1.5271480083465576, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8934789896011353, + "num_tokens": 403650886.0, + "step": 11499 + }, + { + "epoch": 2.1355617455896008, + "grad_norm": 1.4970797300338745, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8944562673568726, + "num_tokens": 403685959.0, + "step": 11500 + }, + { + "epoch": 2.1357474466109565, + "grad_norm": 1.6080076694488525, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.880338191986084, + "num_tokens": 403720380.0, + "step": 11501 + }, + { + "epoch": 2.135933147632312, + "grad_norm": 1.4205601215362549, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8913842439651489, + "num_tokens": 403757874.0, + "step": 11502 + }, + { + "epoch": 2.1361188486536675, + "grad_norm": 1.5043972730636597, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8773484230041504, + "num_tokens": 403796697.0, + "step": 11503 + }, + { + "epoch": 2.1363045496750233, + "grad_norm": 1.5604469776153564, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8865138292312622, + "num_tokens": 403831412.0, + "step": 11504 + }, + { + "epoch": 2.136490250696379, + "grad_norm": 1.4135265350341797, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8970497250556946, + "num_tokens": 403869457.0, + "step": 11505 + }, + { + "epoch": 2.1366759517177343, + "grad_norm": 1.6957190036773682, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8743023872375488, + "num_tokens": 403902557.0, + "step": 11506 + }, + { + "epoch": 2.13686165273909, + "grad_norm": 1.5400794744491577, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8836553692817688, + "num_tokens": 403936948.0, + "step": 11507 + }, + { + "epoch": 2.1370473537604457, + "grad_norm": 1.4897874593734741, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8823035359382629, + "num_tokens": 403973701.0, + "step": 11508 + }, + { + "epoch": 2.1372330547818015, + "grad_norm": 1.5898752212524414, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8910253643989563, + "num_tokens": 404005396.0, + "step": 11509 + }, + { + "epoch": 2.1374187558031568, + "grad_norm": 1.6024101972579956, + "learning_rate": 1e-06, + "loss": 0.2872, + "mean_token_accuracy": 0.8979726433753967, + "num_tokens": 404037020.0, + "step": 11510 + }, + { + "epoch": 2.1376044568245125, + "grad_norm": 1.5735472440719604, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8789150714874268, + "num_tokens": 404073982.0, + "step": 11511 + }, + { + "epoch": 2.1377901578458682, + "grad_norm": 1.6083791255950928, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8889220952987671, + "num_tokens": 404108655.0, + "step": 11512 + }, + { + "epoch": 2.137975858867224, + "grad_norm": 1.677087426185608, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.876331090927124, + "num_tokens": 404141447.0, + "step": 11513 + }, + { + "epoch": 2.1381615598885793, + "grad_norm": 1.5306901931762695, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8838145732879639, + "num_tokens": 404178257.0, + "step": 11514 + }, + { + "epoch": 2.138347260909935, + "grad_norm": 1.5109089612960815, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8918834924697876, + "num_tokens": 404214917.0, + "step": 11515 + }, + { + "epoch": 2.1385329619312907, + "grad_norm": 1.4541075229644775, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8802947998046875, + "num_tokens": 404258826.0, + "step": 11516 + }, + { + "epoch": 2.1387186629526465, + "grad_norm": 1.5437383651733398, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8960001468658447, + "num_tokens": 404292853.0, + "step": 11517 + }, + { + "epoch": 2.1389043639740017, + "grad_norm": 1.4107427597045898, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8958232998847961, + "num_tokens": 404335514.0, + "step": 11518 + }, + { + "epoch": 2.1390900649953575, + "grad_norm": 1.5730382204055786, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8810204863548279, + "num_tokens": 404372425.0, + "step": 11519 + }, + { + "epoch": 2.139275766016713, + "grad_norm": 1.5961405038833618, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8909561634063721, + "num_tokens": 404404715.0, + "step": 11520 + }, + { + "epoch": 2.1394614670380685, + "grad_norm": 1.7556023597717285, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8873131275177002, + "num_tokens": 404435829.0, + "step": 11521 + }, + { + "epoch": 2.1396471680594242, + "grad_norm": 1.5680897235870361, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8728514909744263, + "num_tokens": 404477199.0, + "step": 11522 + }, + { + "epoch": 2.13983286908078, + "grad_norm": 1.525882601737976, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8834412693977356, + "num_tokens": 404510782.0, + "step": 11523 + }, + { + "epoch": 2.1400185701021357, + "grad_norm": 1.4935660362243652, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8864798545837402, + "num_tokens": 404547676.0, + "step": 11524 + }, + { + "epoch": 2.140204271123491, + "grad_norm": 1.447927713394165, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.88338303565979, + "num_tokens": 404589104.0, + "step": 11525 + }, + { + "epoch": 2.1403899721448467, + "grad_norm": 1.5779294967651367, + "learning_rate": 1e-06, + "loss": 0.3934, + "mean_token_accuracy": 0.8621282577514648, + "num_tokens": 404628160.0, + "step": 11526 + }, + { + "epoch": 2.1405756731662025, + "grad_norm": 1.5562043190002441, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8833456635475159, + "num_tokens": 404667917.0, + "step": 11527 + }, + { + "epoch": 2.140761374187558, + "grad_norm": 1.4516679048538208, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.8987275958061218, + "num_tokens": 404705768.0, + "step": 11528 + }, + { + "epoch": 2.1409470752089135, + "grad_norm": 1.5530539751052856, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8766289353370667, + "num_tokens": 404742426.0, + "step": 11529 + }, + { + "epoch": 2.141132776230269, + "grad_norm": 1.572993516921997, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8864878416061401, + "num_tokens": 404777902.0, + "step": 11530 + }, + { + "epoch": 2.141318477251625, + "grad_norm": 1.5225389003753662, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8830059766769409, + "num_tokens": 404816895.0, + "step": 11531 + }, + { + "epoch": 2.1415041782729807, + "grad_norm": 1.6101975440979004, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.8974642157554626, + "num_tokens": 404850939.0, + "step": 11532 + }, + { + "epoch": 2.141689879294336, + "grad_norm": 1.478319764137268, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8955293297767639, + "num_tokens": 404888977.0, + "step": 11533 + }, + { + "epoch": 2.1418755803156917, + "grad_norm": 1.4791909456253052, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8771829605102539, + "num_tokens": 404929655.0, + "step": 11534 + }, + { + "epoch": 2.1420612813370474, + "grad_norm": 1.4471652507781982, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8932772278785706, + "num_tokens": 404971254.0, + "step": 11535 + }, + { + "epoch": 2.142246982358403, + "grad_norm": 1.5141457319259644, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8893566131591797, + "num_tokens": 405006951.0, + "step": 11536 + }, + { + "epoch": 2.1424326833797585, + "grad_norm": 1.6613472700119019, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8734337091445923, + "num_tokens": 405040528.0, + "step": 11537 + }, + { + "epoch": 2.142618384401114, + "grad_norm": 1.4858571290969849, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8892996311187744, + "num_tokens": 405080598.0, + "step": 11538 + }, + { + "epoch": 2.14280408542247, + "grad_norm": 1.4981863498687744, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8959360718727112, + "num_tokens": 405114775.0, + "step": 11539 + }, + { + "epoch": 2.1429897864438257, + "grad_norm": 1.5466848611831665, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.8993593454360962, + "num_tokens": 405146059.0, + "step": 11540 + }, + { + "epoch": 2.143175487465181, + "grad_norm": 1.762873649597168, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8740597367286682, + "num_tokens": 405175509.0, + "step": 11541 + }, + { + "epoch": 2.1433611884865367, + "grad_norm": 1.5230262279510498, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8882861733436584, + "num_tokens": 405213550.0, + "step": 11542 + }, + { + "epoch": 2.1435468895078924, + "grad_norm": 1.5037668943405151, + "learning_rate": 1e-06, + "loss": 0.2991, + "mean_token_accuracy": 0.8928499221801758, + "num_tokens": 405245978.0, + "step": 11543 + }, + { + "epoch": 2.1437325905292477, + "grad_norm": 1.4184626340866089, + "learning_rate": 1e-06, + "loss": 0.2715, + "mean_token_accuracy": 0.9014711380004883, + "num_tokens": 405283552.0, + "step": 11544 + }, + { + "epoch": 2.1439182915506034, + "grad_norm": 1.4922711849212646, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8867543935775757, + "num_tokens": 405322176.0, + "step": 11545 + }, + { + "epoch": 2.144103992571959, + "grad_norm": 1.7523435354232788, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8776825666427612, + "num_tokens": 405352841.0, + "step": 11546 + }, + { + "epoch": 2.144289693593315, + "grad_norm": 1.4898751974105835, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8804124593734741, + "num_tokens": 405394892.0, + "step": 11547 + }, + { + "epoch": 2.14447539461467, + "grad_norm": 1.6178240776062012, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8840458393096924, + "num_tokens": 405429318.0, + "step": 11548 + }, + { + "epoch": 2.144661095636026, + "grad_norm": 1.4979978799819946, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8854336738586426, + "num_tokens": 405469387.0, + "step": 11549 + }, + { + "epoch": 2.1448467966573816, + "grad_norm": 2.4009461402893066, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8755379319190979, + "num_tokens": 405500806.0, + "step": 11550 + }, + { + "epoch": 2.1450324976787374, + "grad_norm": 1.5747690200805664, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8868380784988403, + "num_tokens": 405536397.0, + "step": 11551 + }, + { + "epoch": 2.1452181987000927, + "grad_norm": 1.7861440181732178, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8988227248191833, + "num_tokens": 405563620.0, + "step": 11552 + }, + { + "epoch": 2.1454038997214484, + "grad_norm": 1.53328537940979, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8757140040397644, + "num_tokens": 405608515.0, + "step": 11553 + }, + { + "epoch": 2.145589600742804, + "grad_norm": 1.5766621828079224, + "learning_rate": 1e-06, + "loss": 0.2851, + "mean_token_accuracy": 0.9020653367042542, + "num_tokens": 405640537.0, + "step": 11554 + }, + { + "epoch": 2.14577530176416, + "grad_norm": 1.791305422782898, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8807616829872131, + "num_tokens": 405666103.0, + "step": 11555 + }, + { + "epoch": 2.145961002785515, + "grad_norm": 1.5407865047454834, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8801319599151611, + "num_tokens": 405704155.0, + "step": 11556 + }, + { + "epoch": 2.146146703806871, + "grad_norm": 1.6037715673446655, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.890061616897583, + "num_tokens": 405736470.0, + "step": 11557 + }, + { + "epoch": 2.1463324048282266, + "grad_norm": 1.5390347242355347, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8793107271194458, + "num_tokens": 405776287.0, + "step": 11558 + }, + { + "epoch": 2.1465181058495824, + "grad_norm": 1.7199844121932983, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.8938936591148376, + "num_tokens": 405805206.0, + "step": 11559 + }, + { + "epoch": 2.1467038068709376, + "grad_norm": 1.5971938371658325, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8854881525039673, + "num_tokens": 405840315.0, + "step": 11560 + }, + { + "epoch": 2.1468895078922934, + "grad_norm": 1.571854829788208, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8818167448043823, + "num_tokens": 405874142.0, + "step": 11561 + }, + { + "epoch": 2.147075208913649, + "grad_norm": 1.6695752143859863, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8747652769088745, + "num_tokens": 405912869.0, + "step": 11562 + }, + { + "epoch": 2.147260909935005, + "grad_norm": 1.812671422958374, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8830422163009644, + "num_tokens": 405940597.0, + "step": 11563 + }, + { + "epoch": 2.14744661095636, + "grad_norm": 1.6346325874328613, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8926533460617065, + "num_tokens": 405974230.0, + "step": 11564 + }, + { + "epoch": 2.147632311977716, + "grad_norm": 1.8251978158950806, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8804192543029785, + "num_tokens": 406009362.0, + "step": 11565 + }, + { + "epoch": 2.1478180129990716, + "grad_norm": 1.421811819076538, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8820846080780029, + "num_tokens": 406050482.0, + "step": 11566 + }, + { + "epoch": 2.148003714020427, + "grad_norm": 1.456943154335022, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.90087890625, + "num_tokens": 406085692.0, + "step": 11567 + }, + { + "epoch": 2.1481894150417826, + "grad_norm": 1.7061101198196411, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8864567279815674, + "num_tokens": 406115754.0, + "step": 11568 + }, + { + "epoch": 2.1483751160631384, + "grad_norm": 1.6856350898742676, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8812256455421448, + "num_tokens": 406148231.0, + "step": 11569 + }, + { + "epoch": 2.148560817084494, + "grad_norm": 1.6608502864837646, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8840563893318176, + "num_tokens": 406183212.0, + "step": 11570 + }, + { + "epoch": 2.1487465181058494, + "grad_norm": 1.4433544874191284, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8886967897415161, + "num_tokens": 406224210.0, + "step": 11571 + }, + { + "epoch": 2.148932219127205, + "grad_norm": 1.4472236633300781, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8833203911781311, + "num_tokens": 406268912.0, + "step": 11572 + }, + { + "epoch": 2.149117920148561, + "grad_norm": 1.450109601020813, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8867253065109253, + "num_tokens": 406307253.0, + "step": 11573 + }, + { + "epoch": 2.1493036211699166, + "grad_norm": 1.5955621004104614, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8656007051467896, + "num_tokens": 406348680.0, + "step": 11574 + }, + { + "epoch": 2.149489322191272, + "grad_norm": 1.6338993310928345, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.894332766532898, + "num_tokens": 406381725.0, + "step": 11575 + }, + { + "epoch": 2.1496750232126276, + "grad_norm": 1.6150052547454834, + "learning_rate": 1e-06, + "loss": 0.2884, + "mean_token_accuracy": 0.895525336265564, + "num_tokens": 406413365.0, + "step": 11576 + }, + { + "epoch": 2.1498607242339833, + "grad_norm": 1.6463203430175781, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8798825740814209, + "num_tokens": 406445369.0, + "step": 11577 + }, + { + "epoch": 2.150046425255339, + "grad_norm": 1.5086179971694946, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8898208141326904, + "num_tokens": 406479577.0, + "step": 11578 + }, + { + "epoch": 2.1502321262766944, + "grad_norm": 1.4478086233139038, + "learning_rate": 1e-06, + "loss": 0.2833, + "mean_token_accuracy": 0.8969022631645203, + "num_tokens": 406516703.0, + "step": 11579 + }, + { + "epoch": 2.15041782729805, + "grad_norm": 1.6653465032577515, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8832129240036011, + "num_tokens": 406552819.0, + "step": 11580 + }, + { + "epoch": 2.150603528319406, + "grad_norm": 1.729027509689331, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.885204553604126, + "num_tokens": 406584890.0, + "step": 11581 + }, + { + "epoch": 2.1507892293407616, + "grad_norm": 1.638271689414978, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8802539110183716, + "num_tokens": 406619929.0, + "step": 11582 + }, + { + "epoch": 2.150974930362117, + "grad_norm": 1.3616975545883179, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8910938501358032, + "num_tokens": 406665264.0, + "step": 11583 + }, + { + "epoch": 2.1511606313834726, + "grad_norm": 1.444247841835022, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8922346830368042, + "num_tokens": 406703789.0, + "step": 11584 + }, + { + "epoch": 2.1513463324048283, + "grad_norm": 1.7108800411224365, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8806350231170654, + "num_tokens": 406734889.0, + "step": 11585 + }, + { + "epoch": 2.151532033426184, + "grad_norm": 1.6746350526809692, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8884613513946533, + "num_tokens": 406764169.0, + "step": 11586 + }, + { + "epoch": 2.1517177344475393, + "grad_norm": 1.6442301273345947, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8754987716674805, + "num_tokens": 406801571.0, + "step": 11587 + }, + { + "epoch": 2.151903435468895, + "grad_norm": 1.4595694541931152, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8871763944625854, + "num_tokens": 406836391.0, + "step": 11588 + }, + { + "epoch": 2.152089136490251, + "grad_norm": 1.5302937030792236, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8767237067222595, + "num_tokens": 406873508.0, + "step": 11589 + }, + { + "epoch": 2.152274837511606, + "grad_norm": 1.6321755647659302, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8811460733413696, + "num_tokens": 406908326.0, + "step": 11590 + }, + { + "epoch": 2.152460538532962, + "grad_norm": 1.5257247686386108, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8830262422561646, + "num_tokens": 406945660.0, + "step": 11591 + }, + { + "epoch": 2.1526462395543176, + "grad_norm": 1.7113368511199951, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8900858163833618, + "num_tokens": 406976914.0, + "step": 11592 + }, + { + "epoch": 2.1528319405756733, + "grad_norm": 1.609649658203125, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8887393474578857, + "num_tokens": 407009513.0, + "step": 11593 + }, + { + "epoch": 2.153017641597029, + "grad_norm": 1.3820158243179321, + "learning_rate": 1e-06, + "loss": 0.274, + "mean_token_accuracy": 0.9015480875968933, + "num_tokens": 407052129.0, + "step": 11594 + }, + { + "epoch": 2.1532033426183843, + "grad_norm": 1.5836659669876099, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.882702112197876, + "num_tokens": 407091889.0, + "step": 11595 + }, + { + "epoch": 2.15338904363974, + "grad_norm": 1.6493724584579468, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.88768470287323, + "num_tokens": 407126332.0, + "step": 11596 + }, + { + "epoch": 2.1535747446610958, + "grad_norm": 1.5400606393814087, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8853135704994202, + "num_tokens": 407162194.0, + "step": 11597 + }, + { + "epoch": 2.153760445682451, + "grad_norm": 1.5048060417175293, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8964942693710327, + "num_tokens": 407197029.0, + "step": 11598 + }, + { + "epoch": 2.153946146703807, + "grad_norm": 1.5354464054107666, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8895694613456726, + "num_tokens": 407230817.0, + "step": 11599 + }, + { + "epoch": 2.1541318477251625, + "grad_norm": 1.6463563442230225, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8927643895149231, + "num_tokens": 407263621.0, + "step": 11600 + }, + { + "epoch": 2.1543175487465183, + "grad_norm": 1.4905109405517578, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8840328454971313, + "num_tokens": 407303467.0, + "step": 11601 + }, + { + "epoch": 2.1545032497678736, + "grad_norm": 1.6463472843170166, + "learning_rate": 1e-06, + "loss": 0.2792, + "mean_token_accuracy": 0.8999178409576416, + "num_tokens": 407334357.0, + "step": 11602 + }, + { + "epoch": 2.1546889507892293, + "grad_norm": 1.6315914392471313, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.863315224647522, + "num_tokens": 407369140.0, + "step": 11603 + }, + { + "epoch": 2.154874651810585, + "grad_norm": 1.665803074836731, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8877771496772766, + "num_tokens": 407399967.0, + "step": 11604 + }, + { + "epoch": 2.1550603528319408, + "grad_norm": 1.47859525680542, + "learning_rate": 1e-06, + "loss": 0.2726, + "mean_token_accuracy": 0.9013911485671997, + "num_tokens": 407435530.0, + "step": 11605 + }, + { + "epoch": 2.155246053853296, + "grad_norm": 1.7554243803024292, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8819684982299805, + "num_tokens": 407468406.0, + "step": 11606 + }, + { + "epoch": 2.1554317548746518, + "grad_norm": 1.6020203828811646, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8853515982627869, + "num_tokens": 407504300.0, + "step": 11607 + }, + { + "epoch": 2.1556174558960075, + "grad_norm": 1.5943398475646973, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8868978023529053, + "num_tokens": 407541230.0, + "step": 11608 + }, + { + "epoch": 2.1558031569173632, + "grad_norm": 1.6127151250839233, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8853546380996704, + "num_tokens": 407574072.0, + "step": 11609 + }, + { + "epoch": 2.1559888579387185, + "grad_norm": 1.568747878074646, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8960989713668823, + "num_tokens": 407605357.0, + "step": 11610 + }, + { + "epoch": 2.1561745589600743, + "grad_norm": 1.488361120223999, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8858069181442261, + "num_tokens": 407646695.0, + "step": 11611 + }, + { + "epoch": 2.15636025998143, + "grad_norm": 1.4943054914474487, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8913481831550598, + "num_tokens": 407681381.0, + "step": 11612 + }, + { + "epoch": 2.1565459610027853, + "grad_norm": 1.5419116020202637, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8778548240661621, + "num_tokens": 407722952.0, + "step": 11613 + }, + { + "epoch": 2.156731662024141, + "grad_norm": 1.5168769359588623, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8902449011802673, + "num_tokens": 407758569.0, + "step": 11614 + }, + { + "epoch": 2.1569173630454967, + "grad_norm": 1.6027748584747314, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.866883397102356, + "num_tokens": 407799939.0, + "step": 11615 + }, + { + "epoch": 2.1571030640668525, + "grad_norm": 1.5770115852355957, + "learning_rate": 1e-06, + "loss": 0.2825, + "mean_token_accuracy": 0.9029112458229065, + "num_tokens": 407827473.0, + "step": 11616 + }, + { + "epoch": 2.157288765088208, + "grad_norm": 1.6295287609100342, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8790119290351868, + "num_tokens": 407859617.0, + "step": 11617 + }, + { + "epoch": 2.1574744661095635, + "grad_norm": 1.4799787998199463, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8934775590896606, + "num_tokens": 407897246.0, + "step": 11618 + }, + { + "epoch": 2.1576601671309192, + "grad_norm": 1.5602418184280396, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8838156461715698, + "num_tokens": 407931108.0, + "step": 11619 + }, + { + "epoch": 2.157845868152275, + "grad_norm": 1.495413064956665, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8858096599578857, + "num_tokens": 407968527.0, + "step": 11620 + }, + { + "epoch": 2.1580315691736303, + "grad_norm": 1.4627130031585693, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8802552819252014, + "num_tokens": 408013319.0, + "step": 11621 + }, + { + "epoch": 2.158217270194986, + "grad_norm": 1.646574854850769, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8979480862617493, + "num_tokens": 408045421.0, + "step": 11622 + }, + { + "epoch": 2.1584029712163417, + "grad_norm": 1.499848484992981, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8901890516281128, + "num_tokens": 408082223.0, + "step": 11623 + }, + { + "epoch": 2.1585886722376975, + "grad_norm": 1.4359275102615356, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8753457069396973, + "num_tokens": 408124667.0, + "step": 11624 + }, + { + "epoch": 2.1587743732590527, + "grad_norm": 1.5689969062805176, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8930246829986572, + "num_tokens": 408156057.0, + "step": 11625 + }, + { + "epoch": 2.1589600742804085, + "grad_norm": 1.6744822263717651, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8796373605728149, + "num_tokens": 408186579.0, + "step": 11626 + }, + { + "epoch": 2.159145775301764, + "grad_norm": 1.4414126873016357, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8779877424240112, + "num_tokens": 408226643.0, + "step": 11627 + }, + { + "epoch": 2.15933147632312, + "grad_norm": 1.4945422410964966, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8838261365890503, + "num_tokens": 408264926.0, + "step": 11628 + }, + { + "epoch": 2.1595171773444752, + "grad_norm": 1.440385341644287, + "learning_rate": 1e-06, + "loss": 0.2905, + "mean_token_accuracy": 0.8958112597465515, + "num_tokens": 408301031.0, + "step": 11629 + }, + { + "epoch": 2.159702878365831, + "grad_norm": 1.6623094081878662, + "learning_rate": 1e-06, + "loss": 0.3768, + "mean_token_accuracy": 0.8676543235778809, + "num_tokens": 408334284.0, + "step": 11630 + }, + { + "epoch": 2.1598885793871867, + "grad_norm": 1.5593510866165161, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8852065801620483, + "num_tokens": 408370681.0, + "step": 11631 + }, + { + "epoch": 2.1600742804085424, + "grad_norm": 1.5615673065185547, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8807721138000488, + "num_tokens": 408405256.0, + "step": 11632 + }, + { + "epoch": 2.1602599814298977, + "grad_norm": 1.5317327976226807, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8789293169975281, + "num_tokens": 408440790.0, + "step": 11633 + }, + { + "epoch": 2.1604456824512535, + "grad_norm": 1.4952971935272217, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.887809157371521, + "num_tokens": 408476800.0, + "step": 11634 + }, + { + "epoch": 2.160631383472609, + "grad_norm": 1.5887160301208496, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8787083625793457, + "num_tokens": 408510753.0, + "step": 11635 + }, + { + "epoch": 2.160817084493965, + "grad_norm": 1.5899314880371094, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.883212685585022, + "num_tokens": 408544397.0, + "step": 11636 + }, + { + "epoch": 2.16100278551532, + "grad_norm": 1.5466684103012085, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8836503028869629, + "num_tokens": 408583698.0, + "step": 11637 + }, + { + "epoch": 2.161188486536676, + "grad_norm": 1.4735068082809448, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8862711191177368, + "num_tokens": 408621783.0, + "step": 11638 + }, + { + "epoch": 2.1613741875580317, + "grad_norm": 1.6449888944625854, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8822844624519348, + "num_tokens": 408655363.0, + "step": 11639 + }, + { + "epoch": 2.1615598885793874, + "grad_norm": 1.6593984365463257, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.891826331615448, + "num_tokens": 408685662.0, + "step": 11640 + }, + { + "epoch": 2.1617455896007427, + "grad_norm": 1.5021859407424927, + "learning_rate": 1e-06, + "loss": 0.2789, + "mean_token_accuracy": 0.9005974531173706, + "num_tokens": 408722075.0, + "step": 11641 + }, + { + "epoch": 2.1619312906220984, + "grad_norm": 1.456436038017273, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8883386254310608, + "num_tokens": 408759656.0, + "step": 11642 + }, + { + "epoch": 2.162116991643454, + "grad_norm": 1.4658501148223877, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8952992558479309, + "num_tokens": 408797565.0, + "step": 11643 + }, + { + "epoch": 2.1623026926648095, + "grad_norm": 1.7007930278778076, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.8969849348068237, + "num_tokens": 408825644.0, + "step": 11644 + }, + { + "epoch": 2.162488393686165, + "grad_norm": 1.4615788459777832, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8828118443489075, + "num_tokens": 408864639.0, + "step": 11645 + }, + { + "epoch": 2.162674094707521, + "grad_norm": 1.745973825454712, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8878630995750427, + "num_tokens": 408893892.0, + "step": 11646 + }, + { + "epoch": 2.1628597957288767, + "grad_norm": 1.5966672897338867, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8761283755302429, + "num_tokens": 408930180.0, + "step": 11647 + }, + { + "epoch": 2.163045496750232, + "grad_norm": 1.6302392482757568, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8935744762420654, + "num_tokens": 408960228.0, + "step": 11648 + }, + { + "epoch": 2.1632311977715877, + "grad_norm": 1.464194655418396, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8977328538894653, + "num_tokens": 408993317.0, + "step": 11649 + }, + { + "epoch": 2.1634168987929434, + "grad_norm": 1.4967914819717407, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.8945773839950562, + "num_tokens": 409030356.0, + "step": 11650 + }, + { + "epoch": 2.163602599814299, + "grad_norm": 1.5978425741195679, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8769967555999756, + "num_tokens": 409067356.0, + "step": 11651 + }, + { + "epoch": 2.1637883008356544, + "grad_norm": 1.8104780912399292, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8755888342857361, + "num_tokens": 409101677.0, + "step": 11652 + }, + { + "epoch": 2.16397400185701, + "grad_norm": 1.598272681236267, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8815253376960754, + "num_tokens": 409134145.0, + "step": 11653 + }, + { + "epoch": 2.164159702878366, + "grad_norm": 1.5802621841430664, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8904759287834167, + "num_tokens": 409171210.0, + "step": 11654 + }, + { + "epoch": 2.1643454038997216, + "grad_norm": 1.486371636390686, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.8994836807250977, + "num_tokens": 409203917.0, + "step": 11655 + }, + { + "epoch": 2.164531104921077, + "grad_norm": 1.505541443824768, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8722561597824097, + "num_tokens": 409242276.0, + "step": 11656 + }, + { + "epoch": 2.1647168059424327, + "grad_norm": 1.6340314149856567, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8899263143539429, + "num_tokens": 409274613.0, + "step": 11657 + }, + { + "epoch": 2.1649025069637884, + "grad_norm": 1.6554044485092163, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.878568172454834, + "num_tokens": 409305625.0, + "step": 11658 + }, + { + "epoch": 2.165088207985144, + "grad_norm": 1.6483685970306396, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.881993293762207, + "num_tokens": 409337607.0, + "step": 11659 + }, + { + "epoch": 2.1652739090064994, + "grad_norm": 1.6530795097351074, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8942040801048279, + "num_tokens": 409370267.0, + "step": 11660 + }, + { + "epoch": 2.165459610027855, + "grad_norm": 1.6723947525024414, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.8705447316169739, + "num_tokens": 409408428.0, + "step": 11661 + }, + { + "epoch": 2.165645311049211, + "grad_norm": 1.5056254863739014, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.886976957321167, + "num_tokens": 409447878.0, + "step": 11662 + }, + { + "epoch": 2.1658310120705666, + "grad_norm": 1.6209865808486938, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8738787174224854, + "num_tokens": 409484537.0, + "step": 11663 + }, + { + "epoch": 2.166016713091922, + "grad_norm": 1.5287611484527588, + "learning_rate": 1e-06, + "loss": 0.2821, + "mean_token_accuracy": 0.9027060866355896, + "num_tokens": 409519190.0, + "step": 11664 + }, + { + "epoch": 2.1662024141132776, + "grad_norm": 1.4818062782287598, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8847687244415283, + "num_tokens": 409559449.0, + "step": 11665 + }, + { + "epoch": 2.1663881151346334, + "grad_norm": 1.487119436264038, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.877644956111908, + "num_tokens": 409599943.0, + "step": 11666 + }, + { + "epoch": 2.1665738161559887, + "grad_norm": 1.6458181142807007, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8821074962615967, + "num_tokens": 409633009.0, + "step": 11667 + }, + { + "epoch": 2.1667595171773444, + "grad_norm": 1.5027539730072021, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.9001107215881348, + "num_tokens": 409669617.0, + "step": 11668 + }, + { + "epoch": 2.1669452181987, + "grad_norm": 1.3839040994644165, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8824145793914795, + "num_tokens": 409717317.0, + "step": 11669 + }, + { + "epoch": 2.167130919220056, + "grad_norm": 1.4764282703399658, + "learning_rate": 1e-06, + "loss": 0.2846, + "mean_token_accuracy": 0.8975122570991516, + "num_tokens": 409751803.0, + "step": 11670 + }, + { + "epoch": 2.167316620241411, + "grad_norm": 1.7518806457519531, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8848410248756409, + "num_tokens": 409780067.0, + "step": 11671 + }, + { + "epoch": 2.167502321262767, + "grad_norm": 1.5028586387634277, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8904049396514893, + "num_tokens": 409814533.0, + "step": 11672 + }, + { + "epoch": 2.1676880222841226, + "grad_norm": 1.4756592512130737, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8878974318504333, + "num_tokens": 409851581.0, + "step": 11673 + }, + { + "epoch": 2.1678737233054783, + "grad_norm": 1.8417332172393799, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8794543743133545, + "num_tokens": 409882029.0, + "step": 11674 + }, + { + "epoch": 2.1680594243268336, + "grad_norm": 1.8061432838439941, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8815829157829285, + "num_tokens": 409917716.0, + "step": 11675 + }, + { + "epoch": 2.1682451253481894, + "grad_norm": 1.593385934829712, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8790706396102905, + "num_tokens": 409950593.0, + "step": 11676 + }, + { + "epoch": 2.168430826369545, + "grad_norm": 1.5611845254898071, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.896312415599823, + "num_tokens": 409984931.0, + "step": 11677 + }, + { + "epoch": 2.168616527390901, + "grad_norm": 1.3757240772247314, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.9025350213050842, + "num_tokens": 410025634.0, + "step": 11678 + }, + { + "epoch": 2.168802228412256, + "grad_norm": 1.6409016847610474, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8916782140731812, + "num_tokens": 410056805.0, + "step": 11679 + }, + { + "epoch": 2.168987929433612, + "grad_norm": 1.5954769849777222, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8923337459564209, + "num_tokens": 410088204.0, + "step": 11680 + }, + { + "epoch": 2.1691736304549676, + "grad_norm": 1.5309405326843262, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8795191645622253, + "num_tokens": 410124571.0, + "step": 11681 + }, + { + "epoch": 2.1693593314763233, + "grad_norm": 1.656284213066101, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8841191530227661, + "num_tokens": 410157713.0, + "step": 11682 + }, + { + "epoch": 2.1695450324976786, + "grad_norm": 1.5846368074417114, + "learning_rate": 1e-06, + "loss": 0.2846, + "mean_token_accuracy": 0.898695707321167, + "num_tokens": 410189642.0, + "step": 11683 + }, + { + "epoch": 2.1697307335190343, + "grad_norm": 1.744879961013794, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8864798545837402, + "num_tokens": 410217019.0, + "step": 11684 + }, + { + "epoch": 2.16991643454039, + "grad_norm": 1.5053802728652954, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8872469663619995, + "num_tokens": 410255765.0, + "step": 11685 + }, + { + "epoch": 2.170102135561746, + "grad_norm": 1.5816303491592407, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8882099390029907, + "num_tokens": 410287489.0, + "step": 11686 + }, + { + "epoch": 2.170287836583101, + "grad_norm": 1.6196582317352295, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8930100798606873, + "num_tokens": 410319514.0, + "step": 11687 + }, + { + "epoch": 2.170473537604457, + "grad_norm": 1.6121032238006592, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8852811455726624, + "num_tokens": 410351711.0, + "step": 11688 + }, + { + "epoch": 2.1706592386258126, + "grad_norm": 1.5669231414794922, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8857855200767517, + "num_tokens": 410386332.0, + "step": 11689 + }, + { + "epoch": 2.170844939647168, + "grad_norm": 1.7267624139785767, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8891382217407227, + "num_tokens": 410415895.0, + "step": 11690 + }, + { + "epoch": 2.1710306406685236, + "grad_norm": 1.5465375185012817, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8884884119033813, + "num_tokens": 410453331.0, + "step": 11691 + }, + { + "epoch": 2.1712163416898793, + "grad_norm": 1.5819671154022217, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8894314765930176, + "num_tokens": 410486419.0, + "step": 11692 + }, + { + "epoch": 2.171402042711235, + "grad_norm": 1.6018762588500977, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8803820013999939, + "num_tokens": 410524927.0, + "step": 11693 + }, + { + "epoch": 2.1715877437325903, + "grad_norm": 1.529118299484253, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8826946020126343, + "num_tokens": 410564361.0, + "step": 11694 + }, + { + "epoch": 2.171773444753946, + "grad_norm": 1.6257953643798828, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8782437443733215, + "num_tokens": 410596162.0, + "step": 11695 + }, + { + "epoch": 2.171959145775302, + "grad_norm": 1.6117734909057617, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8811750411987305, + "num_tokens": 410633503.0, + "step": 11696 + }, + { + "epoch": 2.1721448467966575, + "grad_norm": 1.4913402795791626, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.884260356426239, + "num_tokens": 410672266.0, + "step": 11697 + }, + { + "epoch": 2.172330547818013, + "grad_norm": 1.494619607925415, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8930172920227051, + "num_tokens": 410704239.0, + "step": 11698 + }, + { + "epoch": 2.1725162488393686, + "grad_norm": 1.5900813341140747, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8903933763504028, + "num_tokens": 410735424.0, + "step": 11699 + }, + { + "epoch": 2.1727019498607243, + "grad_norm": 1.5034924745559692, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.894783616065979, + "num_tokens": 410774025.0, + "step": 11700 + }, + { + "epoch": 2.17288765088208, + "grad_norm": 1.6882528066635132, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8777244687080383, + "num_tokens": 410802198.0, + "step": 11701 + }, + { + "epoch": 2.1730733519034353, + "grad_norm": 1.5604571104049683, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8829991817474365, + "num_tokens": 410835903.0, + "step": 11702 + }, + { + "epoch": 2.173259052924791, + "grad_norm": 1.695218801498413, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8865505456924438, + "num_tokens": 410867212.0, + "step": 11703 + }, + { + "epoch": 2.173444753946147, + "grad_norm": 1.454055905342102, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8676126003265381, + "num_tokens": 410909966.0, + "step": 11704 + }, + { + "epoch": 2.1736304549675025, + "grad_norm": 1.7146021127700806, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8801932334899902, + "num_tokens": 410940123.0, + "step": 11705 + }, + { + "epoch": 2.173816155988858, + "grad_norm": 1.5441185235977173, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.883583128452301, + "num_tokens": 410975628.0, + "step": 11706 + }, + { + "epoch": 2.1740018570102135, + "grad_norm": 1.6549397706985474, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8768054246902466, + "num_tokens": 411007949.0, + "step": 11707 + }, + { + "epoch": 2.1741875580315693, + "grad_norm": 1.4433610439300537, + "learning_rate": 1e-06, + "loss": 0.2722, + "mean_token_accuracy": 0.9045844078063965, + "num_tokens": 411046947.0, + "step": 11708 + }, + { + "epoch": 2.174373259052925, + "grad_norm": 1.4792741537094116, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8852076530456543, + "num_tokens": 411089928.0, + "step": 11709 + }, + { + "epoch": 2.1745589600742803, + "grad_norm": 1.6653519868850708, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.892237663269043, + "num_tokens": 411121549.0, + "step": 11710 + }, + { + "epoch": 2.174744661095636, + "grad_norm": 1.6952844858169556, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8907985687255859, + "num_tokens": 411152154.0, + "step": 11711 + }, + { + "epoch": 2.1749303621169918, + "grad_norm": 1.7017967700958252, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8727881908416748, + "num_tokens": 411184669.0, + "step": 11712 + }, + { + "epoch": 2.175116063138347, + "grad_norm": 1.7302379608154297, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8821641802787781, + "num_tokens": 411215415.0, + "step": 11713 + }, + { + "epoch": 2.1753017641597028, + "grad_norm": 1.655556321144104, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8806467056274414, + "num_tokens": 411246791.0, + "step": 11714 + }, + { + "epoch": 2.1754874651810585, + "grad_norm": 1.5441303253173828, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8798058032989502, + "num_tokens": 411281597.0, + "step": 11715 + }, + { + "epoch": 2.1756731662024142, + "grad_norm": 1.838489294052124, + "learning_rate": 1e-06, + "loss": 0.409, + "mean_token_accuracy": 0.8570215106010437, + "num_tokens": 411310133.0, + "step": 11716 + }, + { + "epoch": 2.1758588672237695, + "grad_norm": 1.4588696956634521, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8834248781204224, + "num_tokens": 411351910.0, + "step": 11717 + }, + { + "epoch": 2.1760445682451253, + "grad_norm": 1.4751155376434326, + "learning_rate": 1e-06, + "loss": 0.2617, + "mean_token_accuracy": 0.9051935076713562, + "num_tokens": 411383244.0, + "step": 11718 + }, + { + "epoch": 2.176230269266481, + "grad_norm": 1.598280906677246, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8833260536193848, + "num_tokens": 411413508.0, + "step": 11719 + }, + { + "epoch": 2.1764159702878367, + "grad_norm": 1.4884990453720093, + "learning_rate": 1e-06, + "loss": 0.2898, + "mean_token_accuracy": 0.8986347913742065, + "num_tokens": 411451218.0, + "step": 11720 + }, + { + "epoch": 2.176601671309192, + "grad_norm": 1.520987629890442, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8826818466186523, + "num_tokens": 411488042.0, + "step": 11721 + }, + { + "epoch": 2.1767873723305478, + "grad_norm": 1.572766661643982, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8778769969940186, + "num_tokens": 411525429.0, + "step": 11722 + }, + { + "epoch": 2.1769730733519035, + "grad_norm": 1.451019287109375, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8877925276756287, + "num_tokens": 411564596.0, + "step": 11723 + }, + { + "epoch": 2.177158774373259, + "grad_norm": 1.5890064239501953, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.8928724527359009, + "num_tokens": 411598738.0, + "step": 11724 + }, + { + "epoch": 2.1773444753946145, + "grad_norm": 1.421060562133789, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8846377730369568, + "num_tokens": 411641813.0, + "step": 11725 + }, + { + "epoch": 2.1775301764159702, + "grad_norm": 1.591843843460083, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8906980752944946, + "num_tokens": 411675015.0, + "step": 11726 + }, + { + "epoch": 2.177715877437326, + "grad_norm": 1.5855354070663452, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8920287489891052, + "num_tokens": 411711129.0, + "step": 11727 + }, + { + "epoch": 2.1779015784586817, + "grad_norm": 1.6169673204421997, + "learning_rate": 1e-06, + "loss": 0.3921, + "mean_token_accuracy": 0.8670987486839294, + "num_tokens": 411748727.0, + "step": 11728 + }, + { + "epoch": 2.178087279480037, + "grad_norm": 1.4522371292114258, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8867824077606201, + "num_tokens": 411791445.0, + "step": 11729 + }, + { + "epoch": 2.1782729805013927, + "grad_norm": 1.5515059232711792, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8906363248825073, + "num_tokens": 411825972.0, + "step": 11730 + }, + { + "epoch": 2.1784586815227485, + "grad_norm": 1.5069373846054077, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8965070247650146, + "num_tokens": 411861419.0, + "step": 11731 + }, + { + "epoch": 2.178644382544104, + "grad_norm": 1.518371820449829, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8793991208076477, + "num_tokens": 411903048.0, + "step": 11732 + }, + { + "epoch": 2.1788300835654595, + "grad_norm": 1.483816146850586, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8853850960731506, + "num_tokens": 411940917.0, + "step": 11733 + }, + { + "epoch": 2.179015784586815, + "grad_norm": 1.6195435523986816, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8869953155517578, + "num_tokens": 411973353.0, + "step": 11734 + }, + { + "epoch": 2.179201485608171, + "grad_norm": 1.6695293188095093, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8948537111282349, + "num_tokens": 412002364.0, + "step": 11735 + }, + { + "epoch": 2.1793871866295262, + "grad_norm": 1.4660227298736572, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8728302717208862, + "num_tokens": 412045616.0, + "step": 11736 + }, + { + "epoch": 2.179572887650882, + "grad_norm": 1.541764497756958, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8886206150054932, + "num_tokens": 412078221.0, + "step": 11737 + }, + { + "epoch": 2.1797585886722377, + "grad_norm": 1.6692243814468384, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8899814486503601, + "num_tokens": 412109687.0, + "step": 11738 + }, + { + "epoch": 2.1799442896935934, + "grad_norm": 1.8916789293289185, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.886540412902832, + "num_tokens": 412141948.0, + "step": 11739 + }, + { + "epoch": 2.1801299907149487, + "grad_norm": 1.6043899059295654, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8801240921020508, + "num_tokens": 412178518.0, + "step": 11740 + }, + { + "epoch": 2.1803156917363045, + "grad_norm": 1.6257039308547974, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8789075613021851, + "num_tokens": 412215546.0, + "step": 11741 + }, + { + "epoch": 2.18050139275766, + "grad_norm": 1.757824420928955, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8919838666915894, + "num_tokens": 412239991.0, + "step": 11742 + }, + { + "epoch": 2.180687093779016, + "grad_norm": 1.6573443412780762, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.891615629196167, + "num_tokens": 412271478.0, + "step": 11743 + }, + { + "epoch": 2.180872794800371, + "grad_norm": 1.5335161685943604, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.89095538854599, + "num_tokens": 412306879.0, + "step": 11744 + }, + { + "epoch": 2.181058495821727, + "grad_norm": 1.5480002164840698, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8871660828590393, + "num_tokens": 412343434.0, + "step": 11745 + }, + { + "epoch": 2.1812441968430827, + "grad_norm": 1.6991599798202515, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.884268045425415, + "num_tokens": 412378853.0, + "step": 11746 + }, + { + "epoch": 2.1814298978644384, + "grad_norm": 1.5589381456375122, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8750413656234741, + "num_tokens": 412416554.0, + "step": 11747 + }, + { + "epoch": 2.1816155988857937, + "grad_norm": 1.5524863004684448, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8864346146583557, + "num_tokens": 412453023.0, + "step": 11748 + }, + { + "epoch": 2.1818012999071494, + "grad_norm": 1.4710214138031006, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8867990374565125, + "num_tokens": 412490454.0, + "step": 11749 + }, + { + "epoch": 2.181987000928505, + "grad_norm": 1.7092359066009521, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8821816444396973, + "num_tokens": 412519127.0, + "step": 11750 + }, + { + "epoch": 2.182172701949861, + "grad_norm": 1.4566285610198975, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8781533241271973, + "num_tokens": 412561719.0, + "step": 11751 + }, + { + "epoch": 2.182358402971216, + "grad_norm": 1.6055817604064941, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8786740899085999, + "num_tokens": 412597274.0, + "step": 11752 + }, + { + "epoch": 2.182544103992572, + "grad_norm": 1.577567458152771, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8855015635490417, + "num_tokens": 412631165.0, + "step": 11753 + }, + { + "epoch": 2.1827298050139277, + "grad_norm": 1.6274783611297607, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.889545738697052, + "num_tokens": 412663997.0, + "step": 11754 + }, + { + "epoch": 2.1829155060352834, + "grad_norm": 1.567703366279602, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8906973600387573, + "num_tokens": 412705202.0, + "step": 11755 + }, + { + "epoch": 2.1831012070566387, + "grad_norm": 1.7288970947265625, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8572333455085754, + "num_tokens": 412741296.0, + "step": 11756 + }, + { + "epoch": 2.1832869080779944, + "grad_norm": 1.5371803045272827, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8930237293243408, + "num_tokens": 412777504.0, + "step": 11757 + }, + { + "epoch": 2.18347260909935, + "grad_norm": 1.6890015602111816, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8947007060050964, + "num_tokens": 412809767.0, + "step": 11758 + }, + { + "epoch": 2.1836583101207054, + "grad_norm": 1.4478693008422852, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8858098983764648, + "num_tokens": 412850508.0, + "step": 11759 + }, + { + "epoch": 2.183844011142061, + "grad_norm": 1.3877289295196533, + "learning_rate": 1e-06, + "loss": 0.2817, + "mean_token_accuracy": 0.897270917892456, + "num_tokens": 412889656.0, + "step": 11760 + }, + { + "epoch": 2.184029712163417, + "grad_norm": 1.5036017894744873, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8921066522598267, + "num_tokens": 412925801.0, + "step": 11761 + }, + { + "epoch": 2.1842154131847726, + "grad_norm": 1.6653666496276855, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8764359951019287, + "num_tokens": 412960110.0, + "step": 11762 + }, + { + "epoch": 2.1844011142061284, + "grad_norm": 1.520029067993164, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8981116414070129, + "num_tokens": 412994126.0, + "step": 11763 + }, + { + "epoch": 2.1845868152274837, + "grad_norm": 1.710384726524353, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8966688513755798, + "num_tokens": 413025054.0, + "step": 11764 + }, + { + "epoch": 2.1847725162488394, + "grad_norm": 1.446982979774475, + "learning_rate": 1e-06, + "loss": 0.2604, + "mean_token_accuracy": 0.9074667692184448, + "num_tokens": 413059883.0, + "step": 11765 + }, + { + "epoch": 2.184958217270195, + "grad_norm": 1.5565599203109741, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8883788585662842, + "num_tokens": 413095277.0, + "step": 11766 + }, + { + "epoch": 2.1851439182915504, + "grad_norm": 1.4922329187393188, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8811556100845337, + "num_tokens": 413134420.0, + "step": 11767 + }, + { + "epoch": 2.185329619312906, + "grad_norm": 1.5310641527175903, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8849601149559021, + "num_tokens": 413171828.0, + "step": 11768 + }, + { + "epoch": 2.185515320334262, + "grad_norm": 1.5064842700958252, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8824607133865356, + "num_tokens": 413212333.0, + "step": 11769 + }, + { + "epoch": 2.1857010213556176, + "grad_norm": 1.7072755098342896, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8842196464538574, + "num_tokens": 413246500.0, + "step": 11770 + }, + { + "epoch": 2.185886722376973, + "grad_norm": 1.723709225654602, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8929067254066467, + "num_tokens": 413277149.0, + "step": 11771 + }, + { + "epoch": 2.1860724233983286, + "grad_norm": 1.7816680669784546, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8766742944717407, + "num_tokens": 413307369.0, + "step": 11772 + }, + { + "epoch": 2.1862581244196844, + "grad_norm": 1.518674373626709, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.8983200788497925, + "num_tokens": 413341959.0, + "step": 11773 + }, + { + "epoch": 2.18644382544104, + "grad_norm": 1.5380381345748901, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.887292206287384, + "num_tokens": 413377973.0, + "step": 11774 + }, + { + "epoch": 2.1866295264623954, + "grad_norm": 1.5257521867752075, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8827158212661743, + "num_tokens": 413414496.0, + "step": 11775 + }, + { + "epoch": 2.186815227483751, + "grad_norm": 1.5820547342300415, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8837774991989136, + "num_tokens": 413450423.0, + "step": 11776 + }, + { + "epoch": 2.187000928505107, + "grad_norm": 1.4708282947540283, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8947182893753052, + "num_tokens": 413489772.0, + "step": 11777 + }, + { + "epoch": 2.1871866295264626, + "grad_norm": 1.4505590200424194, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8955644965171814, + "num_tokens": 413527522.0, + "step": 11778 + }, + { + "epoch": 2.187372330547818, + "grad_norm": 1.5032963752746582, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8819475173950195, + "num_tokens": 413562595.0, + "step": 11779 + }, + { + "epoch": 2.1875580315691736, + "grad_norm": 1.6203317642211914, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8868244886398315, + "num_tokens": 413596202.0, + "step": 11780 + }, + { + "epoch": 2.1877437325905293, + "grad_norm": 1.6335563659667969, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8763834834098816, + "num_tokens": 413634794.0, + "step": 11781 + }, + { + "epoch": 2.1879294336118846, + "grad_norm": 1.5461533069610596, + "learning_rate": 1e-06, + "loss": 0.2964, + "mean_token_accuracy": 0.8961780667304993, + "num_tokens": 413678156.0, + "step": 11782 + }, + { + "epoch": 2.1881151346332404, + "grad_norm": 1.7205584049224854, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8843450546264648, + "num_tokens": 413707798.0, + "step": 11783 + }, + { + "epoch": 2.188300835654596, + "grad_norm": 1.4972710609436035, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8932244777679443, + "num_tokens": 413743970.0, + "step": 11784 + }, + { + "epoch": 2.188486536675952, + "grad_norm": 1.5905849933624268, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.883893609046936, + "num_tokens": 413781956.0, + "step": 11785 + }, + { + "epoch": 2.1886722376973076, + "grad_norm": 1.6506049633026123, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8916501998901367, + "num_tokens": 413813164.0, + "step": 11786 + }, + { + "epoch": 2.188857938718663, + "grad_norm": 1.6538538932800293, + "learning_rate": 1e-06, + "loss": 0.3955, + "mean_token_accuracy": 0.8640855550765991, + "num_tokens": 413850067.0, + "step": 11787 + }, + { + "epoch": 2.1890436397400186, + "grad_norm": 1.5917507410049438, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8802394270896912, + "num_tokens": 413888163.0, + "step": 11788 + }, + { + "epoch": 2.1892293407613743, + "grad_norm": 1.6175602674484253, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8896794319152832, + "num_tokens": 413921405.0, + "step": 11789 + }, + { + "epoch": 2.1894150417827296, + "grad_norm": 1.6091279983520508, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.895403265953064, + "num_tokens": 413952185.0, + "step": 11790 + }, + { + "epoch": 2.1896007428040853, + "grad_norm": 1.5596215724945068, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8868056535720825, + "num_tokens": 413987730.0, + "step": 11791 + }, + { + "epoch": 2.189786443825441, + "grad_norm": 1.584812045097351, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8888742923736572, + "num_tokens": 414022501.0, + "step": 11792 + }, + { + "epoch": 2.189972144846797, + "grad_norm": 1.5230625867843628, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8908882737159729, + "num_tokens": 414060894.0, + "step": 11793 + }, + { + "epoch": 2.190157845868152, + "grad_norm": 1.7118488550186157, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8773970007896423, + "num_tokens": 414094151.0, + "step": 11794 + }, + { + "epoch": 2.190343546889508, + "grad_norm": 1.5498239994049072, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8952901363372803, + "num_tokens": 414126588.0, + "step": 11795 + }, + { + "epoch": 2.1905292479108636, + "grad_norm": 1.5493682622909546, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.880841851234436, + "num_tokens": 414163645.0, + "step": 11796 + }, + { + "epoch": 2.1907149489322193, + "grad_norm": 1.5943913459777832, + "learning_rate": 1e-06, + "loss": 0.2786, + "mean_token_accuracy": 0.9002618789672852, + "num_tokens": 414200453.0, + "step": 11797 + }, + { + "epoch": 2.1909006499535746, + "grad_norm": 1.5818451642990112, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8804336786270142, + "num_tokens": 414234315.0, + "step": 11798 + }, + { + "epoch": 2.1910863509749303, + "grad_norm": 1.498162865638733, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8892476558685303, + "num_tokens": 414273684.0, + "step": 11799 + }, + { + "epoch": 2.191272051996286, + "grad_norm": 1.3704485893249512, + "learning_rate": 1e-06, + "loss": 0.2791, + "mean_token_accuracy": 0.9025136232376099, + "num_tokens": 414314826.0, + "step": 11800 + }, + { + "epoch": 2.191457753017642, + "grad_norm": 1.5623492002487183, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8847942352294922, + "num_tokens": 414348609.0, + "step": 11801 + }, + { + "epoch": 2.191643454038997, + "grad_norm": 1.5521913766860962, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8853867650032043, + "num_tokens": 414382596.0, + "step": 11802 + }, + { + "epoch": 2.191829155060353, + "grad_norm": 1.6611526012420654, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.891569197177887, + "num_tokens": 414413309.0, + "step": 11803 + }, + { + "epoch": 2.1920148560817085, + "grad_norm": 1.5634267330169678, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8863112926483154, + "num_tokens": 414448392.0, + "step": 11804 + }, + { + "epoch": 2.1922005571030643, + "grad_norm": 1.5087165832519531, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8891624212265015, + "num_tokens": 414487621.0, + "step": 11805 + }, + { + "epoch": 2.1923862581244196, + "grad_norm": 1.6521755456924438, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8932070732116699, + "num_tokens": 414516568.0, + "step": 11806 + }, + { + "epoch": 2.1925719591457753, + "grad_norm": 1.3896865844726562, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8940538167953491, + "num_tokens": 414555451.0, + "step": 11807 + }, + { + "epoch": 2.192757660167131, + "grad_norm": 1.7381654977798462, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8807483911514282, + "num_tokens": 414583993.0, + "step": 11808 + }, + { + "epoch": 2.1929433611884868, + "grad_norm": 1.6811668872833252, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8872190117835999, + "num_tokens": 414615505.0, + "step": 11809 + }, + { + "epoch": 2.193129062209842, + "grad_norm": 1.4402838945388794, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8995925784111023, + "num_tokens": 414652210.0, + "step": 11810 + }, + { + "epoch": 2.193314763231198, + "grad_norm": 1.656869888305664, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8881407976150513, + "num_tokens": 414688431.0, + "step": 11811 + }, + { + "epoch": 2.1935004642525535, + "grad_norm": 1.6099909543991089, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8997935056686401, + "num_tokens": 414718138.0, + "step": 11812 + }, + { + "epoch": 2.193686165273909, + "grad_norm": 1.516400694847107, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8932676315307617, + "num_tokens": 414752198.0, + "step": 11813 + }, + { + "epoch": 2.1938718662952645, + "grad_norm": 1.4510602951049805, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8873243927955627, + "num_tokens": 414796209.0, + "step": 11814 + }, + { + "epoch": 2.1940575673166203, + "grad_norm": 1.5528589487075806, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8699077367782593, + "num_tokens": 414834729.0, + "step": 11815 + }, + { + "epoch": 2.194243268337976, + "grad_norm": 1.5506895780563354, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8902908563613892, + "num_tokens": 414872255.0, + "step": 11816 + }, + { + "epoch": 2.1944289693593313, + "grad_norm": 1.6551097631454468, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8804731369018555, + "num_tokens": 414905618.0, + "step": 11817 + }, + { + "epoch": 2.194614670380687, + "grad_norm": 1.678727626800537, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8816686272621155, + "num_tokens": 414939851.0, + "step": 11818 + }, + { + "epoch": 2.1948003714020428, + "grad_norm": 1.480168342590332, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8948771953582764, + "num_tokens": 414976308.0, + "step": 11819 + }, + { + "epoch": 2.1949860724233985, + "grad_norm": 1.5374114513397217, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8893035054206848, + "num_tokens": 415011817.0, + "step": 11820 + }, + { + "epoch": 2.195171773444754, + "grad_norm": 1.6823022365570068, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8767935633659363, + "num_tokens": 415043235.0, + "step": 11821 + }, + { + "epoch": 2.1953574744661095, + "grad_norm": 1.674918532371521, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.884588360786438, + "num_tokens": 415078031.0, + "step": 11822 + }, + { + "epoch": 2.1955431754874652, + "grad_norm": 1.4370622634887695, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8926296234130859, + "num_tokens": 415118947.0, + "step": 11823 + }, + { + "epoch": 2.195728876508821, + "grad_norm": 1.5259886980056763, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.8924483060836792, + "num_tokens": 415157044.0, + "step": 11824 + }, + { + "epoch": 2.1959145775301763, + "grad_norm": 1.6676278114318848, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8787518739700317, + "num_tokens": 415192051.0, + "step": 11825 + }, + { + "epoch": 2.196100278551532, + "grad_norm": 1.5581254959106445, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8891382813453674, + "num_tokens": 415228373.0, + "step": 11826 + }, + { + "epoch": 2.1962859795728877, + "grad_norm": 1.5091561079025269, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8888731002807617, + "num_tokens": 415266707.0, + "step": 11827 + }, + { + "epoch": 2.1964716805942435, + "grad_norm": 1.5819432735443115, + "learning_rate": 1e-06, + "loss": 0.3018, + "mean_token_accuracy": 0.89399254322052, + "num_tokens": 415300512.0, + "step": 11828 + }, + { + "epoch": 2.1966573816155988, + "grad_norm": 1.5786062479019165, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8710278868675232, + "num_tokens": 415336562.0, + "step": 11829 + }, + { + "epoch": 2.1968430826369545, + "grad_norm": 1.6402827501296997, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.893183708190918, + "num_tokens": 415370094.0, + "step": 11830 + }, + { + "epoch": 2.1970287836583102, + "grad_norm": 1.5078727006912231, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8901689052581787, + "num_tokens": 415405460.0, + "step": 11831 + }, + { + "epoch": 2.197214484679666, + "grad_norm": 1.6577839851379395, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8763321042060852, + "num_tokens": 415437654.0, + "step": 11832 + }, + { + "epoch": 2.1974001857010212, + "grad_norm": 1.4837779998779297, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8871462345123291, + "num_tokens": 415477566.0, + "step": 11833 + }, + { + "epoch": 2.197585886722377, + "grad_norm": 1.4130854606628418, + "learning_rate": 1e-06, + "loss": 0.2725, + "mean_token_accuracy": 0.9011644124984741, + "num_tokens": 415513871.0, + "step": 11834 + }, + { + "epoch": 2.1977715877437327, + "grad_norm": 1.457220196723938, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8963204026222229, + "num_tokens": 415550828.0, + "step": 11835 + }, + { + "epoch": 2.197957288765088, + "grad_norm": 1.6572016477584839, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8772460222244263, + "num_tokens": 415584200.0, + "step": 11836 + }, + { + "epoch": 2.1981429897864437, + "grad_norm": 1.6084561347961426, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8748849034309387, + "num_tokens": 415617788.0, + "step": 11837 + }, + { + "epoch": 2.1983286908077995, + "grad_norm": 1.5379527807235718, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.9010043740272522, + "num_tokens": 415649329.0, + "step": 11838 + }, + { + "epoch": 2.198514391829155, + "grad_norm": 1.473392367362976, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8823071718215942, + "num_tokens": 415692406.0, + "step": 11839 + }, + { + "epoch": 2.1987000928505105, + "grad_norm": 1.5737398862838745, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8881307244300842, + "num_tokens": 415728615.0, + "step": 11840 + }, + { + "epoch": 2.1988857938718662, + "grad_norm": 1.555123209953308, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.9056521654129028, + "num_tokens": 415757850.0, + "step": 11841 + }, + { + "epoch": 2.199071494893222, + "grad_norm": 1.5254995822906494, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8917125463485718, + "num_tokens": 415794224.0, + "step": 11842 + }, + { + "epoch": 2.1992571959145777, + "grad_norm": 1.6734238862991333, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8765983581542969, + "num_tokens": 415828926.0, + "step": 11843 + }, + { + "epoch": 2.199442896935933, + "grad_norm": 1.6662851572036743, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8881412744522095, + "num_tokens": 415859474.0, + "step": 11844 + }, + { + "epoch": 2.1996285979572887, + "grad_norm": 1.4072662591934204, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.893791675567627, + "num_tokens": 415900805.0, + "step": 11845 + }, + { + "epoch": 2.1998142989786444, + "grad_norm": 1.5030142068862915, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8896186351776123, + "num_tokens": 415935157.0, + "step": 11846 + }, + { + "epoch": 2.2, + "grad_norm": 1.6870472431182861, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8760202527046204, + "num_tokens": 415969149.0, + "step": 11847 + }, + { + "epoch": 2.2001857010213555, + "grad_norm": 1.5759270191192627, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8908531665802002, + "num_tokens": 416004360.0, + "step": 11848 + }, + { + "epoch": 2.200371402042711, + "grad_norm": 1.5852817296981812, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8785316944122314, + "num_tokens": 416040656.0, + "step": 11849 + }, + { + "epoch": 2.200557103064067, + "grad_norm": 1.4849587678909302, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8811533451080322, + "num_tokens": 416085170.0, + "step": 11850 + }, + { + "epoch": 2.2007428040854227, + "grad_norm": 1.6143126487731934, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8912827968597412, + "num_tokens": 416118439.0, + "step": 11851 + }, + { + "epoch": 2.200928505106778, + "grad_norm": 1.605355978012085, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.889396071434021, + "num_tokens": 416152997.0, + "step": 11852 + }, + { + "epoch": 2.2011142061281337, + "grad_norm": 1.548039197921753, + "learning_rate": 1e-06, + "loss": 0.2894, + "mean_token_accuracy": 0.8953056335449219, + "num_tokens": 416190951.0, + "step": 11853 + }, + { + "epoch": 2.2012999071494894, + "grad_norm": 1.5399812459945679, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8930381536483765, + "num_tokens": 416225447.0, + "step": 11854 + }, + { + "epoch": 2.201485608170845, + "grad_norm": 1.686437726020813, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8749356269836426, + "num_tokens": 416259095.0, + "step": 11855 + }, + { + "epoch": 2.2016713091922004, + "grad_norm": 1.6378464698791504, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8711097240447998, + "num_tokens": 416292782.0, + "step": 11856 + }, + { + "epoch": 2.201857010213556, + "grad_norm": 1.538508415222168, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.892687201499939, + "num_tokens": 416324661.0, + "step": 11857 + }, + { + "epoch": 2.202042711234912, + "grad_norm": 1.4321856498718262, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8891083002090454, + "num_tokens": 416362392.0, + "step": 11858 + }, + { + "epoch": 2.202228412256267, + "grad_norm": 1.5406136512756348, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8921315670013428, + "num_tokens": 416397986.0, + "step": 11859 + }, + { + "epoch": 2.202414113277623, + "grad_norm": 1.5497777462005615, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8817800879478455, + "num_tokens": 416435331.0, + "step": 11860 + }, + { + "epoch": 2.2025998142989787, + "grad_norm": 1.6138931512832642, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.89057457447052, + "num_tokens": 416464779.0, + "step": 11861 + }, + { + "epoch": 2.2027855153203344, + "grad_norm": 1.3497042655944824, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8953278660774231, + "num_tokens": 416509606.0, + "step": 11862 + }, + { + "epoch": 2.2029712163416897, + "grad_norm": 1.5212035179138184, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8959616422653198, + "num_tokens": 416542539.0, + "step": 11863 + }, + { + "epoch": 2.2031569173630454, + "grad_norm": 1.6737455129623413, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8795198202133179, + "num_tokens": 416575102.0, + "step": 11864 + }, + { + "epoch": 2.203342618384401, + "grad_norm": 1.5052775144577026, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8899475932121277, + "num_tokens": 416614409.0, + "step": 11865 + }, + { + "epoch": 2.203528319405757, + "grad_norm": 1.621890664100647, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8846051692962646, + "num_tokens": 416648806.0, + "step": 11866 + }, + { + "epoch": 2.203714020427112, + "grad_norm": 1.6227823495864868, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8666020035743713, + "num_tokens": 416686386.0, + "step": 11867 + }, + { + "epoch": 2.203899721448468, + "grad_norm": 1.6820859909057617, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8749774098396301, + "num_tokens": 416719922.0, + "step": 11868 + }, + { + "epoch": 2.2040854224698236, + "grad_norm": 1.5580435991287231, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8749321699142456, + "num_tokens": 416760531.0, + "step": 11869 + }, + { + "epoch": 2.2042711234911794, + "grad_norm": 1.4807851314544678, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8699896335601807, + "num_tokens": 416804411.0, + "step": 11870 + }, + { + "epoch": 2.2044568245125347, + "grad_norm": 1.6096056699752808, + "learning_rate": 1e-06, + "loss": 0.2909, + "mean_token_accuracy": 0.8969249725341797, + "num_tokens": 416834245.0, + "step": 11871 + }, + { + "epoch": 2.2046425255338904, + "grad_norm": 1.6628562211990356, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.883470892906189, + "num_tokens": 416869498.0, + "step": 11872 + }, + { + "epoch": 2.204828226555246, + "grad_norm": 1.462631344795227, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8937880396842957, + "num_tokens": 416911649.0, + "step": 11873 + }, + { + "epoch": 2.205013927576602, + "grad_norm": 1.8145090341567993, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8910967111587524, + "num_tokens": 416937472.0, + "step": 11874 + }, + { + "epoch": 2.205199628597957, + "grad_norm": 1.520201325416565, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8945753574371338, + "num_tokens": 416973056.0, + "step": 11875 + }, + { + "epoch": 2.205385329619313, + "grad_norm": 1.5204486846923828, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8857492804527283, + "num_tokens": 417008290.0, + "step": 11876 + }, + { + "epoch": 2.2055710306406686, + "grad_norm": 1.5077542066574097, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8975293636322021, + "num_tokens": 417042402.0, + "step": 11877 + }, + { + "epoch": 2.2057567316620244, + "grad_norm": 1.5556869506835938, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8805849552154541, + "num_tokens": 417076780.0, + "step": 11878 + }, + { + "epoch": 2.2059424326833796, + "grad_norm": 1.5459169149398804, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8874156475067139, + "num_tokens": 417108957.0, + "step": 11879 + }, + { + "epoch": 2.2061281337047354, + "grad_norm": 1.6351224184036255, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8705012798309326, + "num_tokens": 417144598.0, + "step": 11880 + }, + { + "epoch": 2.206313834726091, + "grad_norm": 1.5998173952102661, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.896991491317749, + "num_tokens": 417176612.0, + "step": 11881 + }, + { + "epoch": 2.2064995357474464, + "grad_norm": 1.5508910417556763, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8898018002510071, + "num_tokens": 417212020.0, + "step": 11882 + }, + { + "epoch": 2.206685236768802, + "grad_norm": 1.5555212497711182, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8846133947372437, + "num_tokens": 417247886.0, + "step": 11883 + }, + { + "epoch": 2.206870937790158, + "grad_norm": 1.7093470096588135, + "learning_rate": 1e-06, + "loss": 0.2767, + "mean_token_accuracy": 0.9007923603057861, + "num_tokens": 417275162.0, + "step": 11884 + }, + { + "epoch": 2.2070566388115136, + "grad_norm": 1.468102216720581, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.88136225938797, + "num_tokens": 417319754.0, + "step": 11885 + }, + { + "epoch": 2.207242339832869, + "grad_norm": 1.6104352474212646, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.887570858001709, + "num_tokens": 417353234.0, + "step": 11886 + }, + { + "epoch": 2.2074280408542246, + "grad_norm": 1.5041320323944092, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8907038569450378, + "num_tokens": 417392910.0, + "step": 11887 + }, + { + "epoch": 2.2076137418755803, + "grad_norm": 1.5152668952941895, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8934005498886108, + "num_tokens": 417430183.0, + "step": 11888 + }, + { + "epoch": 2.207799442896936, + "grad_norm": 1.5452806949615479, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8903809785842896, + "num_tokens": 417465572.0, + "step": 11889 + }, + { + "epoch": 2.2079851439182914, + "grad_norm": 1.6248854398727417, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8840820789337158, + "num_tokens": 417498994.0, + "step": 11890 + }, + { + "epoch": 2.208170844939647, + "grad_norm": 1.508333444595337, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.876823365688324, + "num_tokens": 417539218.0, + "step": 11891 + }, + { + "epoch": 2.208356545961003, + "grad_norm": 1.5486781597137451, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8774312734603882, + "num_tokens": 417576824.0, + "step": 11892 + }, + { + "epoch": 2.2085422469823586, + "grad_norm": 1.4881198406219482, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8793191909790039, + "num_tokens": 417616889.0, + "step": 11893 + }, + { + "epoch": 2.208727948003714, + "grad_norm": 1.602485179901123, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8908298015594482, + "num_tokens": 417651822.0, + "step": 11894 + }, + { + "epoch": 2.2089136490250696, + "grad_norm": 1.6101398468017578, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.884398341178894, + "num_tokens": 417689038.0, + "step": 11895 + }, + { + "epoch": 2.2090993500464253, + "grad_norm": 1.7066761255264282, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8781861066818237, + "num_tokens": 417720486.0, + "step": 11896 + }, + { + "epoch": 2.209285051067781, + "grad_norm": 1.6363401412963867, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8803279995918274, + "num_tokens": 417754329.0, + "step": 11897 + }, + { + "epoch": 2.2094707520891363, + "grad_norm": 1.5853049755096436, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8787910342216492, + "num_tokens": 417788503.0, + "step": 11898 + }, + { + "epoch": 2.209656453110492, + "grad_norm": 1.6271241903305054, + "learning_rate": 1e-06, + "loss": 0.2916, + "mean_token_accuracy": 0.8951482772827148, + "num_tokens": 417824253.0, + "step": 11899 + }, + { + "epoch": 2.209842154131848, + "grad_norm": 1.6324933767318726, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8869951963424683, + "num_tokens": 417856511.0, + "step": 11900 + }, + { + "epoch": 2.2100278551532035, + "grad_norm": 1.6229805946350098, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8895635008811951, + "num_tokens": 417886947.0, + "step": 11901 + }, + { + "epoch": 2.210213556174559, + "grad_norm": 1.522830843925476, + "learning_rate": 1e-06, + "loss": 0.2671, + "mean_token_accuracy": 0.9048199653625488, + "num_tokens": 417919621.0, + "step": 11902 + }, + { + "epoch": 2.2103992571959146, + "grad_norm": 1.5206949710845947, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.883548378944397, + "num_tokens": 417956862.0, + "step": 11903 + }, + { + "epoch": 2.2105849582172703, + "grad_norm": 1.616236686706543, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8847116827964783, + "num_tokens": 417992295.0, + "step": 11904 + }, + { + "epoch": 2.2107706592386256, + "grad_norm": 1.6980315446853638, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8815761804580688, + "num_tokens": 418022723.0, + "step": 11905 + }, + { + "epoch": 2.2109563602599813, + "grad_norm": 1.5045998096466064, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8748421669006348, + "num_tokens": 418062999.0, + "step": 11906 + }, + { + "epoch": 2.211142061281337, + "grad_norm": 1.6951673030853271, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8851041197776794, + "num_tokens": 418094689.0, + "step": 11907 + }, + { + "epoch": 2.211327762302693, + "grad_norm": 1.4867441654205322, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8931196928024292, + "num_tokens": 418129736.0, + "step": 11908 + }, + { + "epoch": 2.211513463324048, + "grad_norm": 1.5074423551559448, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8953863382339478, + "num_tokens": 418165895.0, + "step": 11909 + }, + { + "epoch": 2.211699164345404, + "grad_norm": 1.5187656879425049, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8912966251373291, + "num_tokens": 418200621.0, + "step": 11910 + }, + { + "epoch": 2.2118848653667595, + "grad_norm": 1.6335840225219727, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8794584274291992, + "num_tokens": 418232990.0, + "step": 11911 + }, + { + "epoch": 2.2120705663881153, + "grad_norm": 1.5126374959945679, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.881913423538208, + "num_tokens": 418271107.0, + "step": 11912 + }, + { + "epoch": 2.2122562674094706, + "grad_norm": 1.617143154144287, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8764166831970215, + "num_tokens": 418306033.0, + "step": 11913 + }, + { + "epoch": 2.2124419684308263, + "grad_norm": 1.5865063667297363, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.889517068862915, + "num_tokens": 418337980.0, + "step": 11914 + }, + { + "epoch": 2.212627669452182, + "grad_norm": 1.5498008728027344, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8740830421447754, + "num_tokens": 418374200.0, + "step": 11915 + }, + { + "epoch": 2.2128133704735378, + "grad_norm": 1.5373681783676147, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8798038363456726, + "num_tokens": 418410225.0, + "step": 11916 + }, + { + "epoch": 2.212999071494893, + "grad_norm": 1.4218841791152954, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.893452525138855, + "num_tokens": 418451527.0, + "step": 11917 + }, + { + "epoch": 2.213184772516249, + "grad_norm": 1.481399655342102, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8918572664260864, + "num_tokens": 418489823.0, + "step": 11918 + }, + { + "epoch": 2.2133704735376045, + "grad_norm": 1.536739468574524, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.8960146903991699, + "num_tokens": 418526085.0, + "step": 11919 + }, + { + "epoch": 2.2135561745589603, + "grad_norm": 1.6828138828277588, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8907180428504944, + "num_tokens": 418554733.0, + "step": 11920 + }, + { + "epoch": 2.2137418755803155, + "grad_norm": 1.6168755292892456, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8854513168334961, + "num_tokens": 418588659.0, + "step": 11921 + }, + { + "epoch": 2.2139275766016713, + "grad_norm": 1.477388620376587, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8922023773193359, + "num_tokens": 418626617.0, + "step": 11922 + }, + { + "epoch": 2.214113277623027, + "grad_norm": 1.814016342163086, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8822436928749084, + "num_tokens": 418657383.0, + "step": 11923 + }, + { + "epoch": 2.2142989786443827, + "grad_norm": 1.5316346883773804, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8862276077270508, + "num_tokens": 418694742.0, + "step": 11924 + }, + { + "epoch": 2.214484679665738, + "grad_norm": 1.530998945236206, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8921049237251282, + "num_tokens": 418729229.0, + "step": 11925 + }, + { + "epoch": 2.2146703806870938, + "grad_norm": 1.4475880861282349, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8893945813179016, + "num_tokens": 418768806.0, + "step": 11926 + }, + { + "epoch": 2.2148560817084495, + "grad_norm": 1.461756706237793, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8919922709465027, + "num_tokens": 418808200.0, + "step": 11927 + }, + { + "epoch": 2.215041782729805, + "grad_norm": 1.510223627090454, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8928998708724976, + "num_tokens": 418843571.0, + "step": 11928 + }, + { + "epoch": 2.2152274837511605, + "grad_norm": 1.6029419898986816, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8829908967018127, + "num_tokens": 418877297.0, + "step": 11929 + }, + { + "epoch": 2.2154131847725163, + "grad_norm": 1.6645046472549438, + "learning_rate": 1e-06, + "loss": 0.3779, + "mean_token_accuracy": 0.8747215270996094, + "num_tokens": 418911854.0, + "step": 11930 + }, + { + "epoch": 2.215598885793872, + "grad_norm": 1.5573347806930542, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8963096141815186, + "num_tokens": 418945010.0, + "step": 11931 + }, + { + "epoch": 2.2157845868152277, + "grad_norm": 1.5517867803573608, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8935225605964661, + "num_tokens": 418978510.0, + "step": 11932 + }, + { + "epoch": 2.215970287836583, + "grad_norm": 1.469558596611023, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8849161267280579, + "num_tokens": 419015893.0, + "step": 11933 + }, + { + "epoch": 2.2161559888579387, + "grad_norm": 1.5122591257095337, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8801727294921875, + "num_tokens": 419052820.0, + "step": 11934 + }, + { + "epoch": 2.2163416898792945, + "grad_norm": 1.5207240581512451, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8816528916358948, + "num_tokens": 419089747.0, + "step": 11935 + }, + { + "epoch": 2.2165273909006498, + "grad_norm": 1.6981227397918701, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8882772922515869, + "num_tokens": 419120787.0, + "step": 11936 + }, + { + "epoch": 2.2167130919220055, + "grad_norm": 1.7515711784362793, + "learning_rate": 1e-06, + "loss": 0.2991, + "mean_token_accuracy": 0.8955861330032349, + "num_tokens": 419148150.0, + "step": 11937 + }, + { + "epoch": 2.2168987929433612, + "grad_norm": 1.6382789611816406, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8887293338775635, + "num_tokens": 419177496.0, + "step": 11938 + }, + { + "epoch": 2.217084493964717, + "grad_norm": 1.5006449222564697, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8945388793945312, + "num_tokens": 419212135.0, + "step": 11939 + }, + { + "epoch": 2.2172701949860723, + "grad_norm": 1.5722390413284302, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8872687816619873, + "num_tokens": 419248221.0, + "step": 11940 + }, + { + "epoch": 2.217455896007428, + "grad_norm": 1.5090006589889526, + "learning_rate": 1e-06, + "loss": 0.2774, + "mean_token_accuracy": 0.9014929533004761, + "num_tokens": 419281772.0, + "step": 11941 + }, + { + "epoch": 2.2176415970287837, + "grad_norm": 1.6476436853408813, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8883333206176758, + "num_tokens": 419315516.0, + "step": 11942 + }, + { + "epoch": 2.2178272980501395, + "grad_norm": 1.5420211553573608, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8877741098403931, + "num_tokens": 419351176.0, + "step": 11943 + }, + { + "epoch": 2.2180129990714947, + "grad_norm": 1.6709511280059814, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8793508410453796, + "num_tokens": 419385378.0, + "step": 11944 + }, + { + "epoch": 2.2181987000928505, + "grad_norm": 1.482627272605896, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8780354261398315, + "num_tokens": 419423327.0, + "step": 11945 + }, + { + "epoch": 2.218384401114206, + "grad_norm": 1.5103195905685425, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8840726017951965, + "num_tokens": 419461196.0, + "step": 11946 + }, + { + "epoch": 2.218570102135562, + "grad_norm": 1.6474040746688843, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8797511458396912, + "num_tokens": 419493342.0, + "step": 11947 + }, + { + "epoch": 2.2187558031569172, + "grad_norm": 1.7057510614395142, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8868728876113892, + "num_tokens": 419528710.0, + "step": 11948 + }, + { + "epoch": 2.218941504178273, + "grad_norm": 1.6501716375350952, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8790017366409302, + "num_tokens": 419560339.0, + "step": 11949 + }, + { + "epoch": 2.2191272051996287, + "grad_norm": 1.5170795917510986, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8902655839920044, + "num_tokens": 419596737.0, + "step": 11950 + }, + { + "epoch": 2.219312906220984, + "grad_norm": 1.6352648735046387, + "learning_rate": 1e-06, + "loss": 0.2912, + "mean_token_accuracy": 0.8966877460479736, + "num_tokens": 419627252.0, + "step": 11951 + }, + { + "epoch": 2.2194986072423397, + "grad_norm": 1.465587854385376, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8866880536079407, + "num_tokens": 419666673.0, + "step": 11952 + }, + { + "epoch": 2.2196843082636954, + "grad_norm": 1.6440141201019287, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8859014511108398, + "num_tokens": 419697183.0, + "step": 11953 + }, + { + "epoch": 2.219870009285051, + "grad_norm": 1.3698400259017944, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8903713226318359, + "num_tokens": 419742202.0, + "step": 11954 + }, + { + "epoch": 2.220055710306407, + "grad_norm": 1.6062147617340088, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8865500688552856, + "num_tokens": 419774961.0, + "step": 11955 + }, + { + "epoch": 2.220241411327762, + "grad_norm": 1.587834119796753, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8880380392074585, + "num_tokens": 419809590.0, + "step": 11956 + }, + { + "epoch": 2.220427112349118, + "grad_norm": 1.5094877481460571, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8874123096466064, + "num_tokens": 419851481.0, + "step": 11957 + }, + { + "epoch": 2.2206128133704737, + "grad_norm": 1.5118106603622437, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8838999271392822, + "num_tokens": 419888410.0, + "step": 11958 + }, + { + "epoch": 2.220798514391829, + "grad_norm": 1.359074592590332, + "learning_rate": 1e-06, + "loss": 0.2977, + "mean_token_accuracy": 0.8954886198043823, + "num_tokens": 419930879.0, + "step": 11959 + }, + { + "epoch": 2.2209842154131847, + "grad_norm": 1.620390772819519, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8855696320533752, + "num_tokens": 419964601.0, + "step": 11960 + }, + { + "epoch": 2.2211699164345404, + "grad_norm": 1.5473923683166504, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8819530010223389, + "num_tokens": 419998625.0, + "step": 11961 + }, + { + "epoch": 2.221355617455896, + "grad_norm": 1.5712109804153442, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8873604536056519, + "num_tokens": 420032693.0, + "step": 11962 + }, + { + "epoch": 2.2215413184772514, + "grad_norm": 1.556114673614502, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8765243291854858, + "num_tokens": 420070449.0, + "step": 11963 + }, + { + "epoch": 2.221727019498607, + "grad_norm": 1.595621109008789, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8905757665634155, + "num_tokens": 420105702.0, + "step": 11964 + }, + { + "epoch": 2.221912720519963, + "grad_norm": 1.5864742994308472, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.8901076316833496, + "num_tokens": 420137676.0, + "step": 11965 + }, + { + "epoch": 2.2220984215413186, + "grad_norm": 1.3804700374603271, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8878613710403442, + "num_tokens": 420179618.0, + "step": 11966 + }, + { + "epoch": 2.222284122562674, + "grad_norm": 1.5252668857574463, + "learning_rate": 1e-06, + "loss": 0.2873, + "mean_token_accuracy": 0.8967816829681396, + "num_tokens": 420214172.0, + "step": 11967 + }, + { + "epoch": 2.2224698235840297, + "grad_norm": 1.5581260919570923, + "learning_rate": 1e-06, + "loss": 0.2794, + "mean_token_accuracy": 0.8997754454612732, + "num_tokens": 420248262.0, + "step": 11968 + }, + { + "epoch": 2.2226555246053854, + "grad_norm": 1.6446775197982788, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8616631627082825, + "num_tokens": 420288110.0, + "step": 11969 + }, + { + "epoch": 2.222841225626741, + "grad_norm": 1.7485697269439697, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8841177225112915, + "num_tokens": 420319593.0, + "step": 11970 + }, + { + "epoch": 2.2230269266480964, + "grad_norm": 1.5433259010314941, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8879486322402954, + "num_tokens": 420354767.0, + "step": 11971 + }, + { + "epoch": 2.223212627669452, + "grad_norm": 1.7932201623916626, + "learning_rate": 1e-06, + "loss": 0.273, + "mean_token_accuracy": 0.9028042554855347, + "num_tokens": 420390040.0, + "step": 11972 + }, + { + "epoch": 2.223398328690808, + "grad_norm": 1.8347325325012207, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8831596374511719, + "num_tokens": 420413189.0, + "step": 11973 + }, + { + "epoch": 2.2235840297121636, + "grad_norm": 1.586366057395935, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8762902617454529, + "num_tokens": 420453098.0, + "step": 11974 + }, + { + "epoch": 2.223769730733519, + "grad_norm": 1.6313093900680542, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8763107061386108, + "num_tokens": 420487797.0, + "step": 11975 + }, + { + "epoch": 2.2239554317548746, + "grad_norm": 1.56217622756958, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8795360326766968, + "num_tokens": 420527509.0, + "step": 11976 + }, + { + "epoch": 2.2241411327762304, + "grad_norm": 1.6099475622177124, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8917955160140991, + "num_tokens": 420558748.0, + "step": 11977 + }, + { + "epoch": 2.224326833797586, + "grad_norm": 1.6022955179214478, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8733441829681396, + "num_tokens": 420597091.0, + "step": 11978 + }, + { + "epoch": 2.2245125348189414, + "grad_norm": 1.557492971420288, + "learning_rate": 1e-06, + "loss": 0.2809, + "mean_token_accuracy": 0.8970917463302612, + "num_tokens": 420632067.0, + "step": 11979 + }, + { + "epoch": 2.224698235840297, + "grad_norm": 1.5623505115509033, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8815877437591553, + "num_tokens": 420669285.0, + "step": 11980 + }, + { + "epoch": 2.224883936861653, + "grad_norm": 1.7259098291397095, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8858039379119873, + "num_tokens": 420697799.0, + "step": 11981 + }, + { + "epoch": 2.225069637883008, + "grad_norm": 1.5504074096679688, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8814606666564941, + "num_tokens": 420735312.0, + "step": 11982 + }, + { + "epoch": 2.225255338904364, + "grad_norm": 1.5636286735534668, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8754564523696899, + "num_tokens": 420773385.0, + "step": 11983 + }, + { + "epoch": 2.2254410399257196, + "grad_norm": 1.5675857067108154, + "learning_rate": 1e-06, + "loss": 0.2932, + "mean_token_accuracy": 0.8943676948547363, + "num_tokens": 420806132.0, + "step": 11984 + }, + { + "epoch": 2.2256267409470754, + "grad_norm": 1.693859338760376, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.9027600288391113, + "num_tokens": 420834503.0, + "step": 11985 + }, + { + "epoch": 2.2258124419684306, + "grad_norm": 1.5697739124298096, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8769006729125977, + "num_tokens": 420869192.0, + "step": 11986 + }, + { + "epoch": 2.2259981429897864, + "grad_norm": 1.6314842700958252, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.875083863735199, + "num_tokens": 420903132.0, + "step": 11987 + }, + { + "epoch": 2.226183844011142, + "grad_norm": 1.580834984779358, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8663907051086426, + "num_tokens": 420939148.0, + "step": 11988 + }, + { + "epoch": 2.226369545032498, + "grad_norm": 1.6148053407669067, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8899368047714233, + "num_tokens": 420971297.0, + "step": 11989 + }, + { + "epoch": 2.226555246053853, + "grad_norm": 1.5198490619659424, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8870554566383362, + "num_tokens": 421010978.0, + "step": 11990 + }, + { + "epoch": 2.226740947075209, + "grad_norm": 1.7294458150863647, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.8919541239738464, + "num_tokens": 421038718.0, + "step": 11991 + }, + { + "epoch": 2.2269266480965646, + "grad_norm": 1.6327297687530518, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8907632827758789, + "num_tokens": 421072039.0, + "step": 11992 + }, + { + "epoch": 2.2271123491179203, + "grad_norm": 1.5115841627120972, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8900752067565918, + "num_tokens": 421108860.0, + "step": 11993 + }, + { + "epoch": 2.2272980501392756, + "grad_norm": 1.6086390018463135, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8970487713813782, + "num_tokens": 421141631.0, + "step": 11994 + }, + { + "epoch": 2.2274837511606314, + "grad_norm": 1.6314209699630737, + "learning_rate": 1e-06, + "loss": 0.3822, + "mean_token_accuracy": 0.8695157766342163, + "num_tokens": 421179131.0, + "step": 11995 + }, + { + "epoch": 2.227669452181987, + "grad_norm": 1.6611716747283936, + "learning_rate": 1e-06, + "loss": 0.2866, + "mean_token_accuracy": 0.8979107141494751, + "num_tokens": 421206155.0, + "step": 11996 + }, + { + "epoch": 2.227855153203343, + "grad_norm": 1.678206205368042, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8892285227775574, + "num_tokens": 421239547.0, + "step": 11997 + }, + { + "epoch": 2.228040854224698, + "grad_norm": 1.468641996383667, + "learning_rate": 1e-06, + "loss": 0.2656, + "mean_token_accuracy": 0.9054617881774902, + "num_tokens": 421278694.0, + "step": 11998 + }, + { + "epoch": 2.228226555246054, + "grad_norm": 1.7024821043014526, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8744089603424072, + "num_tokens": 421310147.0, + "step": 11999 + }, + { + "epoch": 2.2284122562674096, + "grad_norm": 1.5619566440582275, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8785346746444702, + "num_tokens": 421343599.0, + "step": 12000 + }, + { + "epoch": 2.2285979572887653, + "grad_norm": 1.7314356565475464, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8865671157836914, + "num_tokens": 421376504.0, + "step": 12001 + }, + { + "epoch": 2.2287836583101206, + "grad_norm": 1.5732524394989014, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8827049732208252, + "num_tokens": 421410540.0, + "step": 12002 + }, + { + "epoch": 2.2289693593314763, + "grad_norm": 1.6014896631240845, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8860319256782532, + "num_tokens": 421444296.0, + "step": 12003 + }, + { + "epoch": 2.229155060352832, + "grad_norm": 1.4456015825271606, + "learning_rate": 1e-06, + "loss": 0.2933, + "mean_token_accuracy": 0.8951019048690796, + "num_tokens": 421479726.0, + "step": 12004 + }, + { + "epoch": 2.2293407613741874, + "grad_norm": 1.6504223346710205, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.895266056060791, + "num_tokens": 421511774.0, + "step": 12005 + }, + { + "epoch": 2.229526462395543, + "grad_norm": 1.637744426727295, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8912726044654846, + "num_tokens": 421543270.0, + "step": 12006 + }, + { + "epoch": 2.229712163416899, + "grad_norm": 1.6375540494918823, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8814821839332581, + "num_tokens": 421575331.0, + "step": 12007 + }, + { + "epoch": 2.2298978644382546, + "grad_norm": 1.427775502204895, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8953977823257446, + "num_tokens": 421613199.0, + "step": 12008 + }, + { + "epoch": 2.23008356545961, + "grad_norm": 1.588310956954956, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8846731185913086, + "num_tokens": 421648787.0, + "step": 12009 + }, + { + "epoch": 2.2302692664809656, + "grad_norm": 1.561287522315979, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8871463537216187, + "num_tokens": 421680924.0, + "step": 12010 + }, + { + "epoch": 2.2304549675023213, + "grad_norm": 1.6195399761199951, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8821730613708496, + "num_tokens": 421713368.0, + "step": 12011 + }, + { + "epoch": 2.230640668523677, + "grad_norm": 1.6017143726348877, + "learning_rate": 1e-06, + "loss": 0.3843, + "mean_token_accuracy": 0.8650119304656982, + "num_tokens": 421752369.0, + "step": 12012 + }, + { + "epoch": 2.2308263695450323, + "grad_norm": 1.6863832473754883, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8870295882225037, + "num_tokens": 421783197.0, + "step": 12013 + }, + { + "epoch": 2.231012070566388, + "grad_norm": 1.5746906995773315, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8852744102478027, + "num_tokens": 421816151.0, + "step": 12014 + }, + { + "epoch": 2.231197771587744, + "grad_norm": 1.53767728805542, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8823727369308472, + "num_tokens": 421850793.0, + "step": 12015 + }, + { + "epoch": 2.2313834726090995, + "grad_norm": 1.6465033292770386, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8816674947738647, + "num_tokens": 421883670.0, + "step": 12016 + }, + { + "epoch": 2.231569173630455, + "grad_norm": 1.6751443147659302, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8837001323699951, + "num_tokens": 421916537.0, + "step": 12017 + }, + { + "epoch": 2.2317548746518105, + "grad_norm": 1.4723594188690186, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8805548548698425, + "num_tokens": 421957307.0, + "step": 12018 + }, + { + "epoch": 2.2319405756731663, + "grad_norm": 1.5279768705368042, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.890006422996521, + "num_tokens": 421992050.0, + "step": 12019 + }, + { + "epoch": 2.232126276694522, + "grad_norm": 1.5012701749801636, + "learning_rate": 1e-06, + "loss": 0.3725, + "mean_token_accuracy": 0.868922770023346, + "num_tokens": 422032311.0, + "step": 12020 + }, + { + "epoch": 2.2323119777158773, + "grad_norm": 1.5371984243392944, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8789218068122864, + "num_tokens": 422069140.0, + "step": 12021 + }, + { + "epoch": 2.232497678737233, + "grad_norm": 1.5752527713775635, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8750136494636536, + "num_tokens": 422104810.0, + "step": 12022 + }, + { + "epoch": 2.2326833797585888, + "grad_norm": 1.5744870901107788, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8907641172409058, + "num_tokens": 422136327.0, + "step": 12023 + }, + { + "epoch": 2.2328690807799445, + "grad_norm": 1.5106078386306763, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8851140141487122, + "num_tokens": 422173838.0, + "step": 12024 + }, + { + "epoch": 2.2330547818013, + "grad_norm": 1.6891974210739136, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8808876276016235, + "num_tokens": 422209121.0, + "step": 12025 + }, + { + "epoch": 2.2332404828226555, + "grad_norm": 1.6007012128829956, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8820295929908752, + "num_tokens": 422244949.0, + "step": 12026 + }, + { + "epoch": 2.2334261838440113, + "grad_norm": 1.6713273525238037, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8770090937614441, + "num_tokens": 422278288.0, + "step": 12027 + }, + { + "epoch": 2.2336118848653665, + "grad_norm": 1.564333438873291, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8773067593574524, + "num_tokens": 422313329.0, + "step": 12028 + }, + { + "epoch": 2.2337975858867223, + "grad_norm": 1.4424669742584229, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8856241703033447, + "num_tokens": 422352956.0, + "step": 12029 + }, + { + "epoch": 2.233983286908078, + "grad_norm": 1.5772686004638672, + "learning_rate": 1e-06, + "loss": 0.3686, + "mean_token_accuracy": 0.8711493611335754, + "num_tokens": 422390488.0, + "step": 12030 + }, + { + "epoch": 2.2341689879294337, + "grad_norm": 1.497475266456604, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8884738683700562, + "num_tokens": 422429881.0, + "step": 12031 + }, + { + "epoch": 2.234354688950789, + "grad_norm": 1.5870630741119385, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.8770474791526794, + "num_tokens": 422470383.0, + "step": 12032 + }, + { + "epoch": 2.2345403899721448, + "grad_norm": 1.5409775972366333, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.885879635810852, + "num_tokens": 422508202.0, + "step": 12033 + }, + { + "epoch": 2.2347260909935005, + "grad_norm": 1.5768427848815918, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8776950240135193, + "num_tokens": 422545332.0, + "step": 12034 + }, + { + "epoch": 2.2349117920148562, + "grad_norm": 1.551259160041809, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8888471126556396, + "num_tokens": 422578804.0, + "step": 12035 + }, + { + "epoch": 2.2350974930362115, + "grad_norm": 1.6045753955841064, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8763959407806396, + "num_tokens": 422613342.0, + "step": 12036 + }, + { + "epoch": 2.2352831940575673, + "grad_norm": 1.485417366027832, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.8996268510818481, + "num_tokens": 422650024.0, + "step": 12037 + }, + { + "epoch": 2.235468895078923, + "grad_norm": 1.8220374584197998, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8842458724975586, + "num_tokens": 422676943.0, + "step": 12038 + }, + { + "epoch": 2.2356545961002787, + "grad_norm": 1.45772123336792, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8837974071502686, + "num_tokens": 422714713.0, + "step": 12039 + }, + { + "epoch": 2.235840297121634, + "grad_norm": 1.5420030355453491, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8956830501556396, + "num_tokens": 422748571.0, + "step": 12040 + }, + { + "epoch": 2.2360259981429897, + "grad_norm": 1.6537762880325317, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8823014497756958, + "num_tokens": 422781522.0, + "step": 12041 + }, + { + "epoch": 2.2362116991643455, + "grad_norm": 1.5948240756988525, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8920491337776184, + "num_tokens": 422812735.0, + "step": 12042 + }, + { + "epoch": 2.236397400185701, + "grad_norm": 1.6613441705703735, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8823518753051758, + "num_tokens": 422846230.0, + "step": 12043 + }, + { + "epoch": 2.2365831012070565, + "grad_norm": 1.5901312828063965, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8891333341598511, + "num_tokens": 422880865.0, + "step": 12044 + }, + { + "epoch": 2.2367688022284122, + "grad_norm": 1.562117576599121, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8896346092224121, + "num_tokens": 422915013.0, + "step": 12045 + }, + { + "epoch": 2.236954503249768, + "grad_norm": 1.637529969215393, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8871525526046753, + "num_tokens": 422946858.0, + "step": 12046 + }, + { + "epoch": 2.2371402042711237, + "grad_norm": 1.471196174621582, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.890799343585968, + "num_tokens": 422988771.0, + "step": 12047 + }, + { + "epoch": 2.237325905292479, + "grad_norm": 1.507629156112671, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8901618123054504, + "num_tokens": 423027558.0, + "step": 12048 + }, + { + "epoch": 2.2375116063138347, + "grad_norm": 1.855695128440857, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8836319446563721, + "num_tokens": 423053515.0, + "step": 12049 + }, + { + "epoch": 2.2376973073351905, + "grad_norm": 1.5614969730377197, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8806784152984619, + "num_tokens": 423092940.0, + "step": 12050 + }, + { + "epoch": 2.2378830083565457, + "grad_norm": 1.601993441581726, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8887922167778015, + "num_tokens": 423124945.0, + "step": 12051 + }, + { + "epoch": 2.2380687093779015, + "grad_norm": 1.7057634592056274, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8754990100860596, + "num_tokens": 423155486.0, + "step": 12052 + }, + { + "epoch": 2.238254410399257, + "grad_norm": 1.5786653757095337, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8783477544784546, + "num_tokens": 423189974.0, + "step": 12053 + }, + { + "epoch": 2.238440111420613, + "grad_norm": 1.4265471696853638, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8932628631591797, + "num_tokens": 423228178.0, + "step": 12054 + }, + { + "epoch": 2.2386258124419682, + "grad_norm": 1.5201470851898193, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8954167366027832, + "num_tokens": 423261663.0, + "step": 12055 + }, + { + "epoch": 2.238811513463324, + "grad_norm": 1.506891131401062, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8846772313117981, + "num_tokens": 423298910.0, + "step": 12056 + }, + { + "epoch": 2.2389972144846797, + "grad_norm": 1.4537301063537598, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8892842531204224, + "num_tokens": 423337712.0, + "step": 12057 + }, + { + "epoch": 2.2391829155060354, + "grad_norm": 1.610843539237976, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8669559955596924, + "num_tokens": 423376094.0, + "step": 12058 + }, + { + "epoch": 2.2393686165273907, + "grad_norm": 1.5087528228759766, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8744250535964966, + "num_tokens": 423416045.0, + "step": 12059 + }, + { + "epoch": 2.2395543175487465, + "grad_norm": 1.4980204105377197, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8894197940826416, + "num_tokens": 423455480.0, + "step": 12060 + }, + { + "epoch": 2.239740018570102, + "grad_norm": 1.4398332834243774, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.8977010846138, + "num_tokens": 423490977.0, + "step": 12061 + }, + { + "epoch": 2.239925719591458, + "grad_norm": 1.5353800058364868, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8939545750617981, + "num_tokens": 423527411.0, + "step": 12062 + }, + { + "epoch": 2.240111420612813, + "grad_norm": 1.5936063528060913, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8857822418212891, + "num_tokens": 423560982.0, + "step": 12063 + }, + { + "epoch": 2.240297121634169, + "grad_norm": 1.4806839227676392, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8877279162406921, + "num_tokens": 423603658.0, + "step": 12064 + }, + { + "epoch": 2.2404828226555247, + "grad_norm": 1.6628162860870361, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8683574795722961, + "num_tokens": 423639550.0, + "step": 12065 + }, + { + "epoch": 2.2406685236768804, + "grad_norm": 1.4261871576309204, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8949635624885559, + "num_tokens": 423676303.0, + "step": 12066 + }, + { + "epoch": 2.2408542246982357, + "grad_norm": 1.4802988767623901, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8744043707847595, + "num_tokens": 423715098.0, + "step": 12067 + }, + { + "epoch": 2.2410399257195914, + "grad_norm": 1.5459243059158325, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8699803948402405, + "num_tokens": 423754333.0, + "step": 12068 + }, + { + "epoch": 2.241225626740947, + "grad_norm": 1.4238859415054321, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8882777094841003, + "num_tokens": 423794250.0, + "step": 12069 + }, + { + "epoch": 2.241411327762303, + "grad_norm": 1.6770358085632324, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.8933218121528625, + "num_tokens": 423823254.0, + "step": 12070 + }, + { + "epoch": 2.241597028783658, + "grad_norm": 1.6438908576965332, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8890854120254517, + "num_tokens": 423857023.0, + "step": 12071 + }, + { + "epoch": 2.241782729805014, + "grad_norm": 1.6467794179916382, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8953825235366821, + "num_tokens": 423885734.0, + "step": 12072 + }, + { + "epoch": 2.2419684308263697, + "grad_norm": 1.430436611175537, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8880857825279236, + "num_tokens": 423927463.0, + "step": 12073 + }, + { + "epoch": 2.242154131847725, + "grad_norm": 1.911837100982666, + "learning_rate": 1e-06, + "loss": 0.3974, + "mean_token_accuracy": 0.8599759340286255, + "num_tokens": 423954531.0, + "step": 12074 + }, + { + "epoch": 2.2423398328690807, + "grad_norm": 1.5450172424316406, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8881520628929138, + "num_tokens": 423989317.0, + "step": 12075 + }, + { + "epoch": 2.2425255338904364, + "grad_norm": 1.4791187047958374, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8804736137390137, + "num_tokens": 424031628.0, + "step": 12076 + }, + { + "epoch": 2.242711234911792, + "grad_norm": 1.5720323324203491, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8756991028785706, + "num_tokens": 424067464.0, + "step": 12077 + }, + { + "epoch": 2.2428969359331474, + "grad_norm": 1.5860466957092285, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8780866861343384, + "num_tokens": 424106007.0, + "step": 12078 + }, + { + "epoch": 2.243082636954503, + "grad_norm": 1.6532957553863525, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8791772127151489, + "num_tokens": 424138778.0, + "step": 12079 + }, + { + "epoch": 2.243268337975859, + "grad_norm": 1.5742278099060059, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.884783923625946, + "num_tokens": 424173862.0, + "step": 12080 + }, + { + "epoch": 2.2434540389972146, + "grad_norm": 1.6278395652770996, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8891238570213318, + "num_tokens": 424206381.0, + "step": 12081 + }, + { + "epoch": 2.24363974001857, + "grad_norm": 1.4878160953521729, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8806048035621643, + "num_tokens": 424247152.0, + "step": 12082 + }, + { + "epoch": 2.2438254410399256, + "grad_norm": 1.5244321823120117, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.886442244052887, + "num_tokens": 424286645.0, + "step": 12083 + }, + { + "epoch": 2.2440111420612814, + "grad_norm": 1.428857445716858, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8987032175064087, + "num_tokens": 424325819.0, + "step": 12084 + }, + { + "epoch": 2.244196843082637, + "grad_norm": 1.568043828010559, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8889586925506592, + "num_tokens": 424359516.0, + "step": 12085 + }, + { + "epoch": 2.2443825441039924, + "grad_norm": 1.764832615852356, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8868100643157959, + "num_tokens": 424387842.0, + "step": 12086 + }, + { + "epoch": 2.244568245125348, + "grad_norm": 1.5625786781311035, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8851302862167358, + "num_tokens": 424423516.0, + "step": 12087 + }, + { + "epoch": 2.244753946146704, + "grad_norm": 1.5126901865005493, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8946607708930969, + "num_tokens": 424459105.0, + "step": 12088 + }, + { + "epoch": 2.2449396471680596, + "grad_norm": 1.4793363809585571, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8930209875106812, + "num_tokens": 424499204.0, + "step": 12089 + }, + { + "epoch": 2.245125348189415, + "grad_norm": 1.692266821861267, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8836160898208618, + "num_tokens": 424532791.0, + "step": 12090 + }, + { + "epoch": 2.2453110492107706, + "grad_norm": 1.584473967552185, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8917662501335144, + "num_tokens": 424565608.0, + "step": 12091 + }, + { + "epoch": 2.2454967502321264, + "grad_norm": 1.5224885940551758, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8728847503662109, + "num_tokens": 424604265.0, + "step": 12092 + }, + { + "epoch": 2.245682451253482, + "grad_norm": 1.6167731285095215, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8817155361175537, + "num_tokens": 424642934.0, + "step": 12093 + }, + { + "epoch": 2.2458681522748374, + "grad_norm": 1.5363184213638306, + "learning_rate": 1e-06, + "loss": 0.2942, + "mean_token_accuracy": 0.8947529792785645, + "num_tokens": 424678534.0, + "step": 12094 + }, + { + "epoch": 2.246053853296193, + "grad_norm": 1.3544294834136963, + "learning_rate": 1e-06, + "loss": 0.2687, + "mean_token_accuracy": 0.905877947807312, + "num_tokens": 424718573.0, + "step": 12095 + }, + { + "epoch": 2.246239554317549, + "grad_norm": 1.4237349033355713, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8909405469894409, + "num_tokens": 424760521.0, + "step": 12096 + }, + { + "epoch": 2.246425255338904, + "grad_norm": 1.5505651235580444, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8902819752693176, + "num_tokens": 424793869.0, + "step": 12097 + }, + { + "epoch": 2.24661095636026, + "grad_norm": 1.6137324571609497, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.8758566379547119, + "num_tokens": 424830234.0, + "step": 12098 + }, + { + "epoch": 2.2467966573816156, + "grad_norm": 1.6250635385513306, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8857816457748413, + "num_tokens": 424862041.0, + "step": 12099 + }, + { + "epoch": 2.2469823584029713, + "grad_norm": 1.5413570404052734, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.887000560760498, + "num_tokens": 424899365.0, + "step": 12100 + }, + { + "epoch": 2.247168059424327, + "grad_norm": 1.556505560874939, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8854126930236816, + "num_tokens": 424935460.0, + "step": 12101 + }, + { + "epoch": 2.2473537604456824, + "grad_norm": 1.5491427183151245, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8772573471069336, + "num_tokens": 424971777.0, + "step": 12102 + }, + { + "epoch": 2.247539461467038, + "grad_norm": 1.4006119966506958, + "learning_rate": 1e-06, + "loss": 0.2863, + "mean_token_accuracy": 0.8980651497840881, + "num_tokens": 425009328.0, + "step": 12103 + }, + { + "epoch": 2.247725162488394, + "grad_norm": 1.7614710330963135, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.8961254954338074, + "num_tokens": 425035239.0, + "step": 12104 + }, + { + "epoch": 2.247910863509749, + "grad_norm": 1.5264191627502441, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8873875737190247, + "num_tokens": 425069631.0, + "step": 12105 + }, + { + "epoch": 2.248096564531105, + "grad_norm": 1.6275579929351807, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.8954201936721802, + "num_tokens": 425100672.0, + "step": 12106 + }, + { + "epoch": 2.2482822655524606, + "grad_norm": 1.6695846319198608, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8866226673126221, + "num_tokens": 425131217.0, + "step": 12107 + }, + { + "epoch": 2.2484679665738163, + "grad_norm": 1.6028014421463013, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8846114873886108, + "num_tokens": 425166087.0, + "step": 12108 + }, + { + "epoch": 2.2486536675951716, + "grad_norm": 1.595719814300537, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8844649791717529, + "num_tokens": 425198014.0, + "step": 12109 + }, + { + "epoch": 2.2488393686165273, + "grad_norm": 1.5097600221633911, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8767480254173279, + "num_tokens": 425235000.0, + "step": 12110 + }, + { + "epoch": 2.249025069637883, + "grad_norm": 1.4548429250717163, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8886517882347107, + "num_tokens": 425276647.0, + "step": 12111 + }, + { + "epoch": 2.249210770659239, + "grad_norm": 1.5352510213851929, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8848702907562256, + "num_tokens": 425313541.0, + "step": 12112 + }, + { + "epoch": 2.249396471680594, + "grad_norm": 1.473296880722046, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8971531987190247, + "num_tokens": 425349971.0, + "step": 12113 + }, + { + "epoch": 2.24958217270195, + "grad_norm": 1.7127680778503418, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8839797377586365, + "num_tokens": 425380940.0, + "step": 12114 + }, + { + "epoch": 2.2497678737233056, + "grad_norm": 1.4657440185546875, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8881782293319702, + "num_tokens": 425418554.0, + "step": 12115 + }, + { + "epoch": 2.2499535747446613, + "grad_norm": 1.6637132167816162, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8792365193367004, + "num_tokens": 425450058.0, + "step": 12116 + }, + { + "epoch": 2.2501392757660166, + "grad_norm": 1.6681478023529053, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8830510377883911, + "num_tokens": 425481710.0, + "step": 12117 + }, + { + "epoch": 2.2503249767873723, + "grad_norm": 1.6246910095214844, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.88295578956604, + "num_tokens": 425517620.0, + "step": 12118 + }, + { + "epoch": 2.250510677808728, + "grad_norm": 1.5573086738586426, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8822243213653564, + "num_tokens": 425554905.0, + "step": 12119 + }, + { + "epoch": 2.2506963788300833, + "grad_norm": 1.5077208280563354, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8916391134262085, + "num_tokens": 425590298.0, + "step": 12120 + }, + { + "epoch": 2.250882079851439, + "grad_norm": 1.6625943183898926, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8773394227027893, + "num_tokens": 425626122.0, + "step": 12121 + }, + { + "epoch": 2.251067780872795, + "grad_norm": 1.6589560508728027, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.890099048614502, + "num_tokens": 425654763.0, + "step": 12122 + }, + { + "epoch": 2.2512534818941505, + "grad_norm": 1.4984567165374756, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8956196308135986, + "num_tokens": 425687960.0, + "step": 12123 + }, + { + "epoch": 2.2514391829155063, + "grad_norm": 1.6992727518081665, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8741468787193298, + "num_tokens": 425720342.0, + "step": 12124 + }, + { + "epoch": 2.2516248839368616, + "grad_norm": 1.7580809593200684, + "learning_rate": 1e-06, + "loss": 0.2742, + "mean_token_accuracy": 0.9018882513046265, + "num_tokens": 425752934.0, + "step": 12125 + }, + { + "epoch": 2.2518105849582173, + "grad_norm": 1.4747825860977173, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8870161175727844, + "num_tokens": 425790873.0, + "step": 12126 + }, + { + "epoch": 2.251996285979573, + "grad_norm": 1.5840191841125488, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8904669880867004, + "num_tokens": 425821374.0, + "step": 12127 + }, + { + "epoch": 2.2521819870009283, + "grad_norm": 1.8361973762512207, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8741292953491211, + "num_tokens": 425850009.0, + "step": 12128 + }, + { + "epoch": 2.252367688022284, + "grad_norm": 1.5385829210281372, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8947031497955322, + "num_tokens": 425885173.0, + "step": 12129 + }, + { + "epoch": 2.2525533890436398, + "grad_norm": 1.479048252105713, + "learning_rate": 1e-06, + "loss": 0.2708, + "mean_token_accuracy": 0.8990576267242432, + "num_tokens": 425920053.0, + "step": 12130 + }, + { + "epoch": 2.2527390900649955, + "grad_norm": 1.6737388372421265, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8748394250869751, + "num_tokens": 425958449.0, + "step": 12131 + }, + { + "epoch": 2.252924791086351, + "grad_norm": 1.7047406435012817, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.885162353515625, + "num_tokens": 425991448.0, + "step": 12132 + }, + { + "epoch": 2.2531104921077065, + "grad_norm": 1.6258087158203125, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8856490850448608, + "num_tokens": 426021457.0, + "step": 12133 + }, + { + "epoch": 2.2532961931290623, + "grad_norm": 1.7001153230667114, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8994505405426025, + "num_tokens": 426048838.0, + "step": 12134 + }, + { + "epoch": 2.253481894150418, + "grad_norm": 1.5009214878082275, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8907944560050964, + "num_tokens": 426084431.0, + "step": 12135 + }, + { + "epoch": 2.2536675951717733, + "grad_norm": 1.518064260482788, + "learning_rate": 1e-06, + "loss": 0.2812, + "mean_token_accuracy": 0.9013752341270447, + "num_tokens": 426122069.0, + "step": 12136 + }, + { + "epoch": 2.253853296193129, + "grad_norm": 1.415069341659546, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8823994994163513, + "num_tokens": 426163841.0, + "step": 12137 + }, + { + "epoch": 2.2540389972144848, + "grad_norm": 1.6729661226272583, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8793774843215942, + "num_tokens": 426195145.0, + "step": 12138 + }, + { + "epoch": 2.2542246982358405, + "grad_norm": 1.4786207675933838, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8921398520469666, + "num_tokens": 426233256.0, + "step": 12139 + }, + { + "epoch": 2.2544103992571958, + "grad_norm": 1.5922729969024658, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8914718627929688, + "num_tokens": 426265744.0, + "step": 12140 + }, + { + "epoch": 2.2545961002785515, + "grad_norm": 1.6155112981796265, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8888784646987915, + "num_tokens": 426301834.0, + "step": 12141 + }, + { + "epoch": 2.2547818012999072, + "grad_norm": 1.7166754007339478, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8901597261428833, + "num_tokens": 426335727.0, + "step": 12142 + }, + { + "epoch": 2.2549675023212625, + "grad_norm": 1.6804208755493164, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8771721124649048, + "num_tokens": 426369182.0, + "step": 12143 + }, + { + "epoch": 2.2551532033426183, + "grad_norm": 1.5727016925811768, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8921654224395752, + "num_tokens": 426401028.0, + "step": 12144 + }, + { + "epoch": 2.255338904363974, + "grad_norm": 1.4909543991088867, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.891022801399231, + "num_tokens": 426438549.0, + "step": 12145 + }, + { + "epoch": 2.2555246053853297, + "grad_norm": 1.5508396625518799, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8869656324386597, + "num_tokens": 426475944.0, + "step": 12146 + }, + { + "epoch": 2.2557103064066855, + "grad_norm": 1.646232008934021, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8848275542259216, + "num_tokens": 426509249.0, + "step": 12147 + }, + { + "epoch": 2.2558960074280408, + "grad_norm": 2.105811357498169, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8875426054000854, + "num_tokens": 426538447.0, + "step": 12148 + }, + { + "epoch": 2.2560817084493965, + "grad_norm": 1.5619755983352661, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8928172588348389, + "num_tokens": 426572950.0, + "step": 12149 + }, + { + "epoch": 2.256267409470752, + "grad_norm": 1.7270567417144775, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8784904479980469, + "num_tokens": 426601006.0, + "step": 12150 + }, + { + "epoch": 2.2564531104921075, + "grad_norm": 1.6035867929458618, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8798172473907471, + "num_tokens": 426636947.0, + "step": 12151 + }, + { + "epoch": 2.2566388115134632, + "grad_norm": 1.7275221347808838, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8940857648849487, + "num_tokens": 426667739.0, + "step": 12152 + }, + { + "epoch": 2.256824512534819, + "grad_norm": 1.5386475324630737, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8894314169883728, + "num_tokens": 426703050.0, + "step": 12153 + }, + { + "epoch": 2.2570102135561747, + "grad_norm": 1.518730640411377, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.8957341909408569, + "num_tokens": 426735064.0, + "step": 12154 + }, + { + "epoch": 2.25719591457753, + "grad_norm": 1.4075837135314941, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8902057409286499, + "num_tokens": 426773982.0, + "step": 12155 + }, + { + "epoch": 2.2573816155988857, + "grad_norm": 1.5908035039901733, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8878033757209778, + "num_tokens": 426809732.0, + "step": 12156 + }, + { + "epoch": 2.2575673166202415, + "grad_norm": 1.5951558351516724, + "learning_rate": 1e-06, + "loss": 0.292, + "mean_token_accuracy": 0.8949783444404602, + "num_tokens": 426844908.0, + "step": 12157 + }, + { + "epoch": 2.257753017641597, + "grad_norm": 1.4446417093276978, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8890717029571533, + "num_tokens": 426886104.0, + "step": 12158 + }, + { + "epoch": 2.2579387186629525, + "grad_norm": 1.6718255281448364, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8855083584785461, + "num_tokens": 426920236.0, + "step": 12159 + }, + { + "epoch": 2.258124419684308, + "grad_norm": 1.5397956371307373, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8855138421058655, + "num_tokens": 426955164.0, + "step": 12160 + }, + { + "epoch": 2.258310120705664, + "grad_norm": 1.5500025749206543, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8850889205932617, + "num_tokens": 426990334.0, + "step": 12161 + }, + { + "epoch": 2.2584958217270197, + "grad_norm": 1.6379176378250122, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8917551040649414, + "num_tokens": 427024285.0, + "step": 12162 + }, + { + "epoch": 2.258681522748375, + "grad_norm": 1.7446532249450684, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8909420967102051, + "num_tokens": 427052730.0, + "step": 12163 + }, + { + "epoch": 2.2588672237697307, + "grad_norm": 1.817351222038269, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8927263021469116, + "num_tokens": 427080571.0, + "step": 12164 + }, + { + "epoch": 2.2590529247910864, + "grad_norm": 1.5326608419418335, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.895919919013977, + "num_tokens": 427115893.0, + "step": 12165 + }, + { + "epoch": 2.2592386258124417, + "grad_norm": 1.6169276237487793, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8972641229629517, + "num_tokens": 427149245.0, + "step": 12166 + }, + { + "epoch": 2.2594243268337975, + "grad_norm": 1.6637492179870605, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8725013732910156, + "num_tokens": 427184704.0, + "step": 12167 + }, + { + "epoch": 2.259610027855153, + "grad_norm": 1.5055299997329712, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8891086578369141, + "num_tokens": 427221100.0, + "step": 12168 + }, + { + "epoch": 2.259795728876509, + "grad_norm": 1.6596976518630981, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8925535678863525, + "num_tokens": 427253450.0, + "step": 12169 + }, + { + "epoch": 2.2599814298978647, + "grad_norm": 1.6916357278823853, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8742506504058838, + "num_tokens": 427292133.0, + "step": 12170 + }, + { + "epoch": 2.26016713091922, + "grad_norm": 1.3761731386184692, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.892859160900116, + "num_tokens": 427334444.0, + "step": 12171 + }, + { + "epoch": 2.2603528319405757, + "grad_norm": 1.5446819067001343, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8774678707122803, + "num_tokens": 427368960.0, + "step": 12172 + }, + { + "epoch": 2.2605385329619314, + "grad_norm": 1.6439931392669678, + "learning_rate": 1e-06, + "loss": 0.3001, + "mean_token_accuracy": 0.8926251530647278, + "num_tokens": 427398452.0, + "step": 12173 + }, + { + "epoch": 2.2607242339832867, + "grad_norm": 1.4543989896774292, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8917378783226013, + "num_tokens": 427434939.0, + "step": 12174 + }, + { + "epoch": 2.2609099350046424, + "grad_norm": 1.6667115688323975, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.893220067024231, + "num_tokens": 427464480.0, + "step": 12175 + }, + { + "epoch": 2.261095636025998, + "grad_norm": 1.5983059406280518, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8866473436355591, + "num_tokens": 427500515.0, + "step": 12176 + }, + { + "epoch": 2.261281337047354, + "grad_norm": 1.517259120941162, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8969345092773438, + "num_tokens": 427533910.0, + "step": 12177 + }, + { + "epoch": 2.261467038068709, + "grad_norm": 1.4537814855575562, + "learning_rate": 1e-06, + "loss": 0.2803, + "mean_token_accuracy": 0.8991734981536865, + "num_tokens": 427570735.0, + "step": 12178 + }, + { + "epoch": 2.261652739090065, + "grad_norm": 1.6494104862213135, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8882045149803162, + "num_tokens": 427603190.0, + "step": 12179 + }, + { + "epoch": 2.2618384401114207, + "grad_norm": 1.571170449256897, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8940924406051636, + "num_tokens": 427638451.0, + "step": 12180 + }, + { + "epoch": 2.2620241411327764, + "grad_norm": 1.602232575416565, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8908467292785645, + "num_tokens": 427671233.0, + "step": 12181 + }, + { + "epoch": 2.2622098421541317, + "grad_norm": 1.6781548261642456, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8949941992759705, + "num_tokens": 427703253.0, + "step": 12182 + }, + { + "epoch": 2.2623955431754874, + "grad_norm": 1.6457676887512207, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8690425157546997, + "num_tokens": 427740178.0, + "step": 12183 + }, + { + "epoch": 2.262581244196843, + "grad_norm": 1.8558111190795898, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8865673542022705, + "num_tokens": 427771136.0, + "step": 12184 + }, + { + "epoch": 2.262766945218199, + "grad_norm": 1.7633559703826904, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8903191685676575, + "num_tokens": 427799362.0, + "step": 12185 + }, + { + "epoch": 2.262952646239554, + "grad_norm": 1.498097538948059, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8956148028373718, + "num_tokens": 427833421.0, + "step": 12186 + }, + { + "epoch": 2.26313834726091, + "grad_norm": 1.7618802785873413, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8716333508491516, + "num_tokens": 427864171.0, + "step": 12187 + }, + { + "epoch": 2.2633240482822656, + "grad_norm": 1.7210749387741089, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8753227591514587, + "num_tokens": 427895159.0, + "step": 12188 + }, + { + "epoch": 2.263509749303621, + "grad_norm": 1.6150637865066528, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8883445262908936, + "num_tokens": 427929571.0, + "step": 12189 + }, + { + "epoch": 2.2636954503249767, + "grad_norm": 1.7213283777236938, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8805745840072632, + "num_tokens": 427961497.0, + "step": 12190 + }, + { + "epoch": 2.2638811513463324, + "grad_norm": 1.62384033203125, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8945446014404297, + "num_tokens": 427995582.0, + "step": 12191 + }, + { + "epoch": 2.264066852367688, + "grad_norm": 1.5028661489486694, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8954675197601318, + "num_tokens": 428033950.0, + "step": 12192 + }, + { + "epoch": 2.264252553389044, + "grad_norm": 1.5412845611572266, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8805322647094727, + "num_tokens": 428070988.0, + "step": 12193 + }, + { + "epoch": 2.264438254410399, + "grad_norm": 1.4954882860183716, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8782612085342407, + "num_tokens": 428111482.0, + "step": 12194 + }, + { + "epoch": 2.264623955431755, + "grad_norm": 1.5470812320709229, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8855760097503662, + "num_tokens": 428146685.0, + "step": 12195 + }, + { + "epoch": 2.2648096564531106, + "grad_norm": 1.4860827922821045, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8798489570617676, + "num_tokens": 428186618.0, + "step": 12196 + }, + { + "epoch": 2.264995357474466, + "grad_norm": 1.6206530332565308, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8772448301315308, + "num_tokens": 428226375.0, + "step": 12197 + }, + { + "epoch": 2.2651810584958216, + "grad_norm": 1.5666437149047852, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8779270052909851, + "num_tokens": 428266120.0, + "step": 12198 + }, + { + "epoch": 2.2653667595171774, + "grad_norm": 1.4939907789230347, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8902381658554077, + "num_tokens": 428304932.0, + "step": 12199 + }, + { + "epoch": 2.265552460538533, + "grad_norm": 1.5817214250564575, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8863383531570435, + "num_tokens": 428343617.0, + "step": 12200 + }, + { + "epoch": 2.2657381615598884, + "grad_norm": 1.5539973974227905, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8892101049423218, + "num_tokens": 428377368.0, + "step": 12201 + }, + { + "epoch": 2.265923862581244, + "grad_norm": 1.6086679697036743, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8973637819290161, + "num_tokens": 428409125.0, + "step": 12202 + }, + { + "epoch": 2.2661095636026, + "grad_norm": 1.6664440631866455, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8849930763244629, + "num_tokens": 428445015.0, + "step": 12203 + }, + { + "epoch": 2.2662952646239556, + "grad_norm": 1.5159705877304077, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8734909296035767, + "num_tokens": 428483538.0, + "step": 12204 + }, + { + "epoch": 2.266480965645311, + "grad_norm": 1.5282994508743286, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.894586443901062, + "num_tokens": 428519727.0, + "step": 12205 + }, + { + "epoch": 2.2666666666666666, + "grad_norm": 1.531085729598999, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8913761377334595, + "num_tokens": 428554106.0, + "step": 12206 + }, + { + "epoch": 2.2668523676880223, + "grad_norm": 1.5646659135818481, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8774323463439941, + "num_tokens": 428588766.0, + "step": 12207 + }, + { + "epoch": 2.267038068709378, + "grad_norm": 1.6470073461532593, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8834096789360046, + "num_tokens": 428623168.0, + "step": 12208 + }, + { + "epoch": 2.2672237697307334, + "grad_norm": 1.705379605293274, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8959683775901794, + "num_tokens": 428654066.0, + "step": 12209 + }, + { + "epoch": 2.267409470752089, + "grad_norm": 1.4997962713241577, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8946401476860046, + "num_tokens": 428687180.0, + "step": 12210 + }, + { + "epoch": 2.267595171773445, + "grad_norm": 1.6633625030517578, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8834834694862366, + "num_tokens": 428716573.0, + "step": 12211 + }, + { + "epoch": 2.2677808727948006, + "grad_norm": 1.4544990062713623, + "learning_rate": 1e-06, + "loss": 0.2768, + "mean_token_accuracy": 0.900505006313324, + "num_tokens": 428753392.0, + "step": 12212 + }, + { + "epoch": 2.267966573816156, + "grad_norm": 1.573079228401184, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8821122646331787, + "num_tokens": 428791785.0, + "step": 12213 + }, + { + "epoch": 2.2681522748375116, + "grad_norm": 1.61318039894104, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8931020498275757, + "num_tokens": 428821212.0, + "step": 12214 + }, + { + "epoch": 2.2683379758588673, + "grad_norm": 1.511172890663147, + "learning_rate": 1e-06, + "loss": 0.3064, + "mean_token_accuracy": 0.8908997178077698, + "num_tokens": 428857196.0, + "step": 12215 + }, + { + "epoch": 2.268523676880223, + "grad_norm": 1.5134634971618652, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8866151571273804, + "num_tokens": 428891747.0, + "step": 12216 + }, + { + "epoch": 2.2687093779015783, + "grad_norm": 1.6537622213363647, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8861278295516968, + "num_tokens": 428926427.0, + "step": 12217 + }, + { + "epoch": 2.268895078922934, + "grad_norm": 1.8037041425704956, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8811917901039124, + "num_tokens": 428956284.0, + "step": 12218 + }, + { + "epoch": 2.26908077994429, + "grad_norm": 1.4931917190551758, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8891629576683044, + "num_tokens": 428995089.0, + "step": 12219 + }, + { + "epoch": 2.269266480965645, + "grad_norm": 1.4103820323944092, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8877400755882263, + "num_tokens": 429034681.0, + "step": 12220 + }, + { + "epoch": 2.269452181987001, + "grad_norm": 1.4690574407577515, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8851227164268494, + "num_tokens": 429074475.0, + "step": 12221 + }, + { + "epoch": 2.2696378830083566, + "grad_norm": 1.6357775926589966, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8829294443130493, + "num_tokens": 429107272.0, + "step": 12222 + }, + { + "epoch": 2.2698235840297123, + "grad_norm": 1.520670771598816, + "learning_rate": 1e-06, + "loss": 0.2696, + "mean_token_accuracy": 0.9034984707832336, + "num_tokens": 429140673.0, + "step": 12223 + }, + { + "epoch": 2.270009285051068, + "grad_norm": 1.5532668828964233, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8778071403503418, + "num_tokens": 429178311.0, + "step": 12224 + }, + { + "epoch": 2.2701949860724233, + "grad_norm": 2.173011302947998, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8892045021057129, + "num_tokens": 429210164.0, + "step": 12225 + }, + { + "epoch": 2.270380687093779, + "grad_norm": 1.5177546739578247, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8823570013046265, + "num_tokens": 429247322.0, + "step": 12226 + }, + { + "epoch": 2.270566388115135, + "grad_norm": 1.606881022453308, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.878749668598175, + "num_tokens": 429282193.0, + "step": 12227 + }, + { + "epoch": 2.27075208913649, + "grad_norm": 1.4900740385055542, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8885480761528015, + "num_tokens": 429319165.0, + "step": 12228 + }, + { + "epoch": 2.270937790157846, + "grad_norm": 1.742431879043579, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8813953995704651, + "num_tokens": 429350800.0, + "step": 12229 + }, + { + "epoch": 2.2711234911792015, + "grad_norm": 1.7448034286499023, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8807632923126221, + "num_tokens": 429379928.0, + "step": 12230 + }, + { + "epoch": 2.2713091922005573, + "grad_norm": 1.550249695777893, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8697469830513, + "num_tokens": 429418554.0, + "step": 12231 + }, + { + "epoch": 2.2714948932219126, + "grad_norm": 1.5579047203063965, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8842837810516357, + "num_tokens": 429453002.0, + "step": 12232 + }, + { + "epoch": 2.2716805942432683, + "grad_norm": 1.5381826162338257, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8961671590805054, + "num_tokens": 429488983.0, + "step": 12233 + }, + { + "epoch": 2.271866295264624, + "grad_norm": 1.4848260879516602, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8929952383041382, + "num_tokens": 429527712.0, + "step": 12234 + }, + { + "epoch": 2.2720519962859798, + "grad_norm": 1.6980382204055786, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.886176347732544, + "num_tokens": 429563028.0, + "step": 12235 + }, + { + "epoch": 2.272237697307335, + "grad_norm": 1.405369758605957, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8935908079147339, + "num_tokens": 429601089.0, + "step": 12236 + }, + { + "epoch": 2.272423398328691, + "grad_norm": 1.5578380823135376, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8823113441467285, + "num_tokens": 429637644.0, + "step": 12237 + }, + { + "epoch": 2.2726090993500465, + "grad_norm": 1.5304124355316162, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8907544612884521, + "num_tokens": 429673424.0, + "step": 12238 + }, + { + "epoch": 2.2727948003714022, + "grad_norm": 1.4282020330429077, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8850574493408203, + "num_tokens": 429711829.0, + "step": 12239 + }, + { + "epoch": 2.2729805013927575, + "grad_norm": 1.592235803604126, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8837222456932068, + "num_tokens": 429752084.0, + "step": 12240 + }, + { + "epoch": 2.2731662024141133, + "grad_norm": 1.5329242944717407, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8888252973556519, + "num_tokens": 429787528.0, + "step": 12241 + }, + { + "epoch": 2.273351903435469, + "grad_norm": 1.4161938428878784, + "learning_rate": 1e-06, + "loss": 0.2933, + "mean_token_accuracy": 0.895228922367096, + "num_tokens": 429827136.0, + "step": 12242 + }, + { + "epoch": 2.2735376044568243, + "grad_norm": 1.582996129989624, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8858211040496826, + "num_tokens": 429865554.0, + "step": 12243 + }, + { + "epoch": 2.27372330547818, + "grad_norm": 1.44998037815094, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8860387802124023, + "num_tokens": 429907057.0, + "step": 12244 + }, + { + "epoch": 2.2739090064995358, + "grad_norm": 1.5623894929885864, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8905351758003235, + "num_tokens": 429938042.0, + "step": 12245 + }, + { + "epoch": 2.2740947075208915, + "grad_norm": 1.536603331565857, + "learning_rate": 1e-06, + "loss": 0.2966, + "mean_token_accuracy": 0.8986644148826599, + "num_tokens": 429969290.0, + "step": 12246 + }, + { + "epoch": 2.2742804085422472, + "grad_norm": 1.5749542713165283, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8832961916923523, + "num_tokens": 430007105.0, + "step": 12247 + }, + { + "epoch": 2.2744661095636025, + "grad_norm": 1.7065927982330322, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8880082964897156, + "num_tokens": 430034525.0, + "step": 12248 + }, + { + "epoch": 2.2746518105849582, + "grad_norm": 1.5522518157958984, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.8941014409065247, + "num_tokens": 430070624.0, + "step": 12249 + }, + { + "epoch": 2.274837511606314, + "grad_norm": 1.6175628900527954, + "learning_rate": 1e-06, + "loss": 0.3943, + "mean_token_accuracy": 0.869303286075592, + "num_tokens": 430107743.0, + "step": 12250 + }, + { + "epoch": 2.2750232126276693, + "grad_norm": 1.623608946800232, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8790062665939331, + "num_tokens": 430138857.0, + "step": 12251 + }, + { + "epoch": 2.275208913649025, + "grad_norm": 1.7125407457351685, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8829425573348999, + "num_tokens": 430170093.0, + "step": 12252 + }, + { + "epoch": 2.2753946146703807, + "grad_norm": 1.6194194555282593, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8937678337097168, + "num_tokens": 430203501.0, + "step": 12253 + }, + { + "epoch": 2.2755803156917365, + "grad_norm": 1.5325429439544678, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8850792646408081, + "num_tokens": 430237211.0, + "step": 12254 + }, + { + "epoch": 2.2757660167130918, + "grad_norm": 1.504394292831421, + "learning_rate": 1e-06, + "loss": 0.2712, + "mean_token_accuracy": 0.903079092502594, + "num_tokens": 430273483.0, + "step": 12255 + }, + { + "epoch": 2.2759517177344475, + "grad_norm": 1.5135891437530518, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.8976833820343018, + "num_tokens": 430310228.0, + "step": 12256 + }, + { + "epoch": 2.276137418755803, + "grad_norm": 1.6962497234344482, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8808905482292175, + "num_tokens": 430339570.0, + "step": 12257 + }, + { + "epoch": 2.276323119777159, + "grad_norm": 1.7504600286483765, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8907197117805481, + "num_tokens": 430366936.0, + "step": 12258 + }, + { + "epoch": 2.2765088207985142, + "grad_norm": 1.587257981300354, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8753823041915894, + "num_tokens": 430405027.0, + "step": 12259 + }, + { + "epoch": 2.27669452181987, + "grad_norm": 1.544203519821167, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8857293128967285, + "num_tokens": 430440956.0, + "step": 12260 + }, + { + "epoch": 2.2768802228412257, + "grad_norm": 1.662344217300415, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.8643685579299927, + "num_tokens": 430474369.0, + "step": 12261 + }, + { + "epoch": 2.2770659238625814, + "grad_norm": 1.711890697479248, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8825105428695679, + "num_tokens": 430504531.0, + "step": 12262 + }, + { + "epoch": 2.2772516248839367, + "grad_norm": 1.6253767013549805, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8893040418624878, + "num_tokens": 430536308.0, + "step": 12263 + }, + { + "epoch": 2.2774373259052925, + "grad_norm": 1.6695098876953125, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8857787251472473, + "num_tokens": 430567926.0, + "step": 12264 + }, + { + "epoch": 2.277623026926648, + "grad_norm": 1.666090488433838, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8679929971694946, + "num_tokens": 430602904.0, + "step": 12265 + }, + { + "epoch": 2.2778087279480035, + "grad_norm": 1.7059412002563477, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8811926245689392, + "num_tokens": 430637331.0, + "step": 12266 + }, + { + "epoch": 2.277994428969359, + "grad_norm": 1.5368807315826416, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8723875284194946, + "num_tokens": 430674452.0, + "step": 12267 + }, + { + "epoch": 2.278180129990715, + "grad_norm": 1.5568550825119019, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8747668862342834, + "num_tokens": 430711330.0, + "step": 12268 + }, + { + "epoch": 2.2783658310120707, + "grad_norm": 1.5235891342163086, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8865772485733032, + "num_tokens": 430748762.0, + "step": 12269 + }, + { + "epoch": 2.2785515320334264, + "grad_norm": 1.6710566282272339, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8832944631576538, + "num_tokens": 430780935.0, + "step": 12270 + }, + { + "epoch": 2.2787372330547817, + "grad_norm": 1.536627173423767, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8934720754623413, + "num_tokens": 430819631.0, + "step": 12271 + }, + { + "epoch": 2.2789229340761374, + "grad_norm": 1.402656078338623, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8845446109771729, + "num_tokens": 430861152.0, + "step": 12272 + }, + { + "epoch": 2.279108635097493, + "grad_norm": 1.6061192750930786, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8774285316467285, + "num_tokens": 430901820.0, + "step": 12273 + }, + { + "epoch": 2.2792943361188485, + "grad_norm": 1.5004069805145264, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.901411771774292, + "num_tokens": 430936353.0, + "step": 12274 + }, + { + "epoch": 2.279480037140204, + "grad_norm": 1.558941125869751, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.888857901096344, + "num_tokens": 430968683.0, + "step": 12275 + }, + { + "epoch": 2.27966573816156, + "grad_norm": 1.6590250730514526, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8831824064254761, + "num_tokens": 431000374.0, + "step": 12276 + }, + { + "epoch": 2.2798514391829157, + "grad_norm": 1.6722055673599243, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8857739567756653, + "num_tokens": 431032161.0, + "step": 12277 + }, + { + "epoch": 2.280037140204271, + "grad_norm": 1.478177785873413, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8946330547332764, + "num_tokens": 431070551.0, + "step": 12278 + }, + { + "epoch": 2.2802228412256267, + "grad_norm": 1.520052433013916, + "learning_rate": 1e-06, + "loss": 0.2699, + "mean_token_accuracy": 0.9046993255615234, + "num_tokens": 431101736.0, + "step": 12279 + }, + { + "epoch": 2.2804085422469824, + "grad_norm": 1.5377264022827148, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.9015777111053467, + "num_tokens": 431135602.0, + "step": 12280 + }, + { + "epoch": 2.280594243268338, + "grad_norm": 1.5753206014633179, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8943945169448853, + "num_tokens": 431171743.0, + "step": 12281 + }, + { + "epoch": 2.2807799442896934, + "grad_norm": 1.6049717664718628, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.880887508392334, + "num_tokens": 431205721.0, + "step": 12282 + }, + { + "epoch": 2.280965645311049, + "grad_norm": 1.6232355833053589, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8931521773338318, + "num_tokens": 431237198.0, + "step": 12283 + }, + { + "epoch": 2.281151346332405, + "grad_norm": 1.564820647239685, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8759282827377319, + "num_tokens": 431273044.0, + "step": 12284 + }, + { + "epoch": 2.2813370473537606, + "grad_norm": 1.6031452417373657, + "learning_rate": 1e-06, + "loss": 0.3084, + "mean_token_accuracy": 0.8912947773933411, + "num_tokens": 431306522.0, + "step": 12285 + }, + { + "epoch": 2.281522748375116, + "grad_norm": 1.478391408920288, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8894423246383667, + "num_tokens": 431344480.0, + "step": 12286 + }, + { + "epoch": 2.2817084493964717, + "grad_norm": 1.5391396284103394, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8944121599197388, + "num_tokens": 431380183.0, + "step": 12287 + }, + { + "epoch": 2.2818941504178274, + "grad_norm": 1.4705660343170166, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.883134663105011, + "num_tokens": 431421593.0, + "step": 12288 + }, + { + "epoch": 2.2820798514391827, + "grad_norm": 1.6683367490768433, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8756111860275269, + "num_tokens": 431453770.0, + "step": 12289 + }, + { + "epoch": 2.2822655524605384, + "grad_norm": 1.5221545696258545, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8888041973114014, + "num_tokens": 431492016.0, + "step": 12290 + }, + { + "epoch": 2.282451253481894, + "grad_norm": 1.5993428230285645, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8930788636207581, + "num_tokens": 431521939.0, + "step": 12291 + }, + { + "epoch": 2.28263695450325, + "grad_norm": 1.4044189453125, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8951544761657715, + "num_tokens": 431559624.0, + "step": 12292 + }, + { + "epoch": 2.2828226555246056, + "grad_norm": 1.475150465965271, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8840688467025757, + "num_tokens": 431597115.0, + "step": 12293 + }, + { + "epoch": 2.283008356545961, + "grad_norm": 1.5070260763168335, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.891667902469635, + "num_tokens": 431636121.0, + "step": 12294 + }, + { + "epoch": 2.2831940575673166, + "grad_norm": 1.5371836423873901, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8890095949172974, + "num_tokens": 431673270.0, + "step": 12295 + }, + { + "epoch": 2.2833797585886724, + "grad_norm": 1.5204510688781738, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8833431601524353, + "num_tokens": 431710060.0, + "step": 12296 + }, + { + "epoch": 2.2835654596100277, + "grad_norm": 1.6729402542114258, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8867537379264832, + "num_tokens": 431744720.0, + "step": 12297 + }, + { + "epoch": 2.2837511606313834, + "grad_norm": 1.616838812828064, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8717289566993713, + "num_tokens": 431783727.0, + "step": 12298 + }, + { + "epoch": 2.283936861652739, + "grad_norm": 1.5002440214157104, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8940194249153137, + "num_tokens": 431819798.0, + "step": 12299 + }, + { + "epoch": 2.284122562674095, + "grad_norm": 1.5903723239898682, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8800783753395081, + "num_tokens": 431858207.0, + "step": 12300 + }, + { + "epoch": 2.28430826369545, + "grad_norm": 1.61726713180542, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8906921148300171, + "num_tokens": 431890960.0, + "step": 12301 + }, + { + "epoch": 2.284493964716806, + "grad_norm": 1.6157498359680176, + "learning_rate": 1e-06, + "loss": 0.3814, + "mean_token_accuracy": 0.8717743158340454, + "num_tokens": 431928947.0, + "step": 12302 + }, + { + "epoch": 2.2846796657381616, + "grad_norm": 1.5248217582702637, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8816025257110596, + "num_tokens": 431967548.0, + "step": 12303 + }, + { + "epoch": 2.2848653667595173, + "grad_norm": 1.7386177778244019, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.878825843334198, + "num_tokens": 431996449.0, + "step": 12304 + }, + { + "epoch": 2.2850510677808726, + "grad_norm": 1.753840446472168, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8850083351135254, + "num_tokens": 432030756.0, + "step": 12305 + }, + { + "epoch": 2.2852367688022284, + "grad_norm": 1.528398036956787, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8950316905975342, + "num_tokens": 432065603.0, + "step": 12306 + }, + { + "epoch": 2.285422469823584, + "grad_norm": 1.5671626329421997, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8811242580413818, + "num_tokens": 432101463.0, + "step": 12307 + }, + { + "epoch": 2.28560817084494, + "grad_norm": 1.4815865755081177, + "learning_rate": 1e-06, + "loss": 0.2877, + "mean_token_accuracy": 0.8968467712402344, + "num_tokens": 432139509.0, + "step": 12308 + }, + { + "epoch": 2.285793871866295, + "grad_norm": 1.6159695386886597, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.8910724520683289, + "num_tokens": 432171139.0, + "step": 12309 + }, + { + "epoch": 2.285979572887651, + "grad_norm": 1.5015013217926025, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8924086093902588, + "num_tokens": 432208374.0, + "step": 12310 + }, + { + "epoch": 2.2861652739090066, + "grad_norm": 1.6006007194519043, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8970810174942017, + "num_tokens": 432239727.0, + "step": 12311 + }, + { + "epoch": 2.286350974930362, + "grad_norm": 1.500307321548462, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8785127401351929, + "num_tokens": 432276656.0, + "step": 12312 + }, + { + "epoch": 2.2865366759517176, + "grad_norm": 1.5401524305343628, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.888073205947876, + "num_tokens": 432314158.0, + "step": 12313 + }, + { + "epoch": 2.2867223769730733, + "grad_norm": 1.420615553855896, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.9018245339393616, + "num_tokens": 432350411.0, + "step": 12314 + }, + { + "epoch": 2.286908077994429, + "grad_norm": 1.490576982498169, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8757798671722412, + "num_tokens": 432387586.0, + "step": 12315 + }, + { + "epoch": 2.287093779015785, + "grad_norm": 1.4664697647094727, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8911012411117554, + "num_tokens": 432425738.0, + "step": 12316 + }, + { + "epoch": 2.28727948003714, + "grad_norm": 1.5855134725570679, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.873274564743042, + "num_tokens": 432461619.0, + "step": 12317 + }, + { + "epoch": 2.287465181058496, + "grad_norm": 1.529215931892395, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8813480734825134, + "num_tokens": 432500372.0, + "step": 12318 + }, + { + "epoch": 2.2876508820798516, + "grad_norm": 1.6163556575775146, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8725506067276001, + "num_tokens": 432537154.0, + "step": 12319 + }, + { + "epoch": 2.287836583101207, + "grad_norm": 1.5660040378570557, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.881371021270752, + "num_tokens": 432571392.0, + "step": 12320 + }, + { + "epoch": 2.2880222841225626, + "grad_norm": 1.52176833152771, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8889930248260498, + "num_tokens": 432606558.0, + "step": 12321 + }, + { + "epoch": 2.2882079851439183, + "grad_norm": 1.531218409538269, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.9030987024307251, + "num_tokens": 432637332.0, + "step": 12322 + }, + { + "epoch": 2.288393686165274, + "grad_norm": 1.5552986860275269, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8803608417510986, + "num_tokens": 432674042.0, + "step": 12323 + }, + { + "epoch": 2.2885793871866293, + "grad_norm": 1.8797551393508911, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.9012879133224487, + "num_tokens": 432696758.0, + "step": 12324 + }, + { + "epoch": 2.288765088207985, + "grad_norm": 1.5279619693756104, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8902949094772339, + "num_tokens": 432730960.0, + "step": 12325 + }, + { + "epoch": 2.288950789229341, + "grad_norm": 1.6520637273788452, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8973289132118225, + "num_tokens": 432759325.0, + "step": 12326 + }, + { + "epoch": 2.2891364902506965, + "grad_norm": 1.561722755432129, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8768473863601685, + "num_tokens": 432795424.0, + "step": 12327 + }, + { + "epoch": 2.289322191272052, + "grad_norm": 1.5197627544403076, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8817370533943176, + "num_tokens": 432836313.0, + "step": 12328 + }, + { + "epoch": 2.2895078922934076, + "grad_norm": 1.4892345666885376, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8962187767028809, + "num_tokens": 432872573.0, + "step": 12329 + }, + { + "epoch": 2.2896935933147633, + "grad_norm": 1.5190457105636597, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8794494867324829, + "num_tokens": 432909870.0, + "step": 12330 + }, + { + "epoch": 2.289879294336119, + "grad_norm": 1.4840351343154907, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.883319616317749, + "num_tokens": 432950281.0, + "step": 12331 + }, + { + "epoch": 2.2900649953574743, + "grad_norm": 1.5971627235412598, + "learning_rate": 1e-06, + "loss": 0.3181, + "mean_token_accuracy": 0.8880295753479004, + "num_tokens": 432986703.0, + "step": 12332 + }, + { + "epoch": 2.29025069637883, + "grad_norm": 1.539900541305542, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8870701789855957, + "num_tokens": 433027597.0, + "step": 12333 + }, + { + "epoch": 2.290436397400186, + "grad_norm": 1.6231948137283325, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8813352584838867, + "num_tokens": 433064148.0, + "step": 12334 + }, + { + "epoch": 2.290622098421541, + "grad_norm": 1.554068922996521, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8874579071998596, + "num_tokens": 433102852.0, + "step": 12335 + }, + { + "epoch": 2.290807799442897, + "grad_norm": 1.4981693029403687, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8795331716537476, + "num_tokens": 433143079.0, + "step": 12336 + }, + { + "epoch": 2.2909935004642525, + "grad_norm": 1.5076467990875244, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8910877704620361, + "num_tokens": 433178815.0, + "step": 12337 + }, + { + "epoch": 2.2911792014856083, + "grad_norm": 1.5323059558868408, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8906062841415405, + "num_tokens": 433213301.0, + "step": 12338 + }, + { + "epoch": 2.291364902506964, + "grad_norm": 1.5110689401626587, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8794493675231934, + "num_tokens": 433252119.0, + "step": 12339 + }, + { + "epoch": 2.2915506035283193, + "grad_norm": 1.591916799545288, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8718833923339844, + "num_tokens": 433288410.0, + "step": 12340 + }, + { + "epoch": 2.291736304549675, + "grad_norm": 1.4846339225769043, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8859018683433533, + "num_tokens": 433329641.0, + "step": 12341 + }, + { + "epoch": 2.2919220055710308, + "grad_norm": 1.6657525300979614, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8867964148521423, + "num_tokens": 433359425.0, + "step": 12342 + }, + { + "epoch": 2.292107706592386, + "grad_norm": 1.581620454788208, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8837364912033081, + "num_tokens": 433393746.0, + "step": 12343 + }, + { + "epoch": 2.292293407613742, + "grad_norm": 1.5422409772872925, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8866400122642517, + "num_tokens": 433433771.0, + "step": 12344 + }, + { + "epoch": 2.2924791086350975, + "grad_norm": 1.4822087287902832, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8802555203437805, + "num_tokens": 433472733.0, + "step": 12345 + }, + { + "epoch": 2.2926648096564533, + "grad_norm": 1.5215814113616943, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8910832405090332, + "num_tokens": 433507607.0, + "step": 12346 + }, + { + "epoch": 2.2928505106778085, + "grad_norm": 1.547202229499817, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8934763669967651, + "num_tokens": 433542004.0, + "step": 12347 + }, + { + "epoch": 2.2930362116991643, + "grad_norm": 1.5862337350845337, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8838026523590088, + "num_tokens": 433575730.0, + "step": 12348 + }, + { + "epoch": 2.29322191272052, + "grad_norm": 1.5240178108215332, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.8932446241378784, + "num_tokens": 433612489.0, + "step": 12349 + }, + { + "epoch": 2.2934076137418757, + "grad_norm": 1.479180097579956, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8885780572891235, + "num_tokens": 433649929.0, + "step": 12350 + }, + { + "epoch": 2.293593314763231, + "grad_norm": 1.5554656982421875, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8876701593399048, + "num_tokens": 433684954.0, + "step": 12351 + }, + { + "epoch": 2.2937790157845868, + "grad_norm": 1.4304306507110596, + "learning_rate": 1e-06, + "loss": 0.2551, + "mean_token_accuracy": 0.907251238822937, + "num_tokens": 433717719.0, + "step": 12352 + }, + { + "epoch": 2.2939647168059425, + "grad_norm": 1.5013939142227173, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8855099081993103, + "num_tokens": 433754973.0, + "step": 12353 + }, + { + "epoch": 2.2941504178272982, + "grad_norm": 1.5832653045654297, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8883966207504272, + "num_tokens": 433786418.0, + "step": 12354 + }, + { + "epoch": 2.2943361188486535, + "grad_norm": 1.615146279335022, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8872926235198975, + "num_tokens": 433817321.0, + "step": 12355 + }, + { + "epoch": 2.2945218198700092, + "grad_norm": 1.5671809911727905, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8919111490249634, + "num_tokens": 433850074.0, + "step": 12356 + }, + { + "epoch": 2.294707520891365, + "grad_norm": 1.6287977695465088, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8735096454620361, + "num_tokens": 433887252.0, + "step": 12357 + }, + { + "epoch": 2.2948932219127203, + "grad_norm": 1.5241968631744385, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8895416259765625, + "num_tokens": 433924705.0, + "step": 12358 + }, + { + "epoch": 2.295078922934076, + "grad_norm": 1.5078433752059937, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8924154043197632, + "num_tokens": 433962314.0, + "step": 12359 + }, + { + "epoch": 2.2952646239554317, + "grad_norm": 1.5129367113113403, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8873558044433594, + "num_tokens": 434000360.0, + "step": 12360 + }, + { + "epoch": 2.2954503249767875, + "grad_norm": 1.4226855039596558, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.89049232006073, + "num_tokens": 434042497.0, + "step": 12361 + }, + { + "epoch": 2.295636025998143, + "grad_norm": 1.667351245880127, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8904159069061279, + "num_tokens": 434075553.0, + "step": 12362 + }, + { + "epoch": 2.2958217270194985, + "grad_norm": 1.597035527229309, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8816575407981873, + "num_tokens": 434111368.0, + "step": 12363 + }, + { + "epoch": 2.2960074280408542, + "grad_norm": 1.4842045307159424, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8874322175979614, + "num_tokens": 434150112.0, + "step": 12364 + }, + { + "epoch": 2.29619312906221, + "grad_norm": 1.7844165563583374, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8757920265197754, + "num_tokens": 434176157.0, + "step": 12365 + }, + { + "epoch": 2.2963788300835652, + "grad_norm": 1.6451305150985718, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8811498880386353, + "num_tokens": 434208753.0, + "step": 12366 + }, + { + "epoch": 2.296564531104921, + "grad_norm": 1.3925081491470337, + "learning_rate": 1e-06, + "loss": 0.2716, + "mean_token_accuracy": 0.9049907922744751, + "num_tokens": 434244538.0, + "step": 12367 + }, + { + "epoch": 2.2967502321262767, + "grad_norm": 1.4716811180114746, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8873413801193237, + "num_tokens": 434285509.0, + "step": 12368 + }, + { + "epoch": 2.2969359331476324, + "grad_norm": 1.701630711555481, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8756517171859741, + "num_tokens": 434320233.0, + "step": 12369 + }, + { + "epoch": 2.2971216341689877, + "grad_norm": 1.5201818943023682, + "learning_rate": 1e-06, + "loss": 0.2868, + "mean_token_accuracy": 0.8979663848876953, + "num_tokens": 434352596.0, + "step": 12370 + }, + { + "epoch": 2.2973073351903435, + "grad_norm": 1.686722755432129, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8690676689147949, + "num_tokens": 434386409.0, + "step": 12371 + }, + { + "epoch": 2.297493036211699, + "grad_norm": 1.6438432931900024, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8877396583557129, + "num_tokens": 434421254.0, + "step": 12372 + }, + { + "epoch": 2.297678737233055, + "grad_norm": 1.458950400352478, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8928905725479126, + "num_tokens": 434460859.0, + "step": 12373 + }, + { + "epoch": 2.2978644382544102, + "grad_norm": 1.4517693519592285, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8975424766540527, + "num_tokens": 434498319.0, + "step": 12374 + }, + { + "epoch": 2.298050139275766, + "grad_norm": 1.6012839078903198, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8790508508682251, + "num_tokens": 434533995.0, + "step": 12375 + }, + { + "epoch": 2.2982358402971217, + "grad_norm": 1.4275356531143188, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8910261392593384, + "num_tokens": 434579454.0, + "step": 12376 + }, + { + "epoch": 2.2984215413184774, + "grad_norm": 1.7224887609481812, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8749145269393921, + "num_tokens": 434612065.0, + "step": 12377 + }, + { + "epoch": 2.2986072423398327, + "grad_norm": 1.5549222230911255, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8842104077339172, + "num_tokens": 434646182.0, + "step": 12378 + }, + { + "epoch": 2.2987929433611884, + "grad_norm": 1.708618402481079, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.886475682258606, + "num_tokens": 434676182.0, + "step": 12379 + }, + { + "epoch": 2.298978644382544, + "grad_norm": 1.6992055177688599, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8776042461395264, + "num_tokens": 434707850.0, + "step": 12380 + }, + { + "epoch": 2.2991643454039, + "grad_norm": 1.7081528902053833, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8797379732131958, + "num_tokens": 434738285.0, + "step": 12381 + }, + { + "epoch": 2.299350046425255, + "grad_norm": 1.555888056755066, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8905402421951294, + "num_tokens": 434771784.0, + "step": 12382 + }, + { + "epoch": 2.299535747446611, + "grad_norm": 1.5215126276016235, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8834869861602783, + "num_tokens": 434812887.0, + "step": 12383 + }, + { + "epoch": 2.2997214484679667, + "grad_norm": 1.4341869354248047, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.8965226411819458, + "num_tokens": 434852547.0, + "step": 12384 + }, + { + "epoch": 2.2999071494893224, + "grad_norm": 1.510481834411621, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8766517043113708, + "num_tokens": 434893872.0, + "step": 12385 + }, + { + "epoch": 2.3000928505106777, + "grad_norm": 1.5745493173599243, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8762614727020264, + "num_tokens": 434933482.0, + "step": 12386 + }, + { + "epoch": 2.3002785515320334, + "grad_norm": 1.5378626585006714, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8898072242736816, + "num_tokens": 434969191.0, + "step": 12387 + }, + { + "epoch": 2.300464252553389, + "grad_norm": 1.6871857643127441, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8906883001327515, + "num_tokens": 435001158.0, + "step": 12388 + }, + { + "epoch": 2.3006499535747444, + "grad_norm": 1.626946210861206, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8869757652282715, + "num_tokens": 435034546.0, + "step": 12389 + }, + { + "epoch": 2.3008356545961, + "grad_norm": 1.590893268585205, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8933273553848267, + "num_tokens": 435067994.0, + "step": 12390 + }, + { + "epoch": 2.301021355617456, + "grad_norm": 1.6453453302383423, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8816925287246704, + "num_tokens": 435098526.0, + "step": 12391 + }, + { + "epoch": 2.3012070566388116, + "grad_norm": 1.559533715248108, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8821104764938354, + "num_tokens": 435131229.0, + "step": 12392 + }, + { + "epoch": 2.3013927576601674, + "grad_norm": 1.4968386888504028, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8742440938949585, + "num_tokens": 435170604.0, + "step": 12393 + }, + { + "epoch": 2.3015784586815227, + "grad_norm": 1.5965193510055542, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.894130527973175, + "num_tokens": 435207303.0, + "step": 12394 + }, + { + "epoch": 2.3017641597028784, + "grad_norm": 1.4836454391479492, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8825106620788574, + "num_tokens": 435248054.0, + "step": 12395 + }, + { + "epoch": 2.301949860724234, + "grad_norm": 1.7089953422546387, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8828959465026855, + "num_tokens": 435278139.0, + "step": 12396 + }, + { + "epoch": 2.3021355617455894, + "grad_norm": 1.5068248510360718, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8892621397972107, + "num_tokens": 435313285.0, + "step": 12397 + }, + { + "epoch": 2.302321262766945, + "grad_norm": 1.6904380321502686, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8940085768699646, + "num_tokens": 435343009.0, + "step": 12398 + }, + { + "epoch": 2.302506963788301, + "grad_norm": 1.448563575744629, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.8967558145523071, + "num_tokens": 435380635.0, + "step": 12399 + }, + { + "epoch": 2.3026926648096566, + "grad_norm": 1.520371913909912, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8968921303749084, + "num_tokens": 435418164.0, + "step": 12400 + }, + { + "epoch": 2.302878365831012, + "grad_norm": 1.6688213348388672, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8856347799301147, + "num_tokens": 435450770.0, + "step": 12401 + }, + { + "epoch": 2.3030640668523676, + "grad_norm": 1.499143362045288, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.880283772945404, + "num_tokens": 435493201.0, + "step": 12402 + }, + { + "epoch": 2.3032497678737234, + "grad_norm": 1.44882071018219, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.893596887588501, + "num_tokens": 435534770.0, + "step": 12403 + }, + { + "epoch": 2.303435468895079, + "grad_norm": 1.6303277015686035, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8866547346115112, + "num_tokens": 435568562.0, + "step": 12404 + }, + { + "epoch": 2.3036211699164344, + "grad_norm": 1.7928143739700317, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8714407682418823, + "num_tokens": 435604218.0, + "step": 12405 + }, + { + "epoch": 2.30380687093779, + "grad_norm": 1.6480571031570435, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8885071873664856, + "num_tokens": 435636011.0, + "step": 12406 + }, + { + "epoch": 2.303992571959146, + "grad_norm": 1.7599589824676514, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8817058801651001, + "num_tokens": 435665816.0, + "step": 12407 + }, + { + "epoch": 2.3041782729805016, + "grad_norm": 1.575286865234375, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8851050734519958, + "num_tokens": 435699194.0, + "step": 12408 + }, + { + "epoch": 2.304363974001857, + "grad_norm": 1.641421914100647, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8775880336761475, + "num_tokens": 435733344.0, + "step": 12409 + }, + { + "epoch": 2.3045496750232126, + "grad_norm": 1.672531247138977, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8767443895339966, + "num_tokens": 435770644.0, + "step": 12410 + }, + { + "epoch": 2.3047353760445684, + "grad_norm": 1.5304603576660156, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8862584829330444, + "num_tokens": 435805161.0, + "step": 12411 + }, + { + "epoch": 2.3049210770659236, + "grad_norm": 1.5359615087509155, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8884305357933044, + "num_tokens": 435841129.0, + "step": 12412 + }, + { + "epoch": 2.3051067780872794, + "grad_norm": 1.72126042842865, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.887639582157135, + "num_tokens": 435867769.0, + "step": 12413 + }, + { + "epoch": 2.305292479108635, + "grad_norm": 1.5881052017211914, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8794940710067749, + "num_tokens": 435903059.0, + "step": 12414 + }, + { + "epoch": 2.305478180129991, + "grad_norm": 1.6424416303634644, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8784278035163879, + "num_tokens": 435937956.0, + "step": 12415 + }, + { + "epoch": 2.3056638811513466, + "grad_norm": 1.5295321941375732, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8873367309570312, + "num_tokens": 435977705.0, + "step": 12416 + }, + { + "epoch": 2.305849582172702, + "grad_norm": 1.4346555471420288, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8898440003395081, + "num_tokens": 436015492.0, + "step": 12417 + }, + { + "epoch": 2.3060352831940576, + "grad_norm": 1.533928632736206, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.887060284614563, + "num_tokens": 436053940.0, + "step": 12418 + }, + { + "epoch": 2.3062209842154133, + "grad_norm": 1.5560487508773804, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8932870626449585, + "num_tokens": 436089780.0, + "step": 12419 + }, + { + "epoch": 2.3064066852367686, + "grad_norm": 1.5506386756896973, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.879752516746521, + "num_tokens": 436127681.0, + "step": 12420 + }, + { + "epoch": 2.3065923862581243, + "grad_norm": 1.8797205686569214, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8608860969543457, + "num_tokens": 436156584.0, + "step": 12421 + }, + { + "epoch": 2.30677808727948, + "grad_norm": 1.755289077758789, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.878324568271637, + "num_tokens": 436191327.0, + "step": 12422 + }, + { + "epoch": 2.306963788300836, + "grad_norm": 1.4782073497772217, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8926612734794617, + "num_tokens": 436230094.0, + "step": 12423 + }, + { + "epoch": 2.307149489322191, + "grad_norm": 1.4498857259750366, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8868476748466492, + "num_tokens": 436267404.0, + "step": 12424 + }, + { + "epoch": 2.307335190343547, + "grad_norm": 1.5655183792114258, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8927042484283447, + "num_tokens": 436304351.0, + "step": 12425 + }, + { + "epoch": 2.3075208913649026, + "grad_norm": 1.5274994373321533, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8845100402832031, + "num_tokens": 436345499.0, + "step": 12426 + }, + { + "epoch": 2.3077065923862583, + "grad_norm": 1.5863295793533325, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8771194219589233, + "num_tokens": 436382691.0, + "step": 12427 + }, + { + "epoch": 2.3078922934076136, + "grad_norm": 1.4856137037277222, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.8988525867462158, + "num_tokens": 436422429.0, + "step": 12428 + }, + { + "epoch": 2.3080779944289693, + "grad_norm": 1.6010217666625977, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8904480934143066, + "num_tokens": 436453348.0, + "step": 12429 + }, + { + "epoch": 2.308263695450325, + "grad_norm": 1.500305414199829, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8892810344696045, + "num_tokens": 436488309.0, + "step": 12430 + }, + { + "epoch": 2.308449396471681, + "grad_norm": 1.7028049230575562, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8935474753379822, + "num_tokens": 436516757.0, + "step": 12431 + }, + { + "epoch": 2.308635097493036, + "grad_norm": 1.641106128692627, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8862881660461426, + "num_tokens": 436549763.0, + "step": 12432 + }, + { + "epoch": 2.308820798514392, + "grad_norm": 1.5858815908432007, + "learning_rate": 1e-06, + "loss": 0.2884, + "mean_token_accuracy": 0.8968048095703125, + "num_tokens": 436585492.0, + "step": 12433 + }, + { + "epoch": 2.3090064995357475, + "grad_norm": 1.5515385866165161, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8881322741508484, + "num_tokens": 436620150.0, + "step": 12434 + }, + { + "epoch": 2.309192200557103, + "grad_norm": 1.7143380641937256, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8865485191345215, + "num_tokens": 436653443.0, + "step": 12435 + }, + { + "epoch": 2.3093779015784586, + "grad_norm": 1.5684869289398193, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8879506587982178, + "num_tokens": 436689210.0, + "step": 12436 + }, + { + "epoch": 2.3095636025998143, + "grad_norm": 1.5033550262451172, + "learning_rate": 1e-06, + "loss": 0.2803, + "mean_token_accuracy": 0.9016202688217163, + "num_tokens": 436725193.0, + "step": 12437 + }, + { + "epoch": 2.30974930362117, + "grad_norm": 1.6662423610687256, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8843718767166138, + "num_tokens": 436754779.0, + "step": 12438 + }, + { + "epoch": 2.3099350046425258, + "grad_norm": 1.501313328742981, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8847063183784485, + "num_tokens": 436792275.0, + "step": 12439 + }, + { + "epoch": 2.310120705663881, + "grad_norm": 1.53054940700531, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8828945159912109, + "num_tokens": 436829349.0, + "step": 12440 + }, + { + "epoch": 2.310306406685237, + "grad_norm": 1.7399177551269531, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8795811533927917, + "num_tokens": 436858060.0, + "step": 12441 + }, + { + "epoch": 2.3104921077065925, + "grad_norm": 1.5796376466751099, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8784778118133545, + "num_tokens": 436895595.0, + "step": 12442 + }, + { + "epoch": 2.310677808727948, + "grad_norm": 1.5408108234405518, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8718828558921814, + "num_tokens": 436932783.0, + "step": 12443 + }, + { + "epoch": 2.3108635097493035, + "grad_norm": 1.4524983167648315, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8969378471374512, + "num_tokens": 436972867.0, + "step": 12444 + }, + { + "epoch": 2.3110492107706593, + "grad_norm": 1.4234237670898438, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8939769268035889, + "num_tokens": 437011213.0, + "step": 12445 + }, + { + "epoch": 2.311234911792015, + "grad_norm": 1.5415422916412354, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8865499496459961, + "num_tokens": 437045850.0, + "step": 12446 + }, + { + "epoch": 2.3114206128133703, + "grad_norm": 1.5636053085327148, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.88425213098526, + "num_tokens": 437081893.0, + "step": 12447 + }, + { + "epoch": 2.311606313834726, + "grad_norm": 1.6822437047958374, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8741490244865417, + "num_tokens": 437111552.0, + "step": 12448 + }, + { + "epoch": 2.3117920148560818, + "grad_norm": 1.573744773864746, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8829620480537415, + "num_tokens": 437147227.0, + "step": 12449 + }, + { + "epoch": 2.3119777158774375, + "grad_norm": 1.518730640411377, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8876134157180786, + "num_tokens": 437185163.0, + "step": 12450 + }, + { + "epoch": 2.312163416898793, + "grad_norm": 1.754748821258545, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8921210765838623, + "num_tokens": 437215328.0, + "step": 12451 + }, + { + "epoch": 2.3123491179201485, + "grad_norm": 1.514207124710083, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8927286267280579, + "num_tokens": 437251132.0, + "step": 12452 + }, + { + "epoch": 2.3125348189415043, + "grad_norm": 1.5873236656188965, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8924731016159058, + "num_tokens": 437282095.0, + "step": 12453 + }, + { + "epoch": 2.31272051996286, + "grad_norm": 1.5156970024108887, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8871723413467407, + "num_tokens": 437318032.0, + "step": 12454 + }, + { + "epoch": 2.3129062209842153, + "grad_norm": 1.539253830909729, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8913853168487549, + "num_tokens": 437350875.0, + "step": 12455 + }, + { + "epoch": 2.313091922005571, + "grad_norm": 1.3619521856307983, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8947227001190186, + "num_tokens": 437393840.0, + "step": 12456 + }, + { + "epoch": 2.3132776230269267, + "grad_norm": 1.5085636377334595, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8887949585914612, + "num_tokens": 437431931.0, + "step": 12457 + }, + { + "epoch": 2.313463324048282, + "grad_norm": 1.6380915641784668, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.8871949911117554, + "num_tokens": 437468675.0, + "step": 12458 + }, + { + "epoch": 2.3136490250696378, + "grad_norm": 1.6048153638839722, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8948816061019897, + "num_tokens": 437498903.0, + "step": 12459 + }, + { + "epoch": 2.3138347260909935, + "grad_norm": 1.6252996921539307, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8816838264465332, + "num_tokens": 437532456.0, + "step": 12460 + }, + { + "epoch": 2.3140204271123492, + "grad_norm": 1.3267780542373657, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8951593637466431, + "num_tokens": 437575876.0, + "step": 12461 + }, + { + "epoch": 2.314206128133705, + "grad_norm": 1.5419865846633911, + "learning_rate": 1e-06, + "loss": 0.3607, + "mean_token_accuracy": 0.8746823072433472, + "num_tokens": 437613331.0, + "step": 12462 + }, + { + "epoch": 2.3143918291550603, + "grad_norm": 1.5436404943466187, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8888921737670898, + "num_tokens": 437648817.0, + "step": 12463 + }, + { + "epoch": 2.314577530176416, + "grad_norm": 1.561330795288086, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8917608261108398, + "num_tokens": 437683070.0, + "step": 12464 + }, + { + "epoch": 2.3147632311977717, + "grad_norm": 1.4975144863128662, + "learning_rate": 1e-06, + "loss": 0.2932, + "mean_token_accuracy": 0.896569013595581, + "num_tokens": 437717972.0, + "step": 12465 + }, + { + "epoch": 2.314948932219127, + "grad_norm": 1.4393367767333984, + "learning_rate": 1e-06, + "loss": 0.2819, + "mean_token_accuracy": 0.9008805751800537, + "num_tokens": 437752535.0, + "step": 12466 + }, + { + "epoch": 2.3151346332404827, + "grad_norm": 1.7959554195404053, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8735402822494507, + "num_tokens": 437782012.0, + "step": 12467 + }, + { + "epoch": 2.3153203342618385, + "grad_norm": 1.5812311172485352, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.876530647277832, + "num_tokens": 437819721.0, + "step": 12468 + }, + { + "epoch": 2.315506035283194, + "grad_norm": 1.4455739259719849, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.8889931440353394, + "num_tokens": 437858730.0, + "step": 12469 + }, + { + "epoch": 2.3156917363045495, + "grad_norm": 1.5147504806518555, + "learning_rate": 1e-06, + "loss": 0.282, + "mean_token_accuracy": 0.9020401239395142, + "num_tokens": 437890561.0, + "step": 12470 + }, + { + "epoch": 2.3158774373259052, + "grad_norm": 1.4272031784057617, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8976659774780273, + "num_tokens": 437929646.0, + "step": 12471 + }, + { + "epoch": 2.316063138347261, + "grad_norm": 1.5206842422485352, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8912986516952515, + "num_tokens": 437960107.0, + "step": 12472 + }, + { + "epoch": 2.3162488393686167, + "grad_norm": 1.7168025970458984, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8839777708053589, + "num_tokens": 437989590.0, + "step": 12473 + }, + { + "epoch": 2.316434540389972, + "grad_norm": 1.5370817184448242, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8966251015663147, + "num_tokens": 438025721.0, + "step": 12474 + }, + { + "epoch": 2.3166202414113277, + "grad_norm": 1.4561887979507446, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8929703235626221, + "num_tokens": 438065227.0, + "step": 12475 + }, + { + "epoch": 2.3168059424326835, + "grad_norm": 1.6200977563858032, + "learning_rate": 1e-06, + "loss": 0.2916, + "mean_token_accuracy": 0.8958410620689392, + "num_tokens": 438099881.0, + "step": 12476 + }, + { + "epoch": 2.316991643454039, + "grad_norm": 1.749491572380066, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8779981136322021, + "num_tokens": 438129803.0, + "step": 12477 + }, + { + "epoch": 2.3171773444753945, + "grad_norm": 1.6988861560821533, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8896819949150085, + "num_tokens": 438163434.0, + "step": 12478 + }, + { + "epoch": 2.31736304549675, + "grad_norm": 1.5047434568405151, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8759778738021851, + "num_tokens": 438202973.0, + "step": 12479 + }, + { + "epoch": 2.317548746518106, + "grad_norm": 1.6114150285720825, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8894134163856506, + "num_tokens": 438235954.0, + "step": 12480 + }, + { + "epoch": 2.3177344475394612, + "grad_norm": 1.5292373895645142, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.875755786895752, + "num_tokens": 438275612.0, + "step": 12481 + }, + { + "epoch": 2.317920148560817, + "grad_norm": 1.5781034231185913, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8845308423042297, + "num_tokens": 438311838.0, + "step": 12482 + }, + { + "epoch": 2.3181058495821727, + "grad_norm": 1.6942670345306396, + "learning_rate": 1e-06, + "loss": 0.2863, + "mean_token_accuracy": 0.8949251174926758, + "num_tokens": 438343210.0, + "step": 12483 + }, + { + "epoch": 2.3182915506035284, + "grad_norm": 1.7026562690734863, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8834568858146667, + "num_tokens": 438376774.0, + "step": 12484 + }, + { + "epoch": 2.318477251624884, + "grad_norm": 1.7730427980422974, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8873520493507385, + "num_tokens": 438408420.0, + "step": 12485 + }, + { + "epoch": 2.3186629526462395, + "grad_norm": 1.7639126777648926, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8928745985031128, + "num_tokens": 438436798.0, + "step": 12486 + }, + { + "epoch": 2.318848653667595, + "grad_norm": 1.524580955505371, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8850759267807007, + "num_tokens": 438472426.0, + "step": 12487 + }, + { + "epoch": 2.319034354688951, + "grad_norm": 1.5546553134918213, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8932081460952759, + "num_tokens": 438507353.0, + "step": 12488 + }, + { + "epoch": 2.319220055710306, + "grad_norm": 1.738455057144165, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8854113817214966, + "num_tokens": 438536573.0, + "step": 12489 + }, + { + "epoch": 2.319405756731662, + "grad_norm": 1.5649465322494507, + "learning_rate": 1e-06, + "loss": 0.2907, + "mean_token_accuracy": 0.8960760831832886, + "num_tokens": 438571070.0, + "step": 12490 + }, + { + "epoch": 2.3195914577530177, + "grad_norm": 1.716530203819275, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8875709176063538, + "num_tokens": 438604555.0, + "step": 12491 + }, + { + "epoch": 2.3197771587743734, + "grad_norm": 1.5363456010818481, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8959575295448303, + "num_tokens": 438635696.0, + "step": 12492 + }, + { + "epoch": 2.3199628597957287, + "grad_norm": 1.629960536956787, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8854738473892212, + "num_tokens": 438667742.0, + "step": 12493 + }, + { + "epoch": 2.3201485608170844, + "grad_norm": 1.4427390098571777, + "learning_rate": 1e-06, + "loss": 0.2787, + "mean_token_accuracy": 0.9036718606948853, + "num_tokens": 438705539.0, + "step": 12494 + }, + { + "epoch": 2.32033426183844, + "grad_norm": 1.513314962387085, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8781041502952576, + "num_tokens": 438743021.0, + "step": 12495 + }, + { + "epoch": 2.320519962859796, + "grad_norm": 1.457763433456421, + "learning_rate": 1e-06, + "loss": 0.2947, + "mean_token_accuracy": 0.8931349515914917, + "num_tokens": 438781011.0, + "step": 12496 + }, + { + "epoch": 2.320705663881151, + "grad_norm": 1.5300114154815674, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.88563472032547, + "num_tokens": 438820853.0, + "step": 12497 + }, + { + "epoch": 2.320891364902507, + "grad_norm": 1.5203391313552856, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8885351419448853, + "num_tokens": 438859930.0, + "step": 12498 + }, + { + "epoch": 2.3210770659238626, + "grad_norm": 1.487877607345581, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8902094960212708, + "num_tokens": 438898986.0, + "step": 12499 + }, + { + "epoch": 2.3212627669452184, + "grad_norm": 1.5710909366607666, + "learning_rate": 1e-06, + "loss": 0.2694, + "mean_token_accuracy": 0.9031621813774109, + "num_tokens": 438930992.0, + "step": 12500 + }, + { + "epoch": 2.3214484679665737, + "grad_norm": 1.4097918272018433, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8978921175003052, + "num_tokens": 438970561.0, + "step": 12501 + }, + { + "epoch": 2.3216341689879294, + "grad_norm": 1.5411386489868164, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8857824802398682, + "num_tokens": 439012695.0, + "step": 12502 + }, + { + "epoch": 2.321819870009285, + "grad_norm": 1.5561143159866333, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.878890335559845, + "num_tokens": 439052947.0, + "step": 12503 + }, + { + "epoch": 2.3220055710306404, + "grad_norm": 1.589896321296692, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8895469307899475, + "num_tokens": 439089037.0, + "step": 12504 + }, + { + "epoch": 2.322191272051996, + "grad_norm": 1.5004287958145142, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.8926832675933838, + "num_tokens": 439128793.0, + "step": 12505 + }, + { + "epoch": 2.322376973073352, + "grad_norm": 1.7461618185043335, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8814618587493896, + "num_tokens": 439159487.0, + "step": 12506 + }, + { + "epoch": 2.3225626740947076, + "grad_norm": 1.440622329711914, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8957875967025757, + "num_tokens": 439196627.0, + "step": 12507 + }, + { + "epoch": 2.3227483751160634, + "grad_norm": 1.5119284391403198, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8862212300300598, + "num_tokens": 439231349.0, + "step": 12508 + }, + { + "epoch": 2.3229340761374186, + "grad_norm": 1.6092655658721924, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8946521878242493, + "num_tokens": 439261095.0, + "step": 12509 + }, + { + "epoch": 2.3231197771587744, + "grad_norm": 1.512009859085083, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8969529867172241, + "num_tokens": 439294225.0, + "step": 12510 + }, + { + "epoch": 2.32330547818013, + "grad_norm": 1.6122583150863647, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.891281008720398, + "num_tokens": 439326641.0, + "step": 12511 + }, + { + "epoch": 2.3234911792014854, + "grad_norm": 1.5305792093276978, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8850936889648438, + "num_tokens": 439364779.0, + "step": 12512 + }, + { + "epoch": 2.323676880222841, + "grad_norm": 1.5138911008834839, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.888256311416626, + "num_tokens": 439401704.0, + "step": 12513 + }, + { + "epoch": 2.323862581244197, + "grad_norm": 1.5036184787750244, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8875386714935303, + "num_tokens": 439438100.0, + "step": 12514 + }, + { + "epoch": 2.3240482822655526, + "grad_norm": 1.7172752618789673, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8794625401496887, + "num_tokens": 439473001.0, + "step": 12515 + }, + { + "epoch": 2.324233983286908, + "grad_norm": 1.5303159952163696, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8852534294128418, + "num_tokens": 439511843.0, + "step": 12516 + }, + { + "epoch": 2.3244196843082636, + "grad_norm": 1.5501822233200073, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8954578638076782, + "num_tokens": 439546566.0, + "step": 12517 + }, + { + "epoch": 2.3246053853296194, + "grad_norm": 1.59711754322052, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.871405839920044, + "num_tokens": 439585084.0, + "step": 12518 + }, + { + "epoch": 2.324791086350975, + "grad_norm": 1.4485398530960083, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8882824182510376, + "num_tokens": 439623969.0, + "step": 12519 + }, + { + "epoch": 2.3249767873723304, + "grad_norm": 1.7009499073028564, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8913835287094116, + "num_tokens": 439653334.0, + "step": 12520 + }, + { + "epoch": 2.325162488393686, + "grad_norm": 1.69699227809906, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8869622349739075, + "num_tokens": 439695781.0, + "step": 12521 + }, + { + "epoch": 2.325348189415042, + "grad_norm": 1.5515716075897217, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8783161640167236, + "num_tokens": 439733022.0, + "step": 12522 + }, + { + "epoch": 2.3255338904363976, + "grad_norm": 1.6888986825942993, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8881723880767822, + "num_tokens": 439763275.0, + "step": 12523 + }, + { + "epoch": 2.325719591457753, + "grad_norm": 1.5349704027175903, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8797725439071655, + "num_tokens": 439799413.0, + "step": 12524 + }, + { + "epoch": 2.3259052924791086, + "grad_norm": 1.3781192302703857, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8817244172096252, + "num_tokens": 439842831.0, + "step": 12525 + }, + { + "epoch": 2.3260909935004643, + "grad_norm": 1.6116169691085815, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.884479284286499, + "num_tokens": 439873943.0, + "step": 12526 + }, + { + "epoch": 2.3262766945218196, + "grad_norm": 1.4156317710876465, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8964577913284302, + "num_tokens": 439913029.0, + "step": 12527 + }, + { + "epoch": 2.3264623955431754, + "grad_norm": 1.6288427114486694, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8810908794403076, + "num_tokens": 439944660.0, + "step": 12528 + }, + { + "epoch": 2.326648096564531, + "grad_norm": 1.4603359699249268, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8885796070098877, + "num_tokens": 439985107.0, + "step": 12529 + }, + { + "epoch": 2.326833797585887, + "grad_norm": 1.73385751247406, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8742983341217041, + "num_tokens": 440019005.0, + "step": 12530 + }, + { + "epoch": 2.3270194986072426, + "grad_norm": 1.7101322412490845, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8779971599578857, + "num_tokens": 440048412.0, + "step": 12531 + }, + { + "epoch": 2.327205199628598, + "grad_norm": 1.605562686920166, + "learning_rate": 1e-06, + "loss": 0.3586, + "mean_token_accuracy": 0.8751237392425537, + "num_tokens": 440086302.0, + "step": 12532 + }, + { + "epoch": 2.3273909006499536, + "grad_norm": 1.3981682062149048, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8966219425201416, + "num_tokens": 440129419.0, + "step": 12533 + }, + { + "epoch": 2.3275766016713093, + "grad_norm": 1.6026172637939453, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8888444304466248, + "num_tokens": 440168176.0, + "step": 12534 + }, + { + "epoch": 2.3277623026926646, + "grad_norm": 1.744322419166565, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8843963146209717, + "num_tokens": 440199605.0, + "step": 12535 + }, + { + "epoch": 2.3279480037140203, + "grad_norm": 1.6674020290374756, + "learning_rate": 1e-06, + "loss": 0.271, + "mean_token_accuracy": 0.9042923450469971, + "num_tokens": 440227370.0, + "step": 12536 + }, + { + "epoch": 2.328133704735376, + "grad_norm": 1.4700815677642822, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8947198390960693, + "num_tokens": 440262084.0, + "step": 12537 + }, + { + "epoch": 2.328319405756732, + "grad_norm": 1.718977928161621, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.9002488851547241, + "num_tokens": 440294192.0, + "step": 12538 + }, + { + "epoch": 2.328505106778087, + "grad_norm": 1.4742417335510254, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.8980950117111206, + "num_tokens": 440332146.0, + "step": 12539 + }, + { + "epoch": 2.328690807799443, + "grad_norm": 1.577707290649414, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8936064839363098, + "num_tokens": 440366380.0, + "step": 12540 + }, + { + "epoch": 2.3288765088207986, + "grad_norm": 1.5222088098526, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.8870195150375366, + "num_tokens": 440406748.0, + "step": 12541 + }, + { + "epoch": 2.3290622098421543, + "grad_norm": 1.6404153108596802, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.875893235206604, + "num_tokens": 440437789.0, + "step": 12542 + }, + { + "epoch": 2.3292479108635096, + "grad_norm": 1.435334324836731, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8978029489517212, + "num_tokens": 440473130.0, + "step": 12543 + }, + { + "epoch": 2.3294336118848653, + "grad_norm": 1.6455974578857422, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8868522644042969, + "num_tokens": 440505357.0, + "step": 12544 + }, + { + "epoch": 2.329619312906221, + "grad_norm": 1.45821213722229, + "learning_rate": 1e-06, + "loss": 0.2866, + "mean_token_accuracy": 0.8970628976821899, + "num_tokens": 440543569.0, + "step": 12545 + }, + { + "epoch": 2.3298050139275768, + "grad_norm": 1.502684473991394, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8856523036956787, + "num_tokens": 440580947.0, + "step": 12546 + }, + { + "epoch": 2.329990714948932, + "grad_norm": 1.5830326080322266, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8874801397323608, + "num_tokens": 440613759.0, + "step": 12547 + }, + { + "epoch": 2.330176415970288, + "grad_norm": 1.598236083984375, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8803936243057251, + "num_tokens": 440648414.0, + "step": 12548 + }, + { + "epoch": 2.3303621169916435, + "grad_norm": 1.4447996616363525, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8903480768203735, + "num_tokens": 440685348.0, + "step": 12549 + }, + { + "epoch": 2.3305478180129993, + "grad_norm": 1.7822794914245605, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8918459415435791, + "num_tokens": 440718656.0, + "step": 12550 + }, + { + "epoch": 2.3307335190343546, + "grad_norm": 1.702593207359314, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8733987808227539, + "num_tokens": 440753658.0, + "step": 12551 + }, + { + "epoch": 2.3309192200557103, + "grad_norm": 1.6216378211975098, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8816618323326111, + "num_tokens": 440788796.0, + "step": 12552 + }, + { + "epoch": 2.331104921077066, + "grad_norm": 1.5259701013565063, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8839253187179565, + "num_tokens": 440827084.0, + "step": 12553 + }, + { + "epoch": 2.3312906220984217, + "grad_norm": 1.619158387184143, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8822075128555298, + "num_tokens": 440860690.0, + "step": 12554 + }, + { + "epoch": 2.331476323119777, + "grad_norm": 1.561063289642334, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8892477750778198, + "num_tokens": 440894333.0, + "step": 12555 + }, + { + "epoch": 2.3316620241411328, + "grad_norm": 1.6446540355682373, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8911185264587402, + "num_tokens": 440928417.0, + "step": 12556 + }, + { + "epoch": 2.3318477251624885, + "grad_norm": 1.6019669771194458, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8919322490692139, + "num_tokens": 440964571.0, + "step": 12557 + }, + { + "epoch": 2.332033426183844, + "grad_norm": 1.6861776113510132, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8791534900665283, + "num_tokens": 440997448.0, + "step": 12558 + }, + { + "epoch": 2.3322191272051995, + "grad_norm": 1.5291012525558472, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8778020143508911, + "num_tokens": 441035667.0, + "step": 12559 + }, + { + "epoch": 2.3324048282265553, + "grad_norm": 1.5128625631332397, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8910987377166748, + "num_tokens": 441071927.0, + "step": 12560 + }, + { + "epoch": 2.332590529247911, + "grad_norm": 1.561373233795166, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8939406871795654, + "num_tokens": 441103816.0, + "step": 12561 + }, + { + "epoch": 2.3327762302692667, + "grad_norm": 1.5908013582229614, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8773637413978577, + "num_tokens": 441142129.0, + "step": 12562 + }, + { + "epoch": 2.332961931290622, + "grad_norm": 1.697021484375, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8900878429412842, + "num_tokens": 441171492.0, + "step": 12563 + }, + { + "epoch": 2.3331476323119777, + "grad_norm": 1.4793591499328613, + "learning_rate": 1e-06, + "loss": 0.2933, + "mean_token_accuracy": 0.8960307836532593, + "num_tokens": 441208513.0, + "step": 12564 + }, + { + "epoch": 2.3333333333333335, + "grad_norm": 1.4849604368209839, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8886706829071045, + "num_tokens": 441248476.0, + "step": 12565 + }, + { + "epoch": 2.3335190343546888, + "grad_norm": 1.6128002405166626, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8869709372520447, + "num_tokens": 441279162.0, + "step": 12566 + }, + { + "epoch": 2.3337047353760445, + "grad_norm": 1.4066035747528076, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.896173894405365, + "num_tokens": 441318322.0, + "step": 12567 + }, + { + "epoch": 2.3338904363974002, + "grad_norm": 1.62434720993042, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.896358072757721, + "num_tokens": 441346995.0, + "step": 12568 + }, + { + "epoch": 2.334076137418756, + "grad_norm": 1.6348745822906494, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8859868049621582, + "num_tokens": 441382676.0, + "step": 12569 + }, + { + "epoch": 2.3342618384401113, + "grad_norm": 1.5538908243179321, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8932086825370789, + "num_tokens": 441417375.0, + "step": 12570 + }, + { + "epoch": 2.334447539461467, + "grad_norm": 1.6507831811904907, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.88253253698349, + "num_tokens": 441450064.0, + "step": 12571 + }, + { + "epoch": 2.3346332404828227, + "grad_norm": 1.5361990928649902, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8922651410102844, + "num_tokens": 441486183.0, + "step": 12572 + }, + { + "epoch": 2.3348189415041785, + "grad_norm": 1.9217191934585571, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8818193078041077, + "num_tokens": 441509533.0, + "step": 12573 + }, + { + "epoch": 2.3350046425255337, + "grad_norm": 1.5776360034942627, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8898555040359497, + "num_tokens": 441542828.0, + "step": 12574 + }, + { + "epoch": 2.3351903435468895, + "grad_norm": 1.4832316637039185, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8868861198425293, + "num_tokens": 441582666.0, + "step": 12575 + }, + { + "epoch": 2.335376044568245, + "grad_norm": 1.5207093954086304, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8924182653427124, + "num_tokens": 441619540.0, + "step": 12576 + }, + { + "epoch": 2.335561745589601, + "grad_norm": 1.5924981832504272, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8913114070892334, + "num_tokens": 441651980.0, + "step": 12577 + }, + { + "epoch": 2.3357474466109562, + "grad_norm": 1.608187198638916, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8804605007171631, + "num_tokens": 441687962.0, + "step": 12578 + }, + { + "epoch": 2.335933147632312, + "grad_norm": 1.5141631364822388, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8764591217041016, + "num_tokens": 441727504.0, + "step": 12579 + }, + { + "epoch": 2.3361188486536677, + "grad_norm": 1.6722540855407715, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8789573907852173, + "num_tokens": 441762639.0, + "step": 12580 + }, + { + "epoch": 2.336304549675023, + "grad_norm": 1.4820958375930786, + "learning_rate": 1e-06, + "loss": 0.2868, + "mean_token_accuracy": 0.895627498626709, + "num_tokens": 441801676.0, + "step": 12581 + }, + { + "epoch": 2.3364902506963787, + "grad_norm": 1.5610573291778564, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8875073194503784, + "num_tokens": 441835257.0, + "step": 12582 + }, + { + "epoch": 2.3366759517177345, + "grad_norm": 1.6473019123077393, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8863894939422607, + "num_tokens": 441863350.0, + "step": 12583 + }, + { + "epoch": 2.33686165273909, + "grad_norm": 1.5007938146591187, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8895854949951172, + "num_tokens": 441899014.0, + "step": 12584 + }, + { + "epoch": 2.337047353760446, + "grad_norm": 1.450486421585083, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8989361524581909, + "num_tokens": 441934565.0, + "step": 12585 + }, + { + "epoch": 2.337233054781801, + "grad_norm": 1.6479766368865967, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8848482370376587, + "num_tokens": 441967838.0, + "step": 12586 + }, + { + "epoch": 2.337418755803157, + "grad_norm": 1.5531682968139648, + "learning_rate": 1e-06, + "loss": 0.2893, + "mean_token_accuracy": 0.8975528478622437, + "num_tokens": 442004883.0, + "step": 12587 + }, + { + "epoch": 2.3376044568245127, + "grad_norm": 1.541322112083435, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8763236999511719, + "num_tokens": 442042659.0, + "step": 12588 + }, + { + "epoch": 2.337790157845868, + "grad_norm": 1.8623342514038086, + "learning_rate": 1e-06, + "loss": 0.3911, + "mean_token_accuracy": 0.8644853830337524, + "num_tokens": 442071503.0, + "step": 12589 + }, + { + "epoch": 2.3379758588672237, + "grad_norm": 1.5988726615905762, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.897257924079895, + "num_tokens": 442104245.0, + "step": 12590 + }, + { + "epoch": 2.3381615598885794, + "grad_norm": 1.61322820186615, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8735392093658447, + "num_tokens": 442140262.0, + "step": 12591 + }, + { + "epoch": 2.338347260909935, + "grad_norm": 1.4704314470291138, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8849637508392334, + "num_tokens": 442178661.0, + "step": 12592 + }, + { + "epoch": 2.3385329619312905, + "grad_norm": 1.4174437522888184, + "learning_rate": 1e-06, + "loss": 0.2976, + "mean_token_accuracy": 0.8958994746208191, + "num_tokens": 442216671.0, + "step": 12593 + }, + { + "epoch": 2.338718662952646, + "grad_norm": 1.5025534629821777, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8756455183029175, + "num_tokens": 442257143.0, + "step": 12594 + }, + { + "epoch": 2.338904363974002, + "grad_norm": 1.5814592838287354, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8873475193977356, + "num_tokens": 442290387.0, + "step": 12595 + }, + { + "epoch": 2.3390900649953577, + "grad_norm": 1.6027452945709229, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8892982006072998, + "num_tokens": 442325180.0, + "step": 12596 + }, + { + "epoch": 2.339275766016713, + "grad_norm": 1.5523427724838257, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8815973401069641, + "num_tokens": 442364355.0, + "step": 12597 + }, + { + "epoch": 2.3394614670380687, + "grad_norm": 1.5588511228561401, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8933005332946777, + "num_tokens": 442400649.0, + "step": 12598 + }, + { + "epoch": 2.3396471680594244, + "grad_norm": 1.6215693950653076, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8956280946731567, + "num_tokens": 442431060.0, + "step": 12599 + }, + { + "epoch": 2.33983286908078, + "grad_norm": 1.5014007091522217, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8900431394577026, + "num_tokens": 442469927.0, + "step": 12600 + }, + { + "epoch": 2.3400185701021354, + "grad_norm": 1.568261981010437, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8737616539001465, + "num_tokens": 442503699.0, + "step": 12601 + }, + { + "epoch": 2.340204271123491, + "grad_norm": 1.6117864847183228, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.88035649061203, + "num_tokens": 442537926.0, + "step": 12602 + }, + { + "epoch": 2.340389972144847, + "grad_norm": 1.7665799856185913, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8936627507209778, + "num_tokens": 442563200.0, + "step": 12603 + }, + { + "epoch": 2.340575673166202, + "grad_norm": 1.553722858428955, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8781000375747681, + "num_tokens": 442599322.0, + "step": 12604 + }, + { + "epoch": 2.340761374187558, + "grad_norm": 1.5455124378204346, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.876981258392334, + "num_tokens": 442636852.0, + "step": 12605 + }, + { + "epoch": 2.3409470752089137, + "grad_norm": 1.5047001838684082, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8812913298606873, + "num_tokens": 442672366.0, + "step": 12606 + }, + { + "epoch": 2.3411327762302694, + "grad_norm": 1.6158987283706665, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8843554258346558, + "num_tokens": 442704608.0, + "step": 12607 + }, + { + "epoch": 2.341318477251625, + "grad_norm": 1.5356135368347168, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8825163841247559, + "num_tokens": 442743903.0, + "step": 12608 + }, + { + "epoch": 2.3415041782729804, + "grad_norm": 1.5752143859863281, + "learning_rate": 1e-06, + "loss": 0.359, + "mean_token_accuracy": 0.8753303289413452, + "num_tokens": 442779286.0, + "step": 12609 + }, + { + "epoch": 2.341689879294336, + "grad_norm": 1.4523303508758545, + "learning_rate": 1e-06, + "loss": 0.2976, + "mean_token_accuracy": 0.8924198150634766, + "num_tokens": 442814316.0, + "step": 12610 + }, + { + "epoch": 2.341875580315692, + "grad_norm": 1.50495183467865, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8832718729972839, + "num_tokens": 442850068.0, + "step": 12611 + }, + { + "epoch": 2.342061281337047, + "grad_norm": 1.4363458156585693, + "learning_rate": 1e-06, + "loss": 0.2757, + "mean_token_accuracy": 0.9046350121498108, + "num_tokens": 442885704.0, + "step": 12612 + }, + { + "epoch": 2.342246982358403, + "grad_norm": 1.5632781982421875, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8812330961227417, + "num_tokens": 442924172.0, + "step": 12613 + }, + { + "epoch": 2.3424326833797586, + "grad_norm": 1.470487117767334, + "learning_rate": 1e-06, + "loss": 0.297, + "mean_token_accuracy": 0.8937828540802002, + "num_tokens": 442959650.0, + "step": 12614 + }, + { + "epoch": 2.3426183844011144, + "grad_norm": 1.5303940773010254, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8948290348052979, + "num_tokens": 442992298.0, + "step": 12615 + }, + { + "epoch": 2.3428040854224697, + "grad_norm": 1.4286413192749023, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8967742919921875, + "num_tokens": 443029879.0, + "step": 12616 + }, + { + "epoch": 2.3429897864438254, + "grad_norm": 1.5122276544570923, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8926612138748169, + "num_tokens": 443067437.0, + "step": 12617 + }, + { + "epoch": 2.343175487465181, + "grad_norm": 1.6422263383865356, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8901957273483276, + "num_tokens": 443100566.0, + "step": 12618 + }, + { + "epoch": 2.343361188486537, + "grad_norm": 1.520463228225708, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8794702291488647, + "num_tokens": 443137066.0, + "step": 12619 + }, + { + "epoch": 2.343546889507892, + "grad_norm": 1.6243177652359009, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8883497714996338, + "num_tokens": 443170465.0, + "step": 12620 + }, + { + "epoch": 2.343732590529248, + "grad_norm": 1.586564064025879, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8911076784133911, + "num_tokens": 443204490.0, + "step": 12621 + }, + { + "epoch": 2.3439182915506036, + "grad_norm": 1.393053650856018, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8858992457389832, + "num_tokens": 443247521.0, + "step": 12622 + }, + { + "epoch": 2.3441039925719593, + "grad_norm": 1.4173074960708618, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8927316665649414, + "num_tokens": 443289697.0, + "step": 12623 + }, + { + "epoch": 2.3442896935933146, + "grad_norm": 1.6985925436019897, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8854947090148926, + "num_tokens": 443322111.0, + "step": 12624 + }, + { + "epoch": 2.3444753946146704, + "grad_norm": 1.5364080667495728, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8746265769004822, + "num_tokens": 443360812.0, + "step": 12625 + }, + { + "epoch": 2.344661095636026, + "grad_norm": 1.605881690979004, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.882388710975647, + "num_tokens": 443394540.0, + "step": 12626 + }, + { + "epoch": 2.3448467966573814, + "grad_norm": 1.6379045248031616, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8901553153991699, + "num_tokens": 443429372.0, + "step": 12627 + }, + { + "epoch": 2.345032497678737, + "grad_norm": 1.4752376079559326, + "learning_rate": 1e-06, + "loss": 0.2826, + "mean_token_accuracy": 0.8997398018836975, + "num_tokens": 443460626.0, + "step": 12628 + }, + { + "epoch": 2.345218198700093, + "grad_norm": 1.5659071207046509, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8920677900314331, + "num_tokens": 443492734.0, + "step": 12629 + }, + { + "epoch": 2.3454038997214486, + "grad_norm": 1.6545575857162476, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8813070058822632, + "num_tokens": 443526170.0, + "step": 12630 + }, + { + "epoch": 2.3455896007428043, + "grad_norm": 1.6204397678375244, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8851779699325562, + "num_tokens": 443560750.0, + "step": 12631 + }, + { + "epoch": 2.3457753017641596, + "grad_norm": 1.5218082666397095, + "learning_rate": 1e-06, + "loss": 0.3067, + "mean_token_accuracy": 0.8891779184341431, + "num_tokens": 443595952.0, + "step": 12632 + }, + { + "epoch": 2.3459610027855153, + "grad_norm": 1.5326370000839233, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8816341161727905, + "num_tokens": 443632500.0, + "step": 12633 + }, + { + "epoch": 2.346146703806871, + "grad_norm": 1.375891089439392, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8854103684425354, + "num_tokens": 443675668.0, + "step": 12634 + }, + { + "epoch": 2.3463324048282264, + "grad_norm": 1.7007852792739868, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8735756874084473, + "num_tokens": 443708242.0, + "step": 12635 + }, + { + "epoch": 2.346518105849582, + "grad_norm": 1.6535829305648804, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8841778635978699, + "num_tokens": 443739111.0, + "step": 12636 + }, + { + "epoch": 2.346703806870938, + "grad_norm": 1.5956900119781494, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8867378830909729, + "num_tokens": 443769678.0, + "step": 12637 + }, + { + "epoch": 2.3468895078922936, + "grad_norm": 1.662101149559021, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.8921074867248535, + "num_tokens": 443799221.0, + "step": 12638 + }, + { + "epoch": 2.347075208913649, + "grad_norm": 1.5530574321746826, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.8989404439926147, + "num_tokens": 443832687.0, + "step": 12639 + }, + { + "epoch": 2.3472609099350046, + "grad_norm": 1.5694267749786377, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8907171487808228, + "num_tokens": 443868953.0, + "step": 12640 + }, + { + "epoch": 2.3474466109563603, + "grad_norm": 1.6767442226409912, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8778488039970398, + "num_tokens": 443901641.0, + "step": 12641 + }, + { + "epoch": 2.347632311977716, + "grad_norm": 1.6107338666915894, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8843137621879578, + "num_tokens": 443935666.0, + "step": 12642 + }, + { + "epoch": 2.3478180129990713, + "grad_norm": 1.5698972940444946, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8900964260101318, + "num_tokens": 443967716.0, + "step": 12643 + }, + { + "epoch": 2.348003714020427, + "grad_norm": 1.526566743850708, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8826608061790466, + "num_tokens": 444004125.0, + "step": 12644 + }, + { + "epoch": 2.348189415041783, + "grad_norm": 1.6605029106140137, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8792539834976196, + "num_tokens": 444036845.0, + "step": 12645 + }, + { + "epoch": 2.3483751160631385, + "grad_norm": 1.4149744510650635, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8888500332832336, + "num_tokens": 444078198.0, + "step": 12646 + }, + { + "epoch": 2.348560817084494, + "grad_norm": 1.678104043006897, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8762024641036987, + "num_tokens": 444107586.0, + "step": 12647 + }, + { + "epoch": 2.3487465181058496, + "grad_norm": 1.580621600151062, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8840368390083313, + "num_tokens": 444141797.0, + "step": 12648 + }, + { + "epoch": 2.3489322191272053, + "grad_norm": 1.6321476697921753, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8925349712371826, + "num_tokens": 444169009.0, + "step": 12649 + }, + { + "epoch": 2.3491179201485606, + "grad_norm": 1.7606630325317383, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8760087490081787, + "num_tokens": 444200014.0, + "step": 12650 + }, + { + "epoch": 2.3493036211699163, + "grad_norm": 1.6924697160720825, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8895532488822937, + "num_tokens": 444227915.0, + "step": 12651 + }, + { + "epoch": 2.349489322191272, + "grad_norm": 1.4419726133346558, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8934178948402405, + "num_tokens": 444265061.0, + "step": 12652 + }, + { + "epoch": 2.349675023212628, + "grad_norm": 1.447993516921997, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8959757685661316, + "num_tokens": 444301048.0, + "step": 12653 + }, + { + "epoch": 2.3498607242339835, + "grad_norm": 1.5061448812484741, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8873324990272522, + "num_tokens": 444336954.0, + "step": 12654 + }, + { + "epoch": 2.350046425255339, + "grad_norm": 1.564955472946167, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8836699724197388, + "num_tokens": 444373516.0, + "step": 12655 + }, + { + "epoch": 2.3502321262766945, + "grad_norm": 1.497584342956543, + "learning_rate": 1e-06, + "loss": 0.2766, + "mean_token_accuracy": 0.8989831209182739, + "num_tokens": 444405912.0, + "step": 12656 + }, + { + "epoch": 2.3504178272980503, + "grad_norm": 1.5271154642105103, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8953012228012085, + "num_tokens": 444442517.0, + "step": 12657 + }, + { + "epoch": 2.3506035283194056, + "grad_norm": 1.3951771259307861, + "learning_rate": 1e-06, + "loss": 0.2723, + "mean_token_accuracy": 0.9027438759803772, + "num_tokens": 444480422.0, + "step": 12658 + }, + { + "epoch": 2.3507892293407613, + "grad_norm": 1.6403471231460571, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8816598057746887, + "num_tokens": 444514090.0, + "step": 12659 + }, + { + "epoch": 2.350974930362117, + "grad_norm": 1.5163655281066895, + "learning_rate": 1e-06, + "loss": 0.2615, + "mean_token_accuracy": 0.904822587966919, + "num_tokens": 444546685.0, + "step": 12660 + }, + { + "epoch": 2.3511606313834728, + "grad_norm": 1.4507761001586914, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8914412260055542, + "num_tokens": 444586728.0, + "step": 12661 + }, + { + "epoch": 2.351346332404828, + "grad_norm": 1.4314864873886108, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.8964711427688599, + "num_tokens": 444624616.0, + "step": 12662 + }, + { + "epoch": 2.3515320334261838, + "grad_norm": 1.480339527130127, + "learning_rate": 1e-06, + "loss": 0.2996, + "mean_token_accuracy": 0.8925098776817322, + "num_tokens": 444661554.0, + "step": 12663 + }, + { + "epoch": 2.3517177344475395, + "grad_norm": 1.5243045091629028, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8849225044250488, + "num_tokens": 444697514.0, + "step": 12664 + }, + { + "epoch": 2.3519034354688952, + "grad_norm": 1.6910231113433838, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8900969624519348, + "num_tokens": 444732119.0, + "step": 12665 + }, + { + "epoch": 2.3520891364902505, + "grad_norm": 1.6132690906524658, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8838874101638794, + "num_tokens": 444767857.0, + "step": 12666 + }, + { + "epoch": 2.3522748375116063, + "grad_norm": 1.6168400049209595, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.890610933303833, + "num_tokens": 444799442.0, + "step": 12667 + }, + { + "epoch": 2.352460538532962, + "grad_norm": 1.593134880065918, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8876096606254578, + "num_tokens": 444834851.0, + "step": 12668 + }, + { + "epoch": 2.3526462395543177, + "grad_norm": 1.6200892925262451, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8819577097892761, + "num_tokens": 444869929.0, + "step": 12669 + }, + { + "epoch": 2.352831940575673, + "grad_norm": 1.585742712020874, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8820998668670654, + "num_tokens": 444903540.0, + "step": 12670 + }, + { + "epoch": 2.3530176415970288, + "grad_norm": 1.4852659702301025, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8906234502792358, + "num_tokens": 444943183.0, + "step": 12671 + }, + { + "epoch": 2.3532033426183845, + "grad_norm": 1.5260014533996582, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8971477150917053, + "num_tokens": 444977375.0, + "step": 12672 + }, + { + "epoch": 2.3533890436397398, + "grad_norm": 1.5319232940673828, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8953367471694946, + "num_tokens": 445011963.0, + "step": 12673 + }, + { + "epoch": 2.3535747446610955, + "grad_norm": 1.65212082862854, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8734361529350281, + "num_tokens": 445046668.0, + "step": 12674 + }, + { + "epoch": 2.3537604456824512, + "grad_norm": 1.6891226768493652, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8863726258277893, + "num_tokens": 445077452.0, + "step": 12675 + }, + { + "epoch": 2.353946146703807, + "grad_norm": 1.6126642227172852, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8777807950973511, + "num_tokens": 445110345.0, + "step": 12676 + }, + { + "epoch": 2.3541318477251627, + "grad_norm": 1.6340245008468628, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8771573901176453, + "num_tokens": 445145000.0, + "step": 12677 + }, + { + "epoch": 2.354317548746518, + "grad_norm": 1.5085958242416382, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8790996074676514, + "num_tokens": 445179047.0, + "step": 12678 + }, + { + "epoch": 2.3545032497678737, + "grad_norm": 1.5380488634109497, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8860169649124146, + "num_tokens": 445212518.0, + "step": 12679 + }, + { + "epoch": 2.3546889507892295, + "grad_norm": 1.715336799621582, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8828514814376831, + "num_tokens": 445243615.0, + "step": 12680 + }, + { + "epoch": 2.3548746518105848, + "grad_norm": 1.6175256967544556, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8807615637779236, + "num_tokens": 445275031.0, + "step": 12681 + }, + { + "epoch": 2.3550603528319405, + "grad_norm": 1.5892515182495117, + "learning_rate": 1e-06, + "loss": 0.283, + "mean_token_accuracy": 0.895897626876831, + "num_tokens": 445307946.0, + "step": 12682 + }, + { + "epoch": 2.355246053853296, + "grad_norm": 1.6599338054656982, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8880325555801392, + "num_tokens": 445344629.0, + "step": 12683 + }, + { + "epoch": 2.355431754874652, + "grad_norm": 1.55955970287323, + "learning_rate": 1e-06, + "loss": 0.2821, + "mean_token_accuracy": 0.9018834233283997, + "num_tokens": 445375656.0, + "step": 12684 + }, + { + "epoch": 2.3556174558960072, + "grad_norm": 1.5559242963790894, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8845693469047546, + "num_tokens": 445410621.0, + "step": 12685 + }, + { + "epoch": 2.355803156917363, + "grad_norm": 1.3796024322509766, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8729292750358582, + "num_tokens": 445459818.0, + "step": 12686 + }, + { + "epoch": 2.3559888579387187, + "grad_norm": 1.6970854997634888, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.8948758840560913, + "num_tokens": 445489060.0, + "step": 12687 + }, + { + "epoch": 2.3561745589600744, + "grad_norm": 1.6005995273590088, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8900214433670044, + "num_tokens": 445524091.0, + "step": 12688 + }, + { + "epoch": 2.3563602599814297, + "grad_norm": 1.5788226127624512, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8955247402191162, + "num_tokens": 445555992.0, + "step": 12689 + }, + { + "epoch": 2.3565459610027855, + "grad_norm": 1.49123215675354, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8926751613616943, + "num_tokens": 445594221.0, + "step": 12690 + }, + { + "epoch": 2.356731662024141, + "grad_norm": 1.7351278066635132, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8867104053497314, + "num_tokens": 445625053.0, + "step": 12691 + }, + { + "epoch": 2.356917363045497, + "grad_norm": 1.444840669631958, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8808656930923462, + "num_tokens": 445666444.0, + "step": 12692 + }, + { + "epoch": 2.357103064066852, + "grad_norm": 1.596706748008728, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8910688757896423, + "num_tokens": 445700530.0, + "step": 12693 + }, + { + "epoch": 2.357288765088208, + "grad_norm": 1.8480409383773804, + "learning_rate": 1e-06, + "loss": 0.4012, + "mean_token_accuracy": 0.8654894828796387, + "num_tokens": 445731766.0, + "step": 12694 + }, + { + "epoch": 2.3574744661095637, + "grad_norm": 1.5181891918182373, + "learning_rate": 1e-06, + "loss": 0.3023, + "mean_token_accuracy": 0.8942595720291138, + "num_tokens": 445766979.0, + "step": 12695 + }, + { + "epoch": 2.357660167130919, + "grad_norm": 1.625616192817688, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8796935677528381, + "num_tokens": 445802773.0, + "step": 12696 + }, + { + "epoch": 2.3578458681522747, + "grad_norm": 1.4691423177719116, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8894466161727905, + "num_tokens": 445839634.0, + "step": 12697 + }, + { + "epoch": 2.3580315691736304, + "grad_norm": 1.6789802312850952, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8815581798553467, + "num_tokens": 445874104.0, + "step": 12698 + }, + { + "epoch": 2.358217270194986, + "grad_norm": 1.542473554611206, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8930401802062988, + "num_tokens": 445906666.0, + "step": 12699 + }, + { + "epoch": 2.358402971216342, + "grad_norm": 1.5435913801193237, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.8990642428398132, + "num_tokens": 445939121.0, + "step": 12700 + }, + { + "epoch": 2.358588672237697, + "grad_norm": 1.8431732654571533, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8756231665611267, + "num_tokens": 445970137.0, + "step": 12701 + }, + { + "epoch": 2.358774373259053, + "grad_norm": 1.54308021068573, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.8988273739814758, + "num_tokens": 446001551.0, + "step": 12702 + }, + { + "epoch": 2.3589600742804087, + "grad_norm": 1.5019596815109253, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8960569500923157, + "num_tokens": 446036391.0, + "step": 12703 + }, + { + "epoch": 2.359145775301764, + "grad_norm": 1.5147500038146973, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8827728629112244, + "num_tokens": 446074470.0, + "step": 12704 + }, + { + "epoch": 2.3593314763231197, + "grad_norm": 1.5154774188995361, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8815059661865234, + "num_tokens": 446117978.0, + "step": 12705 + }, + { + "epoch": 2.3595171773444754, + "grad_norm": 1.826682209968567, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.868578314781189, + "num_tokens": 446148768.0, + "step": 12706 + }, + { + "epoch": 2.359702878365831, + "grad_norm": 1.672090768814087, + "learning_rate": 1e-06, + "loss": 0.3047, + "mean_token_accuracy": 0.8904107809066772, + "num_tokens": 446178061.0, + "step": 12707 + }, + { + "epoch": 2.3598885793871864, + "grad_norm": 1.50426185131073, + "learning_rate": 1e-06, + "loss": 0.2802, + "mean_token_accuracy": 0.9003989696502686, + "num_tokens": 446216973.0, + "step": 12708 + }, + { + "epoch": 2.360074280408542, + "grad_norm": 1.6476560831069946, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8756297826766968, + "num_tokens": 446252531.0, + "step": 12709 + }, + { + "epoch": 2.360259981429898, + "grad_norm": 1.4957053661346436, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8860797882080078, + "num_tokens": 446290707.0, + "step": 12710 + }, + { + "epoch": 2.3604456824512536, + "grad_norm": 1.5768914222717285, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8789917826652527, + "num_tokens": 446326694.0, + "step": 12711 + }, + { + "epoch": 2.360631383472609, + "grad_norm": 1.515667200088501, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8905472755432129, + "num_tokens": 446361893.0, + "step": 12712 + }, + { + "epoch": 2.3608170844939647, + "grad_norm": 1.517121434211731, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.88577800989151, + "num_tokens": 446398129.0, + "step": 12713 + }, + { + "epoch": 2.3610027855153204, + "grad_norm": 1.7264385223388672, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.878476619720459, + "num_tokens": 446429636.0, + "step": 12714 + }, + { + "epoch": 2.361188486536676, + "grad_norm": 1.6123543977737427, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.885291337966919, + "num_tokens": 446463163.0, + "step": 12715 + }, + { + "epoch": 2.3613741875580314, + "grad_norm": 1.4396659135818481, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8924888968467712, + "num_tokens": 446503714.0, + "step": 12716 + }, + { + "epoch": 2.361559888579387, + "grad_norm": 1.671919822692871, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8814679384231567, + "num_tokens": 446535660.0, + "step": 12717 + }, + { + "epoch": 2.361745589600743, + "grad_norm": 1.806903600692749, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8748399019241333, + "num_tokens": 446565479.0, + "step": 12718 + }, + { + "epoch": 2.3619312906220986, + "grad_norm": 1.567441701889038, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8771374225616455, + "num_tokens": 446602781.0, + "step": 12719 + }, + { + "epoch": 2.362116991643454, + "grad_norm": 1.5471826791763306, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.879069983959198, + "num_tokens": 446639746.0, + "step": 12720 + }, + { + "epoch": 2.3623026926648096, + "grad_norm": 1.616753101348877, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.9035962820053101, + "num_tokens": 446671583.0, + "step": 12721 + }, + { + "epoch": 2.3624883936861654, + "grad_norm": 1.48080575466156, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8940318822860718, + "num_tokens": 446704476.0, + "step": 12722 + }, + { + "epoch": 2.362674094707521, + "grad_norm": 1.5749738216400146, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8763232231140137, + "num_tokens": 446742444.0, + "step": 12723 + }, + { + "epoch": 2.3628597957288764, + "grad_norm": 1.5602996349334717, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8780561089515686, + "num_tokens": 446778906.0, + "step": 12724 + }, + { + "epoch": 2.363045496750232, + "grad_norm": 1.5689213275909424, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8817470669746399, + "num_tokens": 446814184.0, + "step": 12725 + }, + { + "epoch": 2.363231197771588, + "grad_norm": 1.51541268825531, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8899160623550415, + "num_tokens": 446850419.0, + "step": 12726 + }, + { + "epoch": 2.363416898792943, + "grad_norm": 1.6425327062606812, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8903841972351074, + "num_tokens": 446884025.0, + "step": 12727 + }, + { + "epoch": 2.363602599814299, + "grad_norm": 1.5392810106277466, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8855978846549988, + "num_tokens": 446923415.0, + "step": 12728 + }, + { + "epoch": 2.3637883008356546, + "grad_norm": 1.5502851009368896, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8969117403030396, + "num_tokens": 446957858.0, + "step": 12729 + }, + { + "epoch": 2.3639740018570103, + "grad_norm": 1.9304050207138062, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8686293363571167, + "num_tokens": 446985571.0, + "step": 12730 + }, + { + "epoch": 2.364159702878366, + "grad_norm": 1.5979657173156738, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8888801336288452, + "num_tokens": 447019549.0, + "step": 12731 + }, + { + "epoch": 2.3643454038997214, + "grad_norm": 1.6177302598953247, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.9001473784446716, + "num_tokens": 447048596.0, + "step": 12732 + }, + { + "epoch": 2.364531104921077, + "grad_norm": 1.510508418083191, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8894129991531372, + "num_tokens": 447083769.0, + "step": 12733 + }, + { + "epoch": 2.364716805942433, + "grad_norm": 1.5309349298477173, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.890045166015625, + "num_tokens": 447119724.0, + "step": 12734 + }, + { + "epoch": 2.364902506963788, + "grad_norm": 1.61258065700531, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8958251476287842, + "num_tokens": 447148885.0, + "step": 12735 + }, + { + "epoch": 2.365088207985144, + "grad_norm": 1.446244478225708, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8873800039291382, + "num_tokens": 447187856.0, + "step": 12736 + }, + { + "epoch": 2.3652739090064996, + "grad_norm": 1.4849238395690918, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8903958797454834, + "num_tokens": 447227686.0, + "step": 12737 + }, + { + "epoch": 2.3654596100278553, + "grad_norm": 1.6459027528762817, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8825920820236206, + "num_tokens": 447259364.0, + "step": 12738 + }, + { + "epoch": 2.3656453110492106, + "grad_norm": 1.4774267673492432, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.8978759050369263, + "num_tokens": 447294044.0, + "step": 12739 + }, + { + "epoch": 2.3658310120705663, + "grad_norm": 1.6113935708999634, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.890434205532074, + "num_tokens": 447328332.0, + "step": 12740 + }, + { + "epoch": 2.366016713091922, + "grad_norm": 1.4402669668197632, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8723921775817871, + "num_tokens": 447372338.0, + "step": 12741 + }, + { + "epoch": 2.366202414113278, + "grad_norm": 1.4116874933242798, + "learning_rate": 1e-06, + "loss": 0.2933, + "mean_token_accuracy": 0.896379292011261, + "num_tokens": 447411937.0, + "step": 12742 + }, + { + "epoch": 2.366388115134633, + "grad_norm": 1.628684401512146, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8935122489929199, + "num_tokens": 447440726.0, + "step": 12743 + }, + { + "epoch": 2.366573816155989, + "grad_norm": 1.581337571144104, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8909397125244141, + "num_tokens": 447471285.0, + "step": 12744 + }, + { + "epoch": 2.3667595171773446, + "grad_norm": 1.6077932119369507, + "learning_rate": 1e-06, + "loss": 0.3641, + "mean_token_accuracy": 0.8721394538879395, + "num_tokens": 447507222.0, + "step": 12745 + }, + { + "epoch": 2.3669452181987003, + "grad_norm": 1.469421148300171, + "learning_rate": 1e-06, + "loss": 0.2761, + "mean_token_accuracy": 0.9015676379203796, + "num_tokens": 447538327.0, + "step": 12746 + }, + { + "epoch": 2.3671309192200556, + "grad_norm": 1.6568711996078491, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8933498859405518, + "num_tokens": 447569087.0, + "step": 12747 + }, + { + "epoch": 2.3673166202414113, + "grad_norm": 1.6480711698532104, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8826072812080383, + "num_tokens": 447601906.0, + "step": 12748 + }, + { + "epoch": 2.367502321262767, + "grad_norm": 1.7666782140731812, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8801379203796387, + "num_tokens": 447633128.0, + "step": 12749 + }, + { + "epoch": 2.3676880222841223, + "grad_norm": 1.6659715175628662, + "learning_rate": 1e-06, + "loss": 0.3695, + "mean_token_accuracy": 0.8759666681289673, + "num_tokens": 447668099.0, + "step": 12750 + }, + { + "epoch": 2.367873723305478, + "grad_norm": 1.6199655532836914, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8881980180740356, + "num_tokens": 447700430.0, + "step": 12751 + }, + { + "epoch": 2.368059424326834, + "grad_norm": 1.6811723709106445, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.879055917263031, + "num_tokens": 447735686.0, + "step": 12752 + }, + { + "epoch": 2.3682451253481895, + "grad_norm": 1.6754631996154785, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8772397637367249, + "num_tokens": 447771078.0, + "step": 12753 + }, + { + "epoch": 2.3684308263695453, + "grad_norm": 1.5775096416473389, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8795046806335449, + "num_tokens": 447810754.0, + "step": 12754 + }, + { + "epoch": 2.3686165273909006, + "grad_norm": 1.7949738502502441, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8730311989784241, + "num_tokens": 447840906.0, + "step": 12755 + }, + { + "epoch": 2.3688022284122563, + "grad_norm": 1.5333261489868164, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.871800422668457, + "num_tokens": 447879351.0, + "step": 12756 + }, + { + "epoch": 2.368987929433612, + "grad_norm": 1.680968165397644, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8704493045806885, + "num_tokens": 447917393.0, + "step": 12757 + }, + { + "epoch": 2.3691736304549673, + "grad_norm": 1.595491886138916, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.890105664730072, + "num_tokens": 447951652.0, + "step": 12758 + }, + { + "epoch": 2.369359331476323, + "grad_norm": 1.6916314363479614, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.892066240310669, + "num_tokens": 447982049.0, + "step": 12759 + }, + { + "epoch": 2.369545032497679, + "grad_norm": 1.6150767803192139, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8835159540176392, + "num_tokens": 448017577.0, + "step": 12760 + }, + { + "epoch": 2.3697307335190345, + "grad_norm": 1.4657220840454102, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8905059099197388, + "num_tokens": 448054075.0, + "step": 12761 + }, + { + "epoch": 2.36991643454039, + "grad_norm": 1.5072364807128906, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8885282278060913, + "num_tokens": 448095250.0, + "step": 12762 + }, + { + "epoch": 2.3701021355617455, + "grad_norm": 1.5156837701797485, + "learning_rate": 1e-06, + "loss": 0.2744, + "mean_token_accuracy": 0.900614857673645, + "num_tokens": 448127602.0, + "step": 12763 + }, + { + "epoch": 2.3702878365831013, + "grad_norm": 1.5938247442245483, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8875359892845154, + "num_tokens": 448160349.0, + "step": 12764 + }, + { + "epoch": 2.370473537604457, + "grad_norm": 1.6831011772155762, + "learning_rate": 1e-06, + "loss": 0.3785, + "mean_token_accuracy": 0.8712353110313416, + "num_tokens": 448195898.0, + "step": 12765 + }, + { + "epoch": 2.3706592386258123, + "grad_norm": 1.5584372282028198, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.889173686504364, + "num_tokens": 448231725.0, + "step": 12766 + }, + { + "epoch": 2.370844939647168, + "grad_norm": 1.6526851654052734, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.88138747215271, + "num_tokens": 448264305.0, + "step": 12767 + }, + { + "epoch": 2.3710306406685238, + "grad_norm": 1.4761074781417847, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8954771161079407, + "num_tokens": 448303702.0, + "step": 12768 + }, + { + "epoch": 2.3712163416898795, + "grad_norm": 1.6114602088928223, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8778876066207886, + "num_tokens": 448337951.0, + "step": 12769 + }, + { + "epoch": 2.371402042711235, + "grad_norm": 1.5695465803146362, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.9023861885070801, + "num_tokens": 448370628.0, + "step": 12770 + }, + { + "epoch": 2.3715877437325905, + "grad_norm": 1.5825448036193848, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8926976323127747, + "num_tokens": 448404683.0, + "step": 12771 + }, + { + "epoch": 2.3717734447539462, + "grad_norm": 1.551908254623413, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8785041570663452, + "num_tokens": 448444376.0, + "step": 12772 + }, + { + "epoch": 2.3719591457753015, + "grad_norm": 1.4958093166351318, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.8936737775802612, + "num_tokens": 448481767.0, + "step": 12773 + }, + { + "epoch": 2.3721448467966573, + "grad_norm": 1.6646404266357422, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.8879877328872681, + "num_tokens": 448514440.0, + "step": 12774 + }, + { + "epoch": 2.372330547818013, + "grad_norm": 1.5668376684188843, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8897504806518555, + "num_tokens": 448549654.0, + "step": 12775 + }, + { + "epoch": 2.3725162488393687, + "grad_norm": 1.5435404777526855, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8779311180114746, + "num_tokens": 448584695.0, + "step": 12776 + }, + { + "epoch": 2.3727019498607245, + "grad_norm": 1.4316189289093018, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8865363597869873, + "num_tokens": 448624301.0, + "step": 12777 + }, + { + "epoch": 2.3728876508820798, + "grad_norm": 1.8493326902389526, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8859348893165588, + "num_tokens": 448647960.0, + "step": 12778 + }, + { + "epoch": 2.3730733519034355, + "grad_norm": 1.5225474834442139, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8804161548614502, + "num_tokens": 448689620.0, + "step": 12779 + }, + { + "epoch": 2.3732590529247912, + "grad_norm": 1.4797968864440918, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8944719433784485, + "num_tokens": 448723676.0, + "step": 12780 + }, + { + "epoch": 2.3734447539461465, + "grad_norm": 1.6004890203475952, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8865020871162415, + "num_tokens": 448758030.0, + "step": 12781 + }, + { + "epoch": 2.3736304549675022, + "grad_norm": 1.7973240613937378, + "learning_rate": 1e-06, + "loss": 0.3775, + "mean_token_accuracy": 0.8696163892745972, + "num_tokens": 448786581.0, + "step": 12782 + }, + { + "epoch": 2.373816155988858, + "grad_norm": 1.5787372589111328, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.8911729454994202, + "num_tokens": 448820426.0, + "step": 12783 + }, + { + "epoch": 2.3740018570102137, + "grad_norm": 1.5187207460403442, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8851794004440308, + "num_tokens": 448855613.0, + "step": 12784 + }, + { + "epoch": 2.374187558031569, + "grad_norm": 1.7262710332870483, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8965591192245483, + "num_tokens": 448881808.0, + "step": 12785 + }, + { + "epoch": 2.3743732590529247, + "grad_norm": 1.6123524904251099, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8758713603019714, + "num_tokens": 448917513.0, + "step": 12786 + }, + { + "epoch": 2.3745589600742805, + "grad_norm": 1.6339964866638184, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8893706798553467, + "num_tokens": 448949003.0, + "step": 12787 + }, + { + "epoch": 2.374744661095636, + "grad_norm": 1.474951982498169, + "learning_rate": 1e-06, + "loss": 0.3067, + "mean_token_accuracy": 0.8915225267410278, + "num_tokens": 448986024.0, + "step": 12788 + }, + { + "epoch": 2.3749303621169915, + "grad_norm": 1.4605531692504883, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.885373055934906, + "num_tokens": 449026294.0, + "step": 12789 + }, + { + "epoch": 2.3751160631383472, + "grad_norm": 1.5693080425262451, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8808341026306152, + "num_tokens": 449061209.0, + "step": 12790 + }, + { + "epoch": 2.375301764159703, + "grad_norm": 1.5965642929077148, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8895594477653503, + "num_tokens": 449096171.0, + "step": 12791 + }, + { + "epoch": 2.3754874651810587, + "grad_norm": 1.6661418676376343, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8919659852981567, + "num_tokens": 449127978.0, + "step": 12792 + }, + { + "epoch": 2.375673166202414, + "grad_norm": 1.5471702814102173, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8966943621635437, + "num_tokens": 449161818.0, + "step": 12793 + }, + { + "epoch": 2.3758588672237697, + "grad_norm": 1.8094191551208496, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8814469575881958, + "num_tokens": 449189388.0, + "step": 12794 + }, + { + "epoch": 2.3760445682451254, + "grad_norm": 1.436232089996338, + "learning_rate": 1e-06, + "loss": 0.2762, + "mean_token_accuracy": 0.9007364511489868, + "num_tokens": 449226945.0, + "step": 12795 + }, + { + "epoch": 2.3762302692664807, + "grad_norm": 1.516038179397583, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8909201622009277, + "num_tokens": 449262761.0, + "step": 12796 + }, + { + "epoch": 2.3764159702878365, + "grad_norm": 1.7854804992675781, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.878714919090271, + "num_tokens": 449294414.0, + "step": 12797 + }, + { + "epoch": 2.376601671309192, + "grad_norm": 1.5158079862594604, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8809755444526672, + "num_tokens": 449335801.0, + "step": 12798 + }, + { + "epoch": 2.376787372330548, + "grad_norm": 1.4162734746932983, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8953914642333984, + "num_tokens": 449372249.0, + "step": 12799 + }, + { + "epoch": 2.3769730733519037, + "grad_norm": 1.6706645488739014, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8772314786911011, + "num_tokens": 449402554.0, + "step": 12800 + }, + { + "epoch": 2.377158774373259, + "grad_norm": 1.6199042797088623, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8788102865219116, + "num_tokens": 449434981.0, + "step": 12801 + }, + { + "epoch": 2.3773444753946147, + "grad_norm": 1.6548802852630615, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8826026320457458, + "num_tokens": 449466724.0, + "step": 12802 + }, + { + "epoch": 2.3775301764159704, + "grad_norm": 1.5811201333999634, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8927325010299683, + "num_tokens": 449501711.0, + "step": 12803 + }, + { + "epoch": 2.3777158774373257, + "grad_norm": 1.68304443359375, + "learning_rate": 1e-06, + "loss": 0.2797, + "mean_token_accuracy": 0.897963285446167, + "num_tokens": 449540739.0, + "step": 12804 + }, + { + "epoch": 2.3779015784586814, + "grad_norm": 1.6078588962554932, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8829709887504578, + "num_tokens": 449575358.0, + "step": 12805 + }, + { + "epoch": 2.378087279480037, + "grad_norm": 1.4954670667648315, + "learning_rate": 1e-06, + "loss": 0.2872, + "mean_token_accuracy": 0.9007161855697632, + "num_tokens": 449611621.0, + "step": 12806 + }, + { + "epoch": 2.378272980501393, + "grad_norm": 1.5715210437774658, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8901432156562805, + "num_tokens": 449643513.0, + "step": 12807 + }, + { + "epoch": 2.378458681522748, + "grad_norm": 1.5680583715438843, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8914458751678467, + "num_tokens": 449679926.0, + "step": 12808 + }, + { + "epoch": 2.378644382544104, + "grad_norm": 1.4730782508850098, + "learning_rate": 1e-06, + "loss": 0.2762, + "mean_token_accuracy": 0.9011058807373047, + "num_tokens": 449715563.0, + "step": 12809 + }, + { + "epoch": 2.3788300835654597, + "grad_norm": 1.6293766498565674, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8840393424034119, + "num_tokens": 449748613.0, + "step": 12810 + }, + { + "epoch": 2.3790157845868154, + "grad_norm": 1.6335395574569702, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8755981922149658, + "num_tokens": 449786230.0, + "step": 12811 + }, + { + "epoch": 2.3792014856081707, + "grad_norm": 1.5571231842041016, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8841744065284729, + "num_tokens": 449823751.0, + "step": 12812 + }, + { + "epoch": 2.3793871866295264, + "grad_norm": 1.556348443031311, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8832061886787415, + "num_tokens": 449859422.0, + "step": 12813 + }, + { + "epoch": 2.379572887650882, + "grad_norm": 1.6674036979675293, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8797010183334351, + "num_tokens": 449892448.0, + "step": 12814 + }, + { + "epoch": 2.379758588672238, + "grad_norm": 1.6873254776000977, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8946155309677124, + "num_tokens": 449923132.0, + "step": 12815 + }, + { + "epoch": 2.379944289693593, + "grad_norm": 1.4825479984283447, + "learning_rate": 1e-06, + "loss": 0.2733, + "mean_token_accuracy": 0.9009121656417847, + "num_tokens": 449957464.0, + "step": 12816 + }, + { + "epoch": 2.380129990714949, + "grad_norm": 1.6288970708847046, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8924338817596436, + "num_tokens": 449989914.0, + "step": 12817 + }, + { + "epoch": 2.3803156917363046, + "grad_norm": 1.5400391817092896, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8939514756202698, + "num_tokens": 450024580.0, + "step": 12818 + }, + { + "epoch": 2.38050139275766, + "grad_norm": 1.6580414772033691, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8886545896530151, + "num_tokens": 450055238.0, + "step": 12819 + }, + { + "epoch": 2.3806870937790157, + "grad_norm": 1.5872005224227905, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8861038684844971, + "num_tokens": 450089177.0, + "step": 12820 + }, + { + "epoch": 2.3808727948003714, + "grad_norm": 1.6934144496917725, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8823598027229309, + "num_tokens": 450128373.0, + "step": 12821 + }, + { + "epoch": 2.381058495821727, + "grad_norm": 1.5828603506088257, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8927273750305176, + "num_tokens": 450160456.0, + "step": 12822 + }, + { + "epoch": 2.381244196843083, + "grad_norm": 1.6241300106048584, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8876477479934692, + "num_tokens": 450192868.0, + "step": 12823 + }, + { + "epoch": 2.381429897864438, + "grad_norm": 1.474496603012085, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8871685266494751, + "num_tokens": 450234622.0, + "step": 12824 + }, + { + "epoch": 2.381615598885794, + "grad_norm": 1.6371614933013916, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.884453535079956, + "num_tokens": 450266193.0, + "step": 12825 + }, + { + "epoch": 2.3818012999071496, + "grad_norm": 1.5728996992111206, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8929336071014404, + "num_tokens": 450302577.0, + "step": 12826 + }, + { + "epoch": 2.381987000928505, + "grad_norm": 1.8244333267211914, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8728740811347961, + "num_tokens": 450332363.0, + "step": 12827 + }, + { + "epoch": 2.3821727019498606, + "grad_norm": 1.5086064338684082, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8936442732810974, + "num_tokens": 450367944.0, + "step": 12828 + }, + { + "epoch": 2.3823584029712164, + "grad_norm": 1.6345666646957397, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8981147408485413, + "num_tokens": 450397691.0, + "step": 12829 + }, + { + "epoch": 2.382544103992572, + "grad_norm": 1.4028220176696777, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8882608413696289, + "num_tokens": 450439009.0, + "step": 12830 + }, + { + "epoch": 2.3827298050139274, + "grad_norm": 1.5748026371002197, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8804358243942261, + "num_tokens": 450475627.0, + "step": 12831 + }, + { + "epoch": 2.382915506035283, + "grad_norm": 1.6102302074432373, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.89417964220047, + "num_tokens": 450506503.0, + "step": 12832 + }, + { + "epoch": 2.383101207056639, + "grad_norm": 1.5956333875656128, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8852540254592896, + "num_tokens": 450539147.0, + "step": 12833 + }, + { + "epoch": 2.3832869080779946, + "grad_norm": 1.6713889837265015, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8924316763877869, + "num_tokens": 450567642.0, + "step": 12834 + }, + { + "epoch": 2.38347260909935, + "grad_norm": 1.501081109046936, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8799520134925842, + "num_tokens": 450605172.0, + "step": 12835 + }, + { + "epoch": 2.3836583101207056, + "grad_norm": 1.6230055093765259, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8861464858055115, + "num_tokens": 450639852.0, + "step": 12836 + }, + { + "epoch": 2.3838440111420613, + "grad_norm": 1.788014531135559, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.87940514087677, + "num_tokens": 450670439.0, + "step": 12837 + }, + { + "epoch": 2.384029712163417, + "grad_norm": 1.619982123374939, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8830246329307556, + "num_tokens": 450706549.0, + "step": 12838 + }, + { + "epoch": 2.3842154131847724, + "grad_norm": 1.4301600456237793, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8940134048461914, + "num_tokens": 450746942.0, + "step": 12839 + }, + { + "epoch": 2.384401114206128, + "grad_norm": 1.9345594644546509, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8890710473060608, + "num_tokens": 450772589.0, + "step": 12840 + }, + { + "epoch": 2.384586815227484, + "grad_norm": 1.5926563739776611, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8880195021629333, + "num_tokens": 450807621.0, + "step": 12841 + }, + { + "epoch": 2.384772516248839, + "grad_norm": 1.6325520277023315, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8867413401603699, + "num_tokens": 450840695.0, + "step": 12842 + }, + { + "epoch": 2.384958217270195, + "grad_norm": 1.6847972869873047, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8778173327445984, + "num_tokens": 450876237.0, + "step": 12843 + }, + { + "epoch": 2.3851439182915506, + "grad_norm": 1.5895870923995972, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8932095170021057, + "num_tokens": 450911427.0, + "step": 12844 + }, + { + "epoch": 2.3853296193129063, + "grad_norm": 1.4887990951538086, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8945817947387695, + "num_tokens": 450947302.0, + "step": 12845 + }, + { + "epoch": 2.385515320334262, + "grad_norm": 1.6846176385879517, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8934161067008972, + "num_tokens": 450976656.0, + "step": 12846 + }, + { + "epoch": 2.3857010213556173, + "grad_norm": 1.533814549446106, + "learning_rate": 1e-06, + "loss": 0.3082, + "mean_token_accuracy": 0.8931974172592163, + "num_tokens": 451011177.0, + "step": 12847 + }, + { + "epoch": 2.385886722376973, + "grad_norm": 1.4652678966522217, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8832255005836487, + "num_tokens": 451049894.0, + "step": 12848 + }, + { + "epoch": 2.386072423398329, + "grad_norm": 1.541584849357605, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8827541470527649, + "num_tokens": 451082730.0, + "step": 12849 + }, + { + "epoch": 2.386258124419684, + "grad_norm": 1.5782397985458374, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8985415697097778, + "num_tokens": 451112513.0, + "step": 12850 + }, + { + "epoch": 2.38644382544104, + "grad_norm": 1.4243805408477783, + "learning_rate": 1e-06, + "loss": 0.2783, + "mean_token_accuracy": 0.9026678204536438, + "num_tokens": 451149507.0, + "step": 12851 + }, + { + "epoch": 2.3866295264623956, + "grad_norm": 1.5862860679626465, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.888434886932373, + "num_tokens": 451184024.0, + "step": 12852 + }, + { + "epoch": 2.3868152274837513, + "grad_norm": 1.5158612728118896, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8913681507110596, + "num_tokens": 451222573.0, + "step": 12853 + }, + { + "epoch": 2.3870009285051066, + "grad_norm": 1.6503976583480835, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8741840720176697, + "num_tokens": 451255574.0, + "step": 12854 + }, + { + "epoch": 2.3871866295264623, + "grad_norm": 1.561395525932312, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8883804082870483, + "num_tokens": 451290461.0, + "step": 12855 + }, + { + "epoch": 2.387372330547818, + "grad_norm": 1.5735669136047363, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8893725275993347, + "num_tokens": 451323369.0, + "step": 12856 + }, + { + "epoch": 2.387558031569174, + "grad_norm": 1.4051823616027832, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.8990499973297119, + "num_tokens": 451362890.0, + "step": 12857 + }, + { + "epoch": 2.387743732590529, + "grad_norm": 1.5443235635757446, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8747947812080383, + "num_tokens": 451400835.0, + "step": 12858 + }, + { + "epoch": 2.387929433611885, + "grad_norm": 1.5633125305175781, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8868812322616577, + "num_tokens": 451441715.0, + "step": 12859 + }, + { + "epoch": 2.3881151346332405, + "grad_norm": 1.4720935821533203, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.9018323421478271, + "num_tokens": 451477695.0, + "step": 12860 + }, + { + "epoch": 2.3883008356545963, + "grad_norm": 1.540441870689392, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8894966244697571, + "num_tokens": 451514937.0, + "step": 12861 + }, + { + "epoch": 2.3884865366759516, + "grad_norm": 1.5775623321533203, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.890911340713501, + "num_tokens": 451551084.0, + "step": 12862 + }, + { + "epoch": 2.3886722376973073, + "grad_norm": 1.5463058948516846, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8904532194137573, + "num_tokens": 451584086.0, + "step": 12863 + }, + { + "epoch": 2.388857938718663, + "grad_norm": 1.4332456588745117, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8940715789794922, + "num_tokens": 451619437.0, + "step": 12864 + }, + { + "epoch": 2.3890436397400183, + "grad_norm": 1.5627045631408691, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8805052042007446, + "num_tokens": 451656010.0, + "step": 12865 + }, + { + "epoch": 2.389229340761374, + "grad_norm": 1.44720458984375, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8948370218276978, + "num_tokens": 451695000.0, + "step": 12866 + }, + { + "epoch": 2.38941504178273, + "grad_norm": 1.5151560306549072, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8916802406311035, + "num_tokens": 451729958.0, + "step": 12867 + }, + { + "epoch": 2.3896007428040855, + "grad_norm": 1.5079091787338257, + "learning_rate": 1e-06, + "loss": 0.2697, + "mean_token_accuracy": 0.8993898630142212, + "num_tokens": 451764338.0, + "step": 12868 + }, + { + "epoch": 2.3897864438254413, + "grad_norm": 1.5688104629516602, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8867946267127991, + "num_tokens": 451799511.0, + "step": 12869 + }, + { + "epoch": 2.3899721448467965, + "grad_norm": 1.6565340757369995, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.882609486579895, + "num_tokens": 451831573.0, + "step": 12870 + }, + { + "epoch": 2.3901578458681523, + "grad_norm": 1.49271821975708, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.889806866645813, + "num_tokens": 451871976.0, + "step": 12871 + }, + { + "epoch": 2.390343546889508, + "grad_norm": 1.494368553161621, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8857471942901611, + "num_tokens": 451912421.0, + "step": 12872 + }, + { + "epoch": 2.3905292479108633, + "grad_norm": 1.5154427289962769, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8887544870376587, + "num_tokens": 451950115.0, + "step": 12873 + }, + { + "epoch": 2.390714948932219, + "grad_norm": 1.6438307762145996, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8887927532196045, + "num_tokens": 451982020.0, + "step": 12874 + }, + { + "epoch": 2.3909006499535748, + "grad_norm": 1.620500087738037, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8990486860275269, + "num_tokens": 452012825.0, + "step": 12875 + }, + { + "epoch": 2.3910863509749305, + "grad_norm": 1.6004798412322998, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8893535137176514, + "num_tokens": 452049995.0, + "step": 12876 + }, + { + "epoch": 2.391272051996286, + "grad_norm": 1.6330488920211792, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.87657231092453, + "num_tokens": 452088063.0, + "step": 12877 + }, + { + "epoch": 2.3914577530176415, + "grad_norm": 1.451442003250122, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8780677318572998, + "num_tokens": 452132020.0, + "step": 12878 + }, + { + "epoch": 2.3916434540389973, + "grad_norm": 1.6009184122085571, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.8984268307685852, + "num_tokens": 452164292.0, + "step": 12879 + }, + { + "epoch": 2.391829155060353, + "grad_norm": 1.459753155708313, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8869413137435913, + "num_tokens": 452202121.0, + "step": 12880 + }, + { + "epoch": 2.3920148560817083, + "grad_norm": 1.8221110105514526, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8765828609466553, + "num_tokens": 452228359.0, + "step": 12881 + }, + { + "epoch": 2.392200557103064, + "grad_norm": 1.5803252458572388, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8851048350334167, + "num_tokens": 452264282.0, + "step": 12882 + }, + { + "epoch": 2.3923862581244197, + "grad_norm": 1.4877793788909912, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.9004418849945068, + "num_tokens": 452297946.0, + "step": 12883 + }, + { + "epoch": 2.3925719591457755, + "grad_norm": 1.5680181980133057, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8992452621459961, + "num_tokens": 452332368.0, + "step": 12884 + }, + { + "epoch": 2.3927576601671308, + "grad_norm": 1.5185112953186035, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8828065395355225, + "num_tokens": 452371499.0, + "step": 12885 + }, + { + "epoch": 2.3929433611884865, + "grad_norm": 1.7418491840362549, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8850128054618835, + "num_tokens": 452401863.0, + "step": 12886 + }, + { + "epoch": 2.3931290622098422, + "grad_norm": 1.4590058326721191, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.880815863609314, + "num_tokens": 452442520.0, + "step": 12887 + }, + { + "epoch": 2.393314763231198, + "grad_norm": 1.4611926078796387, + "learning_rate": 1e-06, + "loss": 0.3018, + "mean_token_accuracy": 0.8920506834983826, + "num_tokens": 452481783.0, + "step": 12888 + }, + { + "epoch": 2.3935004642525533, + "grad_norm": 1.579463005065918, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8955875635147095, + "num_tokens": 452514336.0, + "step": 12889 + }, + { + "epoch": 2.393686165273909, + "grad_norm": 1.5466238260269165, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8798986673355103, + "num_tokens": 452547933.0, + "step": 12890 + }, + { + "epoch": 2.3938718662952647, + "grad_norm": 1.5749876499176025, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8789029717445374, + "num_tokens": 452584043.0, + "step": 12891 + }, + { + "epoch": 2.3940575673166204, + "grad_norm": 1.4236184358596802, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.891047477722168, + "num_tokens": 452624852.0, + "step": 12892 + }, + { + "epoch": 2.3942432683379757, + "grad_norm": 1.6733160018920898, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8803185224533081, + "num_tokens": 452658562.0, + "step": 12893 + }, + { + "epoch": 2.3944289693593315, + "grad_norm": 1.7249431610107422, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8715893030166626, + "num_tokens": 452690125.0, + "step": 12894 + }, + { + "epoch": 2.394614670380687, + "grad_norm": 1.5539982318878174, + "learning_rate": 1e-06, + "loss": 0.2985, + "mean_token_accuracy": 0.8941500186920166, + "num_tokens": 452725954.0, + "step": 12895 + }, + { + "epoch": 2.3948003714020425, + "grad_norm": 1.4445635080337524, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8815518021583557, + "num_tokens": 452766279.0, + "step": 12896 + }, + { + "epoch": 2.3949860724233982, + "grad_norm": 1.7821605205535889, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.89274662733078, + "num_tokens": 452789360.0, + "step": 12897 + }, + { + "epoch": 2.395171773444754, + "grad_norm": 1.5200952291488647, + "learning_rate": 1e-06, + "loss": 0.292, + "mean_token_accuracy": 0.8923884630203247, + "num_tokens": 452823753.0, + "step": 12898 + }, + { + "epoch": 2.3953574744661097, + "grad_norm": 1.4534330368041992, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8788823485374451, + "num_tokens": 452862807.0, + "step": 12899 + }, + { + "epoch": 2.3955431754874654, + "grad_norm": 1.5484024286270142, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8843623399734497, + "num_tokens": 452897545.0, + "step": 12900 + }, + { + "epoch": 2.3957288765088207, + "grad_norm": 1.6024943590164185, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8847593069076538, + "num_tokens": 452931460.0, + "step": 12901 + }, + { + "epoch": 2.3959145775301764, + "grad_norm": 1.5929311513900757, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8829278349876404, + "num_tokens": 452966579.0, + "step": 12902 + }, + { + "epoch": 2.396100278551532, + "grad_norm": 1.5937697887420654, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8928847908973694, + "num_tokens": 453001240.0, + "step": 12903 + }, + { + "epoch": 2.3962859795728875, + "grad_norm": 1.6125001907348633, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8751323223114014, + "num_tokens": 453035370.0, + "step": 12904 + }, + { + "epoch": 2.396471680594243, + "grad_norm": 1.4846378564834595, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8915252685546875, + "num_tokens": 453075216.0, + "step": 12905 + }, + { + "epoch": 2.396657381615599, + "grad_norm": 1.6365095376968384, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8927207589149475, + "num_tokens": 453104903.0, + "step": 12906 + }, + { + "epoch": 2.3968430826369547, + "grad_norm": 1.584826111793518, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8895139694213867, + "num_tokens": 453136374.0, + "step": 12907 + }, + { + "epoch": 2.39702878365831, + "grad_norm": 1.6227408647537231, + "learning_rate": 1e-06, + "loss": 0.3873, + "mean_token_accuracy": 0.8689521551132202, + "num_tokens": 453173305.0, + "step": 12908 + }, + { + "epoch": 2.3972144846796657, + "grad_norm": 1.4767059087753296, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8855785131454468, + "num_tokens": 453212915.0, + "step": 12909 + }, + { + "epoch": 2.3974001857010214, + "grad_norm": 1.5471748113632202, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8888911008834839, + "num_tokens": 453250192.0, + "step": 12910 + }, + { + "epoch": 2.397585886722377, + "grad_norm": 1.563644289970398, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.89786696434021, + "num_tokens": 453279905.0, + "step": 12911 + }, + { + "epoch": 2.3977715877437324, + "grad_norm": 1.6782151460647583, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8833492994308472, + "num_tokens": 453312147.0, + "step": 12912 + }, + { + "epoch": 2.397957288765088, + "grad_norm": 1.5502822399139404, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8870410919189453, + "num_tokens": 453342604.0, + "step": 12913 + }, + { + "epoch": 2.398142989786444, + "grad_norm": 1.6323795318603516, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8798238039016724, + "num_tokens": 453379021.0, + "step": 12914 + }, + { + "epoch": 2.3983286908077996, + "grad_norm": 1.5408493280410767, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8901687860488892, + "num_tokens": 453413815.0, + "step": 12915 + }, + { + "epoch": 2.398514391829155, + "grad_norm": 1.6498162746429443, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8791526556015015, + "num_tokens": 453448214.0, + "step": 12916 + }, + { + "epoch": 2.3987000928505107, + "grad_norm": 1.5676926374435425, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.8959224224090576, + "num_tokens": 453482599.0, + "step": 12917 + }, + { + "epoch": 2.3988857938718664, + "grad_norm": 1.6218923330307007, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8832014799118042, + "num_tokens": 453517426.0, + "step": 12918 + }, + { + "epoch": 2.3990714948932217, + "grad_norm": 1.7304521799087524, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8721924424171448, + "num_tokens": 453552832.0, + "step": 12919 + }, + { + "epoch": 2.3992571959145774, + "grad_norm": 1.5483710765838623, + "learning_rate": 1e-06, + "loss": 0.3082, + "mean_token_accuracy": 0.8890458345413208, + "num_tokens": 453590736.0, + "step": 12920 + }, + { + "epoch": 2.399442896935933, + "grad_norm": 1.6355470418930054, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8762185573577881, + "num_tokens": 453623601.0, + "step": 12921 + }, + { + "epoch": 2.399628597957289, + "grad_norm": 1.4595857858657837, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8907028436660767, + "num_tokens": 453666463.0, + "step": 12922 + }, + { + "epoch": 2.3998142989786446, + "grad_norm": 1.418731689453125, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.893919825553894, + "num_tokens": 453704993.0, + "step": 12923 + }, + { + "epoch": 2.4, + "grad_norm": 1.4884246587753296, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8882615566253662, + "num_tokens": 453741264.0, + "step": 12924 + }, + { + "epoch": 2.4001857010213556, + "grad_norm": 1.5175163745880127, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8918106555938721, + "num_tokens": 453775164.0, + "step": 12925 + }, + { + "epoch": 2.4003714020427114, + "grad_norm": 1.649811863899231, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8870536088943481, + "num_tokens": 453809547.0, + "step": 12926 + }, + { + "epoch": 2.4005571030640667, + "grad_norm": 1.6395862102508545, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8910578489303589, + "num_tokens": 453843836.0, + "step": 12927 + }, + { + "epoch": 2.4007428040854224, + "grad_norm": 1.5103652477264404, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8917361497879028, + "num_tokens": 453879458.0, + "step": 12928 + }, + { + "epoch": 2.400928505106778, + "grad_norm": 1.6712297201156616, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8890011310577393, + "num_tokens": 453909634.0, + "step": 12929 + }, + { + "epoch": 2.401114206128134, + "grad_norm": 1.5416439771652222, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8744575381278992, + "num_tokens": 453950317.0, + "step": 12930 + }, + { + "epoch": 2.401299907149489, + "grad_norm": 1.5080386400222778, + "learning_rate": 1e-06, + "loss": 0.2734, + "mean_token_accuracy": 0.9039016962051392, + "num_tokens": 453986068.0, + "step": 12931 + }, + { + "epoch": 2.401485608170845, + "grad_norm": 1.5054982900619507, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8807668089866638, + "num_tokens": 454025021.0, + "step": 12932 + }, + { + "epoch": 2.4016713091922006, + "grad_norm": 1.4670734405517578, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8914695978164673, + "num_tokens": 454061144.0, + "step": 12933 + }, + { + "epoch": 2.4018570102135564, + "grad_norm": 1.5086637735366821, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8913501501083374, + "num_tokens": 454095557.0, + "step": 12934 + }, + { + "epoch": 2.4020427112349116, + "grad_norm": 1.6242027282714844, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8876316547393799, + "num_tokens": 454127118.0, + "step": 12935 + }, + { + "epoch": 2.4022284122562674, + "grad_norm": 1.6702675819396973, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8930535316467285, + "num_tokens": 454156377.0, + "step": 12936 + }, + { + "epoch": 2.402414113277623, + "grad_norm": 1.4758341312408447, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8856884241104126, + "num_tokens": 454196876.0, + "step": 12937 + }, + { + "epoch": 2.402599814298979, + "grad_norm": 1.5312684774398804, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.885664701461792, + "num_tokens": 454232068.0, + "step": 12938 + }, + { + "epoch": 2.402785515320334, + "grad_norm": 1.4830830097198486, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8849604725837708, + "num_tokens": 454270045.0, + "step": 12939 + }, + { + "epoch": 2.40297121634169, + "grad_norm": 1.364883303642273, + "learning_rate": 1e-06, + "loss": 0.2757, + "mean_token_accuracy": 0.9022303819656372, + "num_tokens": 454309347.0, + "step": 12940 + }, + { + "epoch": 2.4031569173630456, + "grad_norm": 1.5271672010421753, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8942543268203735, + "num_tokens": 454343293.0, + "step": 12941 + }, + { + "epoch": 2.403342618384401, + "grad_norm": 1.5429604053497314, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8871797323226929, + "num_tokens": 454379051.0, + "step": 12942 + }, + { + "epoch": 2.4035283194057566, + "grad_norm": 1.6376465559005737, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.890888512134552, + "num_tokens": 454412243.0, + "step": 12943 + }, + { + "epoch": 2.4037140204271124, + "grad_norm": 1.5687345266342163, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8920572996139526, + "num_tokens": 454447452.0, + "step": 12944 + }, + { + "epoch": 2.403899721448468, + "grad_norm": 1.5807428359985352, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8893598318099976, + "num_tokens": 454480076.0, + "step": 12945 + }, + { + "epoch": 2.404085422469824, + "grad_norm": 1.6772292852401733, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8934192061424255, + "num_tokens": 454511645.0, + "step": 12946 + }, + { + "epoch": 2.404271123491179, + "grad_norm": 1.5208510160446167, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8850225210189819, + "num_tokens": 454547123.0, + "step": 12947 + }, + { + "epoch": 2.404456824512535, + "grad_norm": 1.6193139553070068, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8805086612701416, + "num_tokens": 454581802.0, + "step": 12948 + }, + { + "epoch": 2.4046425255338906, + "grad_norm": 1.568149447441101, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8922539949417114, + "num_tokens": 454613915.0, + "step": 12949 + }, + { + "epoch": 2.404828226555246, + "grad_norm": 1.5886471271514893, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8834328651428223, + "num_tokens": 454649007.0, + "step": 12950 + }, + { + "epoch": 2.4050139275766016, + "grad_norm": 1.6187140941619873, + "learning_rate": 1e-06, + "loss": 0.2911, + "mean_token_accuracy": 0.8976756930351257, + "num_tokens": 454683491.0, + "step": 12951 + }, + { + "epoch": 2.4051996285979573, + "grad_norm": 1.5691230297088623, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8729658126831055, + "num_tokens": 454718646.0, + "step": 12952 + }, + { + "epoch": 2.405385329619313, + "grad_norm": 1.4376755952835083, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8907378315925598, + "num_tokens": 454759579.0, + "step": 12953 + }, + { + "epoch": 2.4055710306406684, + "grad_norm": 1.472289800643921, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.883525013923645, + "num_tokens": 454798745.0, + "step": 12954 + }, + { + "epoch": 2.405756731662024, + "grad_norm": 1.6130361557006836, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8786820769309998, + "num_tokens": 454834353.0, + "step": 12955 + }, + { + "epoch": 2.40594243268338, + "grad_norm": 1.5676056146621704, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8983891010284424, + "num_tokens": 454866192.0, + "step": 12956 + }, + { + "epoch": 2.4061281337047355, + "grad_norm": 1.7978180646896362, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8832349181175232, + "num_tokens": 454893827.0, + "step": 12957 + }, + { + "epoch": 2.406313834726091, + "grad_norm": 1.5272947549819946, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8747806549072266, + "num_tokens": 454931247.0, + "step": 12958 + }, + { + "epoch": 2.4064995357474466, + "grad_norm": 1.4755792617797852, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8919419646263123, + "num_tokens": 454967367.0, + "step": 12959 + }, + { + "epoch": 2.4066852367688023, + "grad_norm": 1.608183741569519, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8872624635696411, + "num_tokens": 455002080.0, + "step": 12960 + }, + { + "epoch": 2.406870937790158, + "grad_norm": 1.5411196947097778, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8811254501342773, + "num_tokens": 455037360.0, + "step": 12961 + }, + { + "epoch": 2.4070566388115133, + "grad_norm": 1.7296466827392578, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8761658072471619, + "num_tokens": 455065285.0, + "step": 12962 + }, + { + "epoch": 2.407242339832869, + "grad_norm": 1.4808201789855957, + "learning_rate": 1e-06, + "loss": 0.2858, + "mean_token_accuracy": 0.9012998342514038, + "num_tokens": 455099501.0, + "step": 12963 + }, + { + "epoch": 2.407428040854225, + "grad_norm": 1.483007550239563, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.8988701105117798, + "num_tokens": 455134236.0, + "step": 12964 + }, + { + "epoch": 2.40761374187558, + "grad_norm": 1.546728253364563, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8831734657287598, + "num_tokens": 455172017.0, + "step": 12965 + }, + { + "epoch": 2.407799442896936, + "grad_norm": 1.5594302415847778, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8693199157714844, + "num_tokens": 455210086.0, + "step": 12966 + }, + { + "epoch": 2.4079851439182915, + "grad_norm": 1.539705514907837, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8791062235832214, + "num_tokens": 455247557.0, + "step": 12967 + }, + { + "epoch": 2.4081708449396473, + "grad_norm": 1.708290457725525, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8734009265899658, + "num_tokens": 455282756.0, + "step": 12968 + }, + { + "epoch": 2.408356545961003, + "grad_norm": 1.715180516242981, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8727589845657349, + "num_tokens": 455315410.0, + "step": 12969 + }, + { + "epoch": 2.4085422469823583, + "grad_norm": 1.6402963399887085, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.897800087928772, + "num_tokens": 455349302.0, + "step": 12970 + }, + { + "epoch": 2.408727948003714, + "grad_norm": 1.6164753437042236, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8937784433364868, + "num_tokens": 455381290.0, + "step": 12971 + }, + { + "epoch": 2.4089136490250698, + "grad_norm": 1.6723823547363281, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8949683308601379, + "num_tokens": 455412279.0, + "step": 12972 + }, + { + "epoch": 2.409099350046425, + "grad_norm": 1.4149171113967896, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8875791430473328, + "num_tokens": 455458950.0, + "step": 12973 + }, + { + "epoch": 2.409285051067781, + "grad_norm": 1.4287244081497192, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.892235517501831, + "num_tokens": 455498575.0, + "step": 12974 + }, + { + "epoch": 2.4094707520891365, + "grad_norm": 1.6771374940872192, + "learning_rate": 1e-06, + "loss": 0.3877, + "mean_token_accuracy": 0.8656025528907776, + "num_tokens": 455535572.0, + "step": 12975 + }, + { + "epoch": 2.4096564531104923, + "grad_norm": 1.7078146934509277, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8793214559555054, + "num_tokens": 455568233.0, + "step": 12976 + }, + { + "epoch": 2.4098421541318475, + "grad_norm": 1.5498112440109253, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8969664573669434, + "num_tokens": 455601849.0, + "step": 12977 + }, + { + "epoch": 2.4100278551532033, + "grad_norm": 1.7248530387878418, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8777419328689575, + "num_tokens": 455634411.0, + "step": 12978 + }, + { + "epoch": 2.410213556174559, + "grad_norm": 1.7139616012573242, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8905333280563354, + "num_tokens": 455664462.0, + "step": 12979 + }, + { + "epoch": 2.4103992571959147, + "grad_norm": 1.6784967184066772, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8716678619384766, + "num_tokens": 455698997.0, + "step": 12980 + }, + { + "epoch": 2.41058495821727, + "grad_norm": 1.5016552209854126, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.886802077293396, + "num_tokens": 455735137.0, + "step": 12981 + }, + { + "epoch": 2.4107706592386258, + "grad_norm": 1.5936129093170166, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8868348598480225, + "num_tokens": 455770777.0, + "step": 12982 + }, + { + "epoch": 2.4109563602599815, + "grad_norm": 1.5259661674499512, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8898511528968811, + "num_tokens": 455807247.0, + "step": 12983 + }, + { + "epoch": 2.4111420612813372, + "grad_norm": 1.5898422002792358, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8853846192359924, + "num_tokens": 455841225.0, + "step": 12984 + }, + { + "epoch": 2.4113277623026925, + "grad_norm": 1.672446846961975, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8874362111091614, + "num_tokens": 455867843.0, + "step": 12985 + }, + { + "epoch": 2.4115134633240483, + "grad_norm": 1.5280067920684814, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8862166404724121, + "num_tokens": 455901861.0, + "step": 12986 + }, + { + "epoch": 2.411699164345404, + "grad_norm": 1.681270956993103, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.8746044039726257, + "num_tokens": 455937239.0, + "step": 12987 + }, + { + "epoch": 2.4118848653667593, + "grad_norm": 1.5995216369628906, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8917044997215271, + "num_tokens": 455973802.0, + "step": 12988 + }, + { + "epoch": 2.412070566388115, + "grad_norm": 1.6634515523910522, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8894382119178772, + "num_tokens": 456008025.0, + "step": 12989 + }, + { + "epoch": 2.4122562674094707, + "grad_norm": 1.5511119365692139, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8713083267211914, + "num_tokens": 456047643.0, + "step": 12990 + }, + { + "epoch": 2.4124419684308265, + "grad_norm": 1.5185551643371582, + "learning_rate": 1e-06, + "loss": 0.2847, + "mean_token_accuracy": 0.9010299444198608, + "num_tokens": 456081653.0, + "step": 12991 + }, + { + "epoch": 2.412627669452182, + "grad_norm": 1.6148738861083984, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8836717009544373, + "num_tokens": 456116290.0, + "step": 12992 + }, + { + "epoch": 2.4128133704735375, + "grad_norm": 1.6356174945831299, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8962854146957397, + "num_tokens": 456150574.0, + "step": 12993 + }, + { + "epoch": 2.4129990714948932, + "grad_norm": 1.6518709659576416, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8901478052139282, + "num_tokens": 456187169.0, + "step": 12994 + }, + { + "epoch": 2.413184772516249, + "grad_norm": 1.6809730529785156, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8777475953102112, + "num_tokens": 456223923.0, + "step": 12995 + }, + { + "epoch": 2.4133704735376043, + "grad_norm": 1.6494386196136475, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8716919422149658, + "num_tokens": 456259724.0, + "step": 12996 + }, + { + "epoch": 2.41355617455896, + "grad_norm": 1.521909236907959, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8944710493087769, + "num_tokens": 456298096.0, + "step": 12997 + }, + { + "epoch": 2.4137418755803157, + "grad_norm": 1.6981699466705322, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8812837600708008, + "num_tokens": 456330410.0, + "step": 12998 + }, + { + "epoch": 2.4139275766016715, + "grad_norm": 1.503337025642395, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8815476894378662, + "num_tokens": 456368961.0, + "step": 12999 + }, + { + "epoch": 2.4141132776230267, + "grad_norm": 1.5562617778778076, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.8969072103500366, + "num_tokens": 456403425.0, + "step": 13000 + }, + { + "epoch": 2.4142989786443825, + "grad_norm": 1.49252188205719, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.88494873046875, + "num_tokens": 456442127.0, + "step": 13001 + }, + { + "epoch": 2.414484679665738, + "grad_norm": 1.6915911436080933, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8801061511039734, + "num_tokens": 456471631.0, + "step": 13002 + }, + { + "epoch": 2.414670380687094, + "grad_norm": 1.485010027885437, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8908208608627319, + "num_tokens": 456507639.0, + "step": 13003 + }, + { + "epoch": 2.4148560817084492, + "grad_norm": 1.4496510028839111, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.888112485408783, + "num_tokens": 456545187.0, + "step": 13004 + }, + { + "epoch": 2.415041782729805, + "grad_norm": 1.5215550661087036, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8945284485816956, + "num_tokens": 456579031.0, + "step": 13005 + }, + { + "epoch": 2.4152274837511607, + "grad_norm": 1.4666649103164673, + "learning_rate": 1e-06, + "loss": 0.2798, + "mean_token_accuracy": 0.9006667137145996, + "num_tokens": 456613917.0, + "step": 13006 + }, + { + "epoch": 2.4154131847725164, + "grad_norm": 1.5653678178787231, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8902329802513123, + "num_tokens": 456647729.0, + "step": 13007 + }, + { + "epoch": 2.4155988857938717, + "grad_norm": 1.6875684261322021, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8822453022003174, + "num_tokens": 456680677.0, + "step": 13008 + }, + { + "epoch": 2.4157845868152275, + "grad_norm": 1.5023518800735474, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8887832164764404, + "num_tokens": 456719018.0, + "step": 13009 + }, + { + "epoch": 2.415970287836583, + "grad_norm": 1.9157445430755615, + "learning_rate": 1e-06, + "loss": 0.3647, + "mean_token_accuracy": 0.8757997155189514, + "num_tokens": 456746308.0, + "step": 13010 + }, + { + "epoch": 2.4161559888579385, + "grad_norm": 1.574112057685852, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8813691139221191, + "num_tokens": 456778976.0, + "step": 13011 + }, + { + "epoch": 2.416341689879294, + "grad_norm": 1.558396816253662, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8899346590042114, + "num_tokens": 456815887.0, + "step": 13012 + }, + { + "epoch": 2.41652739090065, + "grad_norm": 1.4491764307022095, + "learning_rate": 1e-06, + "loss": 0.2754, + "mean_token_accuracy": 0.9010589122772217, + "num_tokens": 456853197.0, + "step": 13013 + }, + { + "epoch": 2.4167130919220057, + "grad_norm": 1.5945231914520264, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8792992830276489, + "num_tokens": 456890725.0, + "step": 13014 + }, + { + "epoch": 2.4168987929433614, + "grad_norm": 1.6072490215301514, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8723615407943726, + "num_tokens": 456926615.0, + "step": 13015 + }, + { + "epoch": 2.4170844939647167, + "grad_norm": 1.536773920059204, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.899590015411377, + "num_tokens": 456961358.0, + "step": 13016 + }, + { + "epoch": 2.4172701949860724, + "grad_norm": 1.6053510904312134, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.876763105392456, + "num_tokens": 456999163.0, + "step": 13017 + }, + { + "epoch": 2.417455896007428, + "grad_norm": 1.5109808444976807, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.889211118221283, + "num_tokens": 457035210.0, + "step": 13018 + }, + { + "epoch": 2.4176415970287835, + "grad_norm": 1.572758436203003, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8847025632858276, + "num_tokens": 457071013.0, + "step": 13019 + }, + { + "epoch": 2.417827298050139, + "grad_norm": 1.6423169374465942, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8820744752883911, + "num_tokens": 457104545.0, + "step": 13020 + }, + { + "epoch": 2.418012999071495, + "grad_norm": 1.5246549844741821, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8882969617843628, + "num_tokens": 457142943.0, + "step": 13021 + }, + { + "epoch": 2.4181987000928507, + "grad_norm": 1.5454524755477905, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8842719793319702, + "num_tokens": 457185072.0, + "step": 13022 + }, + { + "epoch": 2.418384401114206, + "grad_norm": 1.7021567821502686, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8836025595664978, + "num_tokens": 457215208.0, + "step": 13023 + }, + { + "epoch": 2.4185701021355617, + "grad_norm": 1.5491808652877808, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8859403133392334, + "num_tokens": 457252342.0, + "step": 13024 + }, + { + "epoch": 2.4187558031569174, + "grad_norm": 1.6934646368026733, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8846352100372314, + "num_tokens": 457286085.0, + "step": 13025 + }, + { + "epoch": 2.418941504178273, + "grad_norm": 1.6311371326446533, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.8952454328536987, + "num_tokens": 457316349.0, + "step": 13026 + }, + { + "epoch": 2.4191272051996284, + "grad_norm": 1.5260977745056152, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8825910091400146, + "num_tokens": 457353812.0, + "step": 13027 + }, + { + "epoch": 2.419312906220984, + "grad_norm": 1.5621305704116821, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8820443749427795, + "num_tokens": 457387908.0, + "step": 13028 + }, + { + "epoch": 2.41949860724234, + "grad_norm": 1.4928531646728516, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8824942708015442, + "num_tokens": 457426648.0, + "step": 13029 + }, + { + "epoch": 2.4196843082636956, + "grad_norm": 1.569808840751648, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8873021602630615, + "num_tokens": 457462268.0, + "step": 13030 + }, + { + "epoch": 2.419870009285051, + "grad_norm": 1.6977704763412476, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8717153072357178, + "num_tokens": 457496073.0, + "step": 13031 + }, + { + "epoch": 2.4200557103064066, + "grad_norm": 1.7238774299621582, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.900673508644104, + "num_tokens": 457523241.0, + "step": 13032 + }, + { + "epoch": 2.4202414113277624, + "grad_norm": 1.5334721803665161, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8949097990989685, + "num_tokens": 457556517.0, + "step": 13033 + }, + { + "epoch": 2.4204271123491177, + "grad_norm": 1.6583027839660645, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8865441679954529, + "num_tokens": 457586475.0, + "step": 13034 + }, + { + "epoch": 2.4206128133704734, + "grad_norm": 1.4600908756256104, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.8957503437995911, + "num_tokens": 457622997.0, + "step": 13035 + }, + { + "epoch": 2.420798514391829, + "grad_norm": 1.7003728151321411, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8928499817848206, + "num_tokens": 457652474.0, + "step": 13036 + }, + { + "epoch": 2.420984215413185, + "grad_norm": 1.633259654045105, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8940469026565552, + "num_tokens": 457683873.0, + "step": 13037 + }, + { + "epoch": 2.4211699164345406, + "grad_norm": 1.518497109413147, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8933669924736023, + "num_tokens": 457717571.0, + "step": 13038 + }, + { + "epoch": 2.421355617455896, + "grad_norm": 1.870108723640442, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8802812695503235, + "num_tokens": 457746324.0, + "step": 13039 + }, + { + "epoch": 2.4215413184772516, + "grad_norm": 1.5673470497131348, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8886569738388062, + "num_tokens": 457777499.0, + "step": 13040 + }, + { + "epoch": 2.4217270194986074, + "grad_norm": 1.7299065589904785, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8822293877601624, + "num_tokens": 457808669.0, + "step": 13041 + }, + { + "epoch": 2.4219127205199626, + "grad_norm": 1.4891550540924072, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8987537622451782, + "num_tokens": 457844259.0, + "step": 13042 + }, + { + "epoch": 2.4220984215413184, + "grad_norm": 1.635504126548767, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8774031400680542, + "num_tokens": 457876946.0, + "step": 13043 + }, + { + "epoch": 2.422284122562674, + "grad_norm": 1.4816073179244995, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8868130445480347, + "num_tokens": 457917836.0, + "step": 13044 + }, + { + "epoch": 2.42246982358403, + "grad_norm": 1.5790690183639526, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8793575763702393, + "num_tokens": 457951247.0, + "step": 13045 + }, + { + "epoch": 2.422655524605385, + "grad_norm": 1.463227391242981, + "learning_rate": 1e-06, + "loss": 0.3151, + "mean_token_accuracy": 0.8900688886642456, + "num_tokens": 457987928.0, + "step": 13046 + }, + { + "epoch": 2.422841225626741, + "grad_norm": 1.4592260122299194, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8830958008766174, + "num_tokens": 458028534.0, + "step": 13047 + }, + { + "epoch": 2.4230269266480966, + "grad_norm": 1.6033339500427246, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8757612705230713, + "num_tokens": 458061264.0, + "step": 13048 + }, + { + "epoch": 2.4232126276694523, + "grad_norm": 1.589296579360962, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8740189075469971, + "num_tokens": 458095752.0, + "step": 13049 + }, + { + "epoch": 2.4233983286908076, + "grad_norm": 1.470432996749878, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8853594064712524, + "num_tokens": 458134856.0, + "step": 13050 + }, + { + "epoch": 2.4235840297121634, + "grad_norm": 1.5358144044876099, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8929527997970581, + "num_tokens": 458168372.0, + "step": 13051 + }, + { + "epoch": 2.423769730733519, + "grad_norm": 1.5623602867126465, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8899770975112915, + "num_tokens": 458203260.0, + "step": 13052 + }, + { + "epoch": 2.423955431754875, + "grad_norm": 1.545188546180725, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8869987726211548, + "num_tokens": 458238776.0, + "step": 13053 + }, + { + "epoch": 2.42414113277623, + "grad_norm": 1.6679264307022095, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8912439346313477, + "num_tokens": 458269306.0, + "step": 13054 + }, + { + "epoch": 2.424326833797586, + "grad_norm": 1.617020606994629, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8823903203010559, + "num_tokens": 458308479.0, + "step": 13055 + }, + { + "epoch": 2.4245125348189416, + "grad_norm": 1.4737945795059204, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8809993863105774, + "num_tokens": 458351337.0, + "step": 13056 + }, + { + "epoch": 2.4246982358402973, + "grad_norm": 1.6383007764816284, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.886803150177002, + "num_tokens": 458382388.0, + "step": 13057 + }, + { + "epoch": 2.4248839368616526, + "grad_norm": 1.5906296968460083, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8726052641868591, + "num_tokens": 458419793.0, + "step": 13058 + }, + { + "epoch": 2.4250696378830083, + "grad_norm": 1.6049528121948242, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.890916109085083, + "num_tokens": 458454044.0, + "step": 13059 + }, + { + "epoch": 2.425255338904364, + "grad_norm": 1.4313379526138306, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8938603401184082, + "num_tokens": 458492571.0, + "step": 13060 + }, + { + "epoch": 2.42544103992572, + "grad_norm": 1.4462494850158691, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8894025087356567, + "num_tokens": 458532398.0, + "step": 13061 + }, + { + "epoch": 2.425626740947075, + "grad_norm": 1.5153495073318481, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8773249387741089, + "num_tokens": 458571021.0, + "step": 13062 + }, + { + "epoch": 2.425812441968431, + "grad_norm": 1.6098805665969849, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8869017958641052, + "num_tokens": 458602321.0, + "step": 13063 + }, + { + "epoch": 2.4259981429897866, + "grad_norm": 1.5141913890838623, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8901586532592773, + "num_tokens": 458636898.0, + "step": 13064 + }, + { + "epoch": 2.426183844011142, + "grad_norm": 1.6270601749420166, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8697654008865356, + "num_tokens": 458671882.0, + "step": 13065 + }, + { + "epoch": 2.4263695450324976, + "grad_norm": 1.547420859336853, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8963829874992371, + "num_tokens": 458703763.0, + "step": 13066 + }, + { + "epoch": 2.4265552460538533, + "grad_norm": 1.5727159976959229, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.889096736907959, + "num_tokens": 458735950.0, + "step": 13067 + }, + { + "epoch": 2.426740947075209, + "grad_norm": 1.5366792678833008, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8906206488609314, + "num_tokens": 458772527.0, + "step": 13068 + }, + { + "epoch": 2.4269266480965648, + "grad_norm": 1.6473873853683472, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8841499090194702, + "num_tokens": 458803439.0, + "step": 13069 + }, + { + "epoch": 2.42711234911792, + "grad_norm": 1.5470861196517944, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8738964796066284, + "num_tokens": 458841750.0, + "step": 13070 + }, + { + "epoch": 2.427298050139276, + "grad_norm": 1.5937614440917969, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.88835608959198, + "num_tokens": 458873049.0, + "step": 13071 + }, + { + "epoch": 2.4274837511606315, + "grad_norm": 1.668506383895874, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8903683423995972, + "num_tokens": 458903929.0, + "step": 13072 + }, + { + "epoch": 2.427669452181987, + "grad_norm": 1.479502558708191, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8933480978012085, + "num_tokens": 458940893.0, + "step": 13073 + }, + { + "epoch": 2.4278551532033426, + "grad_norm": 1.483779788017273, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8860747814178467, + "num_tokens": 458978623.0, + "step": 13074 + }, + { + "epoch": 2.4280408542246983, + "grad_norm": 1.3833590745925903, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8960238099098206, + "num_tokens": 459019156.0, + "step": 13075 + }, + { + "epoch": 2.428226555246054, + "grad_norm": 1.8182631731033325, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8773126602172852, + "num_tokens": 459047278.0, + "step": 13076 + }, + { + "epoch": 2.4284122562674093, + "grad_norm": 1.597069501876831, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8721219897270203, + "num_tokens": 459081061.0, + "step": 13077 + }, + { + "epoch": 2.428597957288765, + "grad_norm": 1.6264214515686035, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8682804107666016, + "num_tokens": 459118887.0, + "step": 13078 + }, + { + "epoch": 2.4287836583101208, + "grad_norm": 1.5001295804977417, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8798502087593079, + "num_tokens": 459158518.0, + "step": 13079 + }, + { + "epoch": 2.4289693593314765, + "grad_norm": 1.6035281419754028, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.879770815372467, + "num_tokens": 459192852.0, + "step": 13080 + }, + { + "epoch": 2.429155060352832, + "grad_norm": 1.5891265869140625, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8954179883003235, + "num_tokens": 459226332.0, + "step": 13081 + }, + { + "epoch": 2.4293407613741875, + "grad_norm": 1.5641189813613892, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8774136900901794, + "num_tokens": 459264272.0, + "step": 13082 + }, + { + "epoch": 2.4295264623955433, + "grad_norm": 1.5501126050949097, + "learning_rate": 1e-06, + "loss": 0.2869, + "mean_token_accuracy": 0.8958700299263, + "num_tokens": 459297482.0, + "step": 13083 + }, + { + "epoch": 2.429712163416899, + "grad_norm": 1.550512671470642, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8831364512443542, + "num_tokens": 459335476.0, + "step": 13084 + }, + { + "epoch": 2.4298978644382543, + "grad_norm": 1.4274169206619263, + "learning_rate": 1e-06, + "loss": 0.3738, + "mean_token_accuracy": 0.8710997104644775, + "num_tokens": 459382718.0, + "step": 13085 + }, + { + "epoch": 2.43008356545961, + "grad_norm": 1.5897003412246704, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8876805305480957, + "num_tokens": 459417266.0, + "step": 13086 + }, + { + "epoch": 2.4302692664809658, + "grad_norm": 1.679067850112915, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8912739157676697, + "num_tokens": 459451918.0, + "step": 13087 + }, + { + "epoch": 2.430454967502321, + "grad_norm": 1.6829822063446045, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8838653564453125, + "num_tokens": 459485161.0, + "step": 13088 + }, + { + "epoch": 2.4306406685236768, + "grad_norm": 1.5077346563339233, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.89751136302948, + "num_tokens": 459522406.0, + "step": 13089 + }, + { + "epoch": 2.4308263695450325, + "grad_norm": 1.674749732017517, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8774991035461426, + "num_tokens": 459557140.0, + "step": 13090 + }, + { + "epoch": 2.4310120705663882, + "grad_norm": 1.5466203689575195, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8921621441841125, + "num_tokens": 459593569.0, + "step": 13091 + }, + { + "epoch": 2.431197771587744, + "grad_norm": 1.7837560176849365, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8765255212783813, + "num_tokens": 459626342.0, + "step": 13092 + }, + { + "epoch": 2.4313834726090993, + "grad_norm": 1.5360134840011597, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8842023015022278, + "num_tokens": 459664721.0, + "step": 13093 + }, + { + "epoch": 2.431569173630455, + "grad_norm": 1.6978765726089478, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8892157077789307, + "num_tokens": 459694475.0, + "step": 13094 + }, + { + "epoch": 2.4317548746518107, + "grad_norm": 1.5201241970062256, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8959255218505859, + "num_tokens": 459729299.0, + "step": 13095 + }, + { + "epoch": 2.431940575673166, + "grad_norm": 1.529096245765686, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8880646228790283, + "num_tokens": 459765526.0, + "step": 13096 + }, + { + "epoch": 2.4321262766945217, + "grad_norm": 1.5579880475997925, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8909931182861328, + "num_tokens": 459800544.0, + "step": 13097 + }, + { + "epoch": 2.4323119777158775, + "grad_norm": 1.5437378883361816, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8781189918518066, + "num_tokens": 459842908.0, + "step": 13098 + }, + { + "epoch": 2.432497678737233, + "grad_norm": 1.6580923795700073, + "learning_rate": 1e-06, + "loss": 0.2903, + "mean_token_accuracy": 0.8997900485992432, + "num_tokens": 459872162.0, + "step": 13099 + }, + { + "epoch": 2.4326833797585885, + "grad_norm": 1.7701494693756104, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8832382559776306, + "num_tokens": 459902084.0, + "step": 13100 + }, + { + "epoch": 2.4328690807799442, + "grad_norm": 1.5765116214752197, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.9017994403839111, + "num_tokens": 459933450.0, + "step": 13101 + }, + { + "epoch": 2.4330547818013, + "grad_norm": 1.4629547595977783, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.897086501121521, + "num_tokens": 459971933.0, + "step": 13102 + }, + { + "epoch": 2.4332404828226557, + "grad_norm": 1.5514661073684692, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8890533447265625, + "num_tokens": 460008499.0, + "step": 13103 + }, + { + "epoch": 2.433426183844011, + "grad_norm": 1.7376869916915894, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8796597123146057, + "num_tokens": 460040928.0, + "step": 13104 + }, + { + "epoch": 2.4336118848653667, + "grad_norm": 1.5635207891464233, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8911128044128418, + "num_tokens": 460073606.0, + "step": 13105 + }, + { + "epoch": 2.4337975858867225, + "grad_norm": 1.5620266199111938, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8852565884590149, + "num_tokens": 460113365.0, + "step": 13106 + }, + { + "epoch": 2.433983286908078, + "grad_norm": 1.560062050819397, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8955867290496826, + "num_tokens": 460147603.0, + "step": 13107 + }, + { + "epoch": 2.4341689879294335, + "grad_norm": 1.5682002305984497, + "learning_rate": 1e-06, + "loss": 0.282, + "mean_token_accuracy": 0.8974233865737915, + "num_tokens": 460178711.0, + "step": 13108 + }, + { + "epoch": 2.434354688950789, + "grad_norm": 1.6369075775146484, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8888294696807861, + "num_tokens": 460212597.0, + "step": 13109 + }, + { + "epoch": 2.434540389972145, + "grad_norm": 1.6459916830062866, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8814497590065002, + "num_tokens": 460244683.0, + "step": 13110 + }, + { + "epoch": 2.4347260909935002, + "grad_norm": 1.5926945209503174, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8851046562194824, + "num_tokens": 460281253.0, + "step": 13111 + }, + { + "epoch": 2.434911792014856, + "grad_norm": 1.6074618101119995, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8743029236793518, + "num_tokens": 460317470.0, + "step": 13112 + }, + { + "epoch": 2.4350974930362117, + "grad_norm": 1.6616421937942505, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8876054883003235, + "num_tokens": 460350926.0, + "step": 13113 + }, + { + "epoch": 2.4352831940575674, + "grad_norm": 1.6077475547790527, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8819470405578613, + "num_tokens": 460387911.0, + "step": 13114 + }, + { + "epoch": 2.435468895078923, + "grad_norm": 1.590753197669983, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8855355381965637, + "num_tokens": 460420678.0, + "step": 13115 + }, + { + "epoch": 2.4356545961002785, + "grad_norm": 1.640605092048645, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8822985887527466, + "num_tokens": 460455052.0, + "step": 13116 + }, + { + "epoch": 2.435840297121634, + "grad_norm": 1.6618905067443848, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8852553367614746, + "num_tokens": 460486497.0, + "step": 13117 + }, + { + "epoch": 2.43602599814299, + "grad_norm": 1.6019322872161865, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8751450777053833, + "num_tokens": 460522761.0, + "step": 13118 + }, + { + "epoch": 2.436211699164345, + "grad_norm": 1.6051057577133179, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8868216872215271, + "num_tokens": 460557079.0, + "step": 13119 + }, + { + "epoch": 2.436397400185701, + "grad_norm": 1.5418487787246704, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.883779764175415, + "num_tokens": 460594127.0, + "step": 13120 + }, + { + "epoch": 2.4365831012070567, + "grad_norm": 1.5666496753692627, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8922433853149414, + "num_tokens": 460628365.0, + "step": 13121 + }, + { + "epoch": 2.4367688022284124, + "grad_norm": 1.5523966550827026, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8778005838394165, + "num_tokens": 460666988.0, + "step": 13122 + }, + { + "epoch": 2.4369545032497677, + "grad_norm": 1.3950005769729614, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8866912126541138, + "num_tokens": 460711127.0, + "step": 13123 + }, + { + "epoch": 2.4371402042711234, + "grad_norm": 1.4586012363433838, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8821821808815002, + "num_tokens": 460752643.0, + "step": 13124 + }, + { + "epoch": 2.437325905292479, + "grad_norm": 1.6734888553619385, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8928014039993286, + "num_tokens": 460781365.0, + "step": 13125 + }, + { + "epoch": 2.437511606313835, + "grad_norm": 1.550721526145935, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8891904354095459, + "num_tokens": 460817346.0, + "step": 13126 + }, + { + "epoch": 2.43769730733519, + "grad_norm": 1.7581819295883179, + "learning_rate": 1e-06, + "loss": 0.2856, + "mean_token_accuracy": 0.8972198367118835, + "num_tokens": 460846263.0, + "step": 13127 + }, + { + "epoch": 2.437883008356546, + "grad_norm": 1.6407021284103394, + "learning_rate": 1e-06, + "loss": 0.295, + "mean_token_accuracy": 0.896225094795227, + "num_tokens": 460875289.0, + "step": 13128 + }, + { + "epoch": 2.4380687093779017, + "grad_norm": 1.6094766855239868, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8749877214431763, + "num_tokens": 460910892.0, + "step": 13129 + }, + { + "epoch": 2.4382544103992574, + "grad_norm": 1.5139203071594238, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8909297585487366, + "num_tokens": 460948579.0, + "step": 13130 + }, + { + "epoch": 2.4384401114206127, + "grad_norm": 1.8775863647460938, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.869895875453949, + "num_tokens": 460979634.0, + "step": 13131 + }, + { + "epoch": 2.4386258124419684, + "grad_norm": 1.6299093961715698, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8812955021858215, + "num_tokens": 461018217.0, + "step": 13132 + }, + { + "epoch": 2.438811513463324, + "grad_norm": 1.7547527551651, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8687760829925537, + "num_tokens": 461052414.0, + "step": 13133 + }, + { + "epoch": 2.4389972144846794, + "grad_norm": 1.5660594701766968, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.88181072473526, + "num_tokens": 461087228.0, + "step": 13134 + }, + { + "epoch": 2.439182915506035, + "grad_norm": 1.545891523361206, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8868552446365356, + "num_tokens": 461126683.0, + "step": 13135 + }, + { + "epoch": 2.439368616527391, + "grad_norm": 1.6476428508758545, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8927829265594482, + "num_tokens": 461156827.0, + "step": 13136 + }, + { + "epoch": 2.4395543175487466, + "grad_norm": 1.4923878908157349, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8918136954307556, + "num_tokens": 461196177.0, + "step": 13137 + }, + { + "epoch": 2.4397400185701024, + "grad_norm": 1.497028112411499, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.887904167175293, + "num_tokens": 461234788.0, + "step": 13138 + }, + { + "epoch": 2.4399257195914577, + "grad_norm": 1.5891764163970947, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8820862770080566, + "num_tokens": 461268723.0, + "step": 13139 + }, + { + "epoch": 2.4401114206128134, + "grad_norm": 1.7306480407714844, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8814224004745483, + "num_tokens": 461298479.0, + "step": 13140 + }, + { + "epoch": 2.440297121634169, + "grad_norm": 1.3573931455612183, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8850584030151367, + "num_tokens": 461344224.0, + "step": 13141 + }, + { + "epoch": 2.4404828226555244, + "grad_norm": 1.6625730991363525, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.894087016582489, + "num_tokens": 461374664.0, + "step": 13142 + }, + { + "epoch": 2.44066852367688, + "grad_norm": 1.5545570850372314, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8911015391349792, + "num_tokens": 461410819.0, + "step": 13143 + }, + { + "epoch": 2.440854224698236, + "grad_norm": 1.7968266010284424, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8801700472831726, + "num_tokens": 461439439.0, + "step": 13144 + }, + { + "epoch": 2.4410399257195916, + "grad_norm": 1.5364081859588623, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8884447813034058, + "num_tokens": 461477629.0, + "step": 13145 + }, + { + "epoch": 2.441225626740947, + "grad_norm": 1.6330909729003906, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8741915225982666, + "num_tokens": 461515226.0, + "step": 13146 + }, + { + "epoch": 2.4414113277623026, + "grad_norm": 1.5526885986328125, + "learning_rate": 1e-06, + "loss": 0.267, + "mean_token_accuracy": 0.9048200249671936, + "num_tokens": 461552357.0, + "step": 13147 + }, + { + "epoch": 2.4415970287836584, + "grad_norm": 1.603786826133728, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.9011678099632263, + "num_tokens": 461584338.0, + "step": 13148 + }, + { + "epoch": 2.441782729805014, + "grad_norm": 1.5256781578063965, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8833285570144653, + "num_tokens": 461623437.0, + "step": 13149 + }, + { + "epoch": 2.4419684308263694, + "grad_norm": 1.6405973434448242, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8915007710456848, + "num_tokens": 461655086.0, + "step": 13150 + }, + { + "epoch": 2.442154131847725, + "grad_norm": 1.569872498512268, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.882693350315094, + "num_tokens": 461687048.0, + "step": 13151 + }, + { + "epoch": 2.442339832869081, + "grad_norm": 1.6590354442596436, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.885079026222229, + "num_tokens": 461719380.0, + "step": 13152 + }, + { + "epoch": 2.4425255338904366, + "grad_norm": 1.455886960029602, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.885273814201355, + "num_tokens": 461758695.0, + "step": 13153 + }, + { + "epoch": 2.442711234911792, + "grad_norm": 1.5199224948883057, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8920301198959351, + "num_tokens": 461793322.0, + "step": 13154 + }, + { + "epoch": 2.4428969359331476, + "grad_norm": 1.5930883884429932, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8921473026275635, + "num_tokens": 461825994.0, + "step": 13155 + }, + { + "epoch": 2.4430826369545033, + "grad_norm": 1.5246039628982544, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8767125606536865, + "num_tokens": 461864076.0, + "step": 13156 + }, + { + "epoch": 2.4432683379758586, + "grad_norm": 1.5459260940551758, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8821903467178345, + "num_tokens": 461905340.0, + "step": 13157 + }, + { + "epoch": 2.4434540389972144, + "grad_norm": 1.5078788995742798, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8893582820892334, + "num_tokens": 461944654.0, + "step": 13158 + }, + { + "epoch": 2.44363974001857, + "grad_norm": 1.5908108949661255, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.8926191926002502, + "num_tokens": 461976056.0, + "step": 13159 + }, + { + "epoch": 2.443825441039926, + "grad_norm": 1.5652480125427246, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8822101354598999, + "num_tokens": 462012151.0, + "step": 13160 + }, + { + "epoch": 2.4440111420612816, + "grad_norm": 1.6224876642227173, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8778447508811951, + "num_tokens": 462048754.0, + "step": 13161 + }, + { + "epoch": 2.444196843082637, + "grad_norm": 1.741028070449829, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8832414150238037, + "num_tokens": 462077317.0, + "step": 13162 + }, + { + "epoch": 2.4443825441039926, + "grad_norm": 1.5696265697479248, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8873807191848755, + "num_tokens": 462114454.0, + "step": 13163 + }, + { + "epoch": 2.4445682451253483, + "grad_norm": 1.7001824378967285, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8822248578071594, + "num_tokens": 462144791.0, + "step": 13164 + }, + { + "epoch": 2.4447539461467036, + "grad_norm": 1.6094032526016235, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8821049928665161, + "num_tokens": 462176873.0, + "step": 13165 + }, + { + "epoch": 2.4449396471680593, + "grad_norm": 1.628523588180542, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8764200210571289, + "num_tokens": 462210486.0, + "step": 13166 + }, + { + "epoch": 2.445125348189415, + "grad_norm": 1.6162185668945312, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.879216194152832, + "num_tokens": 462249870.0, + "step": 13167 + }, + { + "epoch": 2.445311049210771, + "grad_norm": 1.6828709840774536, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8900519609451294, + "num_tokens": 462281849.0, + "step": 13168 + }, + { + "epoch": 2.445496750232126, + "grad_norm": 1.4659440517425537, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.895836353302002, + "num_tokens": 462317088.0, + "step": 13169 + }, + { + "epoch": 2.445682451253482, + "grad_norm": 1.471993088722229, + "learning_rate": 1e-06, + "loss": 0.2898, + "mean_token_accuracy": 0.8972629308700562, + "num_tokens": 462354778.0, + "step": 13170 + }, + { + "epoch": 2.4458681522748376, + "grad_norm": 1.5595847368240356, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8868907690048218, + "num_tokens": 462392219.0, + "step": 13171 + }, + { + "epoch": 2.4460538532961933, + "grad_norm": 1.659374475479126, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8971680402755737, + "num_tokens": 462424525.0, + "step": 13172 + }, + { + "epoch": 2.4462395543175486, + "grad_norm": 1.5018062591552734, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.896173894405365, + "num_tokens": 462460131.0, + "step": 13173 + }, + { + "epoch": 2.4464252553389043, + "grad_norm": 1.5093963146209717, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8833233714103699, + "num_tokens": 462497144.0, + "step": 13174 + }, + { + "epoch": 2.44661095636026, + "grad_norm": 1.5893651247024536, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8981181383132935, + "num_tokens": 462526908.0, + "step": 13175 + }, + { + "epoch": 2.446796657381616, + "grad_norm": 1.460010290145874, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8908414840698242, + "num_tokens": 462562988.0, + "step": 13176 + }, + { + "epoch": 2.446982358402971, + "grad_norm": 1.5863946676254272, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8943604826927185, + "num_tokens": 462596005.0, + "step": 13177 + }, + { + "epoch": 2.447168059424327, + "grad_norm": 1.6507619619369507, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8724451065063477, + "num_tokens": 462629451.0, + "step": 13178 + }, + { + "epoch": 2.4473537604456825, + "grad_norm": 1.5789765119552612, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8893121480941772, + "num_tokens": 462666793.0, + "step": 13179 + }, + { + "epoch": 2.447539461467038, + "grad_norm": 1.4798717498779297, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8855458498001099, + "num_tokens": 462704777.0, + "step": 13180 + }, + { + "epoch": 2.4477251624883936, + "grad_norm": 1.5283321142196655, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.888594925403595, + "num_tokens": 462739928.0, + "step": 13181 + }, + { + "epoch": 2.4479108635097493, + "grad_norm": 1.5718296766281128, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8848312497138977, + "num_tokens": 462770880.0, + "step": 13182 + }, + { + "epoch": 2.448096564531105, + "grad_norm": 1.4897427558898926, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8948777914047241, + "num_tokens": 462807170.0, + "step": 13183 + }, + { + "epoch": 2.4482822655524608, + "grad_norm": 1.443486213684082, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8856496214866638, + "num_tokens": 462848729.0, + "step": 13184 + }, + { + "epoch": 2.448467966573816, + "grad_norm": 1.6173532009124756, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.88807213306427, + "num_tokens": 462884302.0, + "step": 13185 + }, + { + "epoch": 2.448653667595172, + "grad_norm": 1.6399301290512085, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8783451914787292, + "num_tokens": 462918532.0, + "step": 13186 + }, + { + "epoch": 2.4488393686165275, + "grad_norm": 1.6216864585876465, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8861891031265259, + "num_tokens": 462950128.0, + "step": 13187 + }, + { + "epoch": 2.449025069637883, + "grad_norm": 1.585603952407837, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8760073184967041, + "num_tokens": 462986601.0, + "step": 13188 + }, + { + "epoch": 2.4492107706592385, + "grad_norm": 1.5972808599472046, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8965169787406921, + "num_tokens": 463022968.0, + "step": 13189 + }, + { + "epoch": 2.4493964716805943, + "grad_norm": 1.7589759826660156, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8819479942321777, + "num_tokens": 463055744.0, + "step": 13190 + }, + { + "epoch": 2.44958217270195, + "grad_norm": 1.5648852586746216, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8841475248336792, + "num_tokens": 463099185.0, + "step": 13191 + }, + { + "epoch": 2.4497678737233053, + "grad_norm": 1.547920823097229, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8755568265914917, + "num_tokens": 463134599.0, + "step": 13192 + }, + { + "epoch": 2.449953574744661, + "grad_norm": 1.6278964281082153, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8854060173034668, + "num_tokens": 463165134.0, + "step": 13193 + }, + { + "epoch": 2.4501392757660168, + "grad_norm": 1.5490257740020752, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8972152471542358, + "num_tokens": 463197532.0, + "step": 13194 + }, + { + "epoch": 2.4503249767873725, + "grad_norm": 1.432341456413269, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8969706296920776, + "num_tokens": 463235761.0, + "step": 13195 + }, + { + "epoch": 2.4505106778087278, + "grad_norm": 1.5752724409103394, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.88003009557724, + "num_tokens": 463273993.0, + "step": 13196 + }, + { + "epoch": 2.4506963788300835, + "grad_norm": 1.6288906335830688, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8851341009140015, + "num_tokens": 463307160.0, + "step": 13197 + }, + { + "epoch": 2.4508820798514392, + "grad_norm": 1.5269854068756104, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8868175745010376, + "num_tokens": 463343089.0, + "step": 13198 + }, + { + "epoch": 2.451067780872795, + "grad_norm": 1.5617908239364624, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8871448040008545, + "num_tokens": 463376090.0, + "step": 13199 + }, + { + "epoch": 2.4512534818941503, + "grad_norm": 1.6136224269866943, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.876427173614502, + "num_tokens": 463410687.0, + "step": 13200 + }, + { + "epoch": 2.451439182915506, + "grad_norm": 1.4456512928009033, + "learning_rate": 1e-06, + "loss": 0.2966, + "mean_token_accuracy": 0.8927731513977051, + "num_tokens": 463449282.0, + "step": 13201 + }, + { + "epoch": 2.4516248839368617, + "grad_norm": 1.5211480855941772, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8716840147972107, + "num_tokens": 463485317.0, + "step": 13202 + }, + { + "epoch": 2.451810584958217, + "grad_norm": 1.6310267448425293, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8920793533325195, + "num_tokens": 463514426.0, + "step": 13203 + }, + { + "epoch": 2.4519962859795728, + "grad_norm": 1.6155462265014648, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8720851540565491, + "num_tokens": 463549339.0, + "step": 13204 + }, + { + "epoch": 2.4521819870009285, + "grad_norm": 1.4418895244598389, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8935970067977905, + "num_tokens": 463586589.0, + "step": 13205 + }, + { + "epoch": 2.452367688022284, + "grad_norm": 1.5602338314056396, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8749674558639526, + "num_tokens": 463624387.0, + "step": 13206 + }, + { + "epoch": 2.45255338904364, + "grad_norm": 1.4657989740371704, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8728369474411011, + "num_tokens": 463664881.0, + "step": 13207 + }, + { + "epoch": 2.4527390900649952, + "grad_norm": 1.8014764785766602, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8863696455955505, + "num_tokens": 463692387.0, + "step": 13208 + }, + { + "epoch": 2.452924791086351, + "grad_norm": 1.6579841375350952, + "learning_rate": 1e-06, + "loss": 0.2895, + "mean_token_accuracy": 0.8983746767044067, + "num_tokens": 463720767.0, + "step": 13209 + }, + { + "epoch": 2.4531104921077067, + "grad_norm": 1.607371211051941, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8747846484184265, + "num_tokens": 463753703.0, + "step": 13210 + }, + { + "epoch": 2.453296193129062, + "grad_norm": 1.5103282928466797, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8932026624679565, + "num_tokens": 463789170.0, + "step": 13211 + }, + { + "epoch": 2.4534818941504177, + "grad_norm": 1.5300990343093872, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8814881443977356, + "num_tokens": 463823985.0, + "step": 13212 + }, + { + "epoch": 2.4536675951717735, + "grad_norm": 1.4718999862670898, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8828950524330139, + "num_tokens": 463865592.0, + "step": 13213 + }, + { + "epoch": 2.453853296193129, + "grad_norm": 1.639847993850708, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8837435245513916, + "num_tokens": 463897532.0, + "step": 13214 + }, + { + "epoch": 2.4540389972144845, + "grad_norm": 1.6400408744812012, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8836535811424255, + "num_tokens": 463929003.0, + "step": 13215 + }, + { + "epoch": 2.45422469823584, + "grad_norm": 1.6486612558364868, + "learning_rate": 1e-06, + "loss": 0.3613, + "mean_token_accuracy": 0.8739098906517029, + "num_tokens": 463966411.0, + "step": 13216 + }, + { + "epoch": 2.454410399257196, + "grad_norm": 1.6546789407730103, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8859989643096924, + "num_tokens": 463998196.0, + "step": 13217 + }, + { + "epoch": 2.4545961002785517, + "grad_norm": 1.495846152305603, + "learning_rate": 1e-06, + "loss": 0.2623, + "mean_token_accuracy": 0.9047998189926147, + "num_tokens": 464031369.0, + "step": 13218 + }, + { + "epoch": 2.454781801299907, + "grad_norm": 1.647391438484192, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8715847730636597, + "num_tokens": 464066902.0, + "step": 13219 + }, + { + "epoch": 2.4549675023212627, + "grad_norm": 1.6157715320587158, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8800740242004395, + "num_tokens": 464103982.0, + "step": 13220 + }, + { + "epoch": 2.4551532033426184, + "grad_norm": 1.7048847675323486, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8779522180557251, + "num_tokens": 464137274.0, + "step": 13221 + }, + { + "epoch": 2.455338904363974, + "grad_norm": 1.6091374158859253, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8814419507980347, + "num_tokens": 464174148.0, + "step": 13222 + }, + { + "epoch": 2.4555246053853295, + "grad_norm": 1.5791889429092407, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8940577507019043, + "num_tokens": 464210168.0, + "step": 13223 + }, + { + "epoch": 2.455710306406685, + "grad_norm": 1.5015530586242676, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8919622898101807, + "num_tokens": 464245523.0, + "step": 13224 + }, + { + "epoch": 2.455896007428041, + "grad_norm": 1.6174430847167969, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8759193420410156, + "num_tokens": 464279270.0, + "step": 13225 + }, + { + "epoch": 2.4560817084493967, + "grad_norm": 1.6603676080703735, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8852618932723999, + "num_tokens": 464317432.0, + "step": 13226 + }, + { + "epoch": 2.456267409470752, + "grad_norm": 1.481469988822937, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8909929990768433, + "num_tokens": 464354720.0, + "step": 13227 + }, + { + "epoch": 2.4564531104921077, + "grad_norm": 1.483829140663147, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.8996164798736572, + "num_tokens": 464388337.0, + "step": 13228 + }, + { + "epoch": 2.4566388115134634, + "grad_norm": 1.5204203128814697, + "learning_rate": 1e-06, + "loss": 0.2838, + "mean_token_accuracy": 0.8996407985687256, + "num_tokens": 464424244.0, + "step": 13229 + }, + { + "epoch": 2.456824512534819, + "grad_norm": 1.5721027851104736, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.9008192420005798, + "num_tokens": 464456371.0, + "step": 13230 + }, + { + "epoch": 2.4570102135561744, + "grad_norm": 1.5281518697738647, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8893136978149414, + "num_tokens": 464491484.0, + "step": 13231 + }, + { + "epoch": 2.45719591457753, + "grad_norm": 1.6409622430801392, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8755428791046143, + "num_tokens": 464531603.0, + "step": 13232 + }, + { + "epoch": 2.457381615598886, + "grad_norm": 1.70200777053833, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8843739032745361, + "num_tokens": 464559181.0, + "step": 13233 + }, + { + "epoch": 2.457567316620241, + "grad_norm": 1.4363824129104614, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8864526748657227, + "num_tokens": 464598208.0, + "step": 13234 + }, + { + "epoch": 2.457753017641597, + "grad_norm": 1.5734426975250244, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8836566209793091, + "num_tokens": 464639938.0, + "step": 13235 + }, + { + "epoch": 2.4579387186629527, + "grad_norm": 1.6584813594818115, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8798600435256958, + "num_tokens": 464673084.0, + "step": 13236 + }, + { + "epoch": 2.4581244196843084, + "grad_norm": 1.7397404909133911, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8796454668045044, + "num_tokens": 464703227.0, + "step": 13237 + }, + { + "epoch": 2.458310120705664, + "grad_norm": 1.6636217832565308, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8889799118041992, + "num_tokens": 464737189.0, + "step": 13238 + }, + { + "epoch": 2.4584958217270194, + "grad_norm": 1.7169928550720215, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8955477476119995, + "num_tokens": 464766010.0, + "step": 13239 + }, + { + "epoch": 2.458681522748375, + "grad_norm": 1.571937918663025, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8907963037490845, + "num_tokens": 464799365.0, + "step": 13240 + }, + { + "epoch": 2.458867223769731, + "grad_norm": 1.6140800714492798, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.880456805229187, + "num_tokens": 464834178.0, + "step": 13241 + }, + { + "epoch": 2.459052924791086, + "grad_norm": 1.7204327583312988, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8735750913619995, + "num_tokens": 464866257.0, + "step": 13242 + }, + { + "epoch": 2.459238625812442, + "grad_norm": 1.7275315523147583, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8814742565155029, + "num_tokens": 464898832.0, + "step": 13243 + }, + { + "epoch": 2.4594243268337976, + "grad_norm": 1.5906246900558472, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8788135647773743, + "num_tokens": 464933673.0, + "step": 13244 + }, + { + "epoch": 2.4596100278551534, + "grad_norm": 1.7074507474899292, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8865792751312256, + "num_tokens": 464966260.0, + "step": 13245 + }, + { + "epoch": 2.4597957288765087, + "grad_norm": 1.4977985620498657, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8870565891265869, + "num_tokens": 465000631.0, + "step": 13246 + }, + { + "epoch": 2.4599814298978644, + "grad_norm": 1.6775659322738647, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8835645914077759, + "num_tokens": 465031888.0, + "step": 13247 + }, + { + "epoch": 2.46016713091922, + "grad_norm": 1.5705100297927856, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8896617293357849, + "num_tokens": 465066738.0, + "step": 13248 + }, + { + "epoch": 2.460352831940576, + "grad_norm": 1.4852198362350464, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8810396194458008, + "num_tokens": 465106942.0, + "step": 13249 + }, + { + "epoch": 2.460538532961931, + "grad_norm": 1.4680852890014648, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8880907297134399, + "num_tokens": 465146098.0, + "step": 13250 + }, + { + "epoch": 2.460724233983287, + "grad_norm": 1.642411231994629, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.8849669694900513, + "num_tokens": 465175993.0, + "step": 13251 + }, + { + "epoch": 2.4609099350046426, + "grad_norm": 1.5209153890609741, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.889188826084137, + "num_tokens": 465211748.0, + "step": 13252 + }, + { + "epoch": 2.4610956360259983, + "grad_norm": 1.5293114185333252, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8890656232833862, + "num_tokens": 465248934.0, + "step": 13253 + }, + { + "epoch": 2.4612813370473536, + "grad_norm": 1.5915290117263794, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8867440223693848, + "num_tokens": 465282821.0, + "step": 13254 + }, + { + "epoch": 2.4614670380687094, + "grad_norm": 1.4887337684631348, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8780978322029114, + "num_tokens": 465321815.0, + "step": 13255 + }, + { + "epoch": 2.461652739090065, + "grad_norm": 1.5539960861206055, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.886557936668396, + "num_tokens": 465355381.0, + "step": 13256 + }, + { + "epoch": 2.4618384401114204, + "grad_norm": 1.7005747556686401, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8683950304985046, + "num_tokens": 465389458.0, + "step": 13257 + }, + { + "epoch": 2.462024141132776, + "grad_norm": 1.6764870882034302, + "learning_rate": 1e-06, + "loss": 0.3476, + "mean_token_accuracy": 0.8831399083137512, + "num_tokens": 465423022.0, + "step": 13258 + }, + { + "epoch": 2.462209842154132, + "grad_norm": 1.5482184886932373, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8905671834945679, + "num_tokens": 465456262.0, + "step": 13259 + }, + { + "epoch": 2.4623955431754876, + "grad_norm": 1.4063293933868408, + "learning_rate": 1e-06, + "loss": 0.2727, + "mean_token_accuracy": 0.9015922546386719, + "num_tokens": 465493324.0, + "step": 13260 + }, + { + "epoch": 2.4625812441968433, + "grad_norm": 1.6874302625656128, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8789688348770142, + "num_tokens": 465527060.0, + "step": 13261 + }, + { + "epoch": 2.4627669452181986, + "grad_norm": 1.6303961277008057, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.8751741647720337, + "num_tokens": 465560536.0, + "step": 13262 + }, + { + "epoch": 2.4629526462395543, + "grad_norm": 1.6716132164001465, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8946144580841064, + "num_tokens": 465591133.0, + "step": 13263 + }, + { + "epoch": 2.46313834726091, + "grad_norm": 1.625781536102295, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8892756700515747, + "num_tokens": 465628721.0, + "step": 13264 + }, + { + "epoch": 2.4633240482822654, + "grad_norm": 1.562657356262207, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8837485313415527, + "num_tokens": 465663921.0, + "step": 13265 + }, + { + "epoch": 2.463509749303621, + "grad_norm": 1.5071003437042236, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8887124061584473, + "num_tokens": 465703022.0, + "step": 13266 + }, + { + "epoch": 2.463695450324977, + "grad_norm": 1.6094249486923218, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8863564729690552, + "num_tokens": 465735260.0, + "step": 13267 + }, + { + "epoch": 2.4638811513463326, + "grad_norm": 1.6180309057235718, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.878727912902832, + "num_tokens": 465771231.0, + "step": 13268 + }, + { + "epoch": 2.464066852367688, + "grad_norm": 1.5887725353240967, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8852497339248657, + "num_tokens": 465807404.0, + "step": 13269 + }, + { + "epoch": 2.4642525533890436, + "grad_norm": 1.4265329837799072, + "learning_rate": 1e-06, + "loss": 0.2627, + "mean_token_accuracy": 0.905174195766449, + "num_tokens": 465843206.0, + "step": 13270 + }, + { + "epoch": 2.4644382544103993, + "grad_norm": 1.5547056198120117, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8863654136657715, + "num_tokens": 465880013.0, + "step": 13271 + }, + { + "epoch": 2.464623955431755, + "grad_norm": 1.5713021755218506, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.8926055431365967, + "num_tokens": 465915372.0, + "step": 13272 + }, + { + "epoch": 2.4648096564531103, + "grad_norm": 1.6415106058120728, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.887894868850708, + "num_tokens": 465946538.0, + "step": 13273 + }, + { + "epoch": 2.464995357474466, + "grad_norm": 1.532412052154541, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8773380517959595, + "num_tokens": 465986871.0, + "step": 13274 + }, + { + "epoch": 2.465181058495822, + "grad_norm": 1.6128582954406738, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8856507539749146, + "num_tokens": 466020514.0, + "step": 13275 + }, + { + "epoch": 2.4653667595171775, + "grad_norm": 1.5746990442276, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8939356803894043, + "num_tokens": 466059259.0, + "step": 13276 + }, + { + "epoch": 2.465552460538533, + "grad_norm": 1.5741956233978271, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8907696008682251, + "num_tokens": 466099270.0, + "step": 13277 + }, + { + "epoch": 2.4657381615598886, + "grad_norm": 1.6081620454788208, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8873608708381653, + "num_tokens": 466131790.0, + "step": 13278 + }, + { + "epoch": 2.4659238625812443, + "grad_norm": 1.494332194328308, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.892953634262085, + "num_tokens": 466169265.0, + "step": 13279 + }, + { + "epoch": 2.4661095636025996, + "grad_norm": 1.668921709060669, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8824883103370667, + "num_tokens": 466203487.0, + "step": 13280 + }, + { + "epoch": 2.4662952646239553, + "grad_norm": 1.5182031393051147, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8739652633666992, + "num_tokens": 466242178.0, + "step": 13281 + }, + { + "epoch": 2.466480965645311, + "grad_norm": 1.544587254524231, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.89630126953125, + "num_tokens": 466276354.0, + "step": 13282 + }, + { + "epoch": 2.466666666666667, + "grad_norm": 1.660832405090332, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8953643441200256, + "num_tokens": 466309866.0, + "step": 13283 + }, + { + "epoch": 2.4668523676880225, + "grad_norm": 1.6248267889022827, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8940653800964355, + "num_tokens": 466343501.0, + "step": 13284 + }, + { + "epoch": 2.467038068709378, + "grad_norm": 1.5738602876663208, + "learning_rate": 1e-06, + "loss": 0.2911, + "mean_token_accuracy": 0.8965242505073547, + "num_tokens": 466375176.0, + "step": 13285 + }, + { + "epoch": 2.4672237697307335, + "grad_norm": 1.3925607204437256, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8846621513366699, + "num_tokens": 466416606.0, + "step": 13286 + }, + { + "epoch": 2.4674094707520893, + "grad_norm": 1.811874270439148, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8790446519851685, + "num_tokens": 466445905.0, + "step": 13287 + }, + { + "epoch": 2.4675951717734446, + "grad_norm": 1.5631693601608276, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8727215528488159, + "num_tokens": 466485526.0, + "step": 13288 + }, + { + "epoch": 2.4677808727948003, + "grad_norm": 1.8806324005126953, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8653430938720703, + "num_tokens": 466511456.0, + "step": 13289 + }, + { + "epoch": 2.467966573816156, + "grad_norm": 1.73112952709198, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8668724298477173, + "num_tokens": 466547010.0, + "step": 13290 + }, + { + "epoch": 2.4681522748375118, + "grad_norm": 1.5153322219848633, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8766711950302124, + "num_tokens": 466587046.0, + "step": 13291 + }, + { + "epoch": 2.468337975858867, + "grad_norm": 1.5063644647598267, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8877137899398804, + "num_tokens": 466625079.0, + "step": 13292 + }, + { + "epoch": 2.468523676880223, + "grad_norm": 1.5246188640594482, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8909596800804138, + "num_tokens": 466661217.0, + "step": 13293 + }, + { + "epoch": 2.4687093779015785, + "grad_norm": 1.7470455169677734, + "learning_rate": 1e-06, + "loss": 0.3712, + "mean_token_accuracy": 0.8752031326293945, + "num_tokens": 466692613.0, + "step": 13294 + }, + { + "epoch": 2.4688950789229342, + "grad_norm": 1.616757869720459, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8872983455657959, + "num_tokens": 466728192.0, + "step": 13295 + }, + { + "epoch": 2.4690807799442895, + "grad_norm": 1.7064086198806763, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8787829875946045, + "num_tokens": 466767470.0, + "step": 13296 + }, + { + "epoch": 2.4692664809656453, + "grad_norm": 1.581135630607605, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8905727863311768, + "num_tokens": 466804589.0, + "step": 13297 + }, + { + "epoch": 2.469452181987001, + "grad_norm": 1.4179586172103882, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8933465480804443, + "num_tokens": 466843098.0, + "step": 13298 + }, + { + "epoch": 2.4696378830083567, + "grad_norm": 1.3955022096633911, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8898752927780151, + "num_tokens": 466887178.0, + "step": 13299 + }, + { + "epoch": 2.469823584029712, + "grad_norm": 1.4449750185012817, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8904805183410645, + "num_tokens": 466922983.0, + "step": 13300 + }, + { + "epoch": 2.4700092850510678, + "grad_norm": 1.8767859935760498, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8776706457138062, + "num_tokens": 466954035.0, + "step": 13301 + }, + { + "epoch": 2.4701949860724235, + "grad_norm": 1.5931880474090576, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8957399129867554, + "num_tokens": 466987224.0, + "step": 13302 + }, + { + "epoch": 2.470380687093779, + "grad_norm": 1.5500192642211914, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8946704268455505, + "num_tokens": 467020712.0, + "step": 13303 + }, + { + "epoch": 2.4705663881151345, + "grad_norm": 1.5932142734527588, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8870916962623596, + "num_tokens": 467054403.0, + "step": 13304 + }, + { + "epoch": 2.4707520891364902, + "grad_norm": 1.6290132999420166, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.8768414258956909, + "num_tokens": 467087914.0, + "step": 13305 + }, + { + "epoch": 2.470937790157846, + "grad_norm": 1.5643202066421509, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8703932762145996, + "num_tokens": 467124569.0, + "step": 13306 + }, + { + "epoch": 2.4711234911792017, + "grad_norm": 1.7404803037643433, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8911348581314087, + "num_tokens": 467151067.0, + "step": 13307 + }, + { + "epoch": 2.471309192200557, + "grad_norm": 1.704469084739685, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8896264433860779, + "num_tokens": 467181746.0, + "step": 13308 + }, + { + "epoch": 2.4714948932219127, + "grad_norm": 1.5315569639205933, + "learning_rate": 1e-06, + "loss": 0.2905, + "mean_token_accuracy": 0.894866406917572, + "num_tokens": 467216353.0, + "step": 13309 + }, + { + "epoch": 2.4716805942432685, + "grad_norm": 1.566570520401001, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8880285620689392, + "num_tokens": 467253214.0, + "step": 13310 + }, + { + "epoch": 2.4718662952646238, + "grad_norm": 1.5301281213760376, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8932675123214722, + "num_tokens": 467292452.0, + "step": 13311 + }, + { + "epoch": 2.4720519962859795, + "grad_norm": 1.5981388092041016, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8937817811965942, + "num_tokens": 467325188.0, + "step": 13312 + }, + { + "epoch": 2.4722376973073352, + "grad_norm": 1.6889269351959229, + "learning_rate": 1e-06, + "loss": 0.3677, + "mean_token_accuracy": 0.8683582544326782, + "num_tokens": 467355127.0, + "step": 13313 + }, + { + "epoch": 2.472423398328691, + "grad_norm": 1.5056884288787842, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8852506875991821, + "num_tokens": 467394249.0, + "step": 13314 + }, + { + "epoch": 2.4726090993500462, + "grad_norm": 1.5370097160339355, + "learning_rate": 1e-06, + "loss": 0.3705, + "mean_token_accuracy": 0.8701609373092651, + "num_tokens": 467433427.0, + "step": 13315 + }, + { + "epoch": 2.472794800371402, + "grad_norm": 1.4756059646606445, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8983131647109985, + "num_tokens": 467475326.0, + "step": 13316 + }, + { + "epoch": 2.4729805013927577, + "grad_norm": 1.5064181089401245, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.893762469291687, + "num_tokens": 467512487.0, + "step": 13317 + }, + { + "epoch": 2.4731662024141134, + "grad_norm": 1.555059790611267, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8867045640945435, + "num_tokens": 467548868.0, + "step": 13318 + }, + { + "epoch": 2.4733519034354687, + "grad_norm": 1.5132403373718262, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.888675332069397, + "num_tokens": 467585434.0, + "step": 13319 + }, + { + "epoch": 2.4735376044568245, + "grad_norm": 1.624484658241272, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8824334144592285, + "num_tokens": 467618535.0, + "step": 13320 + }, + { + "epoch": 2.47372330547818, + "grad_norm": 1.440085530281067, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8932684659957886, + "num_tokens": 467658326.0, + "step": 13321 + }, + { + "epoch": 2.473909006499536, + "grad_norm": 1.6832542419433594, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8953967094421387, + "num_tokens": 467688663.0, + "step": 13322 + }, + { + "epoch": 2.4740947075208912, + "grad_norm": 1.8704140186309814, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.870272159576416, + "num_tokens": 467720422.0, + "step": 13323 + }, + { + "epoch": 2.474280408542247, + "grad_norm": 1.6621315479278564, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8800104856491089, + "num_tokens": 467755491.0, + "step": 13324 + }, + { + "epoch": 2.4744661095636027, + "grad_norm": 1.4885516166687012, + "learning_rate": 1e-06, + "loss": 0.295, + "mean_token_accuracy": 0.8962347507476807, + "num_tokens": 467794979.0, + "step": 13325 + }, + { + "epoch": 2.474651810584958, + "grad_norm": 1.6546550989151, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.876310408115387, + "num_tokens": 467826686.0, + "step": 13326 + }, + { + "epoch": 2.4748375116063137, + "grad_norm": 1.6177412271499634, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8724257946014404, + "num_tokens": 467863594.0, + "step": 13327 + }, + { + "epoch": 2.4750232126276694, + "grad_norm": 1.5784586668014526, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.9008959531784058, + "num_tokens": 467896082.0, + "step": 13328 + }, + { + "epoch": 2.475208913649025, + "grad_norm": 1.5786898136138916, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8847654461860657, + "num_tokens": 467930242.0, + "step": 13329 + }, + { + "epoch": 2.475394614670381, + "grad_norm": 1.6124728918075562, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8917921185493469, + "num_tokens": 467962040.0, + "step": 13330 + }, + { + "epoch": 2.475580315691736, + "grad_norm": 1.6147621870040894, + "learning_rate": 1e-06, + "loss": 0.2835, + "mean_token_accuracy": 0.9007704257965088, + "num_tokens": 467994963.0, + "step": 13331 + }, + { + "epoch": 2.475766016713092, + "grad_norm": 1.81331205368042, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8796472549438477, + "num_tokens": 468027178.0, + "step": 13332 + }, + { + "epoch": 2.4759517177344477, + "grad_norm": 1.6389931440353394, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8649940490722656, + "num_tokens": 468067706.0, + "step": 13333 + }, + { + "epoch": 2.476137418755803, + "grad_norm": 1.5726603269577026, + "learning_rate": 1e-06, + "loss": 0.2742, + "mean_token_accuracy": 0.9001702666282654, + "num_tokens": 468099472.0, + "step": 13334 + }, + { + "epoch": 2.4763231197771587, + "grad_norm": 1.5823036432266235, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.869701623916626, + "num_tokens": 468138405.0, + "step": 13335 + }, + { + "epoch": 2.4765088207985144, + "grad_norm": 1.771266222000122, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8836712837219238, + "num_tokens": 468170067.0, + "step": 13336 + }, + { + "epoch": 2.47669452181987, + "grad_norm": 1.5894684791564941, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8899359703063965, + "num_tokens": 468204144.0, + "step": 13337 + }, + { + "epoch": 2.4768802228412254, + "grad_norm": 1.552420735359192, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8828555345535278, + "num_tokens": 468237994.0, + "step": 13338 + }, + { + "epoch": 2.477065923862581, + "grad_norm": 1.723719835281372, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.879977822303772, + "num_tokens": 468266714.0, + "step": 13339 + }, + { + "epoch": 2.477251624883937, + "grad_norm": 1.6823238134384155, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8718727827072144, + "num_tokens": 468303774.0, + "step": 13340 + }, + { + "epoch": 2.4774373259052926, + "grad_norm": 1.7695255279541016, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.877998948097229, + "num_tokens": 468332369.0, + "step": 13341 + }, + { + "epoch": 2.477623026926648, + "grad_norm": 1.4969382286071777, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8883405923843384, + "num_tokens": 468368853.0, + "step": 13342 + }, + { + "epoch": 2.4778087279480037, + "grad_norm": 1.6453016996383667, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8695981502532959, + "num_tokens": 468403652.0, + "step": 13343 + }, + { + "epoch": 2.4779944289693594, + "grad_norm": 1.42850923538208, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8968977928161621, + "num_tokens": 468445407.0, + "step": 13344 + }, + { + "epoch": 2.478180129990715, + "grad_norm": 1.6403932571411133, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8796445727348328, + "num_tokens": 468476399.0, + "step": 13345 + }, + { + "epoch": 2.4783658310120704, + "grad_norm": 1.6071637868881226, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8922454714775085, + "num_tokens": 468508779.0, + "step": 13346 + }, + { + "epoch": 2.478551532033426, + "grad_norm": 1.6431834697723389, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8930330872535706, + "num_tokens": 468539754.0, + "step": 13347 + }, + { + "epoch": 2.478737233054782, + "grad_norm": 1.413682460784912, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8920618891716003, + "num_tokens": 468582332.0, + "step": 13348 + }, + { + "epoch": 2.478922934076137, + "grad_norm": 1.538836121559143, + "learning_rate": 1e-06, + "loss": 0.2873, + "mean_token_accuracy": 0.89970463514328, + "num_tokens": 468617221.0, + "step": 13349 + }, + { + "epoch": 2.479108635097493, + "grad_norm": 1.5665361881256104, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8756516575813293, + "num_tokens": 468653221.0, + "step": 13350 + }, + { + "epoch": 2.4792943361188486, + "grad_norm": 1.5124789476394653, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.9036345481872559, + "num_tokens": 468688414.0, + "step": 13351 + }, + { + "epoch": 2.4794800371402044, + "grad_norm": 1.5630375146865845, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.8999000787734985, + "num_tokens": 468721397.0, + "step": 13352 + }, + { + "epoch": 2.47966573816156, + "grad_norm": 1.643510341644287, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8883247375488281, + "num_tokens": 468756477.0, + "step": 13353 + }, + { + "epoch": 2.4798514391829154, + "grad_norm": 1.5811890363693237, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8872712850570679, + "num_tokens": 468793789.0, + "step": 13354 + }, + { + "epoch": 2.480037140204271, + "grad_norm": 1.6632767915725708, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8782092332839966, + "num_tokens": 468826698.0, + "step": 13355 + }, + { + "epoch": 2.480222841225627, + "grad_norm": 1.4699572324752808, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8840179443359375, + "num_tokens": 468868744.0, + "step": 13356 + }, + { + "epoch": 2.480408542246982, + "grad_norm": 1.666495442390442, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.8815549612045288, + "num_tokens": 468905892.0, + "step": 13357 + }, + { + "epoch": 2.480594243268338, + "grad_norm": 1.511274814605713, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8904615044593811, + "num_tokens": 468940097.0, + "step": 13358 + }, + { + "epoch": 2.4807799442896936, + "grad_norm": 1.6333227157592773, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8943196535110474, + "num_tokens": 468970628.0, + "step": 13359 + }, + { + "epoch": 2.4809656453110494, + "grad_norm": 1.6165063381195068, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8854798078536987, + "num_tokens": 469001742.0, + "step": 13360 + }, + { + "epoch": 2.4811513463324046, + "grad_norm": 1.6435331106185913, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8744369149208069, + "num_tokens": 469036683.0, + "step": 13361 + }, + { + "epoch": 2.4813370473537604, + "grad_norm": 1.472997784614563, + "learning_rate": 1e-06, + "loss": 0.3047, + "mean_token_accuracy": 0.8944287300109863, + "num_tokens": 469074558.0, + "step": 13362 + }, + { + "epoch": 2.481522748375116, + "grad_norm": 1.4684652090072632, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8901964426040649, + "num_tokens": 469114209.0, + "step": 13363 + }, + { + "epoch": 2.481708449396472, + "grad_norm": 1.5912233591079712, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8773319125175476, + "num_tokens": 469151084.0, + "step": 13364 + }, + { + "epoch": 2.481894150417827, + "grad_norm": 1.4702385663986206, + "learning_rate": 1e-06, + "loss": 0.2529, + "mean_token_accuracy": 0.9093946218490601, + "num_tokens": 469185145.0, + "step": 13365 + }, + { + "epoch": 2.482079851439183, + "grad_norm": 1.4855409860610962, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8952268362045288, + "num_tokens": 469221788.0, + "step": 13366 + }, + { + "epoch": 2.4822655524605386, + "grad_norm": 1.562193512916565, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8843101859092712, + "num_tokens": 469256431.0, + "step": 13367 + }, + { + "epoch": 2.4824512534818943, + "grad_norm": 1.597339153289795, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8885799050331116, + "num_tokens": 469290675.0, + "step": 13368 + }, + { + "epoch": 2.4826369545032496, + "grad_norm": 1.4739325046539307, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.886023998260498, + "num_tokens": 469327773.0, + "step": 13369 + }, + { + "epoch": 2.4828226555246053, + "grad_norm": 1.7049318552017212, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8823806047439575, + "num_tokens": 469357230.0, + "step": 13370 + }, + { + "epoch": 2.483008356545961, + "grad_norm": 1.4573601484298706, + "learning_rate": 1e-06, + "loss": 0.2709, + "mean_token_accuracy": 0.9042720794677734, + "num_tokens": 469396826.0, + "step": 13371 + }, + { + "epoch": 2.4831940575673164, + "grad_norm": 1.485926628112793, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8865906000137329, + "num_tokens": 469435983.0, + "step": 13372 + }, + { + "epoch": 2.483379758588672, + "grad_norm": 1.462286353111267, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.893553614616394, + "num_tokens": 469473159.0, + "step": 13373 + }, + { + "epoch": 2.483565459610028, + "grad_norm": 1.5373525619506836, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8861774206161499, + "num_tokens": 469509806.0, + "step": 13374 + }, + { + "epoch": 2.4837511606313836, + "grad_norm": 1.741974115371704, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8719149827957153, + "num_tokens": 469542303.0, + "step": 13375 + }, + { + "epoch": 2.4839368616527393, + "grad_norm": 1.4877852201461792, + "learning_rate": 1e-06, + "loss": 0.2869, + "mean_token_accuracy": 0.898251473903656, + "num_tokens": 469577722.0, + "step": 13376 + }, + { + "epoch": 2.4841225626740946, + "grad_norm": 1.4385732412338257, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8971695899963379, + "num_tokens": 469615559.0, + "step": 13377 + }, + { + "epoch": 2.4843082636954503, + "grad_norm": 1.681155800819397, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8864604234695435, + "num_tokens": 469647859.0, + "step": 13378 + }, + { + "epoch": 2.484493964716806, + "grad_norm": 1.679794192314148, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8756253719329834, + "num_tokens": 469682580.0, + "step": 13379 + }, + { + "epoch": 2.4846796657381613, + "grad_norm": 1.4414443969726562, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8928186893463135, + "num_tokens": 469721078.0, + "step": 13380 + }, + { + "epoch": 2.484865366759517, + "grad_norm": 1.5994822978973389, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8856422305107117, + "num_tokens": 469753872.0, + "step": 13381 + }, + { + "epoch": 2.485051067780873, + "grad_norm": 1.5712060928344727, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8881866931915283, + "num_tokens": 469786530.0, + "step": 13382 + }, + { + "epoch": 2.4852367688022285, + "grad_norm": 1.5711901187896729, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8785993456840515, + "num_tokens": 469822445.0, + "step": 13383 + }, + { + "epoch": 2.4854224698235843, + "grad_norm": 1.6197640895843506, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8831624388694763, + "num_tokens": 469860861.0, + "step": 13384 + }, + { + "epoch": 2.4856081708449396, + "grad_norm": 1.3943426609039307, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.882520318031311, + "num_tokens": 469904015.0, + "step": 13385 + }, + { + "epoch": 2.4857938718662953, + "grad_norm": 1.4987339973449707, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.894493579864502, + "num_tokens": 469940303.0, + "step": 13386 + }, + { + "epoch": 2.485979572887651, + "grad_norm": 1.5795859098434448, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8903427124023438, + "num_tokens": 469976498.0, + "step": 13387 + }, + { + "epoch": 2.4861652739090063, + "grad_norm": 1.5713517665863037, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8977193832397461, + "num_tokens": 470007990.0, + "step": 13388 + }, + { + "epoch": 2.486350974930362, + "grad_norm": 1.440106749534607, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8871384859085083, + "num_tokens": 470051007.0, + "step": 13389 + }, + { + "epoch": 2.486536675951718, + "grad_norm": 1.3825525045394897, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8881946802139282, + "num_tokens": 470093810.0, + "step": 13390 + }, + { + "epoch": 2.4867223769730735, + "grad_norm": 1.4357810020446777, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8917044997215271, + "num_tokens": 470133180.0, + "step": 13391 + }, + { + "epoch": 2.486908077994429, + "grad_norm": 1.4686211347579956, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8827800750732422, + "num_tokens": 470172198.0, + "step": 13392 + }, + { + "epoch": 2.4870937790157845, + "grad_norm": 1.5430864095687866, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8779574632644653, + "num_tokens": 470211352.0, + "step": 13393 + }, + { + "epoch": 2.4872794800371403, + "grad_norm": 1.3267912864685059, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8912558555603027, + "num_tokens": 470258651.0, + "step": 13394 + }, + { + "epoch": 2.487465181058496, + "grad_norm": 1.3533830642700195, + "learning_rate": 1e-06, + "loss": 0.2817, + "mean_token_accuracy": 0.8980972766876221, + "num_tokens": 470301068.0, + "step": 13395 + }, + { + "epoch": 2.4876508820798513, + "grad_norm": 1.3974452018737793, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8913707137107849, + "num_tokens": 470341323.0, + "step": 13396 + }, + { + "epoch": 2.487836583101207, + "grad_norm": 1.60002601146698, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.888856828212738, + "num_tokens": 470374274.0, + "step": 13397 + }, + { + "epoch": 2.4880222841225628, + "grad_norm": 1.539446473121643, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8962115049362183, + "num_tokens": 470406853.0, + "step": 13398 + }, + { + "epoch": 2.4882079851439185, + "grad_norm": 1.571374773979187, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8830579519271851, + "num_tokens": 470443779.0, + "step": 13399 + }, + { + "epoch": 2.488393686165274, + "grad_norm": 1.3776867389678955, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8898183107376099, + "num_tokens": 470487489.0, + "step": 13400 + }, + { + "epoch": 2.4885793871866295, + "grad_norm": 1.4023228883743286, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.889671802520752, + "num_tokens": 470527111.0, + "step": 13401 + }, + { + "epoch": 2.4887650882079853, + "grad_norm": 1.5832951068878174, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8833958506584167, + "num_tokens": 470562099.0, + "step": 13402 + }, + { + "epoch": 2.4889507892293405, + "grad_norm": 1.6069422960281372, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8781718611717224, + "num_tokens": 470598744.0, + "step": 13403 + }, + { + "epoch": 2.4891364902506963, + "grad_norm": 1.74262535572052, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8793061971664429, + "num_tokens": 470631910.0, + "step": 13404 + }, + { + "epoch": 2.489322191272052, + "grad_norm": 1.5688769817352295, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8820242881774902, + "num_tokens": 470667359.0, + "step": 13405 + }, + { + "epoch": 2.4895078922934077, + "grad_norm": 1.729680061340332, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8847155570983887, + "num_tokens": 470697072.0, + "step": 13406 + }, + { + "epoch": 2.4896935933147635, + "grad_norm": 1.523990511894226, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8818497657775879, + "num_tokens": 470733997.0, + "step": 13407 + }, + { + "epoch": 2.4898792943361188, + "grad_norm": 1.4491164684295654, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8877516388893127, + "num_tokens": 470773081.0, + "step": 13408 + }, + { + "epoch": 2.4900649953574745, + "grad_norm": 1.4203964471817017, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.892404317855835, + "num_tokens": 470813543.0, + "step": 13409 + }, + { + "epoch": 2.4902506963788302, + "grad_norm": 1.5752668380737305, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8720124363899231, + "num_tokens": 470849291.0, + "step": 13410 + }, + { + "epoch": 2.4904363974001855, + "grad_norm": 1.6274933815002441, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8883117437362671, + "num_tokens": 470882996.0, + "step": 13411 + }, + { + "epoch": 2.4906220984215413, + "grad_norm": 1.6516896486282349, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8841710090637207, + "num_tokens": 470916037.0, + "step": 13412 + }, + { + "epoch": 2.490807799442897, + "grad_norm": 1.6182665824890137, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8858530521392822, + "num_tokens": 470949274.0, + "step": 13413 + }, + { + "epoch": 2.4909935004642527, + "grad_norm": 1.6159874200820923, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8934189081192017, + "num_tokens": 470983556.0, + "step": 13414 + }, + { + "epoch": 2.491179201485608, + "grad_norm": 1.604360580444336, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8767109513282776, + "num_tokens": 471021337.0, + "step": 13415 + }, + { + "epoch": 2.4913649025069637, + "grad_norm": 1.786555290222168, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8763896226882935, + "num_tokens": 471052582.0, + "step": 13416 + }, + { + "epoch": 2.4915506035283195, + "grad_norm": 1.4982147216796875, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.887086033821106, + "num_tokens": 471089030.0, + "step": 13417 + }, + { + "epoch": 2.491736304549675, + "grad_norm": 1.568410038948059, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8901369571685791, + "num_tokens": 471122962.0, + "step": 13418 + }, + { + "epoch": 2.4919220055710305, + "grad_norm": 1.5037552118301392, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8871203660964966, + "num_tokens": 471157394.0, + "step": 13419 + }, + { + "epoch": 2.4921077065923862, + "grad_norm": 1.5928844213485718, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8827470541000366, + "num_tokens": 471192401.0, + "step": 13420 + }, + { + "epoch": 2.492293407613742, + "grad_norm": 1.749804973602295, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8873094320297241, + "num_tokens": 471224333.0, + "step": 13421 + }, + { + "epoch": 2.4924791086350977, + "grad_norm": 1.4729645252227783, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.9015153050422668, + "num_tokens": 471260573.0, + "step": 13422 + }, + { + "epoch": 2.492664809656453, + "grad_norm": 1.6821902990341187, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8973639011383057, + "num_tokens": 471289447.0, + "step": 13423 + }, + { + "epoch": 2.4928505106778087, + "grad_norm": 1.3593233823776245, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.8928675651550293, + "num_tokens": 471332238.0, + "step": 13424 + }, + { + "epoch": 2.4930362116991645, + "grad_norm": 1.5610469579696655, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8835756778717041, + "num_tokens": 471366148.0, + "step": 13425 + }, + { + "epoch": 2.4932219127205197, + "grad_norm": 1.467446208000183, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8766989707946777, + "num_tokens": 471407623.0, + "step": 13426 + }, + { + "epoch": 2.4934076137418755, + "grad_norm": 1.8413242101669312, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8908748030662537, + "num_tokens": 471432300.0, + "step": 13427 + }, + { + "epoch": 2.493593314763231, + "grad_norm": 1.385021448135376, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8989277482032776, + "num_tokens": 471469894.0, + "step": 13428 + }, + { + "epoch": 2.493779015784587, + "grad_norm": 1.6269104480743408, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8803398013114929, + "num_tokens": 471504640.0, + "step": 13429 + }, + { + "epoch": 2.4939647168059427, + "grad_norm": 1.6211458444595337, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8884198665618896, + "num_tokens": 471538712.0, + "step": 13430 + }, + { + "epoch": 2.494150417827298, + "grad_norm": 1.6630194187164307, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8867828845977783, + "num_tokens": 471570586.0, + "step": 13431 + }, + { + "epoch": 2.4943361188486537, + "grad_norm": 1.698053240776062, + "learning_rate": 1e-06, + "loss": 0.3756, + "mean_token_accuracy": 0.8684433698654175, + "num_tokens": 471602756.0, + "step": 13432 + }, + { + "epoch": 2.4945218198700094, + "grad_norm": 1.6668579578399658, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8837077617645264, + "num_tokens": 471634172.0, + "step": 13433 + }, + { + "epoch": 2.4947075208913647, + "grad_norm": 1.758384346961975, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8861113786697388, + "num_tokens": 471665215.0, + "step": 13434 + }, + { + "epoch": 2.4948932219127204, + "grad_norm": 1.5399603843688965, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8828110098838806, + "num_tokens": 471702854.0, + "step": 13435 + }, + { + "epoch": 2.495078922934076, + "grad_norm": 1.734736442565918, + "learning_rate": 1e-06, + "loss": 0.297, + "mean_token_accuracy": 0.8953529596328735, + "num_tokens": 471731543.0, + "step": 13436 + }, + { + "epoch": 2.495264623955432, + "grad_norm": 1.540071725845337, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8872718811035156, + "num_tokens": 471765630.0, + "step": 13437 + }, + { + "epoch": 2.495450324976787, + "grad_norm": 1.6349657773971558, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8816665410995483, + "num_tokens": 471798674.0, + "step": 13438 + }, + { + "epoch": 2.495636025998143, + "grad_norm": 1.4401084184646606, + "learning_rate": 1e-06, + "loss": 0.2878, + "mean_token_accuracy": 0.8977953791618347, + "num_tokens": 471835283.0, + "step": 13439 + }, + { + "epoch": 2.4958217270194987, + "grad_norm": 1.5370064973831177, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8806161284446716, + "num_tokens": 471874244.0, + "step": 13440 + }, + { + "epoch": 2.4960074280408544, + "grad_norm": 1.5204771757125854, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8946045637130737, + "num_tokens": 471907516.0, + "step": 13441 + }, + { + "epoch": 2.4961931290622097, + "grad_norm": 1.6012755632400513, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8914080262184143, + "num_tokens": 471941785.0, + "step": 13442 + }, + { + "epoch": 2.4963788300835654, + "grad_norm": 1.4434064626693726, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8903722763061523, + "num_tokens": 471979123.0, + "step": 13443 + }, + { + "epoch": 2.496564531104921, + "grad_norm": 1.5055981874465942, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8884063959121704, + "num_tokens": 472018015.0, + "step": 13444 + }, + { + "epoch": 2.496750232126277, + "grad_norm": 1.6030046939849854, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.8844101428985596, + "num_tokens": 472053363.0, + "step": 13445 + }, + { + "epoch": 2.496935933147632, + "grad_norm": 1.570311188697815, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8779181241989136, + "num_tokens": 472088401.0, + "step": 13446 + }, + { + "epoch": 2.497121634168988, + "grad_norm": 1.716613531112671, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8814793825149536, + "num_tokens": 472119345.0, + "step": 13447 + }, + { + "epoch": 2.4973073351903436, + "grad_norm": 1.5221965312957764, + "learning_rate": 1e-06, + "loss": 0.3001, + "mean_token_accuracy": 0.8943703174591064, + "num_tokens": 472154454.0, + "step": 13448 + }, + { + "epoch": 2.497493036211699, + "grad_norm": 1.3998103141784668, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8919907212257385, + "num_tokens": 472193111.0, + "step": 13449 + }, + { + "epoch": 2.4976787372330547, + "grad_norm": 1.5612879991531372, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8817397952079773, + "num_tokens": 472228895.0, + "step": 13450 + }, + { + "epoch": 2.4978644382544104, + "grad_norm": 1.4819697141647339, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8906237483024597, + "num_tokens": 472266476.0, + "step": 13451 + }, + { + "epoch": 2.498050139275766, + "grad_norm": 1.7285118103027344, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8909534215927124, + "num_tokens": 472300641.0, + "step": 13452 + }, + { + "epoch": 2.498235840297122, + "grad_norm": 1.4468886852264404, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8903661966323853, + "num_tokens": 472339667.0, + "step": 13453 + }, + { + "epoch": 2.498421541318477, + "grad_norm": 1.6138579845428467, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8901300430297852, + "num_tokens": 472372400.0, + "step": 13454 + }, + { + "epoch": 2.498607242339833, + "grad_norm": 1.590596318244934, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8843063116073608, + "num_tokens": 472409038.0, + "step": 13455 + }, + { + "epoch": 2.4987929433611886, + "grad_norm": 1.6032387018203735, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.8592516183853149, + "num_tokens": 472449510.0, + "step": 13456 + }, + { + "epoch": 2.498978644382544, + "grad_norm": 1.4981751441955566, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.890697717666626, + "num_tokens": 472486598.0, + "step": 13457 + }, + { + "epoch": 2.4991643454038996, + "grad_norm": 1.7621164321899414, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8829340934753418, + "num_tokens": 472514907.0, + "step": 13458 + }, + { + "epoch": 2.4993500464252554, + "grad_norm": 1.6308320760726929, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.88499915599823, + "num_tokens": 472543476.0, + "step": 13459 + }, + { + "epoch": 2.499535747446611, + "grad_norm": 1.432674527168274, + "learning_rate": 1e-06, + "loss": 0.275, + "mean_token_accuracy": 0.9009995460510254, + "num_tokens": 472577874.0, + "step": 13460 + }, + { + "epoch": 2.4997214484679664, + "grad_norm": 1.4810230731964111, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8956084847450256, + "num_tokens": 472616084.0, + "step": 13461 + }, + { + "epoch": 2.499907149489322, + "grad_norm": 1.5303980112075806, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.8968162536621094, + "num_tokens": 472649168.0, + "step": 13462 + }, + { + "epoch": 2.500092850510678, + "grad_norm": 1.5898641347885132, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8806807994842529, + "num_tokens": 472683912.0, + "step": 13463 + }, + { + "epoch": 2.500278551532033, + "grad_norm": 1.5423026084899902, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8923711776733398, + "num_tokens": 472717128.0, + "step": 13464 + }, + { + "epoch": 2.500464252553389, + "grad_norm": 1.463823914527893, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8937801122665405, + "num_tokens": 472754208.0, + "step": 13465 + }, + { + "epoch": 2.5006499535747446, + "grad_norm": 1.4681390523910522, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.881037712097168, + "num_tokens": 472794184.0, + "step": 13466 + }, + { + "epoch": 2.5008356545961004, + "grad_norm": 1.6450928449630737, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.8844494819641113, + "num_tokens": 472828442.0, + "step": 13467 + }, + { + "epoch": 2.501021355617456, + "grad_norm": 2.498128652572632, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.884183406829834, + "num_tokens": 472859020.0, + "step": 13468 + }, + { + "epoch": 2.5012070566388114, + "grad_norm": 1.527483344078064, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8940418362617493, + "num_tokens": 472898861.0, + "step": 13469 + }, + { + "epoch": 2.501392757660167, + "grad_norm": 1.6351269483566284, + "learning_rate": 1e-06, + "loss": 0.3915, + "mean_token_accuracy": 0.8683900237083435, + "num_tokens": 472936924.0, + "step": 13470 + }, + { + "epoch": 2.501578458681523, + "grad_norm": 1.6219024658203125, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8821781873703003, + "num_tokens": 472971858.0, + "step": 13471 + }, + { + "epoch": 2.501764159702878, + "grad_norm": 1.4701441526412964, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8868297338485718, + "num_tokens": 473010223.0, + "step": 13472 + }, + { + "epoch": 2.501949860724234, + "grad_norm": 1.5112193822860718, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.9010401368141174, + "num_tokens": 473042699.0, + "step": 13473 + }, + { + "epoch": 2.5021355617455896, + "grad_norm": 1.6908049583435059, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.897358775138855, + "num_tokens": 473070833.0, + "step": 13474 + }, + { + "epoch": 2.5023212627669453, + "grad_norm": 1.4866408109664917, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8874018788337708, + "num_tokens": 473107118.0, + "step": 13475 + }, + { + "epoch": 2.502506963788301, + "grad_norm": 1.5177732706069946, + "learning_rate": 1e-06, + "loss": 0.2659, + "mean_token_accuracy": 0.9021393060684204, + "num_tokens": 473140446.0, + "step": 13476 + }, + { + "epoch": 2.5026926648096564, + "grad_norm": 1.5116915702819824, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8807080388069153, + "num_tokens": 473180351.0, + "step": 13477 + }, + { + "epoch": 2.502878365831012, + "grad_norm": 1.5245287418365479, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8922932744026184, + "num_tokens": 473214645.0, + "step": 13478 + }, + { + "epoch": 2.503064066852368, + "grad_norm": 1.6796571016311646, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8906636238098145, + "num_tokens": 473243309.0, + "step": 13479 + }, + { + "epoch": 2.503249767873723, + "grad_norm": 1.8134870529174805, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8813234567642212, + "num_tokens": 473269117.0, + "step": 13480 + }, + { + "epoch": 2.503435468895079, + "grad_norm": 1.5897539854049683, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8839527368545532, + "num_tokens": 473306398.0, + "step": 13481 + }, + { + "epoch": 2.5036211699164346, + "grad_norm": 1.6299551725387573, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.89326012134552, + "num_tokens": 473339399.0, + "step": 13482 + }, + { + "epoch": 2.5038068709377903, + "grad_norm": 1.4567787647247314, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8921281695365906, + "num_tokens": 473374796.0, + "step": 13483 + }, + { + "epoch": 2.503992571959146, + "grad_norm": 1.6211401224136353, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.893274188041687, + "num_tokens": 473405662.0, + "step": 13484 + }, + { + "epoch": 2.5041782729805013, + "grad_norm": 1.5155513286590576, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8841285705566406, + "num_tokens": 473445446.0, + "step": 13485 + }, + { + "epoch": 2.504363974001857, + "grad_norm": 1.6315622329711914, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.8840898871421814, + "num_tokens": 473475252.0, + "step": 13486 + }, + { + "epoch": 2.5045496750232124, + "grad_norm": 1.5469214916229248, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8828237056732178, + "num_tokens": 473516086.0, + "step": 13487 + }, + { + "epoch": 2.504735376044568, + "grad_norm": 1.5053709745407104, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8810558319091797, + "num_tokens": 473556072.0, + "step": 13488 + }, + { + "epoch": 2.504921077065924, + "grad_norm": 1.5404868125915527, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8896652460098267, + "num_tokens": 473592020.0, + "step": 13489 + }, + { + "epoch": 2.5051067780872796, + "grad_norm": 1.552808403968811, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8963052034378052, + "num_tokens": 473624780.0, + "step": 13490 + }, + { + "epoch": 2.5052924791086353, + "grad_norm": 1.7113757133483887, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8837029933929443, + "num_tokens": 473654849.0, + "step": 13491 + }, + { + "epoch": 2.5054781801299906, + "grad_norm": 1.591327428817749, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8852579593658447, + "num_tokens": 473693422.0, + "step": 13492 + }, + { + "epoch": 2.5056638811513463, + "grad_norm": 1.6037724018096924, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8746376037597656, + "num_tokens": 473728736.0, + "step": 13493 + }, + { + "epoch": 2.505849582172702, + "grad_norm": 1.577234148979187, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8901413083076477, + "num_tokens": 473761099.0, + "step": 13494 + }, + { + "epoch": 2.5060352831940573, + "grad_norm": 1.6032640933990479, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8963104486465454, + "num_tokens": 473791911.0, + "step": 13495 + }, + { + "epoch": 2.506220984215413, + "grad_norm": 1.4729324579238892, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8764908313751221, + "num_tokens": 473829899.0, + "step": 13496 + }, + { + "epoch": 2.506406685236769, + "grad_norm": 1.5133230686187744, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8782618045806885, + "num_tokens": 473868045.0, + "step": 13497 + }, + { + "epoch": 2.5065923862581245, + "grad_norm": 1.6293449401855469, + "learning_rate": 1e-06, + "loss": 0.3235, + "mean_token_accuracy": 0.88432776927948, + "num_tokens": 473899372.0, + "step": 13498 + }, + { + "epoch": 2.5067780872794803, + "grad_norm": 1.5377521514892578, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8944411277770996, + "num_tokens": 473931937.0, + "step": 13499 + }, + { + "epoch": 2.5069637883008355, + "grad_norm": 1.4909430742263794, + "learning_rate": 1e-06, + "loss": 0.2835, + "mean_token_accuracy": 0.8975249528884888, + "num_tokens": 473966052.0, + "step": 13500 + }, + { + "epoch": 2.5071494893221913, + "grad_norm": 1.4512220621109009, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8953477740287781, + "num_tokens": 474005902.0, + "step": 13501 + }, + { + "epoch": 2.507335190343547, + "grad_norm": 1.7139701843261719, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8796745538711548, + "num_tokens": 474037561.0, + "step": 13502 + }, + { + "epoch": 2.5075208913649023, + "grad_norm": 1.62997305393219, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8946141004562378, + "num_tokens": 474071161.0, + "step": 13503 + }, + { + "epoch": 2.507706592386258, + "grad_norm": 1.464781641960144, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8876297473907471, + "num_tokens": 474111289.0, + "step": 13504 + }, + { + "epoch": 2.5078922934076138, + "grad_norm": 1.5708322525024414, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8941413164138794, + "num_tokens": 474144544.0, + "step": 13505 + }, + { + "epoch": 2.5080779944289695, + "grad_norm": 1.4437551498413086, + "learning_rate": 1e-06, + "loss": 0.2679, + "mean_token_accuracy": 0.9061116576194763, + "num_tokens": 474183159.0, + "step": 13506 + }, + { + "epoch": 2.5082636954503252, + "grad_norm": 1.522227168083191, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8948474526405334, + "num_tokens": 474220860.0, + "step": 13507 + }, + { + "epoch": 2.5084493964716805, + "grad_norm": 1.5288363695144653, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8872261643409729, + "num_tokens": 474255511.0, + "step": 13508 + }, + { + "epoch": 2.5086350974930363, + "grad_norm": 1.487252950668335, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8848649263381958, + "num_tokens": 474291534.0, + "step": 13509 + }, + { + "epoch": 2.508820798514392, + "grad_norm": 1.4890992641448975, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8810861110687256, + "num_tokens": 474329252.0, + "step": 13510 + }, + { + "epoch": 2.5090064995357473, + "grad_norm": 1.5939273834228516, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8948442339897156, + "num_tokens": 474363894.0, + "step": 13511 + }, + { + "epoch": 2.509192200557103, + "grad_norm": 1.3878461122512817, + "learning_rate": 1e-06, + "loss": 0.3047, + "mean_token_accuracy": 0.8917404413223267, + "num_tokens": 474404929.0, + "step": 13512 + }, + { + "epoch": 2.5093779015784587, + "grad_norm": 1.4299404621124268, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8834713101387024, + "num_tokens": 474450093.0, + "step": 13513 + }, + { + "epoch": 2.5095636025998145, + "grad_norm": 1.4496922492980957, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8903190493583679, + "num_tokens": 474491652.0, + "step": 13514 + }, + { + "epoch": 2.5097493036211698, + "grad_norm": 1.6792711019515991, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.888671875, + "num_tokens": 474521116.0, + "step": 13515 + }, + { + "epoch": 2.5099350046425255, + "grad_norm": 1.7463525533676147, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8806555867195129, + "num_tokens": 474551327.0, + "step": 13516 + }, + { + "epoch": 2.5101207056638812, + "grad_norm": 1.7183839082717896, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8695111274719238, + "num_tokens": 474584779.0, + "step": 13517 + }, + { + "epoch": 2.5103064066852365, + "grad_norm": 1.483702540397644, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8861796855926514, + "num_tokens": 474622503.0, + "step": 13518 + }, + { + "epoch": 2.5104921077065923, + "grad_norm": 1.5001678466796875, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.896071195602417, + "num_tokens": 474659106.0, + "step": 13519 + }, + { + "epoch": 2.510677808727948, + "grad_norm": 1.479331374168396, + "learning_rate": 1e-06, + "loss": 0.2857, + "mean_token_accuracy": 0.8974395990371704, + "num_tokens": 474693730.0, + "step": 13520 + }, + { + "epoch": 2.5108635097493037, + "grad_norm": 1.5076185464859009, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8913019895553589, + "num_tokens": 474731135.0, + "step": 13521 + }, + { + "epoch": 2.5110492107706595, + "grad_norm": 1.6128755807876587, + "learning_rate": 1e-06, + "loss": 0.2838, + "mean_token_accuracy": 0.8970273733139038, + "num_tokens": 474762593.0, + "step": 13522 + }, + { + "epoch": 2.5112349117920147, + "grad_norm": 1.514504313468933, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8841632604598999, + "num_tokens": 474801834.0, + "step": 13523 + }, + { + "epoch": 2.5114206128133705, + "grad_norm": 1.5917550325393677, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8901500701904297, + "num_tokens": 474833585.0, + "step": 13524 + }, + { + "epoch": 2.511606313834726, + "grad_norm": 1.6698590517044067, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8767545223236084, + "num_tokens": 474870289.0, + "step": 13525 + }, + { + "epoch": 2.5117920148560815, + "grad_norm": 1.5235704183578491, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.889178991317749, + "num_tokens": 474910131.0, + "step": 13526 + }, + { + "epoch": 2.5119777158774372, + "grad_norm": 1.4435291290283203, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8921605944633484, + "num_tokens": 474947841.0, + "step": 13527 + }, + { + "epoch": 2.512163416898793, + "grad_norm": 1.669616937637329, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8752626776695251, + "num_tokens": 474980030.0, + "step": 13528 + }, + { + "epoch": 2.5123491179201487, + "grad_norm": 1.5544588565826416, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.892578661441803, + "num_tokens": 475017968.0, + "step": 13529 + }, + { + "epoch": 2.5125348189415044, + "grad_norm": 1.5796376466751099, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8775360584259033, + "num_tokens": 475059815.0, + "step": 13530 + }, + { + "epoch": 2.5127205199628597, + "grad_norm": 1.5989784002304077, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.884634792804718, + "num_tokens": 475093581.0, + "step": 13531 + }, + { + "epoch": 2.5129062209842155, + "grad_norm": 1.6732242107391357, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8838355541229248, + "num_tokens": 475127025.0, + "step": 13532 + }, + { + "epoch": 2.513091922005571, + "grad_norm": 1.6946756839752197, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8861175179481506, + "num_tokens": 475157036.0, + "step": 13533 + }, + { + "epoch": 2.5132776230269265, + "grad_norm": 1.5432623624801636, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8889065980911255, + "num_tokens": 475199390.0, + "step": 13534 + }, + { + "epoch": 2.513463324048282, + "grad_norm": 1.5325802564620972, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8890058398246765, + "num_tokens": 475233425.0, + "step": 13535 + }, + { + "epoch": 2.513649025069638, + "grad_norm": 1.6195344924926758, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8724343180656433, + "num_tokens": 475270051.0, + "step": 13536 + }, + { + "epoch": 2.5138347260909937, + "grad_norm": 1.6388931274414062, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8903046250343323, + "num_tokens": 475301508.0, + "step": 13537 + }, + { + "epoch": 2.514020427112349, + "grad_norm": 1.7788865566253662, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8797482252120972, + "num_tokens": 475330667.0, + "step": 13538 + }, + { + "epoch": 2.5142061281337047, + "grad_norm": 1.5344853401184082, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8816449046134949, + "num_tokens": 475369011.0, + "step": 13539 + }, + { + "epoch": 2.5143918291550604, + "grad_norm": 1.5733716487884521, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8858843445777893, + "num_tokens": 475405684.0, + "step": 13540 + }, + { + "epoch": 2.5145775301764157, + "grad_norm": 1.659226417541504, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8822159767150879, + "num_tokens": 475438296.0, + "step": 13541 + }, + { + "epoch": 2.5147632311977715, + "grad_norm": 1.5547336339950562, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8818473815917969, + "num_tokens": 475475298.0, + "step": 13542 + }, + { + "epoch": 2.514948932219127, + "grad_norm": 1.5731794834136963, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8764303922653198, + "num_tokens": 475513756.0, + "step": 13543 + }, + { + "epoch": 2.515134633240483, + "grad_norm": 1.602319598197937, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8887047171592712, + "num_tokens": 475546487.0, + "step": 13544 + }, + { + "epoch": 2.5153203342618387, + "grad_norm": 1.4152690172195435, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8914278745651245, + "num_tokens": 475586542.0, + "step": 13545 + }, + { + "epoch": 2.515506035283194, + "grad_norm": 1.5775572061538696, + "learning_rate": 1e-06, + "loss": 0.2869, + "mean_token_accuracy": 0.8973798751831055, + "num_tokens": 475617615.0, + "step": 13546 + }, + { + "epoch": 2.5156917363045497, + "grad_norm": 1.5286226272583008, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8843645453453064, + "num_tokens": 475653009.0, + "step": 13547 + }, + { + "epoch": 2.5158774373259054, + "grad_norm": 1.590254545211792, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8742564916610718, + "num_tokens": 475687477.0, + "step": 13548 + }, + { + "epoch": 2.5160631383472607, + "grad_norm": 1.5072404146194458, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8851855993270874, + "num_tokens": 475726091.0, + "step": 13549 + }, + { + "epoch": 2.5162488393686164, + "grad_norm": 1.762525200843811, + "learning_rate": 1e-06, + "loss": 0.3729, + "mean_token_accuracy": 0.8734057545661926, + "num_tokens": 475755454.0, + "step": 13550 + }, + { + "epoch": 2.516434540389972, + "grad_norm": 1.5464789867401123, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8844244480133057, + "num_tokens": 475791033.0, + "step": 13551 + }, + { + "epoch": 2.516620241411328, + "grad_norm": 1.7054617404937744, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8755602836608887, + "num_tokens": 475825157.0, + "step": 13552 + }, + { + "epoch": 2.5168059424326836, + "grad_norm": 1.5307955741882324, + "learning_rate": 1e-06, + "loss": 0.3446, + "mean_token_accuracy": 0.8799864053726196, + "num_tokens": 475861158.0, + "step": 13553 + }, + { + "epoch": 2.516991643454039, + "grad_norm": 1.5972458124160767, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.889401376247406, + "num_tokens": 475894682.0, + "step": 13554 + }, + { + "epoch": 2.5171773444753947, + "grad_norm": 1.6012128591537476, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8762900233268738, + "num_tokens": 475928604.0, + "step": 13555 + }, + { + "epoch": 2.5173630454967504, + "grad_norm": 1.6196221113204956, + "learning_rate": 1e-06, + "loss": 0.3486, + "mean_token_accuracy": 0.8777239918708801, + "num_tokens": 475965553.0, + "step": 13556 + }, + { + "epoch": 2.5175487465181057, + "grad_norm": 1.6516064405441284, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8766017556190491, + "num_tokens": 475999602.0, + "step": 13557 + }, + { + "epoch": 2.5177344475394614, + "grad_norm": 1.6845166683197021, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.874727725982666, + "num_tokens": 476036077.0, + "step": 13558 + }, + { + "epoch": 2.517920148560817, + "grad_norm": 1.5807740688323975, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8818190097808838, + "num_tokens": 476073105.0, + "step": 13559 + }, + { + "epoch": 2.518105849582173, + "grad_norm": 1.5453397035598755, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8916609883308411, + "num_tokens": 476104974.0, + "step": 13560 + }, + { + "epoch": 2.518291550603528, + "grad_norm": 1.5670139789581299, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8686370849609375, + "num_tokens": 476141456.0, + "step": 13561 + }, + { + "epoch": 2.518477251624884, + "grad_norm": 1.5373146533966064, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.888506293296814, + "num_tokens": 476177224.0, + "step": 13562 + }, + { + "epoch": 2.5186629526462396, + "grad_norm": 1.4751291275024414, + "learning_rate": 1e-06, + "loss": 0.2795, + "mean_token_accuracy": 0.9013247489929199, + "num_tokens": 476214110.0, + "step": 13563 + }, + { + "epoch": 2.518848653667595, + "grad_norm": 1.6683777570724487, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8899886608123779, + "num_tokens": 476244819.0, + "step": 13564 + }, + { + "epoch": 2.5190343546889506, + "grad_norm": 1.585446834564209, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8874364495277405, + "num_tokens": 476277896.0, + "step": 13565 + }, + { + "epoch": 2.5192200557103064, + "grad_norm": 1.7762432098388672, + "learning_rate": 1e-06, + "loss": 0.3825, + "mean_token_accuracy": 0.8707476854324341, + "num_tokens": 476311875.0, + "step": 13566 + }, + { + "epoch": 2.519405756731662, + "grad_norm": 1.5735987424850464, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8867883682250977, + "num_tokens": 476346807.0, + "step": 13567 + }, + { + "epoch": 2.519591457753018, + "grad_norm": 1.6942925453186035, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8679598569869995, + "num_tokens": 476378233.0, + "step": 13568 + }, + { + "epoch": 2.519777158774373, + "grad_norm": 1.5057464838027954, + "learning_rate": 1e-06, + "loss": 0.271, + "mean_token_accuracy": 0.9018467664718628, + "num_tokens": 476412834.0, + "step": 13569 + }, + { + "epoch": 2.519962859795729, + "grad_norm": 1.6903467178344727, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8746058940887451, + "num_tokens": 476447618.0, + "step": 13570 + }, + { + "epoch": 2.5201485608170846, + "grad_norm": 1.6370251178741455, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8781834840774536, + "num_tokens": 476483309.0, + "step": 13571 + }, + { + "epoch": 2.52033426183844, + "grad_norm": 1.5885009765625, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8863279819488525, + "num_tokens": 476515936.0, + "step": 13572 + }, + { + "epoch": 2.5205199628597956, + "grad_norm": 1.4448446035385132, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8871212005615234, + "num_tokens": 476560362.0, + "step": 13573 + }, + { + "epoch": 2.5207056638811514, + "grad_norm": 1.6087344884872437, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8846174478530884, + "num_tokens": 476594266.0, + "step": 13574 + }, + { + "epoch": 2.520891364902507, + "grad_norm": 1.6257050037384033, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8860756754875183, + "num_tokens": 476625086.0, + "step": 13575 + }, + { + "epoch": 2.521077065923863, + "grad_norm": 1.706580638885498, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8851161003112793, + "num_tokens": 476656935.0, + "step": 13576 + }, + { + "epoch": 2.521262766945218, + "grad_norm": 1.5878889560699463, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8901679515838623, + "num_tokens": 476691932.0, + "step": 13577 + }, + { + "epoch": 2.521448467966574, + "grad_norm": 1.462807059288025, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8992255330085754, + "num_tokens": 476726228.0, + "step": 13578 + }, + { + "epoch": 2.5216341689879296, + "grad_norm": 1.659659743309021, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.879942774772644, + "num_tokens": 476757607.0, + "step": 13579 + }, + { + "epoch": 2.521819870009285, + "grad_norm": 1.7574338912963867, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8864476680755615, + "num_tokens": 476785401.0, + "step": 13580 + }, + { + "epoch": 2.5220055710306406, + "grad_norm": 1.5784685611724854, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8842294216156006, + "num_tokens": 476822284.0, + "step": 13581 + }, + { + "epoch": 2.5221912720519963, + "grad_norm": 1.4362082481384277, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8928921222686768, + "num_tokens": 476862210.0, + "step": 13582 + }, + { + "epoch": 2.522376973073352, + "grad_norm": 1.5103636980056763, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.886368989944458, + "num_tokens": 476899152.0, + "step": 13583 + }, + { + "epoch": 2.5225626740947074, + "grad_norm": 1.5725294351577759, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.881359338760376, + "num_tokens": 476936359.0, + "step": 13584 + }, + { + "epoch": 2.522748375116063, + "grad_norm": 1.514594554901123, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8937609195709229, + "num_tokens": 476969773.0, + "step": 13585 + }, + { + "epoch": 2.522934076137419, + "grad_norm": 1.442015528678894, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8954793214797974, + "num_tokens": 477005295.0, + "step": 13586 + }, + { + "epoch": 2.523119777158774, + "grad_norm": 1.5011144876480103, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.8960179090499878, + "num_tokens": 477036690.0, + "step": 13587 + }, + { + "epoch": 2.52330547818013, + "grad_norm": 1.6441292762756348, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8919923305511475, + "num_tokens": 477074461.0, + "step": 13588 + }, + { + "epoch": 2.5234911792014856, + "grad_norm": 1.5112632513046265, + "learning_rate": 1e-06, + "loss": 0.2735, + "mean_token_accuracy": 0.901436448097229, + "num_tokens": 477108715.0, + "step": 13589 + }, + { + "epoch": 2.5236768802228413, + "grad_norm": 1.5111228227615356, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8900923728942871, + "num_tokens": 477143938.0, + "step": 13590 + }, + { + "epoch": 2.523862581244197, + "grad_norm": 1.5430771112442017, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8819098472595215, + "num_tokens": 477179978.0, + "step": 13591 + }, + { + "epoch": 2.5240482822655523, + "grad_norm": 1.5993636846542358, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8740485310554504, + "num_tokens": 477215405.0, + "step": 13592 + }, + { + "epoch": 2.524233983286908, + "grad_norm": 1.6781994104385376, + "learning_rate": 1e-06, + "loss": 0.2728, + "mean_token_accuracy": 0.9017540216445923, + "num_tokens": 477244240.0, + "step": 13593 + }, + { + "epoch": 2.524419684308264, + "grad_norm": 1.602508783340454, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8876380920410156, + "num_tokens": 477278317.0, + "step": 13594 + }, + { + "epoch": 2.524605385329619, + "grad_norm": 1.5982213020324707, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8940051794052124, + "num_tokens": 477317552.0, + "step": 13595 + }, + { + "epoch": 2.524791086350975, + "grad_norm": 1.5896507501602173, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8931995034217834, + "num_tokens": 477350388.0, + "step": 13596 + }, + { + "epoch": 2.5249767873723306, + "grad_norm": 1.479024887084961, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8911608457565308, + "num_tokens": 477386889.0, + "step": 13597 + }, + { + "epoch": 2.5251624883936863, + "grad_norm": 1.6428217887878418, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8862268924713135, + "num_tokens": 477421610.0, + "step": 13598 + }, + { + "epoch": 2.525348189415042, + "grad_norm": 1.6309149265289307, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.8917536735534668, + "num_tokens": 477449370.0, + "step": 13599 + }, + { + "epoch": 2.5255338904363973, + "grad_norm": 1.5271865129470825, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.9003757238388062, + "num_tokens": 477481200.0, + "step": 13600 + }, + { + "epoch": 2.525719591457753, + "grad_norm": 1.603442668914795, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8738009929656982, + "num_tokens": 477516579.0, + "step": 13601 + }, + { + "epoch": 2.5259052924791088, + "grad_norm": 1.4665931463241577, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.8938716650009155, + "num_tokens": 477551658.0, + "step": 13602 + }, + { + "epoch": 2.526090993500464, + "grad_norm": 1.529874324798584, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8829338550567627, + "num_tokens": 477590109.0, + "step": 13603 + }, + { + "epoch": 2.52627669452182, + "grad_norm": 1.4230608940124512, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8950119018554688, + "num_tokens": 477631134.0, + "step": 13604 + }, + { + "epoch": 2.5264623955431755, + "grad_norm": 1.5604044198989868, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.8834985494613647, + "num_tokens": 477667768.0, + "step": 13605 + }, + { + "epoch": 2.5266480965645313, + "grad_norm": 1.5657267570495605, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8904387354850769, + "num_tokens": 477702648.0, + "step": 13606 + }, + { + "epoch": 2.526833797585887, + "grad_norm": 1.399396300315857, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8920005559921265, + "num_tokens": 477743169.0, + "step": 13607 + }, + { + "epoch": 2.5270194986072423, + "grad_norm": 1.7151103019714355, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8817877769470215, + "num_tokens": 477776966.0, + "step": 13608 + }, + { + "epoch": 2.527205199628598, + "grad_norm": 1.6943774223327637, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8768306970596313, + "num_tokens": 477810298.0, + "step": 13609 + }, + { + "epoch": 2.5273909006499533, + "grad_norm": 1.5222703218460083, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8949356079101562, + "num_tokens": 477845336.0, + "step": 13610 + }, + { + "epoch": 2.527576601671309, + "grad_norm": 1.4376705884933472, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8850710391998291, + "num_tokens": 477886360.0, + "step": 13611 + }, + { + "epoch": 2.5277623026926648, + "grad_norm": 1.6111245155334473, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8776541352272034, + "num_tokens": 477922930.0, + "step": 13612 + }, + { + "epoch": 2.5279480037140205, + "grad_norm": 1.5478588342666626, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8788966536521912, + "num_tokens": 477956395.0, + "step": 13613 + }, + { + "epoch": 2.5281337047353762, + "grad_norm": 1.659210205078125, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8677022457122803, + "num_tokens": 477999093.0, + "step": 13614 + }, + { + "epoch": 2.5283194057567315, + "grad_norm": 1.520358681678772, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8823891878128052, + "num_tokens": 478035571.0, + "step": 13615 + }, + { + "epoch": 2.5285051067780873, + "grad_norm": 1.5697972774505615, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8805535435676575, + "num_tokens": 478073959.0, + "step": 13616 + }, + { + "epoch": 2.528690807799443, + "grad_norm": 1.5396924018859863, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8917437791824341, + "num_tokens": 478108071.0, + "step": 13617 + }, + { + "epoch": 2.5288765088207983, + "grad_norm": 1.4931334257125854, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8886273503303528, + "num_tokens": 478148140.0, + "step": 13618 + }, + { + "epoch": 2.529062209842154, + "grad_norm": 1.5287283658981323, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8800104260444641, + "num_tokens": 478184367.0, + "step": 13619 + }, + { + "epoch": 2.5292479108635098, + "grad_norm": 1.6294995546340942, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8868054151535034, + "num_tokens": 478218040.0, + "step": 13620 + }, + { + "epoch": 2.5294336118848655, + "grad_norm": 1.5411789417266846, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8798868060112, + "num_tokens": 478254392.0, + "step": 13621 + }, + { + "epoch": 2.529619312906221, + "grad_norm": 1.5580508708953857, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8963383436203003, + "num_tokens": 478287554.0, + "step": 13622 + }, + { + "epoch": 2.5298050139275765, + "grad_norm": 1.5737125873565674, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8869059681892395, + "num_tokens": 478326493.0, + "step": 13623 + }, + { + "epoch": 2.5299907149489322, + "grad_norm": 1.6991350650787354, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8800919055938721, + "num_tokens": 478358453.0, + "step": 13624 + }, + { + "epoch": 2.530176415970288, + "grad_norm": 1.5443445444107056, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8848730325698853, + "num_tokens": 478394173.0, + "step": 13625 + }, + { + "epoch": 2.5303621169916433, + "grad_norm": 1.5824308395385742, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8934475183486938, + "num_tokens": 478426257.0, + "step": 13626 + }, + { + "epoch": 2.530547818012999, + "grad_norm": 1.4629766941070557, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.9007222652435303, + "num_tokens": 478462943.0, + "step": 13627 + }, + { + "epoch": 2.5307335190343547, + "grad_norm": 1.821103572845459, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8880441188812256, + "num_tokens": 478490318.0, + "step": 13628 + }, + { + "epoch": 2.5309192200557105, + "grad_norm": 1.4588042497634888, + "learning_rate": 1e-06, + "loss": 0.2563, + "mean_token_accuracy": 0.9078339338302612, + "num_tokens": 478525261.0, + "step": 13629 + }, + { + "epoch": 2.531104921077066, + "grad_norm": 1.7107409238815308, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.880285382270813, + "num_tokens": 478554770.0, + "step": 13630 + }, + { + "epoch": 2.5312906220984215, + "grad_norm": 1.549975037574768, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8904764652252197, + "num_tokens": 478593216.0, + "step": 13631 + }, + { + "epoch": 2.531476323119777, + "grad_norm": 1.5801947116851807, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8902004957199097, + "num_tokens": 478626009.0, + "step": 13632 + }, + { + "epoch": 2.5316620241411325, + "grad_norm": 1.42819082736969, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8881327509880066, + "num_tokens": 478670617.0, + "step": 13633 + }, + { + "epoch": 2.5318477251624882, + "grad_norm": 1.5356303453445435, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.8937041759490967, + "num_tokens": 478703224.0, + "step": 13634 + }, + { + "epoch": 2.532033426183844, + "grad_norm": 1.7313083410263062, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8674657344818115, + "num_tokens": 478733582.0, + "step": 13635 + }, + { + "epoch": 2.5322191272051997, + "grad_norm": 1.4542529582977295, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8880151510238647, + "num_tokens": 478770366.0, + "step": 13636 + }, + { + "epoch": 2.5324048282265554, + "grad_norm": 1.6181912422180176, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8884788751602173, + "num_tokens": 478809093.0, + "step": 13637 + }, + { + "epoch": 2.5325905292479107, + "grad_norm": 1.7227836847305298, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8844320774078369, + "num_tokens": 478839078.0, + "step": 13638 + }, + { + "epoch": 2.5327762302692665, + "grad_norm": 1.4350886344909668, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8945951461791992, + "num_tokens": 478875715.0, + "step": 13639 + }, + { + "epoch": 2.532961931290622, + "grad_norm": 1.4213967323303223, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8898736834526062, + "num_tokens": 478914230.0, + "step": 13640 + }, + { + "epoch": 2.5331476323119775, + "grad_norm": 1.5295219421386719, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8921716213226318, + "num_tokens": 478947871.0, + "step": 13641 + }, + { + "epoch": 2.533333333333333, + "grad_norm": 1.5716794729232788, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8905856609344482, + "num_tokens": 478982415.0, + "step": 13642 + }, + { + "epoch": 2.533519034354689, + "grad_norm": 1.5251970291137695, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8952016830444336, + "num_tokens": 479015252.0, + "step": 13643 + }, + { + "epoch": 2.5337047353760447, + "grad_norm": 1.568648099899292, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.876840353012085, + "num_tokens": 479055283.0, + "step": 13644 + }, + { + "epoch": 2.5338904363974004, + "grad_norm": 1.8228726387023926, + "learning_rate": 1e-06, + "loss": 0.3668, + "mean_token_accuracy": 0.8728969097137451, + "num_tokens": 479083945.0, + "step": 13645 + }, + { + "epoch": 2.5340761374187557, + "grad_norm": 1.5686699151992798, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8857824206352234, + "num_tokens": 479122939.0, + "step": 13646 + }, + { + "epoch": 2.5342618384401114, + "grad_norm": 1.5258175134658813, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8941126465797424, + "num_tokens": 479158159.0, + "step": 13647 + }, + { + "epoch": 2.534447539461467, + "grad_norm": 1.6032122373580933, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8830251693725586, + "num_tokens": 479191784.0, + "step": 13648 + }, + { + "epoch": 2.5346332404828225, + "grad_norm": 1.5865728855133057, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8887803554534912, + "num_tokens": 479225269.0, + "step": 13649 + }, + { + "epoch": 2.534818941504178, + "grad_norm": 1.634852409362793, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8891540765762329, + "num_tokens": 479260420.0, + "step": 13650 + }, + { + "epoch": 2.535004642525534, + "grad_norm": 1.5473613739013672, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8853885531425476, + "num_tokens": 479301678.0, + "step": 13651 + }, + { + "epoch": 2.5351903435468897, + "grad_norm": 1.719200849533081, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8931727409362793, + "num_tokens": 479332101.0, + "step": 13652 + }, + { + "epoch": 2.5353760445682454, + "grad_norm": 1.7136679887771606, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.9000405073165894, + "num_tokens": 479358823.0, + "step": 13653 + }, + { + "epoch": 2.5355617455896007, + "grad_norm": 1.669488549232483, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.884833574295044, + "num_tokens": 479391402.0, + "step": 13654 + }, + { + "epoch": 2.5357474466109564, + "grad_norm": 1.5799367427825928, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8806246519088745, + "num_tokens": 479429219.0, + "step": 13655 + }, + { + "epoch": 2.5359331476323117, + "grad_norm": 1.597535490989685, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8696945905685425, + "num_tokens": 479465222.0, + "step": 13656 + }, + { + "epoch": 2.5361188486536674, + "grad_norm": 1.8033617734909058, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8844931721687317, + "num_tokens": 479495197.0, + "step": 13657 + }, + { + "epoch": 2.536304549675023, + "grad_norm": 1.559217929840088, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8935644626617432, + "num_tokens": 479525171.0, + "step": 13658 + }, + { + "epoch": 2.536490250696379, + "grad_norm": 1.5798360109329224, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8842209577560425, + "num_tokens": 479559556.0, + "step": 13659 + }, + { + "epoch": 2.5366759517177346, + "grad_norm": 1.5241895914077759, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.9020750522613525, + "num_tokens": 479592809.0, + "step": 13660 + }, + { + "epoch": 2.53686165273909, + "grad_norm": 1.6999685764312744, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8796408772468567, + "num_tokens": 479622756.0, + "step": 13661 + }, + { + "epoch": 2.5370473537604457, + "grad_norm": 1.5471887588500977, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8805283904075623, + "num_tokens": 479657003.0, + "step": 13662 + }, + { + "epoch": 2.5372330547818014, + "grad_norm": 1.5798985958099365, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8848716020584106, + "num_tokens": 479691834.0, + "step": 13663 + }, + { + "epoch": 2.5374187558031567, + "grad_norm": 1.6348357200622559, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.873751699924469, + "num_tokens": 479726886.0, + "step": 13664 + }, + { + "epoch": 2.5376044568245124, + "grad_norm": 1.4833756685256958, + "learning_rate": 1e-06, + "loss": 0.2578, + "mean_token_accuracy": 0.9054886102676392, + "num_tokens": 479760796.0, + "step": 13665 + }, + { + "epoch": 2.537790157845868, + "grad_norm": 1.531246542930603, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8899317979812622, + "num_tokens": 479796468.0, + "step": 13666 + }, + { + "epoch": 2.537975858867224, + "grad_norm": 1.572353482246399, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8801712989807129, + "num_tokens": 479833647.0, + "step": 13667 + }, + { + "epoch": 2.5381615598885796, + "grad_norm": 1.4837437868118286, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8843194246292114, + "num_tokens": 479869237.0, + "step": 13668 + }, + { + "epoch": 2.538347260909935, + "grad_norm": 1.475719928741455, + "learning_rate": 1e-06, + "loss": 0.3902, + "mean_token_accuracy": 0.8695975542068481, + "num_tokens": 479913775.0, + "step": 13669 + }, + { + "epoch": 2.5385329619312906, + "grad_norm": 1.504945993423462, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8977274894714355, + "num_tokens": 479947872.0, + "step": 13670 + }, + { + "epoch": 2.5387186629526464, + "grad_norm": 1.5705512762069702, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8810986280441284, + "num_tokens": 479985777.0, + "step": 13671 + }, + { + "epoch": 2.5389043639740017, + "grad_norm": 1.69414222240448, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8848936557769775, + "num_tokens": 480013498.0, + "step": 13672 + }, + { + "epoch": 2.5390900649953574, + "grad_norm": 1.4152880907058716, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8936973810195923, + "num_tokens": 480054893.0, + "step": 13673 + }, + { + "epoch": 2.539275766016713, + "grad_norm": 1.604274868965149, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8826792240142822, + "num_tokens": 480088354.0, + "step": 13674 + }, + { + "epoch": 2.539461467038069, + "grad_norm": 1.6154996156692505, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.887354850769043, + "num_tokens": 480119533.0, + "step": 13675 + }, + { + "epoch": 2.5396471680594246, + "grad_norm": 1.6197611093521118, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8815864324569702, + "num_tokens": 480153312.0, + "step": 13676 + }, + { + "epoch": 2.53983286908078, + "grad_norm": 1.6928032636642456, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8710892796516418, + "num_tokens": 480188481.0, + "step": 13677 + }, + { + "epoch": 2.5400185701021356, + "grad_norm": 1.6177072525024414, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.887519359588623, + "num_tokens": 480222179.0, + "step": 13678 + }, + { + "epoch": 2.5402042711234913, + "grad_norm": 1.6874067783355713, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8908464908599854, + "num_tokens": 480253589.0, + "step": 13679 + }, + { + "epoch": 2.5403899721448466, + "grad_norm": 1.6062288284301758, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8802329897880554, + "num_tokens": 480288335.0, + "step": 13680 + }, + { + "epoch": 2.5405756731662024, + "grad_norm": 1.5191612243652344, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8891819715499878, + "num_tokens": 480326290.0, + "step": 13681 + }, + { + "epoch": 2.540761374187558, + "grad_norm": 1.6755211353302002, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8921245336532593, + "num_tokens": 480359328.0, + "step": 13682 + }, + { + "epoch": 2.540947075208914, + "grad_norm": 1.5723400115966797, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8875573873519897, + "num_tokens": 480394430.0, + "step": 13683 + }, + { + "epoch": 2.541132776230269, + "grad_norm": 1.769087553024292, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8851488828659058, + "num_tokens": 480425947.0, + "step": 13684 + }, + { + "epoch": 2.541318477251625, + "grad_norm": 1.5174040794372559, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8855379819869995, + "num_tokens": 480462249.0, + "step": 13685 + }, + { + "epoch": 2.5415041782729806, + "grad_norm": 1.4695929288864136, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8932638168334961, + "num_tokens": 480502258.0, + "step": 13686 + }, + { + "epoch": 2.541689879294336, + "grad_norm": 1.4727331399917603, + "learning_rate": 1e-06, + "loss": 0.3726, + "mean_token_accuracy": 0.8701339960098267, + "num_tokens": 480545865.0, + "step": 13687 + }, + { + "epoch": 2.5418755803156916, + "grad_norm": 1.4671469926834106, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8875715732574463, + "num_tokens": 480587375.0, + "step": 13688 + }, + { + "epoch": 2.5420612813370473, + "grad_norm": 1.5990716218948364, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8842136263847351, + "num_tokens": 480621332.0, + "step": 13689 + }, + { + "epoch": 2.542246982358403, + "grad_norm": 1.5803108215332031, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8927391171455383, + "num_tokens": 480654805.0, + "step": 13690 + }, + { + "epoch": 2.542432683379759, + "grad_norm": 1.484649658203125, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8799849152565002, + "num_tokens": 480692811.0, + "step": 13691 + }, + { + "epoch": 2.542618384401114, + "grad_norm": 1.6048749685287476, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8838810920715332, + "num_tokens": 480725262.0, + "step": 13692 + }, + { + "epoch": 2.54280408542247, + "grad_norm": 1.4300308227539062, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8804517984390259, + "num_tokens": 480765981.0, + "step": 13693 + }, + { + "epoch": 2.5429897864438256, + "grad_norm": 1.6921947002410889, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8898817896842957, + "num_tokens": 480795282.0, + "step": 13694 + }, + { + "epoch": 2.543175487465181, + "grad_norm": 1.539841890335083, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.886091947555542, + "num_tokens": 480831339.0, + "step": 13695 + }, + { + "epoch": 2.5433611884865366, + "grad_norm": 1.543778419494629, + "learning_rate": 1e-06, + "loss": 0.2984, + "mean_token_accuracy": 0.892268180847168, + "num_tokens": 480861902.0, + "step": 13696 + }, + { + "epoch": 2.5435468895078923, + "grad_norm": 1.6156086921691895, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8745523691177368, + "num_tokens": 480897386.0, + "step": 13697 + }, + { + "epoch": 2.543732590529248, + "grad_norm": 1.5442705154418945, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8868093490600586, + "num_tokens": 480934124.0, + "step": 13698 + }, + { + "epoch": 2.543918291550604, + "grad_norm": 1.5198802947998047, + "learning_rate": 1e-06, + "loss": 0.2898, + "mean_token_accuracy": 0.9000811576843262, + "num_tokens": 480968987.0, + "step": 13699 + }, + { + "epoch": 2.544103992571959, + "grad_norm": 1.6076072454452515, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8820838928222656, + "num_tokens": 481004527.0, + "step": 13700 + }, + { + "epoch": 2.544289693593315, + "grad_norm": 1.6326069831848145, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8900578618049622, + "num_tokens": 481036695.0, + "step": 13701 + }, + { + "epoch": 2.5444753946146705, + "grad_norm": 1.7369211912155151, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8826523423194885, + "num_tokens": 481073297.0, + "step": 13702 + }, + { + "epoch": 2.544661095636026, + "grad_norm": 1.636342167854309, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8842582106590271, + "num_tokens": 481106200.0, + "step": 13703 + }, + { + "epoch": 2.5448467966573816, + "grad_norm": 1.685874581336975, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8901802897453308, + "num_tokens": 481134787.0, + "step": 13704 + }, + { + "epoch": 2.5450324976787373, + "grad_norm": 1.5087915658950806, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8913439512252808, + "num_tokens": 481171547.0, + "step": 13705 + }, + { + "epoch": 2.545218198700093, + "grad_norm": 1.6289894580841064, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8894871473312378, + "num_tokens": 481208317.0, + "step": 13706 + }, + { + "epoch": 2.5454038997214483, + "grad_norm": 1.4865769147872925, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8972353935241699, + "num_tokens": 481241446.0, + "step": 13707 + }, + { + "epoch": 2.545589600742804, + "grad_norm": 1.4257690906524658, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8847224116325378, + "num_tokens": 481284794.0, + "step": 13708 + }, + { + "epoch": 2.54577530176416, + "grad_norm": 1.6691577434539795, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.873747706413269, + "num_tokens": 481316862.0, + "step": 13709 + }, + { + "epoch": 2.545961002785515, + "grad_norm": 1.7240713834762573, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8769232630729675, + "num_tokens": 481349813.0, + "step": 13710 + }, + { + "epoch": 2.546146703806871, + "grad_norm": 1.5433435440063477, + "learning_rate": 1e-06, + "loss": 0.2942, + "mean_token_accuracy": 0.8947674036026001, + "num_tokens": 481386298.0, + "step": 13711 + }, + { + "epoch": 2.5463324048282265, + "grad_norm": 1.6397929191589355, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8947960734367371, + "num_tokens": 481415580.0, + "step": 13712 + }, + { + "epoch": 2.5465181058495823, + "grad_norm": 1.7481589317321777, + "learning_rate": 1e-06, + "loss": 0.4046, + "mean_token_accuracy": 0.8707396984100342, + "num_tokens": 481452380.0, + "step": 13713 + }, + { + "epoch": 2.546703806870938, + "grad_norm": 1.515523910522461, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8840048313140869, + "num_tokens": 481487828.0, + "step": 13714 + }, + { + "epoch": 2.5468895078922933, + "grad_norm": 1.5399678945541382, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8807385563850403, + "num_tokens": 481523334.0, + "step": 13715 + }, + { + "epoch": 2.547075208913649, + "grad_norm": 1.563280463218689, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.89023357629776, + "num_tokens": 481557441.0, + "step": 13716 + }, + { + "epoch": 2.5472609099350048, + "grad_norm": 1.6299173831939697, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8887485265731812, + "num_tokens": 481590558.0, + "step": 13717 + }, + { + "epoch": 2.54744661095636, + "grad_norm": 1.6896188259124756, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.8813340663909912, + "num_tokens": 481620309.0, + "step": 13718 + }, + { + "epoch": 2.547632311977716, + "grad_norm": 1.5160717964172363, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8907468914985657, + "num_tokens": 481653961.0, + "step": 13719 + }, + { + "epoch": 2.5478180129990715, + "grad_norm": 1.483134150505066, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8862147927284241, + "num_tokens": 481691708.0, + "step": 13720 + }, + { + "epoch": 2.5480037140204272, + "grad_norm": 1.5172362327575684, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8838140964508057, + "num_tokens": 481732422.0, + "step": 13721 + }, + { + "epoch": 2.548189415041783, + "grad_norm": 1.5075970888137817, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8919124007225037, + "num_tokens": 481769579.0, + "step": 13722 + }, + { + "epoch": 2.5483751160631383, + "grad_norm": 1.566628336906433, + "learning_rate": 1e-06, + "loss": 0.2943, + "mean_token_accuracy": 0.8945335149765015, + "num_tokens": 481804797.0, + "step": 13723 + }, + { + "epoch": 2.548560817084494, + "grad_norm": 1.6277040243148804, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8770226240158081, + "num_tokens": 481837631.0, + "step": 13724 + }, + { + "epoch": 2.5487465181058497, + "grad_norm": 1.4624830484390259, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8990800380706787, + "num_tokens": 481873597.0, + "step": 13725 + }, + { + "epoch": 2.548932219127205, + "grad_norm": 1.5801845788955688, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8751528859138489, + "num_tokens": 481908714.0, + "step": 13726 + }, + { + "epoch": 2.5491179201485608, + "grad_norm": 1.6002719402313232, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8905029892921448, + "num_tokens": 481939586.0, + "step": 13727 + }, + { + "epoch": 2.5493036211699165, + "grad_norm": 1.4772199392318726, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8962498903274536, + "num_tokens": 481972686.0, + "step": 13728 + }, + { + "epoch": 2.5494893221912722, + "grad_norm": 1.560823678970337, + "learning_rate": 1e-06, + "loss": 0.3064, + "mean_token_accuracy": 0.890109658241272, + "num_tokens": 482007950.0, + "step": 13729 + }, + { + "epoch": 2.5496750232126275, + "grad_norm": 1.492477536201477, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.886185884475708, + "num_tokens": 482046860.0, + "step": 13730 + }, + { + "epoch": 2.5498607242339832, + "grad_norm": 1.5336735248565674, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8842451572418213, + "num_tokens": 482085112.0, + "step": 13731 + }, + { + "epoch": 2.550046425255339, + "grad_norm": 1.4623920917510986, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8816410303115845, + "num_tokens": 482127611.0, + "step": 13732 + }, + { + "epoch": 2.5502321262766943, + "grad_norm": 1.5778096914291382, + "learning_rate": 1e-06, + "loss": 0.2929, + "mean_token_accuracy": 0.8998112678527832, + "num_tokens": 482158758.0, + "step": 13733 + }, + { + "epoch": 2.55041782729805, + "grad_norm": 1.6819978952407837, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8856467008590698, + "num_tokens": 482187276.0, + "step": 13734 + }, + { + "epoch": 2.5506035283194057, + "grad_norm": 1.5406568050384521, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8966293931007385, + "num_tokens": 482222042.0, + "step": 13735 + }, + { + "epoch": 2.5507892293407615, + "grad_norm": 1.4720741510391235, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.8965707421302795, + "num_tokens": 482259262.0, + "step": 13736 + }, + { + "epoch": 2.550974930362117, + "grad_norm": 1.5340665578842163, + "learning_rate": 1e-06, + "loss": 0.2835, + "mean_token_accuracy": 0.8979246020317078, + "num_tokens": 482290687.0, + "step": 13737 + }, + { + "epoch": 2.5511606313834725, + "grad_norm": 1.5132195949554443, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8777366876602173, + "num_tokens": 482326077.0, + "step": 13738 + }, + { + "epoch": 2.551346332404828, + "grad_norm": 1.6044639348983765, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.885069727897644, + "num_tokens": 482365748.0, + "step": 13739 + }, + { + "epoch": 2.551532033426184, + "grad_norm": 1.6274834871292114, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8917901515960693, + "num_tokens": 482398196.0, + "step": 13740 + }, + { + "epoch": 2.5517177344475392, + "grad_norm": 2.109405517578125, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8825974464416504, + "num_tokens": 482434715.0, + "step": 13741 + }, + { + "epoch": 2.551903435468895, + "grad_norm": 1.7101011276245117, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8872372508049011, + "num_tokens": 482465070.0, + "step": 13742 + }, + { + "epoch": 2.5520891364902507, + "grad_norm": 1.5813425779342651, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8864153623580933, + "num_tokens": 482498641.0, + "step": 13743 + }, + { + "epoch": 2.5522748375116064, + "grad_norm": 1.498504400253296, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8892269134521484, + "num_tokens": 482533179.0, + "step": 13744 + }, + { + "epoch": 2.552460538532962, + "grad_norm": 1.5655457973480225, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8850039839744568, + "num_tokens": 482569285.0, + "step": 13745 + }, + { + "epoch": 2.5526462395543175, + "grad_norm": 1.6432000398635864, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.884175181388855, + "num_tokens": 482599324.0, + "step": 13746 + }, + { + "epoch": 2.552831940575673, + "grad_norm": 1.5102883577346802, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.890823483467102, + "num_tokens": 482634345.0, + "step": 13747 + }, + { + "epoch": 2.553017641597029, + "grad_norm": 1.5600367784500122, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8835569620132446, + "num_tokens": 482668327.0, + "step": 13748 + }, + { + "epoch": 2.553203342618384, + "grad_norm": 1.4707386493682861, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8913243412971497, + "num_tokens": 482702580.0, + "step": 13749 + }, + { + "epoch": 2.55338904363974, + "grad_norm": 1.596213698387146, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.879671573638916, + "num_tokens": 482734195.0, + "step": 13750 + }, + { + "epoch": 2.5535747446610957, + "grad_norm": 1.5296002626419067, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.8938831090927124, + "num_tokens": 482768165.0, + "step": 13751 + }, + { + "epoch": 2.5537604456824514, + "grad_norm": 1.3631271123886108, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8938135504722595, + "num_tokens": 482812349.0, + "step": 13752 + }, + { + "epoch": 2.5539461467038067, + "grad_norm": 1.5525330305099487, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8964675068855286, + "num_tokens": 482847688.0, + "step": 13753 + }, + { + "epoch": 2.5541318477251624, + "grad_norm": 1.5202761888504028, + "learning_rate": 1e-06, + "loss": 0.2688, + "mean_token_accuracy": 0.9042998552322388, + "num_tokens": 482882552.0, + "step": 13754 + }, + { + "epoch": 2.554317548746518, + "grad_norm": 1.6071014404296875, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8831751346588135, + "num_tokens": 482918635.0, + "step": 13755 + }, + { + "epoch": 2.5545032497678735, + "grad_norm": 1.361877679824829, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8894873857498169, + "num_tokens": 482962995.0, + "step": 13756 + }, + { + "epoch": 2.554688950789229, + "grad_norm": 1.691596508026123, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8828641176223755, + "num_tokens": 482993982.0, + "step": 13757 + }, + { + "epoch": 2.554874651810585, + "grad_norm": 1.6984034776687622, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8677005767822266, + "num_tokens": 483031205.0, + "step": 13758 + }, + { + "epoch": 2.5550603528319407, + "grad_norm": 1.4858492612838745, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8846303224563599, + "num_tokens": 483068253.0, + "step": 13759 + }, + { + "epoch": 2.5552460538532964, + "grad_norm": 1.5645161867141724, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8918760418891907, + "num_tokens": 483101579.0, + "step": 13760 + }, + { + "epoch": 2.5554317548746517, + "grad_norm": 1.5492428541183472, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8805217742919922, + "num_tokens": 483141302.0, + "step": 13761 + }, + { + "epoch": 2.5556174558960074, + "grad_norm": 1.5517330169677734, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8759639859199524, + "num_tokens": 483177949.0, + "step": 13762 + }, + { + "epoch": 2.555803156917363, + "grad_norm": 1.5546982288360596, + "learning_rate": 1e-06, + "loss": 0.2942, + "mean_token_accuracy": 0.8942378759384155, + "num_tokens": 483212951.0, + "step": 13763 + }, + { + "epoch": 2.5559888579387184, + "grad_norm": 1.5606393814086914, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8899847865104675, + "num_tokens": 483245891.0, + "step": 13764 + }, + { + "epoch": 2.556174558960074, + "grad_norm": 1.5617610216140747, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8886532783508301, + "num_tokens": 483283019.0, + "step": 13765 + }, + { + "epoch": 2.55636025998143, + "grad_norm": 1.6052151918411255, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8941012620925903, + "num_tokens": 483314090.0, + "step": 13766 + }, + { + "epoch": 2.5565459610027856, + "grad_norm": 1.8714715242385864, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8943783044815063, + "num_tokens": 483338683.0, + "step": 13767 + }, + { + "epoch": 2.5567316620241414, + "grad_norm": 1.6308656930923462, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8803882598876953, + "num_tokens": 483373823.0, + "step": 13768 + }, + { + "epoch": 2.5569173630454967, + "grad_norm": 1.5022211074829102, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8849314451217651, + "num_tokens": 483409727.0, + "step": 13769 + }, + { + "epoch": 2.5571030640668524, + "grad_norm": 1.518721580505371, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8891830444335938, + "num_tokens": 483443557.0, + "step": 13770 + }, + { + "epoch": 2.557288765088208, + "grad_norm": 1.4295283555984497, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8858652114868164, + "num_tokens": 483483207.0, + "step": 13771 + }, + { + "epoch": 2.5574744661095634, + "grad_norm": 1.6443727016448975, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8746296763420105, + "num_tokens": 483517510.0, + "step": 13772 + }, + { + "epoch": 2.557660167130919, + "grad_norm": 1.5316686630249023, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8771016597747803, + "num_tokens": 483554132.0, + "step": 13773 + }, + { + "epoch": 2.557845868152275, + "grad_norm": 1.8758437633514404, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8837285041809082, + "num_tokens": 483580134.0, + "step": 13774 + }, + { + "epoch": 2.5580315691736306, + "grad_norm": 1.6279850006103516, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8873882293701172, + "num_tokens": 483612741.0, + "step": 13775 + }, + { + "epoch": 2.5582172701949863, + "grad_norm": 1.5653613805770874, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8867451548576355, + "num_tokens": 483650891.0, + "step": 13776 + }, + { + "epoch": 2.5584029712163416, + "grad_norm": 1.6124989986419678, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8904169797897339, + "num_tokens": 483688802.0, + "step": 13777 + }, + { + "epoch": 2.5585886722376974, + "grad_norm": 1.5517010688781738, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8815617561340332, + "num_tokens": 483724882.0, + "step": 13778 + }, + { + "epoch": 2.5587743732590527, + "grad_norm": 1.5178683996200562, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8912698030471802, + "num_tokens": 483761092.0, + "step": 13779 + }, + { + "epoch": 2.5589600742804084, + "grad_norm": 1.49299955368042, + "learning_rate": 1e-06, + "loss": 0.2708, + "mean_token_accuracy": 0.9030656814575195, + "num_tokens": 483795074.0, + "step": 13780 + }, + { + "epoch": 2.559145775301764, + "grad_norm": 1.6453883647918701, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8809918761253357, + "num_tokens": 483827421.0, + "step": 13781 + }, + { + "epoch": 2.55933147632312, + "grad_norm": 1.6199195384979248, + "learning_rate": 1e-06, + "loss": 0.3527, + "mean_token_accuracy": 0.8791928291320801, + "num_tokens": 483862423.0, + "step": 13782 + }, + { + "epoch": 2.5595171773444756, + "grad_norm": 1.4806791543960571, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8891453146934509, + "num_tokens": 483898705.0, + "step": 13783 + }, + { + "epoch": 2.559702878365831, + "grad_norm": 1.4526100158691406, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8948108553886414, + "num_tokens": 483936919.0, + "step": 13784 + }, + { + "epoch": 2.5598885793871866, + "grad_norm": 1.6572073698043823, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8795434236526489, + "num_tokens": 483970889.0, + "step": 13785 + }, + { + "epoch": 2.5600742804085423, + "grad_norm": 1.5564534664154053, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8950728178024292, + "num_tokens": 484006202.0, + "step": 13786 + }, + { + "epoch": 2.5602599814298976, + "grad_norm": 1.4638986587524414, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.88321852684021, + "num_tokens": 484046799.0, + "step": 13787 + }, + { + "epoch": 2.5604456824512534, + "grad_norm": 1.3847090005874634, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8851782083511353, + "num_tokens": 484090781.0, + "step": 13788 + }, + { + "epoch": 2.560631383472609, + "grad_norm": 1.429918646812439, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.882211446762085, + "num_tokens": 484133646.0, + "step": 13789 + }, + { + "epoch": 2.560817084493965, + "grad_norm": 1.4152106046676636, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.897284746170044, + "num_tokens": 484175563.0, + "step": 13790 + }, + { + "epoch": 2.5610027855153206, + "grad_norm": 1.588098168373108, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.892123818397522, + "num_tokens": 484211078.0, + "step": 13791 + }, + { + "epoch": 2.561188486536676, + "grad_norm": 1.6922144889831543, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8719854950904846, + "num_tokens": 484242429.0, + "step": 13792 + }, + { + "epoch": 2.5613741875580316, + "grad_norm": 1.5743969678878784, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8880033493041992, + "num_tokens": 484277427.0, + "step": 13793 + }, + { + "epoch": 2.5615598885793873, + "grad_norm": 1.544600486755371, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.889250636100769, + "num_tokens": 484310278.0, + "step": 13794 + }, + { + "epoch": 2.5617455896007426, + "grad_norm": 1.7477329969406128, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.888069748878479, + "num_tokens": 484337632.0, + "step": 13795 + }, + { + "epoch": 2.5619312906220983, + "grad_norm": 1.789117455482483, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8869227170944214, + "num_tokens": 484368413.0, + "step": 13796 + }, + { + "epoch": 2.562116991643454, + "grad_norm": 1.4876539707183838, + "learning_rate": 1e-06, + "loss": 0.2844, + "mean_token_accuracy": 0.9012131690979004, + "num_tokens": 484403395.0, + "step": 13797 + }, + { + "epoch": 2.56230269266481, + "grad_norm": 1.6509708166122437, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8840851783752441, + "num_tokens": 484433957.0, + "step": 13798 + }, + { + "epoch": 2.5624883936861655, + "grad_norm": 1.6289983987808228, + "learning_rate": 1e-06, + "loss": 0.2891, + "mean_token_accuracy": 0.8972710371017456, + "num_tokens": 484469371.0, + "step": 13799 + }, + { + "epoch": 2.562674094707521, + "grad_norm": 1.5809634923934937, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8926764726638794, + "num_tokens": 484502482.0, + "step": 13800 + }, + { + "epoch": 2.5628597957288766, + "grad_norm": 1.5160489082336426, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8846259117126465, + "num_tokens": 484541283.0, + "step": 13801 + }, + { + "epoch": 2.563045496750232, + "grad_norm": 1.7318556308746338, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.883009135723114, + "num_tokens": 484571679.0, + "step": 13802 + }, + { + "epoch": 2.5632311977715876, + "grad_norm": 1.419898509979248, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.894626259803772, + "num_tokens": 484608206.0, + "step": 13803 + }, + { + "epoch": 2.5634168987929433, + "grad_norm": 1.6932824850082397, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8877968788146973, + "num_tokens": 484641216.0, + "step": 13804 + }, + { + "epoch": 2.563602599814299, + "grad_norm": 1.5651835203170776, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8981059193611145, + "num_tokens": 484673583.0, + "step": 13805 + }, + { + "epoch": 2.563788300835655, + "grad_norm": 1.6471012830734253, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8967190384864807, + "num_tokens": 484705582.0, + "step": 13806 + }, + { + "epoch": 2.56397400185701, + "grad_norm": 1.429133653640747, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.8988027572631836, + "num_tokens": 484743956.0, + "step": 13807 + }, + { + "epoch": 2.564159702878366, + "grad_norm": 1.4840283393859863, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.887947142124176, + "num_tokens": 484779451.0, + "step": 13808 + }, + { + "epoch": 2.5643454038997215, + "grad_norm": 1.8647408485412598, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8951073884963989, + "num_tokens": 484813597.0, + "step": 13809 + }, + { + "epoch": 2.564531104921077, + "grad_norm": 1.6673698425292969, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8813336491584778, + "num_tokens": 484847863.0, + "step": 13810 + }, + { + "epoch": 2.5647168059424326, + "grad_norm": 1.543053388595581, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8836686611175537, + "num_tokens": 484888388.0, + "step": 13811 + }, + { + "epoch": 2.5649025069637883, + "grad_norm": 1.5924545526504517, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8854524493217468, + "num_tokens": 484921564.0, + "step": 13812 + }, + { + "epoch": 2.565088207985144, + "grad_norm": 1.63267183303833, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8792514801025391, + "num_tokens": 484961232.0, + "step": 13813 + }, + { + "epoch": 2.5652739090064998, + "grad_norm": 1.610201120376587, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8742930293083191, + "num_tokens": 484998147.0, + "step": 13814 + }, + { + "epoch": 2.565459610027855, + "grad_norm": 1.678087830543518, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.880220890045166, + "num_tokens": 485029916.0, + "step": 13815 + }, + { + "epoch": 2.565645311049211, + "grad_norm": 1.6378862857818604, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8836053609848022, + "num_tokens": 485064166.0, + "step": 13816 + }, + { + "epoch": 2.5658310120705665, + "grad_norm": 1.5529582500457764, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.88871169090271, + "num_tokens": 485098093.0, + "step": 13817 + }, + { + "epoch": 2.566016713091922, + "grad_norm": 1.5033314228057861, + "learning_rate": 1e-06, + "loss": 0.3181, + "mean_token_accuracy": 0.8875232338905334, + "num_tokens": 485135044.0, + "step": 13818 + }, + { + "epoch": 2.5662024141132775, + "grad_norm": 1.6049041748046875, + "learning_rate": 1e-06, + "loss": 0.2844, + "mean_token_accuracy": 0.8979505300521851, + "num_tokens": 485165105.0, + "step": 13819 + }, + { + "epoch": 2.5663881151346333, + "grad_norm": 1.5205446481704712, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8923458456993103, + "num_tokens": 485198987.0, + "step": 13820 + }, + { + "epoch": 2.566573816155989, + "grad_norm": 1.3957821130752563, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8806252479553223, + "num_tokens": 485240644.0, + "step": 13821 + }, + { + "epoch": 2.5667595171773447, + "grad_norm": 1.6097489595413208, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8786591291427612, + "num_tokens": 485274400.0, + "step": 13822 + }, + { + "epoch": 2.5669452181987, + "grad_norm": 1.508488416671753, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8860152959823608, + "num_tokens": 485311058.0, + "step": 13823 + }, + { + "epoch": 2.5671309192200558, + "grad_norm": 1.7196699380874634, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8854935169219971, + "num_tokens": 485342703.0, + "step": 13824 + }, + { + "epoch": 2.567316620241411, + "grad_norm": 1.6467713117599487, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8891035318374634, + "num_tokens": 485373032.0, + "step": 13825 + }, + { + "epoch": 2.567502321262767, + "grad_norm": 1.6440750360488892, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8910228610038757, + "num_tokens": 485401755.0, + "step": 13826 + }, + { + "epoch": 2.5676880222841225, + "grad_norm": 1.6896212100982666, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8801698684692383, + "num_tokens": 485434998.0, + "step": 13827 + }, + { + "epoch": 2.5678737233054783, + "grad_norm": 1.5775916576385498, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8810526132583618, + "num_tokens": 485468728.0, + "step": 13828 + }, + { + "epoch": 2.568059424326834, + "grad_norm": 1.7098382711410522, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.8869093656539917, + "num_tokens": 485498391.0, + "step": 13829 + }, + { + "epoch": 2.5682451253481893, + "grad_norm": 1.5744853019714355, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8836458921432495, + "num_tokens": 485533461.0, + "step": 13830 + }, + { + "epoch": 2.568430826369545, + "grad_norm": 1.7232658863067627, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8910530805587769, + "num_tokens": 485564113.0, + "step": 13831 + }, + { + "epoch": 2.5686165273909007, + "grad_norm": 1.6733726263046265, + "learning_rate": 1e-06, + "loss": 0.3804, + "mean_token_accuracy": 0.8710998296737671, + "num_tokens": 485597860.0, + "step": 13832 + }, + { + "epoch": 2.568802228412256, + "grad_norm": 1.6410270929336548, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8820937275886536, + "num_tokens": 485628756.0, + "step": 13833 + }, + { + "epoch": 2.5689879294336118, + "grad_norm": 1.6800743341445923, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8747028708457947, + "num_tokens": 485661377.0, + "step": 13834 + }, + { + "epoch": 2.5691736304549675, + "grad_norm": 1.6084603071212769, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8821414709091187, + "num_tokens": 485692809.0, + "step": 13835 + }, + { + "epoch": 2.5693593314763232, + "grad_norm": 1.5286498069763184, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8874685168266296, + "num_tokens": 485729289.0, + "step": 13836 + }, + { + "epoch": 2.569545032497679, + "grad_norm": 1.6850721836090088, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8823347091674805, + "num_tokens": 485762948.0, + "step": 13837 + }, + { + "epoch": 2.5697307335190342, + "grad_norm": 1.6257964372634888, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8768556714057922, + "num_tokens": 485795493.0, + "step": 13838 + }, + { + "epoch": 2.56991643454039, + "grad_norm": 1.5205798149108887, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8859042525291443, + "num_tokens": 485829272.0, + "step": 13839 + }, + { + "epoch": 2.5701021355617457, + "grad_norm": 1.7249137163162231, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8763497471809387, + "num_tokens": 485858486.0, + "step": 13840 + }, + { + "epoch": 2.570287836583101, + "grad_norm": 1.5804073810577393, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8916901350021362, + "num_tokens": 485895356.0, + "step": 13841 + }, + { + "epoch": 2.5704735376044567, + "grad_norm": 1.4219788312911987, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8846494555473328, + "num_tokens": 485936541.0, + "step": 13842 + }, + { + "epoch": 2.5706592386258125, + "grad_norm": 1.6088021993637085, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8904667496681213, + "num_tokens": 485971011.0, + "step": 13843 + }, + { + "epoch": 2.570844939647168, + "grad_norm": 1.395998239517212, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8958179950714111, + "num_tokens": 486009455.0, + "step": 13844 + }, + { + "epoch": 2.571030640668524, + "grad_norm": 1.6931356191635132, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8853867650032043, + "num_tokens": 486040350.0, + "step": 13845 + }, + { + "epoch": 2.5712163416898792, + "grad_norm": 1.7550277709960938, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8763072490692139, + "num_tokens": 486070653.0, + "step": 13846 + }, + { + "epoch": 2.571402042711235, + "grad_norm": 1.6615209579467773, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8866962194442749, + "num_tokens": 486101531.0, + "step": 13847 + }, + { + "epoch": 2.5715877437325907, + "grad_norm": 1.6098294258117676, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8724304437637329, + "num_tokens": 486138178.0, + "step": 13848 + }, + { + "epoch": 2.571773444753946, + "grad_norm": 1.6159801483154297, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.89415442943573, + "num_tokens": 486172822.0, + "step": 13849 + }, + { + "epoch": 2.5719591457753017, + "grad_norm": 1.5370479822158813, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8747479319572449, + "num_tokens": 486211417.0, + "step": 13850 + }, + { + "epoch": 2.5721448467966574, + "grad_norm": 1.6751861572265625, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8702028393745422, + "num_tokens": 486244048.0, + "step": 13851 + }, + { + "epoch": 2.572330547818013, + "grad_norm": 1.8135097026824951, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8773837089538574, + "num_tokens": 486271844.0, + "step": 13852 + }, + { + "epoch": 2.5725162488393685, + "grad_norm": 1.6469193696975708, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8707976341247559, + "num_tokens": 486306521.0, + "step": 13853 + }, + { + "epoch": 2.572701949860724, + "grad_norm": 1.4728388786315918, + "learning_rate": 1e-06, + "loss": 0.2936, + "mean_token_accuracy": 0.893362283706665, + "num_tokens": 486344288.0, + "step": 13854 + }, + { + "epoch": 2.57288765088208, + "grad_norm": 1.5579007863998413, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.8866982460021973, + "num_tokens": 486380020.0, + "step": 13855 + }, + { + "epoch": 2.5730733519034352, + "grad_norm": 1.47300124168396, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8871293067932129, + "num_tokens": 486422289.0, + "step": 13856 + }, + { + "epoch": 2.573259052924791, + "grad_norm": 1.5235947370529175, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.873140275478363, + "num_tokens": 486461919.0, + "step": 13857 + }, + { + "epoch": 2.5734447539461467, + "grad_norm": 1.6451795101165771, + "learning_rate": 1e-06, + "loss": 0.4147, + "mean_token_accuracy": 0.8620346784591675, + "num_tokens": 486499347.0, + "step": 13858 + }, + { + "epoch": 2.5736304549675024, + "grad_norm": 1.4879614114761353, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8954296112060547, + "num_tokens": 486537207.0, + "step": 13859 + }, + { + "epoch": 2.573816155988858, + "grad_norm": 1.7110055685043335, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8948521614074707, + "num_tokens": 486564760.0, + "step": 13860 + }, + { + "epoch": 2.5740018570102134, + "grad_norm": 1.5615415573120117, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8832319974899292, + "num_tokens": 486599456.0, + "step": 13861 + }, + { + "epoch": 2.574187558031569, + "grad_norm": 1.7106120586395264, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8832521438598633, + "num_tokens": 486630234.0, + "step": 13862 + }, + { + "epoch": 2.574373259052925, + "grad_norm": 1.441987156867981, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8985947370529175, + "num_tokens": 486664724.0, + "step": 13863 + }, + { + "epoch": 2.57455896007428, + "grad_norm": 1.5271199941635132, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8895711898803711, + "num_tokens": 486699565.0, + "step": 13864 + }, + { + "epoch": 2.574744661095636, + "grad_norm": 1.5511751174926758, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8897119760513306, + "num_tokens": 486733586.0, + "step": 13865 + }, + { + "epoch": 2.5749303621169917, + "grad_norm": 1.52833092212677, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8835986852645874, + "num_tokens": 486773482.0, + "step": 13866 + }, + { + "epoch": 2.5751160631383474, + "grad_norm": 1.561923623085022, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.873155951499939, + "num_tokens": 486810654.0, + "step": 13867 + }, + { + "epoch": 2.575301764159703, + "grad_norm": 1.426066279411316, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8940064907073975, + "num_tokens": 486847623.0, + "step": 13868 + }, + { + "epoch": 2.5754874651810584, + "grad_norm": 1.7582429647445679, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8799061179161072, + "num_tokens": 486886159.0, + "step": 13869 + }, + { + "epoch": 2.575673166202414, + "grad_norm": 1.3680442571640015, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.8973528742790222, + "num_tokens": 486928490.0, + "step": 13870 + }, + { + "epoch": 2.57585886722377, + "grad_norm": 1.6289139986038208, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8874858617782593, + "num_tokens": 486961463.0, + "step": 13871 + }, + { + "epoch": 2.576044568245125, + "grad_norm": 1.5816892385482788, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8830697536468506, + "num_tokens": 486994401.0, + "step": 13872 + }, + { + "epoch": 2.576230269266481, + "grad_norm": 1.603977084159851, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.894729733467102, + "num_tokens": 487024341.0, + "step": 13873 + }, + { + "epoch": 2.5764159702878366, + "grad_norm": 1.541456937789917, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8741356134414673, + "num_tokens": 487063261.0, + "step": 13874 + }, + { + "epoch": 2.5766016713091924, + "grad_norm": 1.5103274583816528, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8883033394813538, + "num_tokens": 487097893.0, + "step": 13875 + }, + { + "epoch": 2.5767873723305477, + "grad_norm": 1.63339102268219, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8887619972229004, + "num_tokens": 487127496.0, + "step": 13876 + }, + { + "epoch": 2.5769730733519034, + "grad_norm": 1.5931507349014282, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8700690269470215, + "num_tokens": 487160291.0, + "step": 13877 + }, + { + "epoch": 2.577158774373259, + "grad_norm": 1.5771056413650513, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8806530237197876, + "num_tokens": 487199566.0, + "step": 13878 + }, + { + "epoch": 2.5773444753946144, + "grad_norm": 1.5568314790725708, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8906885385513306, + "num_tokens": 487234625.0, + "step": 13879 + }, + { + "epoch": 2.57753017641597, + "grad_norm": 1.5047051906585693, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8881233334541321, + "num_tokens": 487270348.0, + "step": 13880 + }, + { + "epoch": 2.577715877437326, + "grad_norm": 1.5919864177703857, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8804855346679688, + "num_tokens": 487301637.0, + "step": 13881 + }, + { + "epoch": 2.5779015784586816, + "grad_norm": 1.5192937850952148, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8888986110687256, + "num_tokens": 487335860.0, + "step": 13882 + }, + { + "epoch": 2.5780872794800374, + "grad_norm": 1.772697925567627, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8945289850234985, + "num_tokens": 487365885.0, + "step": 13883 + }, + { + "epoch": 2.5782729805013926, + "grad_norm": 1.6064468622207642, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8793622851371765, + "num_tokens": 487403875.0, + "step": 13884 + }, + { + "epoch": 2.5784586815227484, + "grad_norm": 1.6636936664581299, + "learning_rate": 1e-06, + "loss": 0.2976, + "mean_token_accuracy": 0.8922815322875977, + "num_tokens": 487433666.0, + "step": 13885 + }, + { + "epoch": 2.578644382544104, + "grad_norm": 1.5756168365478516, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.8946235775947571, + "num_tokens": 487468595.0, + "step": 13886 + }, + { + "epoch": 2.5788300835654594, + "grad_norm": 1.5207335948944092, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8851091861724854, + "num_tokens": 487507296.0, + "step": 13887 + }, + { + "epoch": 2.579015784586815, + "grad_norm": 1.6769471168518066, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.8957200646400452, + "num_tokens": 487536535.0, + "step": 13888 + }, + { + "epoch": 2.579201485608171, + "grad_norm": 1.531866431236267, + "learning_rate": 1e-06, + "loss": 0.2747, + "mean_token_accuracy": 0.901117205619812, + "num_tokens": 487567715.0, + "step": 13889 + }, + { + "epoch": 2.5793871866295266, + "grad_norm": 1.600978136062622, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8871222138404846, + "num_tokens": 487603609.0, + "step": 13890 + }, + { + "epoch": 2.5795728876508823, + "grad_norm": 1.5464773178100586, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.888342022895813, + "num_tokens": 487638826.0, + "step": 13891 + }, + { + "epoch": 2.5797585886722376, + "grad_norm": 1.5279152393341064, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8916224241256714, + "num_tokens": 487674631.0, + "step": 13892 + }, + { + "epoch": 2.5799442896935934, + "grad_norm": 1.6651759147644043, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.9015892148017883, + "num_tokens": 487706821.0, + "step": 13893 + }, + { + "epoch": 2.580129990714949, + "grad_norm": 1.5489779710769653, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8834920525550842, + "num_tokens": 487745284.0, + "step": 13894 + }, + { + "epoch": 2.5803156917363044, + "grad_norm": 1.5628312826156616, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8851947784423828, + "num_tokens": 487779265.0, + "step": 13895 + }, + { + "epoch": 2.58050139275766, + "grad_norm": 1.4220424890518188, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8968925476074219, + "num_tokens": 487821129.0, + "step": 13896 + }, + { + "epoch": 2.580687093779016, + "grad_norm": 1.6846569776535034, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.887389063835144, + "num_tokens": 487857965.0, + "step": 13897 + }, + { + "epoch": 2.5808727948003716, + "grad_norm": 1.5805710554122925, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8902322053909302, + "num_tokens": 487893456.0, + "step": 13898 + }, + { + "epoch": 2.581058495821727, + "grad_norm": 1.53623366355896, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8820095062255859, + "num_tokens": 487935166.0, + "step": 13899 + }, + { + "epoch": 2.5812441968430826, + "grad_norm": 1.636889100074768, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8869304656982422, + "num_tokens": 487967276.0, + "step": 13900 + }, + { + "epoch": 2.5814298978644383, + "grad_norm": 1.446370244026184, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8926453590393066, + "num_tokens": 488004484.0, + "step": 13901 + }, + { + "epoch": 2.5816155988857936, + "grad_norm": 1.5510289669036865, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8878166675567627, + "num_tokens": 488043910.0, + "step": 13902 + }, + { + "epoch": 2.5818012999071493, + "grad_norm": 1.5023517608642578, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8888025879859924, + "num_tokens": 488081897.0, + "step": 13903 + }, + { + "epoch": 2.581987000928505, + "grad_norm": 1.706889033317566, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.8666175007820129, + "num_tokens": 488115110.0, + "step": 13904 + }, + { + "epoch": 2.582172701949861, + "grad_norm": 1.790053129196167, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8877782821655273, + "num_tokens": 488149600.0, + "step": 13905 + }, + { + "epoch": 2.5823584029712165, + "grad_norm": 1.9165678024291992, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8767831325531006, + "num_tokens": 488178231.0, + "step": 13906 + }, + { + "epoch": 2.582544103992572, + "grad_norm": 1.609287142753601, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8950037956237793, + "num_tokens": 488210240.0, + "step": 13907 + }, + { + "epoch": 2.5827298050139276, + "grad_norm": 1.5013182163238525, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8873536586761475, + "num_tokens": 488245253.0, + "step": 13908 + }, + { + "epoch": 2.5829155060352833, + "grad_norm": 1.7504067420959473, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.881940484046936, + "num_tokens": 488278044.0, + "step": 13909 + }, + { + "epoch": 2.5831012070566386, + "grad_norm": 1.6003868579864502, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8929932713508606, + "num_tokens": 488309991.0, + "step": 13910 + }, + { + "epoch": 2.5832869080779943, + "grad_norm": 1.6016380786895752, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.875314474105835, + "num_tokens": 488344929.0, + "step": 13911 + }, + { + "epoch": 2.58347260909935, + "grad_norm": 1.6752959489822388, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8964380025863647, + "num_tokens": 488375584.0, + "step": 13912 + }, + { + "epoch": 2.583658310120706, + "grad_norm": 1.7565040588378906, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8832278847694397, + "num_tokens": 488406053.0, + "step": 13913 + }, + { + "epoch": 2.5838440111420615, + "grad_norm": 1.6167188882827759, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.871748685836792, + "num_tokens": 488439104.0, + "step": 13914 + }, + { + "epoch": 2.584029712163417, + "grad_norm": 1.7565544843673706, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.895927369594574, + "num_tokens": 488463099.0, + "step": 13915 + }, + { + "epoch": 2.5842154131847725, + "grad_norm": 1.6003559827804565, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8859409689903259, + "num_tokens": 488499091.0, + "step": 13916 + }, + { + "epoch": 2.5844011142061283, + "grad_norm": 1.7550848722457886, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8763997554779053, + "num_tokens": 488528381.0, + "step": 13917 + }, + { + "epoch": 2.5845868152274836, + "grad_norm": 1.6722712516784668, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8821196556091309, + "num_tokens": 488561031.0, + "step": 13918 + }, + { + "epoch": 2.5847725162488393, + "grad_norm": 1.4446988105773926, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8914130330085754, + "num_tokens": 488598267.0, + "step": 13919 + }, + { + "epoch": 2.584958217270195, + "grad_norm": 1.5618393421173096, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8853970766067505, + "num_tokens": 488629344.0, + "step": 13920 + }, + { + "epoch": 2.5851439182915508, + "grad_norm": 1.7277573347091675, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8749526739120483, + "num_tokens": 488659925.0, + "step": 13921 + }, + { + "epoch": 2.585329619312906, + "grad_norm": 1.4364482164382935, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8917573690414429, + "num_tokens": 488700567.0, + "step": 13922 + }, + { + "epoch": 2.585515320334262, + "grad_norm": 1.4994412660598755, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8934984803199768, + "num_tokens": 488738686.0, + "step": 13923 + }, + { + "epoch": 2.5857010213556175, + "grad_norm": 1.464581847190857, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8799521923065186, + "num_tokens": 488780058.0, + "step": 13924 + }, + { + "epoch": 2.585886722376973, + "grad_norm": 1.567030429840088, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8931870460510254, + "num_tokens": 488811772.0, + "step": 13925 + }, + { + "epoch": 2.5860724233983285, + "grad_norm": 1.5995479822158813, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.892266035079956, + "num_tokens": 488846872.0, + "step": 13926 + }, + { + "epoch": 2.5862581244196843, + "grad_norm": 1.485709309577942, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.8890093564987183, + "num_tokens": 488885770.0, + "step": 13927 + }, + { + "epoch": 2.58644382544104, + "grad_norm": 1.5975242853164673, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8904048204421997, + "num_tokens": 488916948.0, + "step": 13928 + }, + { + "epoch": 2.5866295264623957, + "grad_norm": 1.50151526927948, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8935703039169312, + "num_tokens": 488953355.0, + "step": 13929 + }, + { + "epoch": 2.586815227483751, + "grad_norm": 1.5454174280166626, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8952916264533997, + "num_tokens": 488986145.0, + "step": 13930 + }, + { + "epoch": 2.5870009285051068, + "grad_norm": 1.558922529220581, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8893972039222717, + "num_tokens": 489019433.0, + "step": 13931 + }, + { + "epoch": 2.5871866295264625, + "grad_norm": 1.4968485832214355, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.891109824180603, + "num_tokens": 489058621.0, + "step": 13932 + }, + { + "epoch": 2.587372330547818, + "grad_norm": 1.4390003681182861, + "learning_rate": 1e-06, + "loss": 0.2826, + "mean_token_accuracy": 0.8977483510971069, + "num_tokens": 489094867.0, + "step": 13933 + }, + { + "epoch": 2.5875580315691735, + "grad_norm": 1.6741971969604492, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.884979248046875, + "num_tokens": 489126592.0, + "step": 13934 + }, + { + "epoch": 2.5877437325905293, + "grad_norm": 1.5554745197296143, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8815335035324097, + "num_tokens": 489162066.0, + "step": 13935 + }, + { + "epoch": 2.587929433611885, + "grad_norm": 1.6122504472732544, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8793078660964966, + "num_tokens": 489197589.0, + "step": 13936 + }, + { + "epoch": 2.5881151346332407, + "grad_norm": 1.588687539100647, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8903231024742126, + "num_tokens": 489231164.0, + "step": 13937 + }, + { + "epoch": 2.588300835654596, + "grad_norm": 1.5625277757644653, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8850811719894409, + "num_tokens": 489266681.0, + "step": 13938 + }, + { + "epoch": 2.5884865366759517, + "grad_norm": 1.6560711860656738, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8773144483566284, + "num_tokens": 489304461.0, + "step": 13939 + }, + { + "epoch": 2.5886722376973075, + "grad_norm": 1.5989789962768555, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.882598876953125, + "num_tokens": 489335361.0, + "step": 13940 + }, + { + "epoch": 2.5888579387186628, + "grad_norm": 1.5922771692276, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8865026235580444, + "num_tokens": 489370620.0, + "step": 13941 + }, + { + "epoch": 2.5890436397400185, + "grad_norm": 1.5635401010513306, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8957661390304565, + "num_tokens": 489404432.0, + "step": 13942 + }, + { + "epoch": 2.5892293407613742, + "grad_norm": 1.56290602684021, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.888421893119812, + "num_tokens": 489443183.0, + "step": 13943 + }, + { + "epoch": 2.58941504178273, + "grad_norm": 1.493540644645691, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.8934137225151062, + "num_tokens": 489479515.0, + "step": 13944 + }, + { + "epoch": 2.5896007428040857, + "grad_norm": 1.440210223197937, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.888508141040802, + "num_tokens": 489517848.0, + "step": 13945 + }, + { + "epoch": 2.589786443825441, + "grad_norm": 1.6785082817077637, + "learning_rate": 1e-06, + "loss": 0.3817, + "mean_token_accuracy": 0.8686544895172119, + "num_tokens": 489551981.0, + "step": 13946 + }, + { + "epoch": 2.5899721448467967, + "grad_norm": 1.600219488143921, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.8990979790687561, + "num_tokens": 489581992.0, + "step": 13947 + }, + { + "epoch": 2.590157845868152, + "grad_norm": 1.4960167407989502, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.882623553276062, + "num_tokens": 489618314.0, + "step": 13948 + }, + { + "epoch": 2.5903435468895077, + "grad_norm": 1.5351771116256714, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8764073252677917, + "num_tokens": 489656108.0, + "step": 13949 + }, + { + "epoch": 2.5905292479108635, + "grad_norm": 1.6532701253890991, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8679386377334595, + "num_tokens": 489691202.0, + "step": 13950 + }, + { + "epoch": 2.590714948932219, + "grad_norm": 1.4778122901916504, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8879470825195312, + "num_tokens": 489728776.0, + "step": 13951 + }, + { + "epoch": 2.590900649953575, + "grad_norm": 1.7525368928909302, + "learning_rate": 1e-06, + "loss": 0.3625, + "mean_token_accuracy": 0.8762910962104797, + "num_tokens": 489759384.0, + "step": 13952 + }, + { + "epoch": 2.5910863509749302, + "grad_norm": 1.4280188083648682, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8948109149932861, + "num_tokens": 489799135.0, + "step": 13953 + }, + { + "epoch": 2.591272051996286, + "grad_norm": 1.6448554992675781, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8903017044067383, + "num_tokens": 489831587.0, + "step": 13954 + }, + { + "epoch": 2.5914577530176417, + "grad_norm": 1.4824919700622559, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8885374665260315, + "num_tokens": 489868369.0, + "step": 13955 + }, + { + "epoch": 2.591643454038997, + "grad_norm": 1.6335197687149048, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8918944597244263, + "num_tokens": 489899914.0, + "step": 13956 + }, + { + "epoch": 2.5918291550603527, + "grad_norm": 1.5580792427062988, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8954480290412903, + "num_tokens": 489932581.0, + "step": 13957 + }, + { + "epoch": 2.5920148560817085, + "grad_norm": 1.535593867301941, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8846793174743652, + "num_tokens": 489967415.0, + "step": 13958 + }, + { + "epoch": 2.592200557103064, + "grad_norm": 1.600852370262146, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.8769996166229248, + "num_tokens": 490002248.0, + "step": 13959 + }, + { + "epoch": 2.59238625812442, + "grad_norm": 1.5528671741485596, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8908580541610718, + "num_tokens": 490034146.0, + "step": 13960 + }, + { + "epoch": 2.592571959145775, + "grad_norm": 1.5579454898834229, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8843089938163757, + "num_tokens": 490070984.0, + "step": 13961 + }, + { + "epoch": 2.592757660167131, + "grad_norm": 1.6129286289215088, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8806809186935425, + "num_tokens": 490102580.0, + "step": 13962 + }, + { + "epoch": 2.5929433611884867, + "grad_norm": 1.6207822561264038, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8814858794212341, + "num_tokens": 490137596.0, + "step": 13963 + }, + { + "epoch": 2.593129062209842, + "grad_norm": 1.4392017126083374, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8761126399040222, + "num_tokens": 490182201.0, + "step": 13964 + }, + { + "epoch": 2.5933147632311977, + "grad_norm": 1.4681390523910522, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8934890627861023, + "num_tokens": 490218066.0, + "step": 13965 + }, + { + "epoch": 2.5935004642525534, + "grad_norm": 1.5819050073623657, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8817707896232605, + "num_tokens": 490253112.0, + "step": 13966 + }, + { + "epoch": 2.593686165273909, + "grad_norm": 1.4527472257614136, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8776020407676697, + "num_tokens": 490293242.0, + "step": 13967 + }, + { + "epoch": 2.593871866295265, + "grad_norm": 1.7147740125656128, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8931229710578918, + "num_tokens": 490320171.0, + "step": 13968 + }, + { + "epoch": 2.59405756731662, + "grad_norm": 1.6414281129837036, + "learning_rate": 1e-06, + "loss": 0.2751, + "mean_token_accuracy": 0.8988516926765442, + "num_tokens": 490351243.0, + "step": 13969 + }, + { + "epoch": 2.594243268337976, + "grad_norm": 1.545097827911377, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8876262903213501, + "num_tokens": 490387518.0, + "step": 13970 + }, + { + "epoch": 2.594428969359331, + "grad_norm": 1.6817182302474976, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8843614459037781, + "num_tokens": 490419197.0, + "step": 13971 + }, + { + "epoch": 2.594614670380687, + "grad_norm": 1.5116115808486938, + "learning_rate": 1e-06, + "loss": 0.2828, + "mean_token_accuracy": 0.901591420173645, + "num_tokens": 490457975.0, + "step": 13972 + }, + { + "epoch": 2.5948003714020427, + "grad_norm": 1.5078197717666626, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8860214352607727, + "num_tokens": 490496932.0, + "step": 13973 + }, + { + "epoch": 2.5949860724233984, + "grad_norm": 1.6467413902282715, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8828529119491577, + "num_tokens": 490528930.0, + "step": 13974 + }, + { + "epoch": 2.595171773444754, + "grad_norm": 1.5411498546600342, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.8958934545516968, + "num_tokens": 490560718.0, + "step": 13975 + }, + { + "epoch": 2.5953574744661094, + "grad_norm": 1.7197331190109253, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8861171007156372, + "num_tokens": 490589574.0, + "step": 13976 + }, + { + "epoch": 2.595543175487465, + "grad_norm": 1.5573978424072266, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8832487463951111, + "num_tokens": 490628904.0, + "step": 13977 + }, + { + "epoch": 2.595728876508821, + "grad_norm": 1.6014941930770874, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8932377696037292, + "num_tokens": 490659979.0, + "step": 13978 + }, + { + "epoch": 2.595914577530176, + "grad_norm": 1.6451514959335327, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8756966590881348, + "num_tokens": 490695639.0, + "step": 13979 + }, + { + "epoch": 2.596100278551532, + "grad_norm": 1.726286768913269, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.882944643497467, + "num_tokens": 490723889.0, + "step": 13980 + }, + { + "epoch": 2.5962859795728876, + "grad_norm": 1.575499176979065, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8864301443099976, + "num_tokens": 490762406.0, + "step": 13981 + }, + { + "epoch": 2.5964716805942434, + "grad_norm": 1.596358299255371, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.881217360496521, + "num_tokens": 490795887.0, + "step": 13982 + }, + { + "epoch": 2.596657381615599, + "grad_norm": 1.467691421508789, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8881500363349915, + "num_tokens": 490836038.0, + "step": 13983 + }, + { + "epoch": 2.5968430826369544, + "grad_norm": 1.5776636600494385, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8890788555145264, + "num_tokens": 490872628.0, + "step": 13984 + }, + { + "epoch": 2.59702878365831, + "grad_norm": 1.481590747833252, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.882986843585968, + "num_tokens": 490910298.0, + "step": 13985 + }, + { + "epoch": 2.597214484679666, + "grad_norm": 1.6044565439224243, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.886422872543335, + "num_tokens": 490942152.0, + "step": 13986 + }, + { + "epoch": 2.597400185701021, + "grad_norm": 1.715093731880188, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8892541527748108, + "num_tokens": 490971688.0, + "step": 13987 + }, + { + "epoch": 2.597585886722377, + "grad_norm": 1.561067819595337, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8904709815979004, + "num_tokens": 491007065.0, + "step": 13988 + }, + { + "epoch": 2.5977715877437326, + "grad_norm": 1.3895728588104248, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8931115865707397, + "num_tokens": 491049263.0, + "step": 13989 + }, + { + "epoch": 2.5979572887650884, + "grad_norm": 1.5653431415557861, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8795387744903564, + "num_tokens": 491087513.0, + "step": 13990 + }, + { + "epoch": 2.598142989786444, + "grad_norm": 1.524102807044983, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8822987079620361, + "num_tokens": 491120973.0, + "step": 13991 + }, + { + "epoch": 2.5983286908077994, + "grad_norm": 1.595025658607483, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8830549716949463, + "num_tokens": 491155777.0, + "step": 13992 + }, + { + "epoch": 2.598514391829155, + "grad_norm": 1.669077754020691, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8833906650543213, + "num_tokens": 491187464.0, + "step": 13993 + }, + { + "epoch": 2.5987000928505104, + "grad_norm": 1.5723706483840942, + "learning_rate": 1e-06, + "loss": 0.3624, + "mean_token_accuracy": 0.8763604164123535, + "num_tokens": 491225309.0, + "step": 13994 + }, + { + "epoch": 2.598885793871866, + "grad_norm": 1.675659418106079, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8827347755432129, + "num_tokens": 491257918.0, + "step": 13995 + }, + { + "epoch": 2.599071494893222, + "grad_norm": 1.6343631744384766, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8889328241348267, + "num_tokens": 491289650.0, + "step": 13996 + }, + { + "epoch": 2.5992571959145776, + "grad_norm": 1.4650691747665405, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8807196617126465, + "num_tokens": 491330172.0, + "step": 13997 + }, + { + "epoch": 2.5994428969359333, + "grad_norm": 1.624464511871338, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8900559544563293, + "num_tokens": 491365000.0, + "step": 13998 + }, + { + "epoch": 2.5996285979572886, + "grad_norm": 1.6025205850601196, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8728512525558472, + "num_tokens": 491401307.0, + "step": 13999 + }, + { + "epoch": 2.5998142989786444, + "grad_norm": 1.4552594423294067, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.880097508430481, + "num_tokens": 491444438.0, + "step": 14000 + }, + { + "epoch": 2.6, + "grad_norm": 1.6189717054367065, + "learning_rate": 1e-06, + "loss": 0.3561, + "mean_token_accuracy": 0.8749757409095764, + "num_tokens": 491480841.0, + "step": 14001 + }, + { + "epoch": 2.6001857010213554, + "grad_norm": 1.5251383781433105, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8961752653121948, + "num_tokens": 491513754.0, + "step": 14002 + }, + { + "epoch": 2.600371402042711, + "grad_norm": 1.681391716003418, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8840553164482117, + "num_tokens": 491547896.0, + "step": 14003 + }, + { + "epoch": 2.600557103064067, + "grad_norm": 1.61723792552948, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8905898332595825, + "num_tokens": 491579376.0, + "step": 14004 + }, + { + "epoch": 2.6007428040854226, + "grad_norm": 1.4800947904586792, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8848847150802612, + "num_tokens": 491618951.0, + "step": 14005 + }, + { + "epoch": 2.6009285051067783, + "grad_norm": 1.4901204109191895, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8861644268035889, + "num_tokens": 491656079.0, + "step": 14006 + }, + { + "epoch": 2.6011142061281336, + "grad_norm": 1.458112359046936, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8858922719955444, + "num_tokens": 491697886.0, + "step": 14007 + }, + { + "epoch": 2.6012999071494893, + "grad_norm": 1.5027337074279785, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8765301704406738, + "num_tokens": 491740034.0, + "step": 14008 + }, + { + "epoch": 2.601485608170845, + "grad_norm": 1.6676092147827148, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.8953152298927307, + "num_tokens": 491771047.0, + "step": 14009 + }, + { + "epoch": 2.6016713091922004, + "grad_norm": 1.67721688747406, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8924418687820435, + "num_tokens": 491799926.0, + "step": 14010 + }, + { + "epoch": 2.601857010213556, + "grad_norm": 1.511242151260376, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8774992823600769, + "num_tokens": 491840653.0, + "step": 14011 + }, + { + "epoch": 2.602042711234912, + "grad_norm": 1.649226188659668, + "learning_rate": 1e-06, + "loss": 0.3929, + "mean_token_accuracy": 0.8621882200241089, + "num_tokens": 491874033.0, + "step": 14012 + }, + { + "epoch": 2.6022284122562676, + "grad_norm": 1.448095679283142, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8807387351989746, + "num_tokens": 491916905.0, + "step": 14013 + }, + { + "epoch": 2.6024141132776233, + "grad_norm": 1.5889378786087036, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8920745849609375, + "num_tokens": 491949890.0, + "step": 14014 + }, + { + "epoch": 2.6025998142989786, + "grad_norm": 1.6822303533554077, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8825728297233582, + "num_tokens": 491984042.0, + "step": 14015 + }, + { + "epoch": 2.6027855153203343, + "grad_norm": 1.7573411464691162, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8761008381843567, + "num_tokens": 492015436.0, + "step": 14016 + }, + { + "epoch": 2.60297121634169, + "grad_norm": 1.4720453023910522, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8939658999443054, + "num_tokens": 492049730.0, + "step": 14017 + }, + { + "epoch": 2.6031569173630453, + "grad_norm": 1.5982136726379395, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8958513736724854, + "num_tokens": 492082870.0, + "step": 14018 + }, + { + "epoch": 2.603342618384401, + "grad_norm": 1.5147546529769897, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8786896467208862, + "num_tokens": 492121933.0, + "step": 14019 + }, + { + "epoch": 2.603528319405757, + "grad_norm": 1.6236850023269653, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8894049525260925, + "num_tokens": 492151563.0, + "step": 14020 + }, + { + "epoch": 2.6037140204271125, + "grad_norm": 1.3795934915542603, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.8964794874191284, + "num_tokens": 492188580.0, + "step": 14021 + }, + { + "epoch": 2.603899721448468, + "grad_norm": 1.5307526588439941, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8966431021690369, + "num_tokens": 492223582.0, + "step": 14022 + }, + { + "epoch": 2.6040854224698236, + "grad_norm": 1.5631431341171265, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8829032778739929, + "num_tokens": 492257829.0, + "step": 14023 + }, + { + "epoch": 2.6042711234911793, + "grad_norm": 1.6307742595672607, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8779613971710205, + "num_tokens": 492290875.0, + "step": 14024 + }, + { + "epoch": 2.6044568245125346, + "grad_norm": 1.4569751024246216, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8859872221946716, + "num_tokens": 492329539.0, + "step": 14025 + }, + { + "epoch": 2.6046425255338903, + "grad_norm": 1.4793827533721924, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8928025960922241, + "num_tokens": 492370101.0, + "step": 14026 + }, + { + "epoch": 2.604828226555246, + "grad_norm": 1.4862931966781616, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8836577534675598, + "num_tokens": 492409448.0, + "step": 14027 + }, + { + "epoch": 2.6050139275766018, + "grad_norm": 1.6500016450881958, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8952824473381042, + "num_tokens": 492438389.0, + "step": 14028 + }, + { + "epoch": 2.6051996285979575, + "grad_norm": 1.517637848854065, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.9023264646530151, + "num_tokens": 492471982.0, + "step": 14029 + }, + { + "epoch": 2.605385329619313, + "grad_norm": 1.6073086261749268, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8955270051956177, + "num_tokens": 492506662.0, + "step": 14030 + }, + { + "epoch": 2.6055710306406685, + "grad_norm": 1.804656744003296, + "learning_rate": 1e-06, + "loss": 0.3805, + "mean_token_accuracy": 0.8697349429130554, + "num_tokens": 492543245.0, + "step": 14031 + }, + { + "epoch": 2.6057567316620243, + "grad_norm": 1.628132700920105, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8783851861953735, + "num_tokens": 492579779.0, + "step": 14032 + }, + { + "epoch": 2.6059424326833796, + "grad_norm": 1.560320496559143, + "learning_rate": 1e-06, + "loss": 0.321, + "mean_token_accuracy": 0.8866572380065918, + "num_tokens": 492613234.0, + "step": 14033 + }, + { + "epoch": 2.6061281337047353, + "grad_norm": 1.6551896333694458, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8851066827774048, + "num_tokens": 492646653.0, + "step": 14034 + }, + { + "epoch": 2.606313834726091, + "grad_norm": 1.4771528244018555, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8892436027526855, + "num_tokens": 492683087.0, + "step": 14035 + }, + { + "epoch": 2.6064995357474467, + "grad_norm": 1.5062001943588257, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8755344152450562, + "num_tokens": 492722738.0, + "step": 14036 + }, + { + "epoch": 2.6066852367688025, + "grad_norm": 1.4891996383666992, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8854551911354065, + "num_tokens": 492761901.0, + "step": 14037 + }, + { + "epoch": 2.6068709377901578, + "grad_norm": 1.4376837015151978, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8780598640441895, + "num_tokens": 492803493.0, + "step": 14038 + }, + { + "epoch": 2.6070566388115135, + "grad_norm": 1.4975860118865967, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8942750692367554, + "num_tokens": 492838469.0, + "step": 14039 + }, + { + "epoch": 2.6072423398328692, + "grad_norm": 1.4662139415740967, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8982900977134705, + "num_tokens": 492875328.0, + "step": 14040 + }, + { + "epoch": 2.6074280408542245, + "grad_norm": 1.4479897022247314, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8901520371437073, + "num_tokens": 492912598.0, + "step": 14041 + }, + { + "epoch": 2.6076137418755803, + "grad_norm": 1.4928202629089355, + "learning_rate": 1e-06, + "loss": 0.2723, + "mean_token_accuracy": 0.9035486578941345, + "num_tokens": 492946878.0, + "step": 14042 + }, + { + "epoch": 2.607799442896936, + "grad_norm": 1.4983110427856445, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8784660696983337, + "num_tokens": 492985399.0, + "step": 14043 + }, + { + "epoch": 2.6079851439182917, + "grad_norm": 1.5474716424942017, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8863446116447449, + "num_tokens": 493022096.0, + "step": 14044 + }, + { + "epoch": 2.608170844939647, + "grad_norm": 1.589594841003418, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8883993625640869, + "num_tokens": 493055003.0, + "step": 14045 + }, + { + "epoch": 2.6083565459610027, + "grad_norm": 1.540723204612732, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.884924054145813, + "num_tokens": 493090217.0, + "step": 14046 + }, + { + "epoch": 2.6085422469823585, + "grad_norm": 1.6489425897598267, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8897147178649902, + "num_tokens": 493118384.0, + "step": 14047 + }, + { + "epoch": 2.6087279480037138, + "grad_norm": 1.5922306776046753, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8748075366020203, + "num_tokens": 493156900.0, + "step": 14048 + }, + { + "epoch": 2.6089136490250695, + "grad_norm": 1.6398502588272095, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8923368453979492, + "num_tokens": 493187707.0, + "step": 14049 + }, + { + "epoch": 2.6090993500464252, + "grad_norm": 1.7655532360076904, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.886027991771698, + "num_tokens": 493216210.0, + "step": 14050 + }, + { + "epoch": 2.609285051067781, + "grad_norm": 1.5443105697631836, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8926008939743042, + "num_tokens": 493249740.0, + "step": 14051 + }, + { + "epoch": 2.6094707520891367, + "grad_norm": 1.4740259647369385, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8940292596817017, + "num_tokens": 493289307.0, + "step": 14052 + }, + { + "epoch": 2.609656453110492, + "grad_norm": 1.401438593864441, + "learning_rate": 1e-06, + "loss": 0.2714, + "mean_token_accuracy": 0.9037590622901917, + "num_tokens": 493325462.0, + "step": 14053 + }, + { + "epoch": 2.6098421541318477, + "grad_norm": 1.6714123487472534, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8782080411911011, + "num_tokens": 493356936.0, + "step": 14054 + }, + { + "epoch": 2.6100278551532035, + "grad_norm": 1.5362460613250732, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.888294517993927, + "num_tokens": 493392481.0, + "step": 14055 + }, + { + "epoch": 2.6102135561745587, + "grad_norm": 1.6244169473648071, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8861786127090454, + "num_tokens": 493424909.0, + "step": 14056 + }, + { + "epoch": 2.6103992571959145, + "grad_norm": 1.515932321548462, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8944545984268188, + "num_tokens": 493458828.0, + "step": 14057 + }, + { + "epoch": 2.61058495821727, + "grad_norm": 1.5530225038528442, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.881981611251831, + "num_tokens": 493493028.0, + "step": 14058 + }, + { + "epoch": 2.610770659238626, + "grad_norm": 1.4124987125396729, + "learning_rate": 1e-06, + "loss": 0.2732, + "mean_token_accuracy": 0.9032486081123352, + "num_tokens": 493531825.0, + "step": 14059 + }, + { + "epoch": 2.6109563602599817, + "grad_norm": 1.5373482704162598, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8924872875213623, + "num_tokens": 493564743.0, + "step": 14060 + }, + { + "epoch": 2.611142061281337, + "grad_norm": 1.5490158796310425, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8782721161842346, + "num_tokens": 493602534.0, + "step": 14061 + }, + { + "epoch": 2.6113277623026927, + "grad_norm": 1.6047881841659546, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.886676549911499, + "num_tokens": 493634682.0, + "step": 14062 + }, + { + "epoch": 2.6115134633240484, + "grad_norm": 1.562940239906311, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8912720680236816, + "num_tokens": 493670401.0, + "step": 14063 + }, + { + "epoch": 2.6116991643454037, + "grad_norm": 1.6258879899978638, + "learning_rate": 1e-06, + "loss": 0.3644, + "mean_token_accuracy": 0.8751440644264221, + "num_tokens": 493705797.0, + "step": 14064 + }, + { + "epoch": 2.6118848653667595, + "grad_norm": 1.6007131338119507, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8742815852165222, + "num_tokens": 493741468.0, + "step": 14065 + }, + { + "epoch": 2.612070566388115, + "grad_norm": 1.4219427108764648, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8836477398872375, + "num_tokens": 493785696.0, + "step": 14066 + }, + { + "epoch": 2.612256267409471, + "grad_norm": 1.494279146194458, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.881199061870575, + "num_tokens": 493823684.0, + "step": 14067 + }, + { + "epoch": 2.612441968430826, + "grad_norm": 1.4408299922943115, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8813123106956482, + "num_tokens": 493865147.0, + "step": 14068 + }, + { + "epoch": 2.612627669452182, + "grad_norm": 1.6210683584213257, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8898759484291077, + "num_tokens": 493896297.0, + "step": 14069 + }, + { + "epoch": 2.6128133704735377, + "grad_norm": 1.6274527311325073, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8891363143920898, + "num_tokens": 493930791.0, + "step": 14070 + }, + { + "epoch": 2.612999071494893, + "grad_norm": 1.645372986793518, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8788122534751892, + "num_tokens": 493964667.0, + "step": 14071 + }, + { + "epoch": 2.6131847725162487, + "grad_norm": 1.486567735671997, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8914189338684082, + "num_tokens": 494002690.0, + "step": 14072 + }, + { + "epoch": 2.6133704735376044, + "grad_norm": 1.3920093774795532, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8938905000686646, + "num_tokens": 494046060.0, + "step": 14073 + }, + { + "epoch": 2.61355617455896, + "grad_norm": 1.4507544040679932, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8760602474212646, + "num_tokens": 494086564.0, + "step": 14074 + }, + { + "epoch": 2.613741875580316, + "grad_norm": 1.562420129776001, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.892219603061676, + "num_tokens": 494121802.0, + "step": 14075 + }, + { + "epoch": 2.613927576601671, + "grad_norm": 1.5252400636672974, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8843859434127808, + "num_tokens": 494164086.0, + "step": 14076 + }, + { + "epoch": 2.614113277623027, + "grad_norm": 1.5345515012741089, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.8985211253166199, + "num_tokens": 494198856.0, + "step": 14077 + }, + { + "epoch": 2.6142989786443827, + "grad_norm": 1.5856157541275024, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8840799331665039, + "num_tokens": 494233416.0, + "step": 14078 + }, + { + "epoch": 2.614484679665738, + "grad_norm": 1.6391466856002808, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.8780845403671265, + "num_tokens": 494269061.0, + "step": 14079 + }, + { + "epoch": 2.6146703806870937, + "grad_norm": 1.6134992837905884, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8864003419876099, + "num_tokens": 494298352.0, + "step": 14080 + }, + { + "epoch": 2.6148560817084494, + "grad_norm": 1.6010349988937378, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8886765241622925, + "num_tokens": 494332294.0, + "step": 14081 + }, + { + "epoch": 2.615041782729805, + "grad_norm": 1.6326996088027954, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8777506351470947, + "num_tokens": 494367476.0, + "step": 14082 + }, + { + "epoch": 2.615227483751161, + "grad_norm": 1.4375098943710327, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8969739079475403, + "num_tokens": 494405044.0, + "step": 14083 + }, + { + "epoch": 2.615413184772516, + "grad_norm": 1.8666540384292603, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8961545825004578, + "num_tokens": 494434840.0, + "step": 14084 + }, + { + "epoch": 2.615598885793872, + "grad_norm": 1.5661672353744507, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8862872123718262, + "num_tokens": 494470126.0, + "step": 14085 + }, + { + "epoch": 2.6157845868152276, + "grad_norm": 1.575682282447815, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8834555149078369, + "num_tokens": 494506650.0, + "step": 14086 + }, + { + "epoch": 2.615970287836583, + "grad_norm": 1.6062114238739014, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8915203809738159, + "num_tokens": 494539113.0, + "step": 14087 + }, + { + "epoch": 2.6161559888579387, + "grad_norm": 1.6293901205062866, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8854994773864746, + "num_tokens": 494572566.0, + "step": 14088 + }, + { + "epoch": 2.6163416898792944, + "grad_norm": 1.6229304075241089, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8831025958061218, + "num_tokens": 494608568.0, + "step": 14089 + }, + { + "epoch": 2.61652739090065, + "grad_norm": 1.609045147895813, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8835043907165527, + "num_tokens": 494643191.0, + "step": 14090 + }, + { + "epoch": 2.6167130919220054, + "grad_norm": 1.3539742231369019, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8912234902381897, + "num_tokens": 494683560.0, + "step": 14091 + }, + { + "epoch": 2.616898792943361, + "grad_norm": 1.5887168645858765, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8713381290435791, + "num_tokens": 494721329.0, + "step": 14092 + }, + { + "epoch": 2.617084493964717, + "grad_norm": 1.4859498739242554, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.888034999370575, + "num_tokens": 494759742.0, + "step": 14093 + }, + { + "epoch": 2.617270194986072, + "grad_norm": 1.6076700687408447, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8854914903640747, + "num_tokens": 494793882.0, + "step": 14094 + }, + { + "epoch": 2.617455896007428, + "grad_norm": 1.4833282232284546, + "learning_rate": 1e-06, + "loss": 0.2625, + "mean_token_accuracy": 0.9055452346801758, + "num_tokens": 494825671.0, + "step": 14095 + }, + { + "epoch": 2.6176415970287836, + "grad_norm": 1.594056248664856, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8890237808227539, + "num_tokens": 494857081.0, + "step": 14096 + }, + { + "epoch": 2.6178272980501394, + "grad_norm": 1.839080810546875, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8633937835693359, + "num_tokens": 494886124.0, + "step": 14097 + }, + { + "epoch": 2.618012999071495, + "grad_norm": 1.6751976013183594, + "learning_rate": 1e-06, + "loss": 0.3408, + "mean_token_accuracy": 0.8762987852096558, + "num_tokens": 494920665.0, + "step": 14098 + }, + { + "epoch": 2.6181987000928504, + "grad_norm": 1.4579658508300781, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8851988911628723, + "num_tokens": 494961054.0, + "step": 14099 + }, + { + "epoch": 2.618384401114206, + "grad_norm": 1.5940930843353271, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8779934048652649, + "num_tokens": 494997717.0, + "step": 14100 + }, + { + "epoch": 2.618570102135562, + "grad_norm": 1.560296654701233, + "learning_rate": 1e-06, + "loss": 0.3599, + "mean_token_accuracy": 0.8771673440933228, + "num_tokens": 495036679.0, + "step": 14101 + }, + { + "epoch": 2.618755803156917, + "grad_norm": 1.567569613456726, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8973984718322754, + "num_tokens": 495068026.0, + "step": 14102 + }, + { + "epoch": 2.618941504178273, + "grad_norm": 1.6835856437683105, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8871887922286987, + "num_tokens": 495100309.0, + "step": 14103 + }, + { + "epoch": 2.6191272051996286, + "grad_norm": 1.4610741138458252, + "learning_rate": 1e-06, + "loss": 0.2835, + "mean_token_accuracy": 0.8963460922241211, + "num_tokens": 495136063.0, + "step": 14104 + }, + { + "epoch": 2.6193129062209843, + "grad_norm": 1.4883203506469727, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8955028653144836, + "num_tokens": 495171549.0, + "step": 14105 + }, + { + "epoch": 2.61949860724234, + "grad_norm": 1.5324589014053345, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.8971532583236694, + "num_tokens": 495205490.0, + "step": 14106 + }, + { + "epoch": 2.6196843082636954, + "grad_norm": 1.9221643209457397, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.8656158447265625, + "num_tokens": 495233974.0, + "step": 14107 + }, + { + "epoch": 2.619870009285051, + "grad_norm": 1.5384495258331299, + "learning_rate": 1e-06, + "loss": 0.2781, + "mean_token_accuracy": 0.8996472358703613, + "num_tokens": 495265360.0, + "step": 14108 + }, + { + "epoch": 2.620055710306407, + "grad_norm": 1.5572211742401123, + "learning_rate": 1e-06, + "loss": 0.2798, + "mean_token_accuracy": 0.8997194766998291, + "num_tokens": 495295134.0, + "step": 14109 + }, + { + "epoch": 2.620241411327762, + "grad_norm": 1.7165954113006592, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8803366422653198, + "num_tokens": 495324379.0, + "step": 14110 + }, + { + "epoch": 2.620427112349118, + "grad_norm": 1.491650104522705, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8900902271270752, + "num_tokens": 495360816.0, + "step": 14111 + }, + { + "epoch": 2.6206128133704736, + "grad_norm": 1.5493601560592651, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8780946731567383, + "num_tokens": 495397750.0, + "step": 14112 + }, + { + "epoch": 2.6207985143918293, + "grad_norm": 1.5002493858337402, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8727885484695435, + "num_tokens": 495437356.0, + "step": 14113 + }, + { + "epoch": 2.620984215413185, + "grad_norm": 1.679782509803772, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8795562982559204, + "num_tokens": 495466589.0, + "step": 14114 + }, + { + "epoch": 2.6211699164345403, + "grad_norm": 1.422875165939331, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8890930414199829, + "num_tokens": 495503230.0, + "step": 14115 + }, + { + "epoch": 2.621355617455896, + "grad_norm": 1.3413455486297607, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8874710202217102, + "num_tokens": 495547093.0, + "step": 14116 + }, + { + "epoch": 2.6215413184772514, + "grad_norm": 1.663353681564331, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8888633847236633, + "num_tokens": 495577546.0, + "step": 14117 + }, + { + "epoch": 2.621727019498607, + "grad_norm": 1.734635829925537, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8793325424194336, + "num_tokens": 495609895.0, + "step": 14118 + }, + { + "epoch": 2.621912720519963, + "grad_norm": 1.500185251235962, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8856703639030457, + "num_tokens": 495647733.0, + "step": 14119 + }, + { + "epoch": 2.6220984215413186, + "grad_norm": 1.6162391901016235, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8790113925933838, + "num_tokens": 495681606.0, + "step": 14120 + }, + { + "epoch": 2.6222841225626743, + "grad_norm": 1.4126522541046143, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8909494876861572, + "num_tokens": 495723867.0, + "step": 14121 + }, + { + "epoch": 2.6224698235840296, + "grad_norm": 1.589868187904358, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8930103778839111, + "num_tokens": 495759932.0, + "step": 14122 + }, + { + "epoch": 2.6226555246053853, + "grad_norm": 1.602563738822937, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.889771580696106, + "num_tokens": 495791722.0, + "step": 14123 + }, + { + "epoch": 2.622841225626741, + "grad_norm": 1.592302918434143, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8811508417129517, + "num_tokens": 495828156.0, + "step": 14124 + }, + { + "epoch": 2.6230269266480963, + "grad_norm": 1.5433915853500366, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8915790915489197, + "num_tokens": 495863889.0, + "step": 14125 + }, + { + "epoch": 2.623212627669452, + "grad_norm": 1.5040388107299805, + "learning_rate": 1e-06, + "loss": 0.3698, + "mean_token_accuracy": 0.8753329515457153, + "num_tokens": 495904920.0, + "step": 14126 + }, + { + "epoch": 2.623398328690808, + "grad_norm": 1.6471202373504639, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8959454894065857, + "num_tokens": 495933876.0, + "step": 14127 + }, + { + "epoch": 2.6235840297121635, + "grad_norm": 1.5120199918746948, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8774300217628479, + "num_tokens": 495975863.0, + "step": 14128 + }, + { + "epoch": 2.6237697307335193, + "grad_norm": 1.6178165674209595, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.878146767616272, + "num_tokens": 496010079.0, + "step": 14129 + }, + { + "epoch": 2.6239554317548746, + "grad_norm": 1.5677523612976074, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8944069147109985, + "num_tokens": 496045820.0, + "step": 14130 + }, + { + "epoch": 2.6241411327762303, + "grad_norm": 1.5968382358551025, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8967288136482239, + "num_tokens": 496078439.0, + "step": 14131 + }, + { + "epoch": 2.624326833797586, + "grad_norm": 1.6380146741867065, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8823719024658203, + "num_tokens": 496112228.0, + "step": 14132 + }, + { + "epoch": 2.6245125348189413, + "grad_norm": 1.6010479927062988, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8840173482894897, + "num_tokens": 496146004.0, + "step": 14133 + }, + { + "epoch": 2.624698235840297, + "grad_norm": 1.5996557474136353, + "learning_rate": 1e-06, + "loss": 0.3567, + "mean_token_accuracy": 0.8754191398620605, + "num_tokens": 496185463.0, + "step": 14134 + }, + { + "epoch": 2.6248839368616528, + "grad_norm": 1.6158127784729004, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8952520489692688, + "num_tokens": 496216883.0, + "step": 14135 + }, + { + "epoch": 2.6250696378830085, + "grad_norm": 1.4838911294937134, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.8960692882537842, + "num_tokens": 496253079.0, + "step": 14136 + }, + { + "epoch": 2.6252553389043642, + "grad_norm": 1.5641573667526245, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8850662112236023, + "num_tokens": 496288919.0, + "step": 14137 + }, + { + "epoch": 2.6254410399257195, + "grad_norm": 2.0491042137145996, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8939685821533203, + "num_tokens": 496325807.0, + "step": 14138 + }, + { + "epoch": 2.6256267409470753, + "grad_norm": 1.6210414171218872, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8827295303344727, + "num_tokens": 496359576.0, + "step": 14139 + }, + { + "epoch": 2.6258124419684306, + "grad_norm": 1.7533470392227173, + "learning_rate": 1e-06, + "loss": 0.403, + "mean_token_accuracy": 0.8636607527732849, + "num_tokens": 496394490.0, + "step": 14140 + }, + { + "epoch": 2.6259981429897863, + "grad_norm": 1.5974360704421997, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8766106367111206, + "num_tokens": 496429982.0, + "step": 14141 + }, + { + "epoch": 2.626183844011142, + "grad_norm": 1.5268882513046265, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8771299123764038, + "num_tokens": 496468569.0, + "step": 14142 + }, + { + "epoch": 2.6263695450324978, + "grad_norm": 1.6749013662338257, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.89008629322052, + "num_tokens": 496500377.0, + "step": 14143 + }, + { + "epoch": 2.6265552460538535, + "grad_norm": 1.5150490999221802, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.894217848777771, + "num_tokens": 496536191.0, + "step": 14144 + }, + { + "epoch": 2.6267409470752088, + "grad_norm": 1.6506606340408325, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8804589509963989, + "num_tokens": 496571130.0, + "step": 14145 + }, + { + "epoch": 2.6269266480965645, + "grad_norm": 1.5360543727874756, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8849081993103027, + "num_tokens": 496611025.0, + "step": 14146 + }, + { + "epoch": 2.6271123491179202, + "grad_norm": 1.5820635557174683, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.876069188117981, + "num_tokens": 496650031.0, + "step": 14147 + }, + { + "epoch": 2.6272980501392755, + "grad_norm": 1.5412306785583496, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8851309418678284, + "num_tokens": 496688371.0, + "step": 14148 + }, + { + "epoch": 2.6274837511606313, + "grad_norm": 1.6174981594085693, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8731564283370972, + "num_tokens": 496725505.0, + "step": 14149 + }, + { + "epoch": 2.627669452181987, + "grad_norm": 1.456716537475586, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8971154689788818, + "num_tokens": 496764446.0, + "step": 14150 + }, + { + "epoch": 2.6278551532033427, + "grad_norm": 1.5348087549209595, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8810060024261475, + "num_tokens": 496801800.0, + "step": 14151 + }, + { + "epoch": 2.6280408542246985, + "grad_norm": 1.4906481504440308, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8900367021560669, + "num_tokens": 496837398.0, + "step": 14152 + }, + { + "epoch": 2.6282265552460538, + "grad_norm": 1.566498875617981, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.89337158203125, + "num_tokens": 496870285.0, + "step": 14153 + }, + { + "epoch": 2.6284122562674095, + "grad_norm": 1.596347451210022, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8797515034675598, + "num_tokens": 496908620.0, + "step": 14154 + }, + { + "epoch": 2.628597957288765, + "grad_norm": 1.6226948499679565, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.8894503712654114, + "num_tokens": 496939580.0, + "step": 14155 + }, + { + "epoch": 2.6287836583101205, + "grad_norm": 1.46303129196167, + "learning_rate": 1e-06, + "loss": 0.3151, + "mean_token_accuracy": 0.8872171640396118, + "num_tokens": 496978507.0, + "step": 14156 + }, + { + "epoch": 2.6289693593314762, + "grad_norm": 1.6346784830093384, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8826397657394409, + "num_tokens": 497010367.0, + "step": 14157 + }, + { + "epoch": 2.629155060352832, + "grad_norm": 1.4507147073745728, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8913142681121826, + "num_tokens": 497048247.0, + "step": 14158 + }, + { + "epoch": 2.6293407613741877, + "grad_norm": 1.5163788795471191, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8836760520935059, + "num_tokens": 497087512.0, + "step": 14159 + }, + { + "epoch": 2.6295264623955434, + "grad_norm": 1.5522520542144775, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8768332600593567, + "num_tokens": 497125559.0, + "step": 14160 + }, + { + "epoch": 2.6297121634168987, + "grad_norm": 1.4289917945861816, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8884707093238831, + "num_tokens": 497162880.0, + "step": 14161 + }, + { + "epoch": 2.6298978644382545, + "grad_norm": 1.6093193292617798, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8807491064071655, + "num_tokens": 497197117.0, + "step": 14162 + }, + { + "epoch": 2.6300835654596098, + "grad_norm": 1.6369751691818237, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8802915215492249, + "num_tokens": 497230763.0, + "step": 14163 + }, + { + "epoch": 2.6302692664809655, + "grad_norm": 1.5831586122512817, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8903913497924805, + "num_tokens": 497262437.0, + "step": 14164 + }, + { + "epoch": 2.630454967502321, + "grad_norm": 1.7893368005752563, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8845333456993103, + "num_tokens": 497291640.0, + "step": 14165 + }, + { + "epoch": 2.630640668523677, + "grad_norm": 1.674594521522522, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.884894847869873, + "num_tokens": 497324755.0, + "step": 14166 + }, + { + "epoch": 2.6308263695450327, + "grad_norm": 1.7391748428344727, + "learning_rate": 1e-06, + "loss": 0.3965, + "mean_token_accuracy": 0.8693947196006775, + "num_tokens": 497357575.0, + "step": 14167 + }, + { + "epoch": 2.631012070566388, + "grad_norm": 1.6538827419281006, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8879976272583008, + "num_tokens": 497389856.0, + "step": 14168 + }, + { + "epoch": 2.6311977715877437, + "grad_norm": 1.5218453407287598, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.886895477771759, + "num_tokens": 497425893.0, + "step": 14169 + }, + { + "epoch": 2.6313834726090994, + "grad_norm": 1.6258981227874756, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.882532000541687, + "num_tokens": 497461621.0, + "step": 14170 + }, + { + "epoch": 2.6315691736304547, + "grad_norm": 1.540310025215149, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8972373008728027, + "num_tokens": 497497159.0, + "step": 14171 + }, + { + "epoch": 2.6317548746518105, + "grad_norm": 1.4465011358261108, + "learning_rate": 1e-06, + "loss": 0.2796, + "mean_token_accuracy": 0.9016250967979431, + "num_tokens": 497531980.0, + "step": 14172 + }, + { + "epoch": 2.631940575673166, + "grad_norm": 1.5457056760787964, + "learning_rate": 1e-06, + "loss": 0.3001, + "mean_token_accuracy": 0.8918784856796265, + "num_tokens": 497566899.0, + "step": 14173 + }, + { + "epoch": 2.632126276694522, + "grad_norm": 1.6105018854141235, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8881833553314209, + "num_tokens": 497600050.0, + "step": 14174 + }, + { + "epoch": 2.6323119777158777, + "grad_norm": 1.5289746522903442, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8842315077781677, + "num_tokens": 497636101.0, + "step": 14175 + }, + { + "epoch": 2.632497678737233, + "grad_norm": 1.5922036170959473, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8840248584747314, + "num_tokens": 497669957.0, + "step": 14176 + }, + { + "epoch": 2.6326833797585887, + "grad_norm": 1.5414602756500244, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8796600103378296, + "num_tokens": 497705138.0, + "step": 14177 + }, + { + "epoch": 2.6328690807799444, + "grad_norm": 1.5107710361480713, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8795784115791321, + "num_tokens": 497744847.0, + "step": 14178 + }, + { + "epoch": 2.6330547818012997, + "grad_norm": 1.3720811605453491, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8908858895301819, + "num_tokens": 497792602.0, + "step": 14179 + }, + { + "epoch": 2.6332404828226554, + "grad_norm": 1.5311176776885986, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8845345973968506, + "num_tokens": 497830161.0, + "step": 14180 + }, + { + "epoch": 2.633426183844011, + "grad_norm": 1.5773149728775024, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8667589426040649, + "num_tokens": 497867496.0, + "step": 14181 + }, + { + "epoch": 2.633611884865367, + "grad_norm": 1.5813472270965576, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8990936279296875, + "num_tokens": 497897342.0, + "step": 14182 + }, + { + "epoch": 2.6337975858867226, + "grad_norm": 1.591447114944458, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.889190137386322, + "num_tokens": 497931809.0, + "step": 14183 + }, + { + "epoch": 2.633983286908078, + "grad_norm": 1.4466336965560913, + "learning_rate": 1e-06, + "loss": 0.2791, + "mean_token_accuracy": 0.9003664255142212, + "num_tokens": 497967840.0, + "step": 14184 + }, + { + "epoch": 2.6341689879294337, + "grad_norm": 1.491058111190796, + "learning_rate": 1e-06, + "loss": 0.2826, + "mean_token_accuracy": 0.8990738987922668, + "num_tokens": 498002573.0, + "step": 14185 + }, + { + "epoch": 2.6343546889507894, + "grad_norm": 1.7008476257324219, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8787161707878113, + "num_tokens": 498033849.0, + "step": 14186 + }, + { + "epoch": 2.6345403899721447, + "grad_norm": 1.4973937273025513, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8894537091255188, + "num_tokens": 498074503.0, + "step": 14187 + }, + { + "epoch": 2.6347260909935004, + "grad_norm": 1.4571359157562256, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8846653699874878, + "num_tokens": 498112835.0, + "step": 14188 + }, + { + "epoch": 2.634911792014856, + "grad_norm": 1.4825581312179565, + "learning_rate": 1e-06, + "loss": 0.2681, + "mean_token_accuracy": 0.9053558707237244, + "num_tokens": 498144431.0, + "step": 14189 + }, + { + "epoch": 2.635097493036212, + "grad_norm": 1.515141487121582, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8827651143074036, + "num_tokens": 498181577.0, + "step": 14190 + }, + { + "epoch": 2.635283194057567, + "grad_norm": 1.6518815755844116, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.8776625394821167, + "num_tokens": 498212860.0, + "step": 14191 + }, + { + "epoch": 2.635468895078923, + "grad_norm": 1.7030130624771118, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.876091718673706, + "num_tokens": 498243139.0, + "step": 14192 + }, + { + "epoch": 2.6356545961002786, + "grad_norm": 1.4891502857208252, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.9005893468856812, + "num_tokens": 498281008.0, + "step": 14193 + }, + { + "epoch": 2.635840297121634, + "grad_norm": 1.515475869178772, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.892403781414032, + "num_tokens": 498315002.0, + "step": 14194 + }, + { + "epoch": 2.6360259981429897, + "grad_norm": 1.4584683179855347, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8857134580612183, + "num_tokens": 498351670.0, + "step": 14195 + }, + { + "epoch": 2.6362116991643454, + "grad_norm": 1.7634328603744507, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8671708106994629, + "num_tokens": 498383428.0, + "step": 14196 + }, + { + "epoch": 2.636397400185701, + "grad_norm": 1.5066261291503906, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.875790536403656, + "num_tokens": 498419722.0, + "step": 14197 + }, + { + "epoch": 2.636583101207057, + "grad_norm": 1.5044920444488525, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8884725570678711, + "num_tokens": 498456189.0, + "step": 14198 + }, + { + "epoch": 2.636768802228412, + "grad_norm": 1.5379821062088013, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8745670318603516, + "num_tokens": 498492302.0, + "step": 14199 + }, + { + "epoch": 2.636954503249768, + "grad_norm": 1.4851703643798828, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8928075432777405, + "num_tokens": 498528478.0, + "step": 14200 + }, + { + "epoch": 2.6371402042711236, + "grad_norm": 1.5196784734725952, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.883184552192688, + "num_tokens": 498565065.0, + "step": 14201 + }, + { + "epoch": 2.637325905292479, + "grad_norm": 1.4684672355651855, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8835663795471191, + "num_tokens": 498602895.0, + "step": 14202 + }, + { + "epoch": 2.6375116063138346, + "grad_norm": 1.4943739175796509, + "learning_rate": 1e-06, + "loss": 0.2795, + "mean_token_accuracy": 0.9010481834411621, + "num_tokens": 498636767.0, + "step": 14203 + }, + { + "epoch": 2.6376973073351904, + "grad_norm": 1.6591215133666992, + "learning_rate": 1e-06, + "loss": 0.4219, + "mean_token_accuracy": 0.8628720045089722, + "num_tokens": 498673939.0, + "step": 14204 + }, + { + "epoch": 2.637883008356546, + "grad_norm": 1.447500228881836, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8854279518127441, + "num_tokens": 498712896.0, + "step": 14205 + }, + { + "epoch": 2.638068709377902, + "grad_norm": 1.6081151962280273, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.87928307056427, + "num_tokens": 498744865.0, + "step": 14206 + }, + { + "epoch": 2.638254410399257, + "grad_norm": 1.559585690498352, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8960146307945251, + "num_tokens": 498774938.0, + "step": 14207 + }, + { + "epoch": 2.638440111420613, + "grad_norm": 1.588335394859314, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.9110592603683472, + "num_tokens": 498806140.0, + "step": 14208 + }, + { + "epoch": 2.6386258124419686, + "grad_norm": 1.438974380493164, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8829246163368225, + "num_tokens": 498848204.0, + "step": 14209 + }, + { + "epoch": 2.638811513463324, + "grad_norm": 1.5031541585922241, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8832607865333557, + "num_tokens": 498893654.0, + "step": 14210 + }, + { + "epoch": 2.6389972144846796, + "grad_norm": 1.536583423614502, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8848479986190796, + "num_tokens": 498931598.0, + "step": 14211 + }, + { + "epoch": 2.6391829155060353, + "grad_norm": 1.5997579097747803, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8808152675628662, + "num_tokens": 498970787.0, + "step": 14212 + }, + { + "epoch": 2.639368616527391, + "grad_norm": 1.7035770416259766, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8939236402511597, + "num_tokens": 499005150.0, + "step": 14213 + }, + { + "epoch": 2.6395543175487464, + "grad_norm": 1.462663173675537, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8987602591514587, + "num_tokens": 499041912.0, + "step": 14214 + }, + { + "epoch": 2.639740018570102, + "grad_norm": 1.6079037189483643, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8907383680343628, + "num_tokens": 499071735.0, + "step": 14215 + }, + { + "epoch": 2.639925719591458, + "grad_norm": 1.5115472078323364, + "learning_rate": 1e-06, + "loss": 0.3208, + "mean_token_accuracy": 0.8865231871604919, + "num_tokens": 499110233.0, + "step": 14216 + }, + { + "epoch": 2.640111420612813, + "grad_norm": 1.6051180362701416, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8851212859153748, + "num_tokens": 499146338.0, + "step": 14217 + }, + { + "epoch": 2.640297121634169, + "grad_norm": 1.6380490064620972, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.8935786485671997, + "num_tokens": 499180947.0, + "step": 14218 + }, + { + "epoch": 2.6404828226555246, + "grad_norm": 1.6754263639450073, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8717061281204224, + "num_tokens": 499214187.0, + "step": 14219 + }, + { + "epoch": 2.6406685236768803, + "grad_norm": 1.4833966493606567, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8802109956741333, + "num_tokens": 499253341.0, + "step": 14220 + }, + { + "epoch": 2.640854224698236, + "grad_norm": 1.47303307056427, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8888721466064453, + "num_tokens": 499294127.0, + "step": 14221 + }, + { + "epoch": 2.6410399257195913, + "grad_norm": 1.4807313680648804, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8837153315544128, + "num_tokens": 499330677.0, + "step": 14222 + }, + { + "epoch": 2.641225626740947, + "grad_norm": 1.527415156364441, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8918048143386841, + "num_tokens": 499368818.0, + "step": 14223 + }, + { + "epoch": 2.641411327762303, + "grad_norm": 1.7171494960784912, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8909642696380615, + "num_tokens": 499400743.0, + "step": 14224 + }, + { + "epoch": 2.641597028783658, + "grad_norm": 1.5153065919876099, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8904001116752625, + "num_tokens": 499437748.0, + "step": 14225 + }, + { + "epoch": 2.641782729805014, + "grad_norm": 1.5900747776031494, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8700178861618042, + "num_tokens": 499481913.0, + "step": 14226 + }, + { + "epoch": 2.6419684308263696, + "grad_norm": 1.5967429876327515, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8910199403762817, + "num_tokens": 499513943.0, + "step": 14227 + }, + { + "epoch": 2.6421541318477253, + "grad_norm": 1.5096275806427002, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8966269493103027, + "num_tokens": 499547953.0, + "step": 14228 + }, + { + "epoch": 2.642339832869081, + "grad_norm": 1.5122171640396118, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.8956693410873413, + "num_tokens": 499581325.0, + "step": 14229 + }, + { + "epoch": 2.6425255338904363, + "grad_norm": 1.5491023063659668, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8881148099899292, + "num_tokens": 499619127.0, + "step": 14230 + }, + { + "epoch": 2.642711234911792, + "grad_norm": 1.579808235168457, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8874755501747131, + "num_tokens": 499653832.0, + "step": 14231 + }, + { + "epoch": 2.642896935933148, + "grad_norm": 1.6515952348709106, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8783735632896423, + "num_tokens": 499684359.0, + "step": 14232 + }, + { + "epoch": 2.643082636954503, + "grad_norm": 1.759901523590088, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8901810646057129, + "num_tokens": 499719118.0, + "step": 14233 + }, + { + "epoch": 2.643268337975859, + "grad_norm": 1.5418314933776855, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8822259902954102, + "num_tokens": 499755489.0, + "step": 14234 + }, + { + "epoch": 2.6434540389972145, + "grad_norm": 1.6127198934555054, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.889348566532135, + "num_tokens": 499788200.0, + "step": 14235 + }, + { + "epoch": 2.6436397400185703, + "grad_norm": 1.6046429872512817, + "learning_rate": 1e-06, + "loss": 0.2811, + "mean_token_accuracy": 0.8980375528335571, + "num_tokens": 499817672.0, + "step": 14236 + }, + { + "epoch": 2.6438254410399256, + "grad_norm": 1.5865095853805542, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8831061720848083, + "num_tokens": 499851526.0, + "step": 14237 + }, + { + "epoch": 2.6440111420612813, + "grad_norm": 1.6383726596832275, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8851404786109924, + "num_tokens": 499886406.0, + "step": 14238 + }, + { + "epoch": 2.644196843082637, + "grad_norm": 1.5293841361999512, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8897144794464111, + "num_tokens": 499921649.0, + "step": 14239 + }, + { + "epoch": 2.6443825441039923, + "grad_norm": 1.5449672937393188, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8892495036125183, + "num_tokens": 499960305.0, + "step": 14240 + }, + { + "epoch": 2.644568245125348, + "grad_norm": 1.5832197666168213, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8950036764144897, + "num_tokens": 499994165.0, + "step": 14241 + }, + { + "epoch": 2.644753946146704, + "grad_norm": 1.467881202697754, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8741399645805359, + "num_tokens": 500035158.0, + "step": 14242 + }, + { + "epoch": 2.6449396471680595, + "grad_norm": 1.458504319190979, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8846885561943054, + "num_tokens": 500075966.0, + "step": 14243 + }, + { + "epoch": 2.6451253481894152, + "grad_norm": 1.5419989824295044, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.8933537602424622, + "num_tokens": 500112767.0, + "step": 14244 + }, + { + "epoch": 2.6453110492107705, + "grad_norm": 1.505149483680725, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8790626525878906, + "num_tokens": 500149151.0, + "step": 14245 + }, + { + "epoch": 2.6454967502321263, + "grad_norm": 1.7026278972625732, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8807957172393799, + "num_tokens": 500182827.0, + "step": 14246 + }, + { + "epoch": 2.645682451253482, + "grad_norm": 1.5601685047149658, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.8947195410728455, + "num_tokens": 500215040.0, + "step": 14247 + }, + { + "epoch": 2.6458681522748373, + "grad_norm": 1.4799333810806274, + "learning_rate": 1e-06, + "loss": 0.2727, + "mean_token_accuracy": 0.9016777873039246, + "num_tokens": 500250092.0, + "step": 14248 + }, + { + "epoch": 2.646053853296193, + "grad_norm": 1.6670862436294556, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8957980871200562, + "num_tokens": 500285608.0, + "step": 14249 + }, + { + "epoch": 2.6462395543175488, + "grad_norm": 1.9799576997756958, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8769073486328125, + "num_tokens": 500321060.0, + "step": 14250 + }, + { + "epoch": 2.6464252553389045, + "grad_norm": 1.6558377742767334, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8739268779754639, + "num_tokens": 500355504.0, + "step": 14251 + }, + { + "epoch": 2.6466109563602602, + "grad_norm": 1.5244793891906738, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8746681809425354, + "num_tokens": 500393096.0, + "step": 14252 + }, + { + "epoch": 2.6467966573816155, + "grad_norm": 1.48838472366333, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8926011919975281, + "num_tokens": 500428777.0, + "step": 14253 + }, + { + "epoch": 2.6469823584029712, + "grad_norm": 1.5007569789886475, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8900240659713745, + "num_tokens": 500464196.0, + "step": 14254 + }, + { + "epoch": 2.647168059424327, + "grad_norm": 1.7049263715744019, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.871160626411438, + "num_tokens": 500495920.0, + "step": 14255 + }, + { + "epoch": 2.6473537604456823, + "grad_norm": 1.7524160146713257, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8813745379447937, + "num_tokens": 500525528.0, + "step": 14256 + }, + { + "epoch": 2.647539461467038, + "grad_norm": 1.4819680452346802, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8886557817459106, + "num_tokens": 500561328.0, + "step": 14257 + }, + { + "epoch": 2.6477251624883937, + "grad_norm": 1.6821209192276, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8786678910255432, + "num_tokens": 500595433.0, + "step": 14258 + }, + { + "epoch": 2.6479108635097495, + "grad_norm": 1.5285996198654175, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8914300203323364, + "num_tokens": 500634538.0, + "step": 14259 + }, + { + "epoch": 2.6480965645311048, + "grad_norm": 1.6138242483139038, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8797183036804199, + "num_tokens": 500670458.0, + "step": 14260 + }, + { + "epoch": 2.6482822655524605, + "grad_norm": 1.6305221319198608, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8924299478530884, + "num_tokens": 500703079.0, + "step": 14261 + }, + { + "epoch": 2.6484679665738162, + "grad_norm": 1.4782084226608276, + "learning_rate": 1e-06, + "loss": 0.362, + "mean_token_accuracy": 0.8744983077049255, + "num_tokens": 500741671.0, + "step": 14262 + }, + { + "epoch": 2.6486536675951715, + "grad_norm": 1.4027173519134521, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8849678039550781, + "num_tokens": 500779588.0, + "step": 14263 + }, + { + "epoch": 2.6488393686165272, + "grad_norm": 1.597957968711853, + "learning_rate": 1e-06, + "loss": 0.3674, + "mean_token_accuracy": 0.8723595142364502, + "num_tokens": 500815670.0, + "step": 14264 + }, + { + "epoch": 2.649025069637883, + "grad_norm": 1.558031439781189, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8827909827232361, + "num_tokens": 500854455.0, + "step": 14265 + }, + { + "epoch": 2.6492107706592387, + "grad_norm": 1.5447978973388672, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8958535194396973, + "num_tokens": 500887434.0, + "step": 14266 + }, + { + "epoch": 2.6493964716805944, + "grad_norm": 1.6693757772445679, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8822275996208191, + "num_tokens": 500919982.0, + "step": 14267 + }, + { + "epoch": 2.6495821727019497, + "grad_norm": 1.5121400356292725, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8822558522224426, + "num_tokens": 500959221.0, + "step": 14268 + }, + { + "epoch": 2.6497678737233055, + "grad_norm": 1.6770823001861572, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8876407146453857, + "num_tokens": 500994069.0, + "step": 14269 + }, + { + "epoch": 2.649953574744661, + "grad_norm": 1.6080175638198853, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8738136291503906, + "num_tokens": 501026315.0, + "step": 14270 + }, + { + "epoch": 2.6501392757660165, + "grad_norm": 1.3910949230194092, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8927749395370483, + "num_tokens": 501064674.0, + "step": 14271 + }, + { + "epoch": 2.650324976787372, + "grad_norm": 1.741809368133545, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8743183612823486, + "num_tokens": 501094942.0, + "step": 14272 + }, + { + "epoch": 2.650510677808728, + "grad_norm": 1.6826716661453247, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8818148374557495, + "num_tokens": 501124519.0, + "step": 14273 + }, + { + "epoch": 2.6506963788300837, + "grad_norm": 1.5733603239059448, + "learning_rate": 1e-06, + "loss": 0.2723, + "mean_token_accuracy": 0.9021687507629395, + "num_tokens": 501153341.0, + "step": 14274 + }, + { + "epoch": 2.6508820798514394, + "grad_norm": 1.39348304271698, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8946139216423035, + "num_tokens": 501197283.0, + "step": 14275 + }, + { + "epoch": 2.6510677808727947, + "grad_norm": 1.5634149312973022, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8757233619689941, + "num_tokens": 501233559.0, + "step": 14276 + }, + { + "epoch": 2.6512534818941504, + "grad_norm": 1.4821345806121826, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.890141487121582, + "num_tokens": 501269190.0, + "step": 14277 + }, + { + "epoch": 2.651439182915506, + "grad_norm": 1.5105303525924683, + "learning_rate": 1e-06, + "loss": 0.2326, + "mean_token_accuracy": 0.9139487147331238, + "num_tokens": 501298741.0, + "step": 14278 + }, + { + "epoch": 2.6516248839368615, + "grad_norm": 1.532563328742981, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8940303325653076, + "num_tokens": 501330405.0, + "step": 14279 + }, + { + "epoch": 2.651810584958217, + "grad_norm": 1.5399051904678345, + "learning_rate": 1e-06, + "loss": 0.328, + "mean_token_accuracy": 0.8835539817810059, + "num_tokens": 501367611.0, + "step": 14280 + }, + { + "epoch": 2.651996285979573, + "grad_norm": 1.5133534669876099, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8814451098442078, + "num_tokens": 501407032.0, + "step": 14281 + }, + { + "epoch": 2.6521819870009287, + "grad_norm": 1.449862003326416, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.883702278137207, + "num_tokens": 501448848.0, + "step": 14282 + }, + { + "epoch": 2.6523676880222844, + "grad_norm": 1.538840651512146, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8852565288543701, + "num_tokens": 501485841.0, + "step": 14283 + }, + { + "epoch": 2.6525533890436397, + "grad_norm": 1.512650728225708, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.892475962638855, + "num_tokens": 501521794.0, + "step": 14284 + }, + { + "epoch": 2.6527390900649954, + "grad_norm": 1.50510573387146, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8781644105911255, + "num_tokens": 501562645.0, + "step": 14285 + }, + { + "epoch": 2.6529247910863507, + "grad_norm": 1.5461698770523071, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8813931941986084, + "num_tokens": 501598068.0, + "step": 14286 + }, + { + "epoch": 2.6531104921077064, + "grad_norm": 1.5907065868377686, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8855583667755127, + "num_tokens": 501633518.0, + "step": 14287 + }, + { + "epoch": 2.653296193129062, + "grad_norm": 1.7218096256256104, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8695831298828125, + "num_tokens": 501666278.0, + "step": 14288 + }, + { + "epoch": 2.653481894150418, + "grad_norm": 1.6959624290466309, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8902983069419861, + "num_tokens": 501700803.0, + "step": 14289 + }, + { + "epoch": 2.6536675951717736, + "grad_norm": 1.6530876159667969, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8860384225845337, + "num_tokens": 501733068.0, + "step": 14290 + }, + { + "epoch": 2.653853296193129, + "grad_norm": 1.5508413314819336, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8872838020324707, + "num_tokens": 501768160.0, + "step": 14291 + }, + { + "epoch": 2.6540389972144847, + "grad_norm": 1.642196536064148, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8856812715530396, + "num_tokens": 501800392.0, + "step": 14292 + }, + { + "epoch": 2.6542246982358404, + "grad_norm": 1.4471203088760376, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8883904218673706, + "num_tokens": 501837806.0, + "step": 14293 + }, + { + "epoch": 2.6544103992571957, + "grad_norm": 1.4843542575836182, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8771378993988037, + "num_tokens": 501875808.0, + "step": 14294 + }, + { + "epoch": 2.6545961002785514, + "grad_norm": 1.7000411748886108, + "learning_rate": 1e-06, + "loss": 0.3533, + "mean_token_accuracy": 0.8778791427612305, + "num_tokens": 501907579.0, + "step": 14295 + }, + { + "epoch": 2.654781801299907, + "grad_norm": 1.5988082885742188, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8829996585845947, + "num_tokens": 501942172.0, + "step": 14296 + }, + { + "epoch": 2.654967502321263, + "grad_norm": 1.5905117988586426, + "learning_rate": 1e-06, + "loss": 0.2947, + "mean_token_accuracy": 0.8928921222686768, + "num_tokens": 501976225.0, + "step": 14297 + }, + { + "epoch": 2.6551532033426186, + "grad_norm": 1.628875970840454, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.9034963846206665, + "num_tokens": 502006820.0, + "step": 14298 + }, + { + "epoch": 2.655338904363974, + "grad_norm": 1.5401171445846558, + "learning_rate": 1e-06, + "loss": 0.3663, + "mean_token_accuracy": 0.8801531195640564, + "num_tokens": 502045268.0, + "step": 14299 + }, + { + "epoch": 2.6555246053853296, + "grad_norm": 1.4140217304229736, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8897750377655029, + "num_tokens": 502084125.0, + "step": 14300 + }, + { + "epoch": 2.6557103064066854, + "grad_norm": 1.6458933353424072, + "learning_rate": 1e-06, + "loss": 0.348, + "mean_token_accuracy": 0.8798090219497681, + "num_tokens": 502115443.0, + "step": 14301 + }, + { + "epoch": 2.6558960074280407, + "grad_norm": 1.6847517490386963, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8870033621788025, + "num_tokens": 502144025.0, + "step": 14302 + }, + { + "epoch": 2.6560817084493964, + "grad_norm": 1.814234733581543, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8887338638305664, + "num_tokens": 502172620.0, + "step": 14303 + }, + { + "epoch": 2.656267409470752, + "grad_norm": 1.641206979751587, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8940714597702026, + "num_tokens": 502199483.0, + "step": 14304 + }, + { + "epoch": 2.656453110492108, + "grad_norm": 1.6555589437484741, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8863413333892822, + "num_tokens": 502231251.0, + "step": 14305 + }, + { + "epoch": 2.6566388115134636, + "grad_norm": 1.5444552898406982, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8934123516082764, + "num_tokens": 502264796.0, + "step": 14306 + }, + { + "epoch": 2.656824512534819, + "grad_norm": 1.5400453805923462, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8797913193702698, + "num_tokens": 502303426.0, + "step": 14307 + }, + { + "epoch": 2.6570102135561746, + "grad_norm": 1.5439281463623047, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.8914968967437744, + "num_tokens": 502338132.0, + "step": 14308 + }, + { + "epoch": 2.65719591457753, + "grad_norm": 1.602566123008728, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8812891244888306, + "num_tokens": 502371423.0, + "step": 14309 + }, + { + "epoch": 2.6573816155988856, + "grad_norm": 1.5944125652313232, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8905143141746521, + "num_tokens": 502404572.0, + "step": 14310 + }, + { + "epoch": 2.6575673166202414, + "grad_norm": 1.6310304403305054, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8916379809379578, + "num_tokens": 502435511.0, + "step": 14311 + }, + { + "epoch": 2.657753017641597, + "grad_norm": 1.5729613304138184, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8836537599563599, + "num_tokens": 502473074.0, + "step": 14312 + }, + { + "epoch": 2.657938718662953, + "grad_norm": 1.3804552555084229, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8944012522697449, + "num_tokens": 502515334.0, + "step": 14313 + }, + { + "epoch": 2.658124419684308, + "grad_norm": 1.5968092679977417, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8832340240478516, + "num_tokens": 502550290.0, + "step": 14314 + }, + { + "epoch": 2.658310120705664, + "grad_norm": 1.7027933597564697, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8842416405677795, + "num_tokens": 502581355.0, + "step": 14315 + }, + { + "epoch": 2.6584958217270196, + "grad_norm": 1.4873496294021606, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8811497688293457, + "num_tokens": 502622752.0, + "step": 14316 + }, + { + "epoch": 2.658681522748375, + "grad_norm": 1.5372343063354492, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.889404833316803, + "num_tokens": 502660054.0, + "step": 14317 + }, + { + "epoch": 2.6588672237697306, + "grad_norm": 1.583674669265747, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.877518892288208, + "num_tokens": 502696422.0, + "step": 14318 + }, + { + "epoch": 2.6590529247910863, + "grad_norm": 1.588525414466858, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.9007807374000549, + "num_tokens": 502732322.0, + "step": 14319 + }, + { + "epoch": 2.659238625812442, + "grad_norm": 1.4608708620071411, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8842571973800659, + "num_tokens": 502771769.0, + "step": 14320 + }, + { + "epoch": 2.659424326833798, + "grad_norm": 1.7843061685562134, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8634104132652283, + "num_tokens": 502804568.0, + "step": 14321 + }, + { + "epoch": 2.659610027855153, + "grad_norm": 1.6632009744644165, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.875793993473053, + "num_tokens": 502841110.0, + "step": 14322 + }, + { + "epoch": 2.659795728876509, + "grad_norm": 1.665258526802063, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8807933926582336, + "num_tokens": 502872323.0, + "step": 14323 + }, + { + "epoch": 2.6599814298978646, + "grad_norm": 1.547737717628479, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.886458158493042, + "num_tokens": 502910859.0, + "step": 14324 + }, + { + "epoch": 2.66016713091922, + "grad_norm": 1.5555959939956665, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8789292573928833, + "num_tokens": 502948929.0, + "step": 14325 + }, + { + "epoch": 2.6603528319405756, + "grad_norm": 1.4734504222869873, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8889352083206177, + "num_tokens": 502985066.0, + "step": 14326 + }, + { + "epoch": 2.6605385329619313, + "grad_norm": 1.4657543897628784, + "learning_rate": 1e-06, + "loss": 0.2848, + "mean_token_accuracy": 0.8966084718704224, + "num_tokens": 503020256.0, + "step": 14327 + }, + { + "epoch": 2.660724233983287, + "grad_norm": 1.5263887643814087, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8831762075424194, + "num_tokens": 503056657.0, + "step": 14328 + }, + { + "epoch": 2.660909935004643, + "grad_norm": 1.5893990993499756, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8769571781158447, + "num_tokens": 503091859.0, + "step": 14329 + }, + { + "epoch": 2.661095636025998, + "grad_norm": 1.5348187685012817, + "learning_rate": 1e-06, + "loss": 0.2752, + "mean_token_accuracy": 0.8981218338012695, + "num_tokens": 503123388.0, + "step": 14330 + }, + { + "epoch": 2.661281337047354, + "grad_norm": 1.4539167881011963, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8907487392425537, + "num_tokens": 503160990.0, + "step": 14331 + }, + { + "epoch": 2.661467038068709, + "grad_norm": 1.5165637731552124, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8926834464073181, + "num_tokens": 503194799.0, + "step": 14332 + }, + { + "epoch": 2.661652739090065, + "grad_norm": 1.4939355850219727, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8820772767066956, + "num_tokens": 503233237.0, + "step": 14333 + }, + { + "epoch": 2.6618384401114206, + "grad_norm": 1.6745240688323975, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8917055130004883, + "num_tokens": 503262563.0, + "step": 14334 + }, + { + "epoch": 2.6620241411327763, + "grad_norm": 1.5233259201049805, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8865849375724792, + "num_tokens": 503302457.0, + "step": 14335 + }, + { + "epoch": 2.662209842154132, + "grad_norm": 1.6908836364746094, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8826884627342224, + "num_tokens": 503335501.0, + "step": 14336 + }, + { + "epoch": 2.6623955431754873, + "grad_norm": 1.6843359470367432, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.886399507522583, + "num_tokens": 503365893.0, + "step": 14337 + }, + { + "epoch": 2.662581244196843, + "grad_norm": 1.6010010242462158, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8824726343154907, + "num_tokens": 503403494.0, + "step": 14338 + }, + { + "epoch": 2.662766945218199, + "grad_norm": 1.5897650718688965, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8819151520729065, + "num_tokens": 503437510.0, + "step": 14339 + }, + { + "epoch": 2.662952646239554, + "grad_norm": 1.5291613340377808, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8863461017608643, + "num_tokens": 503473824.0, + "step": 14340 + }, + { + "epoch": 2.66313834726091, + "grad_norm": 1.4893237352371216, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.891226053237915, + "num_tokens": 503511558.0, + "step": 14341 + }, + { + "epoch": 2.6633240482822655, + "grad_norm": 1.6265414953231812, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8762913942337036, + "num_tokens": 503544429.0, + "step": 14342 + }, + { + "epoch": 2.6635097493036213, + "grad_norm": 1.7341898679733276, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8879945278167725, + "num_tokens": 503582965.0, + "step": 14343 + }, + { + "epoch": 2.663695450324977, + "grad_norm": 1.3944945335388184, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8931818008422852, + "num_tokens": 503626204.0, + "step": 14344 + }, + { + "epoch": 2.6638811513463323, + "grad_norm": 1.5055828094482422, + "learning_rate": 1e-06, + "loss": 0.2735, + "mean_token_accuracy": 0.9024167060852051, + "num_tokens": 503659886.0, + "step": 14345 + }, + { + "epoch": 2.664066852367688, + "grad_norm": 1.4693866968154907, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8920437097549438, + "num_tokens": 503699486.0, + "step": 14346 + }, + { + "epoch": 2.6642525533890438, + "grad_norm": 1.5705078840255737, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8868743181228638, + "num_tokens": 503736569.0, + "step": 14347 + }, + { + "epoch": 2.664438254410399, + "grad_norm": 1.5309251546859741, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8861716389656067, + "num_tokens": 503774230.0, + "step": 14348 + }, + { + "epoch": 2.664623955431755, + "grad_norm": 1.5439457893371582, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.8940668702125549, + "num_tokens": 503803387.0, + "step": 14349 + }, + { + "epoch": 2.6648096564531105, + "grad_norm": 1.3913800716400146, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.8994277715682983, + "num_tokens": 503840157.0, + "step": 14350 + }, + { + "epoch": 2.6649953574744663, + "grad_norm": 1.5552078485488892, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8902431130409241, + "num_tokens": 503877205.0, + "step": 14351 + }, + { + "epoch": 2.665181058495822, + "grad_norm": 1.6663761138916016, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8869147300720215, + "num_tokens": 503912742.0, + "step": 14352 + }, + { + "epoch": 2.6653667595171773, + "grad_norm": 1.596152901649475, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8923563361167908, + "num_tokens": 503946207.0, + "step": 14353 + }, + { + "epoch": 2.665552460538533, + "grad_norm": 1.6033875942230225, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8880863189697266, + "num_tokens": 503979647.0, + "step": 14354 + }, + { + "epoch": 2.6657381615598887, + "grad_norm": 1.5812833309173584, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8829330801963806, + "num_tokens": 504015679.0, + "step": 14355 + }, + { + "epoch": 2.665923862581244, + "grad_norm": 1.5260614156723022, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8895999193191528, + "num_tokens": 504048827.0, + "step": 14356 + }, + { + "epoch": 2.6661095636025998, + "grad_norm": 1.705888032913208, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8797990083694458, + "num_tokens": 504080923.0, + "step": 14357 + }, + { + "epoch": 2.6662952646239555, + "grad_norm": 1.5868593454360962, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8893042802810669, + "num_tokens": 504114818.0, + "step": 14358 + }, + { + "epoch": 2.6664809656453112, + "grad_norm": 1.4286837577819824, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8963760733604431, + "num_tokens": 504153903.0, + "step": 14359 + }, + { + "epoch": 2.6666666666666665, + "grad_norm": 1.4643077850341797, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8904647827148438, + "num_tokens": 504191951.0, + "step": 14360 + }, + { + "epoch": 2.6668523676880223, + "grad_norm": 1.5787960290908813, + "learning_rate": 1e-06, + "loss": 0.3508, + "mean_token_accuracy": 0.8753795623779297, + "num_tokens": 504225824.0, + "step": 14361 + }, + { + "epoch": 2.667038068709378, + "grad_norm": 1.5438683032989502, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8940187096595764, + "num_tokens": 504260938.0, + "step": 14362 + }, + { + "epoch": 2.6672237697307333, + "grad_norm": 1.59294855594635, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8828139305114746, + "num_tokens": 504291527.0, + "step": 14363 + }, + { + "epoch": 2.667409470752089, + "grad_norm": 1.7496713399887085, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8817028403282166, + "num_tokens": 504319794.0, + "step": 14364 + }, + { + "epoch": 2.6675951717734447, + "grad_norm": 1.4916744232177734, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8882421255111694, + "num_tokens": 504359524.0, + "step": 14365 + }, + { + "epoch": 2.6677808727948005, + "grad_norm": 1.5484832525253296, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8872661590576172, + "num_tokens": 504398919.0, + "step": 14366 + }, + { + "epoch": 2.667966573816156, + "grad_norm": 1.7103172540664673, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8923540115356445, + "num_tokens": 504430042.0, + "step": 14367 + }, + { + "epoch": 2.6681522748375115, + "grad_norm": 1.535499095916748, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.891097366809845, + "num_tokens": 504464751.0, + "step": 14368 + }, + { + "epoch": 2.6683379758588672, + "grad_norm": 1.6254158020019531, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8887416124343872, + "num_tokens": 504495895.0, + "step": 14369 + }, + { + "epoch": 2.668523676880223, + "grad_norm": 1.5202112197875977, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8859928846359253, + "num_tokens": 504530061.0, + "step": 14370 + }, + { + "epoch": 2.6687093779015783, + "grad_norm": 1.611350655555725, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8743705749511719, + "num_tokens": 504567574.0, + "step": 14371 + }, + { + "epoch": 2.668895078922934, + "grad_norm": 1.5726028680801392, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8861212730407715, + "num_tokens": 504603255.0, + "step": 14372 + }, + { + "epoch": 2.6690807799442897, + "grad_norm": 1.5643692016601562, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8897494077682495, + "num_tokens": 504637095.0, + "step": 14373 + }, + { + "epoch": 2.6692664809656454, + "grad_norm": 1.5937656164169312, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.8807127475738525, + "num_tokens": 504670044.0, + "step": 14374 + }, + { + "epoch": 2.669452181987001, + "grad_norm": 1.6084957122802734, + "learning_rate": 1e-06, + "loss": 0.3342, + "mean_token_accuracy": 0.8824961185455322, + "num_tokens": 504703709.0, + "step": 14375 + }, + { + "epoch": 2.6696378830083565, + "grad_norm": 1.5991179943084717, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8861761093139648, + "num_tokens": 504734079.0, + "step": 14376 + }, + { + "epoch": 2.669823584029712, + "grad_norm": 1.6032373905181885, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8854738473892212, + "num_tokens": 504765613.0, + "step": 14377 + }, + { + "epoch": 2.670009285051068, + "grad_norm": 1.6535661220550537, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8733314275741577, + "num_tokens": 504803501.0, + "step": 14378 + }, + { + "epoch": 2.6701949860724232, + "grad_norm": 1.4740413427352905, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8854711651802063, + "num_tokens": 504842812.0, + "step": 14379 + }, + { + "epoch": 2.670380687093779, + "grad_norm": 1.4560760259628296, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8936998248100281, + "num_tokens": 504881600.0, + "step": 14380 + }, + { + "epoch": 2.6705663881151347, + "grad_norm": 1.5607556104660034, + "learning_rate": 1e-06, + "loss": 0.2763, + "mean_token_accuracy": 0.9031001329421997, + "num_tokens": 504912216.0, + "step": 14381 + }, + { + "epoch": 2.6707520891364904, + "grad_norm": 1.6577714681625366, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8878068923950195, + "num_tokens": 504942572.0, + "step": 14382 + }, + { + "epoch": 2.6709377901578457, + "grad_norm": 1.6460449695587158, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8816543817520142, + "num_tokens": 504977800.0, + "step": 14383 + }, + { + "epoch": 2.6711234911792014, + "grad_norm": 1.6917320489883423, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8779750466346741, + "num_tokens": 505010049.0, + "step": 14384 + }, + { + "epoch": 2.671309192200557, + "grad_norm": 1.4812911748886108, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8924741744995117, + "num_tokens": 505045288.0, + "step": 14385 + }, + { + "epoch": 2.6714948932219125, + "grad_norm": 1.518096685409546, + "learning_rate": 1e-06, + "loss": 0.2682, + "mean_token_accuracy": 0.9022537469863892, + "num_tokens": 505078938.0, + "step": 14386 + }, + { + "epoch": 2.671680594243268, + "grad_norm": 1.694708228111267, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8820585012435913, + "num_tokens": 505109362.0, + "step": 14387 + }, + { + "epoch": 2.671866295264624, + "grad_norm": 1.5869814157485962, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8898462653160095, + "num_tokens": 505143601.0, + "step": 14388 + }, + { + "epoch": 2.6720519962859797, + "grad_norm": 1.5043795108795166, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8904505968093872, + "num_tokens": 505182232.0, + "step": 14389 + }, + { + "epoch": 2.6722376973073354, + "grad_norm": 1.837066650390625, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8861922025680542, + "num_tokens": 505217405.0, + "step": 14390 + }, + { + "epoch": 2.6724233983286907, + "grad_norm": 1.583549976348877, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.883782148361206, + "num_tokens": 505253045.0, + "step": 14391 + }, + { + "epoch": 2.6726090993500464, + "grad_norm": 1.6342664957046509, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8950009942054749, + "num_tokens": 505285442.0, + "step": 14392 + }, + { + "epoch": 2.672794800371402, + "grad_norm": 1.5964723825454712, + "learning_rate": 1e-06, + "loss": 0.2677, + "mean_token_accuracy": 0.9043596982955933, + "num_tokens": 505315027.0, + "step": 14393 + }, + { + "epoch": 2.6729805013927574, + "grad_norm": 1.5235587358474731, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8846107721328735, + "num_tokens": 505352279.0, + "step": 14394 + }, + { + "epoch": 2.673166202414113, + "grad_norm": 1.4674750566482544, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8856556415557861, + "num_tokens": 505394274.0, + "step": 14395 + }, + { + "epoch": 2.673351903435469, + "grad_norm": 1.6754822731018066, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8921288847923279, + "num_tokens": 505428198.0, + "step": 14396 + }, + { + "epoch": 2.6735376044568246, + "grad_norm": 1.5552955865859985, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8919621706008911, + "num_tokens": 505465060.0, + "step": 14397 + }, + { + "epoch": 2.6737233054781804, + "grad_norm": 1.815314531326294, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8912279605865479, + "num_tokens": 505493789.0, + "step": 14398 + }, + { + "epoch": 2.6739090064995357, + "grad_norm": 1.5904277563095093, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8824499845504761, + "num_tokens": 505530496.0, + "step": 14399 + }, + { + "epoch": 2.6740947075208914, + "grad_norm": 1.6542433500289917, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8700001239776611, + "num_tokens": 505567139.0, + "step": 14400 + }, + { + "epoch": 2.674280408542247, + "grad_norm": 1.6306430101394653, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.885733962059021, + "num_tokens": 505601106.0, + "step": 14401 + }, + { + "epoch": 2.6744661095636024, + "grad_norm": 1.4837316274642944, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8881015181541443, + "num_tokens": 505638875.0, + "step": 14402 + }, + { + "epoch": 2.674651810584958, + "grad_norm": 1.7662098407745361, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8765097856521606, + "num_tokens": 505669576.0, + "step": 14403 + }, + { + "epoch": 2.674837511606314, + "grad_norm": 1.485674500465393, + "learning_rate": 1e-06, + "loss": 0.2763, + "mean_token_accuracy": 0.9022836685180664, + "num_tokens": 505704583.0, + "step": 14404 + }, + { + "epoch": 2.6750232126276696, + "grad_norm": 1.5724127292633057, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.8990025520324707, + "num_tokens": 505737695.0, + "step": 14405 + }, + { + "epoch": 2.675208913649025, + "grad_norm": 1.8116254806518555, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8796207904815674, + "num_tokens": 505767711.0, + "step": 14406 + }, + { + "epoch": 2.6753946146703806, + "grad_norm": 1.5609641075134277, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8846179842948914, + "num_tokens": 505802056.0, + "step": 14407 + }, + { + "epoch": 2.6755803156917364, + "grad_norm": 1.5207183361053467, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8834375143051147, + "num_tokens": 505838276.0, + "step": 14408 + }, + { + "epoch": 2.6757660167130917, + "grad_norm": 1.463865876197815, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8969170451164246, + "num_tokens": 505878789.0, + "step": 14409 + }, + { + "epoch": 2.6759517177344474, + "grad_norm": 1.6988940238952637, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8717222213745117, + "num_tokens": 505912763.0, + "step": 14410 + }, + { + "epoch": 2.676137418755803, + "grad_norm": 1.5157963037490845, + "learning_rate": 1e-06, + "loss": 0.2703, + "mean_token_accuracy": 0.9008657932281494, + "num_tokens": 505946860.0, + "step": 14411 + }, + { + "epoch": 2.676323119777159, + "grad_norm": 1.5375785827636719, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8946661949157715, + "num_tokens": 505980425.0, + "step": 14412 + }, + { + "epoch": 2.6765088207985146, + "grad_norm": 1.5150830745697021, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8852403163909912, + "num_tokens": 506015382.0, + "step": 14413 + }, + { + "epoch": 2.67669452181987, + "grad_norm": 1.559552550315857, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8820078372955322, + "num_tokens": 506049772.0, + "step": 14414 + }, + { + "epoch": 2.6768802228412256, + "grad_norm": 1.5941420793533325, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8874428868293762, + "num_tokens": 506082811.0, + "step": 14415 + }, + { + "epoch": 2.6770659238625814, + "grad_norm": 1.6282731294631958, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8867599964141846, + "num_tokens": 506115255.0, + "step": 14416 + }, + { + "epoch": 2.6772516248839366, + "grad_norm": 1.4709060192108154, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8875625133514404, + "num_tokens": 506154708.0, + "step": 14417 + }, + { + "epoch": 2.6774373259052924, + "grad_norm": 1.613357424736023, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8861958980560303, + "num_tokens": 506189062.0, + "step": 14418 + }, + { + "epoch": 2.677623026926648, + "grad_norm": 1.5816199779510498, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8693561553955078, + "num_tokens": 506227920.0, + "step": 14419 + }, + { + "epoch": 2.677808727948004, + "grad_norm": 1.493187427520752, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8741928935050964, + "num_tokens": 506270942.0, + "step": 14420 + }, + { + "epoch": 2.6779944289693596, + "grad_norm": 1.5022908449172974, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8860169649124146, + "num_tokens": 506308384.0, + "step": 14421 + }, + { + "epoch": 2.678180129990715, + "grad_norm": 1.545244812965393, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.894853949546814, + "num_tokens": 506342703.0, + "step": 14422 + }, + { + "epoch": 2.6783658310120706, + "grad_norm": 1.5906492471694946, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8931362628936768, + "num_tokens": 506381194.0, + "step": 14423 + }, + { + "epoch": 2.6785515320334263, + "grad_norm": 1.661299467086792, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8938770294189453, + "num_tokens": 506410757.0, + "step": 14424 + }, + { + "epoch": 2.6787372330547816, + "grad_norm": 1.553006649017334, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8897894024848938, + "num_tokens": 506445260.0, + "step": 14425 + }, + { + "epoch": 2.6789229340761374, + "grad_norm": 1.7522273063659668, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8957105875015259, + "num_tokens": 506473873.0, + "step": 14426 + }, + { + "epoch": 2.679108635097493, + "grad_norm": 1.6973742246627808, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8836574554443359, + "num_tokens": 506510944.0, + "step": 14427 + }, + { + "epoch": 2.679294336118849, + "grad_norm": 1.4718445539474487, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8930209279060364, + "num_tokens": 506549766.0, + "step": 14428 + }, + { + "epoch": 2.679480037140204, + "grad_norm": 1.4878902435302734, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.899673581123352, + "num_tokens": 506584765.0, + "step": 14429 + }, + { + "epoch": 2.67966573816156, + "grad_norm": 1.5552774667739868, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8875267505645752, + "num_tokens": 506619548.0, + "step": 14430 + }, + { + "epoch": 2.6798514391829156, + "grad_norm": 1.6206618547439575, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8950273990631104, + "num_tokens": 506651343.0, + "step": 14431 + }, + { + "epoch": 2.680037140204271, + "grad_norm": 1.5096983909606934, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8949087858200073, + "num_tokens": 506684168.0, + "step": 14432 + }, + { + "epoch": 2.6802228412256266, + "grad_norm": 1.5179312229156494, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8925639390945435, + "num_tokens": 506722168.0, + "step": 14433 + }, + { + "epoch": 2.6804085422469823, + "grad_norm": 1.5335360765457153, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.8951025009155273, + "num_tokens": 506760005.0, + "step": 14434 + }, + { + "epoch": 2.680594243268338, + "grad_norm": 1.6903578042984009, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8931723833084106, + "num_tokens": 506789875.0, + "step": 14435 + }, + { + "epoch": 2.680779944289694, + "grad_norm": 1.5610637664794922, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.8961195945739746, + "num_tokens": 506822048.0, + "step": 14436 + }, + { + "epoch": 2.680965645311049, + "grad_norm": 1.4546067714691162, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8952518701553345, + "num_tokens": 506860109.0, + "step": 14437 + }, + { + "epoch": 2.681151346332405, + "grad_norm": 1.3911319971084595, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8798409700393677, + "num_tokens": 506902399.0, + "step": 14438 + }, + { + "epoch": 2.6813370473537605, + "grad_norm": 1.812616229057312, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8760815858840942, + "num_tokens": 506930748.0, + "step": 14439 + }, + { + "epoch": 2.681522748375116, + "grad_norm": 1.596315860748291, + "learning_rate": 1e-06, + "loss": 0.3164, + "mean_token_accuracy": 0.8871130347251892, + "num_tokens": 506966406.0, + "step": 14440 + }, + { + "epoch": 2.6817084493964716, + "grad_norm": 1.620208978652954, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8866007328033447, + "num_tokens": 507002588.0, + "step": 14441 + }, + { + "epoch": 2.6818941504178273, + "grad_norm": 1.5893505811691284, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.874464213848114, + "num_tokens": 507040354.0, + "step": 14442 + }, + { + "epoch": 2.682079851439183, + "grad_norm": 1.536674976348877, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8787098526954651, + "num_tokens": 507075488.0, + "step": 14443 + }, + { + "epoch": 2.6822655524605388, + "grad_norm": 1.5017666816711426, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8893307447433472, + "num_tokens": 507112158.0, + "step": 14444 + }, + { + "epoch": 2.682451253481894, + "grad_norm": 1.5861622095108032, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.882493257522583, + "num_tokens": 507145751.0, + "step": 14445 + }, + { + "epoch": 2.68263695450325, + "grad_norm": 1.5711028575897217, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8796947002410889, + "num_tokens": 507184136.0, + "step": 14446 + }, + { + "epoch": 2.6828226555246055, + "grad_norm": 1.6476444005966187, + "learning_rate": 1e-06, + "loss": 0.2817, + "mean_token_accuracy": 0.8994667530059814, + "num_tokens": 507213367.0, + "step": 14447 + }, + { + "epoch": 2.683008356545961, + "grad_norm": 1.4229764938354492, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8845621943473816, + "num_tokens": 507258165.0, + "step": 14448 + }, + { + "epoch": 2.6831940575673165, + "grad_norm": 1.5647847652435303, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8832858800888062, + "num_tokens": 507294281.0, + "step": 14449 + }, + { + "epoch": 2.6833797585886723, + "grad_norm": 1.5465692281723022, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8897026777267456, + "num_tokens": 507327652.0, + "step": 14450 + }, + { + "epoch": 2.683565459610028, + "grad_norm": 1.3782552480697632, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8905247449874878, + "num_tokens": 507371422.0, + "step": 14451 + }, + { + "epoch": 2.6837511606313837, + "grad_norm": 1.486373782157898, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8925319314002991, + "num_tokens": 507405748.0, + "step": 14452 + }, + { + "epoch": 2.683936861652739, + "grad_norm": 1.4964386224746704, + "learning_rate": 1e-06, + "loss": 0.2875, + "mean_token_accuracy": 0.8974641561508179, + "num_tokens": 507441927.0, + "step": 14453 + }, + { + "epoch": 2.6841225626740948, + "grad_norm": 1.6547303199768066, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8954547047615051, + "num_tokens": 507470774.0, + "step": 14454 + }, + { + "epoch": 2.68430826369545, + "grad_norm": 1.4952001571655273, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8785748481750488, + "num_tokens": 507508688.0, + "step": 14455 + }, + { + "epoch": 2.684493964716806, + "grad_norm": 1.6480106115341187, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.886509120464325, + "num_tokens": 507543397.0, + "step": 14456 + }, + { + "epoch": 2.6846796657381615, + "grad_norm": 1.7473117113113403, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8860876560211182, + "num_tokens": 507571086.0, + "step": 14457 + }, + { + "epoch": 2.6848653667595173, + "grad_norm": 1.7309263944625854, + "learning_rate": 1e-06, + "loss": 0.3425, + "mean_token_accuracy": 0.8842539191246033, + "num_tokens": 507598185.0, + "step": 14458 + }, + { + "epoch": 2.685051067780873, + "grad_norm": 1.4955894947052002, + "learning_rate": 1e-06, + "loss": 0.2635, + "mean_token_accuracy": 0.9050657749176025, + "num_tokens": 507633244.0, + "step": 14459 + }, + { + "epoch": 2.6852367688022283, + "grad_norm": 1.6312257051467896, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8897573351860046, + "num_tokens": 507663147.0, + "step": 14460 + }, + { + "epoch": 2.685422469823584, + "grad_norm": 1.4592151641845703, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8842768669128418, + "num_tokens": 507707562.0, + "step": 14461 + }, + { + "epoch": 2.6856081708449397, + "grad_norm": 1.5081744194030762, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8885871171951294, + "num_tokens": 507740729.0, + "step": 14462 + }, + { + "epoch": 2.685793871866295, + "grad_norm": 1.6666847467422485, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8760815858840942, + "num_tokens": 507772467.0, + "step": 14463 + }, + { + "epoch": 2.6859795728876508, + "grad_norm": 1.5017093420028687, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8899283409118652, + "num_tokens": 507813090.0, + "step": 14464 + }, + { + "epoch": 2.6861652739090065, + "grad_norm": 1.4169834852218628, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8937168121337891, + "num_tokens": 507853379.0, + "step": 14465 + }, + { + "epoch": 2.6863509749303622, + "grad_norm": 1.5512195825576782, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8811784982681274, + "num_tokens": 507889094.0, + "step": 14466 + }, + { + "epoch": 2.686536675951718, + "grad_norm": 1.4485584497451782, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8942573666572571, + "num_tokens": 507926894.0, + "step": 14467 + }, + { + "epoch": 2.6867223769730733, + "grad_norm": 1.6686514616012573, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8703866004943848, + "num_tokens": 507961669.0, + "step": 14468 + }, + { + "epoch": 2.686908077994429, + "grad_norm": 1.5124263763427734, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8857715725898743, + "num_tokens": 507998931.0, + "step": 14469 + }, + { + "epoch": 2.6870937790157847, + "grad_norm": 1.504006028175354, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.893822193145752, + "num_tokens": 508038306.0, + "step": 14470 + }, + { + "epoch": 2.68727948003714, + "grad_norm": 1.493641972541809, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8824945688247681, + "num_tokens": 508077993.0, + "step": 14471 + }, + { + "epoch": 2.6874651810584957, + "grad_norm": 1.6044929027557373, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8894123435020447, + "num_tokens": 508112821.0, + "step": 14472 + }, + { + "epoch": 2.6876508820798515, + "grad_norm": 1.7067761421203613, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8840073347091675, + "num_tokens": 508143515.0, + "step": 14473 + }, + { + "epoch": 2.687836583101207, + "grad_norm": 1.6633652448654175, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8809278011322021, + "num_tokens": 508176764.0, + "step": 14474 + }, + { + "epoch": 2.688022284122563, + "grad_norm": 1.568320631980896, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8828660845756531, + "num_tokens": 508215436.0, + "step": 14475 + }, + { + "epoch": 2.6882079851439182, + "grad_norm": 1.6212241649627686, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8857976794242859, + "num_tokens": 508252002.0, + "step": 14476 + }, + { + "epoch": 2.688393686165274, + "grad_norm": 1.6339478492736816, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8785635232925415, + "num_tokens": 508284070.0, + "step": 14477 + }, + { + "epoch": 2.6885793871866293, + "grad_norm": 1.6098419427871704, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8837950229644775, + "num_tokens": 508314138.0, + "step": 14478 + }, + { + "epoch": 2.688765088207985, + "grad_norm": 1.5084995031356812, + "learning_rate": 1e-06, + "loss": 0.2984, + "mean_token_accuracy": 0.893271803855896, + "num_tokens": 508348837.0, + "step": 14479 + }, + { + "epoch": 2.6889507892293407, + "grad_norm": 1.6645524501800537, + "learning_rate": 1e-06, + "loss": 0.2384, + "mean_token_accuracy": 0.9144433736801147, + "num_tokens": 508374929.0, + "step": 14480 + }, + { + "epoch": 2.6891364902506965, + "grad_norm": 1.5666438341140747, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8762689828872681, + "num_tokens": 508412733.0, + "step": 14481 + }, + { + "epoch": 2.689322191272052, + "grad_norm": 1.4035707712173462, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8831163048744202, + "num_tokens": 508454410.0, + "step": 14482 + }, + { + "epoch": 2.6895078922934075, + "grad_norm": 1.4976727962493896, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.872755765914917, + "num_tokens": 508496049.0, + "step": 14483 + }, + { + "epoch": 2.689693593314763, + "grad_norm": 1.4900453090667725, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8813568353652954, + "num_tokens": 508533585.0, + "step": 14484 + }, + { + "epoch": 2.689879294336119, + "grad_norm": 1.4583473205566406, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8807891607284546, + "num_tokens": 508575975.0, + "step": 14485 + }, + { + "epoch": 2.6900649953574742, + "grad_norm": 1.6216561794281006, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8861010670661926, + "num_tokens": 508606236.0, + "step": 14486 + }, + { + "epoch": 2.69025069637883, + "grad_norm": 1.4558995962142944, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8982571363449097, + "num_tokens": 508643825.0, + "step": 14487 + }, + { + "epoch": 2.6904363974001857, + "grad_norm": 1.499413013458252, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8962928652763367, + "num_tokens": 508680557.0, + "step": 14488 + }, + { + "epoch": 2.6906220984215414, + "grad_norm": 1.7093874216079712, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8935841917991638, + "num_tokens": 508710383.0, + "step": 14489 + }, + { + "epoch": 2.690807799442897, + "grad_norm": 1.5281803607940674, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8906586766242981, + "num_tokens": 508741898.0, + "step": 14490 + }, + { + "epoch": 2.6909935004642525, + "grad_norm": 1.6385297775268555, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.894909143447876, + "num_tokens": 508769332.0, + "step": 14491 + }, + { + "epoch": 2.691179201485608, + "grad_norm": 1.5184745788574219, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8838779926300049, + "num_tokens": 508807422.0, + "step": 14492 + }, + { + "epoch": 2.691364902506964, + "grad_norm": 1.7023775577545166, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8909901976585388, + "num_tokens": 508834631.0, + "step": 14493 + }, + { + "epoch": 2.691550603528319, + "grad_norm": 1.4858522415161133, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8838722705841064, + "num_tokens": 508870707.0, + "step": 14494 + }, + { + "epoch": 2.691736304549675, + "grad_norm": 1.4684537649154663, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8883336186408997, + "num_tokens": 508908284.0, + "step": 14495 + }, + { + "epoch": 2.6919220055710307, + "grad_norm": 1.612833023071289, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8926031589508057, + "num_tokens": 508942883.0, + "step": 14496 + }, + { + "epoch": 2.6921077065923864, + "grad_norm": 1.6696531772613525, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8869335651397705, + "num_tokens": 508974564.0, + "step": 14497 + }, + { + "epoch": 2.692293407613742, + "grad_norm": 1.6081396341323853, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8756328821182251, + "num_tokens": 509011105.0, + "step": 14498 + }, + { + "epoch": 2.6924791086350974, + "grad_norm": 1.508988618850708, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.88844895362854, + "num_tokens": 509045981.0, + "step": 14499 + }, + { + "epoch": 2.692664809656453, + "grad_norm": 1.423309564590454, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8922492265701294, + "num_tokens": 509082420.0, + "step": 14500 + }, + { + "epoch": 2.6928505106778085, + "grad_norm": 1.5327554941177368, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8772798776626587, + "num_tokens": 509122461.0, + "step": 14501 + }, + { + "epoch": 2.693036211699164, + "grad_norm": 1.6017508506774902, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8872295022010803, + "num_tokens": 509155585.0, + "step": 14502 + }, + { + "epoch": 2.69322191272052, + "grad_norm": 1.5372189283370972, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8946027755737305, + "num_tokens": 509191043.0, + "step": 14503 + }, + { + "epoch": 2.6934076137418757, + "grad_norm": 1.4793956279754639, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8919191360473633, + "num_tokens": 509231066.0, + "step": 14504 + }, + { + "epoch": 2.6935933147632314, + "grad_norm": 1.5010137557983398, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.8956369161605835, + "num_tokens": 509264852.0, + "step": 14505 + }, + { + "epoch": 2.6937790157845867, + "grad_norm": 1.4981671571731567, + "learning_rate": 1e-06, + "loss": 0.274, + "mean_token_accuracy": 0.9009020924568176, + "num_tokens": 509297229.0, + "step": 14506 + }, + { + "epoch": 2.6939647168059424, + "grad_norm": 1.5153719186782837, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8894373178482056, + "num_tokens": 509336808.0, + "step": 14507 + }, + { + "epoch": 2.694150417827298, + "grad_norm": 1.5876396894454956, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8976761698722839, + "num_tokens": 509370320.0, + "step": 14508 + }, + { + "epoch": 2.6943361188486534, + "grad_norm": 1.5954183340072632, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8960578441619873, + "num_tokens": 509402040.0, + "step": 14509 + }, + { + "epoch": 2.694521819870009, + "grad_norm": 1.57696533203125, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8819485306739807, + "num_tokens": 509440373.0, + "step": 14510 + }, + { + "epoch": 2.694707520891365, + "grad_norm": 1.6283059120178223, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8844193816184998, + "num_tokens": 509476086.0, + "step": 14511 + }, + { + "epoch": 2.6948932219127206, + "grad_norm": 1.536712646484375, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8906785249710083, + "num_tokens": 509512797.0, + "step": 14512 + }, + { + "epoch": 2.6950789229340764, + "grad_norm": 1.6226013898849487, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8771131038665771, + "num_tokens": 509549737.0, + "step": 14513 + }, + { + "epoch": 2.6952646239554316, + "grad_norm": 1.666502594947815, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8896046876907349, + "num_tokens": 509581430.0, + "step": 14514 + }, + { + "epoch": 2.6954503249767874, + "grad_norm": 1.606013536453247, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.8965398073196411, + "num_tokens": 509613880.0, + "step": 14515 + }, + { + "epoch": 2.695636025998143, + "grad_norm": 1.5620636940002441, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.8900803327560425, + "num_tokens": 509650672.0, + "step": 14516 + }, + { + "epoch": 2.6958217270194984, + "grad_norm": 1.5305676460266113, + "learning_rate": 1e-06, + "loss": 0.3394, + "mean_token_accuracy": 0.8810794353485107, + "num_tokens": 509688410.0, + "step": 14517 + }, + { + "epoch": 2.696007428040854, + "grad_norm": 1.6500321626663208, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8828990459442139, + "num_tokens": 509719406.0, + "step": 14518 + }, + { + "epoch": 2.69619312906221, + "grad_norm": 1.4318811893463135, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8921166062355042, + "num_tokens": 509760500.0, + "step": 14519 + }, + { + "epoch": 2.6963788300835656, + "grad_norm": 1.688299536705017, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8746446371078491, + "num_tokens": 509788997.0, + "step": 14520 + }, + { + "epoch": 2.6965645311049213, + "grad_norm": 1.5334651470184326, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8854420185089111, + "num_tokens": 509825904.0, + "step": 14521 + }, + { + "epoch": 2.6967502321262766, + "grad_norm": 1.4585191011428833, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8791549205780029, + "num_tokens": 509867192.0, + "step": 14522 + }, + { + "epoch": 2.6969359331476324, + "grad_norm": 1.789414405822754, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8842371106147766, + "num_tokens": 509895692.0, + "step": 14523 + }, + { + "epoch": 2.697121634168988, + "grad_norm": 1.4983075857162476, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8985124826431274, + "num_tokens": 509932834.0, + "step": 14524 + }, + { + "epoch": 2.6973073351903434, + "grad_norm": 1.4992494583129883, + "learning_rate": 1e-06, + "loss": 0.2747, + "mean_token_accuracy": 0.9011852741241455, + "num_tokens": 509968585.0, + "step": 14525 + }, + { + "epoch": 2.697493036211699, + "grad_norm": 1.6725800037384033, + "learning_rate": 1e-06, + "loss": 0.3737, + "mean_token_accuracy": 0.8755789995193481, + "num_tokens": 510004191.0, + "step": 14526 + }, + { + "epoch": 2.697678737233055, + "grad_norm": 1.5126274824142456, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8829033374786377, + "num_tokens": 510041830.0, + "step": 14527 + }, + { + "epoch": 2.6978644382544106, + "grad_norm": 1.6420748233795166, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8916418552398682, + "num_tokens": 510078749.0, + "step": 14528 + }, + { + "epoch": 2.698050139275766, + "grad_norm": 1.7325875759124756, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8945012092590332, + "num_tokens": 510105020.0, + "step": 14529 + }, + { + "epoch": 2.6982358402971216, + "grad_norm": 1.612381935119629, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8811919093132019, + "num_tokens": 510142249.0, + "step": 14530 + }, + { + "epoch": 2.6984215413184773, + "grad_norm": 1.4578726291656494, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.8945602774620056, + "num_tokens": 510182515.0, + "step": 14531 + }, + { + "epoch": 2.6986072423398326, + "grad_norm": 1.6487153768539429, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8880155682563782, + "num_tokens": 510216633.0, + "step": 14532 + }, + { + "epoch": 2.6987929433611884, + "grad_norm": 1.4441629648208618, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8822090029716492, + "num_tokens": 510260367.0, + "step": 14533 + }, + { + "epoch": 2.698978644382544, + "grad_norm": 1.5236587524414062, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8849985599517822, + "num_tokens": 510300328.0, + "step": 14534 + }, + { + "epoch": 2.6991643454039, + "grad_norm": 1.6713916063308716, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8827888369560242, + "num_tokens": 510338638.0, + "step": 14535 + }, + { + "epoch": 2.6993500464252556, + "grad_norm": 1.669346809387207, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8753183484077454, + "num_tokens": 510373906.0, + "step": 14536 + }, + { + "epoch": 2.699535747446611, + "grad_norm": 1.485392689704895, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8921265602111816, + "num_tokens": 510408670.0, + "step": 14537 + }, + { + "epoch": 2.6997214484679666, + "grad_norm": 1.6428337097167969, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8803007006645203, + "num_tokens": 510443534.0, + "step": 14538 + }, + { + "epoch": 2.6999071494893223, + "grad_norm": 1.443350076675415, + "learning_rate": 1e-06, + "loss": 0.2672, + "mean_token_accuracy": 0.9055468440055847, + "num_tokens": 510477975.0, + "step": 14539 + }, + { + "epoch": 2.7000928505106776, + "grad_norm": 1.6558748483657837, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8777250647544861, + "num_tokens": 510514979.0, + "step": 14540 + }, + { + "epoch": 2.7002785515320333, + "grad_norm": 1.593531847000122, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8776721358299255, + "num_tokens": 510549441.0, + "step": 14541 + }, + { + "epoch": 2.700464252553389, + "grad_norm": 1.557494878768921, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8895266056060791, + "num_tokens": 510585702.0, + "step": 14542 + }, + { + "epoch": 2.700649953574745, + "grad_norm": 1.7449458837509155, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8916501998901367, + "num_tokens": 510613597.0, + "step": 14543 + }, + { + "epoch": 2.7008356545961005, + "grad_norm": 1.6195420026779175, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.882709264755249, + "num_tokens": 510645678.0, + "step": 14544 + }, + { + "epoch": 2.701021355617456, + "grad_norm": 1.711145281791687, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8847696781158447, + "num_tokens": 510674386.0, + "step": 14545 + }, + { + "epoch": 2.7012070566388116, + "grad_norm": 1.5936962366104126, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8854242563247681, + "num_tokens": 510715609.0, + "step": 14546 + }, + { + "epoch": 2.7013927576601673, + "grad_norm": 1.636608362197876, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8762969970703125, + "num_tokens": 510750278.0, + "step": 14547 + }, + { + "epoch": 2.7015784586815226, + "grad_norm": 1.795074701309204, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8869577646255493, + "num_tokens": 510780789.0, + "step": 14548 + }, + { + "epoch": 2.7017641597028783, + "grad_norm": 1.5517666339874268, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.8934825658798218, + "num_tokens": 510814774.0, + "step": 14549 + }, + { + "epoch": 2.701949860724234, + "grad_norm": 1.5647269487380981, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.8964401483535767, + "num_tokens": 510848760.0, + "step": 14550 + }, + { + "epoch": 2.7021355617455898, + "grad_norm": 1.575644850730896, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8936922550201416, + "num_tokens": 510880924.0, + "step": 14551 + }, + { + "epoch": 2.702321262766945, + "grad_norm": 1.604134202003479, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8828443884849548, + "num_tokens": 510916600.0, + "step": 14552 + }, + { + "epoch": 2.702506963788301, + "grad_norm": 1.5669946670532227, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8915693759918213, + "num_tokens": 510950833.0, + "step": 14553 + }, + { + "epoch": 2.7026926648096565, + "grad_norm": 1.479445219039917, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8838330507278442, + "num_tokens": 510996752.0, + "step": 14554 + }, + { + "epoch": 2.702878365831012, + "grad_norm": 1.490369200706482, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8796508312225342, + "num_tokens": 511038455.0, + "step": 14555 + }, + { + "epoch": 2.7030640668523676, + "grad_norm": 1.5302056074142456, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8920174837112427, + "num_tokens": 511077646.0, + "step": 14556 + }, + { + "epoch": 2.7032497678737233, + "grad_norm": 1.6254069805145264, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8758557438850403, + "num_tokens": 511113494.0, + "step": 14557 + }, + { + "epoch": 2.703435468895079, + "grad_norm": 1.605916976928711, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.889533519744873, + "num_tokens": 511144090.0, + "step": 14558 + }, + { + "epoch": 2.7036211699164348, + "grad_norm": 1.5457119941711426, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8656587600708008, + "num_tokens": 511183805.0, + "step": 14559 + }, + { + "epoch": 2.70380687093779, + "grad_norm": 1.6575279235839844, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.872002363204956, + "num_tokens": 511215869.0, + "step": 14560 + }, + { + "epoch": 2.7039925719591458, + "grad_norm": 1.474319338798523, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8816425204277039, + "num_tokens": 511253783.0, + "step": 14561 + }, + { + "epoch": 2.7041782729805015, + "grad_norm": 1.721990704536438, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8827831745147705, + "num_tokens": 511286502.0, + "step": 14562 + }, + { + "epoch": 2.704363974001857, + "grad_norm": 1.7520660161972046, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8776770830154419, + "num_tokens": 511317599.0, + "step": 14563 + }, + { + "epoch": 2.7045496750232125, + "grad_norm": 1.4854466915130615, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8925111293792725, + "num_tokens": 511357924.0, + "step": 14564 + }, + { + "epoch": 2.7047353760445683, + "grad_norm": 1.400715947151184, + "learning_rate": 1e-06, + "loss": 0.2828, + "mean_token_accuracy": 0.8966525793075562, + "num_tokens": 511394380.0, + "step": 14565 + }, + { + "epoch": 2.704921077065924, + "grad_norm": 1.6105048656463623, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.889957845211029, + "num_tokens": 511424909.0, + "step": 14566 + }, + { + "epoch": 2.7051067780872797, + "grad_norm": 1.4699618816375732, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8832347989082336, + "num_tokens": 511466239.0, + "step": 14567 + }, + { + "epoch": 2.705292479108635, + "grad_norm": 1.4842396974563599, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8825436234474182, + "num_tokens": 511507445.0, + "step": 14568 + }, + { + "epoch": 2.7054781801299908, + "grad_norm": 1.4932231903076172, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8930193185806274, + "num_tokens": 511541852.0, + "step": 14569 + }, + { + "epoch": 2.7056638811513465, + "grad_norm": 1.45601487159729, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.888898491859436, + "num_tokens": 511580209.0, + "step": 14570 + }, + { + "epoch": 2.7058495821727018, + "grad_norm": 1.5923765897750854, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8704801797866821, + "num_tokens": 511618636.0, + "step": 14571 + }, + { + "epoch": 2.7060352831940575, + "grad_norm": 1.6792858839035034, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8836073875427246, + "num_tokens": 511651989.0, + "step": 14572 + }, + { + "epoch": 2.7062209842154132, + "grad_norm": 1.5018237829208374, + "learning_rate": 1e-06, + "loss": 0.3135, + "mean_token_accuracy": 0.8902983069419861, + "num_tokens": 511686487.0, + "step": 14573 + }, + { + "epoch": 2.706406685236769, + "grad_norm": 1.513398289680481, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8820981979370117, + "num_tokens": 511723939.0, + "step": 14574 + }, + { + "epoch": 2.7065923862581243, + "grad_norm": 1.6624805927276611, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8812063336372375, + "num_tokens": 511759318.0, + "step": 14575 + }, + { + "epoch": 2.70677808727948, + "grad_norm": 1.601792573928833, + "learning_rate": 1e-06, + "loss": 0.279, + "mean_token_accuracy": 0.9021072387695312, + "num_tokens": 511788102.0, + "step": 14576 + }, + { + "epoch": 2.7069637883008357, + "grad_norm": 1.612187385559082, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8805952072143555, + "num_tokens": 511821647.0, + "step": 14577 + }, + { + "epoch": 2.707149489322191, + "grad_norm": 1.5646580457687378, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8819072246551514, + "num_tokens": 511857483.0, + "step": 14578 + }, + { + "epoch": 2.7073351903435467, + "grad_norm": 1.511736512184143, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8878927230834961, + "num_tokens": 511893585.0, + "step": 14579 + }, + { + "epoch": 2.7075208913649025, + "grad_norm": 1.6200666427612305, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8839371204376221, + "num_tokens": 511929200.0, + "step": 14580 + }, + { + "epoch": 2.707706592386258, + "grad_norm": 1.489654541015625, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8989359140396118, + "num_tokens": 511964228.0, + "step": 14581 + }, + { + "epoch": 2.707892293407614, + "grad_norm": 1.5024954080581665, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8872436881065369, + "num_tokens": 512003061.0, + "step": 14582 + }, + { + "epoch": 2.7080779944289692, + "grad_norm": 1.5998183488845825, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8900012373924255, + "num_tokens": 512035664.0, + "step": 14583 + }, + { + "epoch": 2.708263695450325, + "grad_norm": 1.456882357597351, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8788397312164307, + "num_tokens": 512078888.0, + "step": 14584 + }, + { + "epoch": 2.7084493964716807, + "grad_norm": 1.7799896001815796, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8815060257911682, + "num_tokens": 512105668.0, + "step": 14585 + }, + { + "epoch": 2.708635097493036, + "grad_norm": 1.6477891206741333, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8889952898025513, + "num_tokens": 512141398.0, + "step": 14586 + }, + { + "epoch": 2.7088207985143917, + "grad_norm": 1.6609047651290894, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8846564292907715, + "num_tokens": 512171938.0, + "step": 14587 + }, + { + "epoch": 2.7090064995357475, + "grad_norm": 1.5208181142807007, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8780452013015747, + "num_tokens": 512211853.0, + "step": 14588 + }, + { + "epoch": 2.709192200557103, + "grad_norm": 1.7111718654632568, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8850754499435425, + "num_tokens": 512242683.0, + "step": 14589 + }, + { + "epoch": 2.709377901578459, + "grad_norm": 1.4984654188156128, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8910558819770813, + "num_tokens": 512278554.0, + "step": 14590 + }, + { + "epoch": 2.709563602599814, + "grad_norm": 1.4662519693374634, + "learning_rate": 1e-06, + "loss": 0.249, + "mean_token_accuracy": 0.9109005331993103, + "num_tokens": 512312219.0, + "step": 14591 + }, + { + "epoch": 2.70974930362117, + "grad_norm": 1.594824194908142, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8895732164382935, + "num_tokens": 512344369.0, + "step": 14592 + }, + { + "epoch": 2.7099350046425257, + "grad_norm": 2.006239652633667, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8782563805580139, + "num_tokens": 512378565.0, + "step": 14593 + }, + { + "epoch": 2.710120705663881, + "grad_norm": 1.4347001314163208, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8975335955619812, + "num_tokens": 512415889.0, + "step": 14594 + }, + { + "epoch": 2.7103064066852367, + "grad_norm": 1.5085018873214722, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8873244524002075, + "num_tokens": 512450386.0, + "step": 14595 + }, + { + "epoch": 2.7104921077065924, + "grad_norm": 1.4452694654464722, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.894191563129425, + "num_tokens": 512485583.0, + "step": 14596 + }, + { + "epoch": 2.710677808727948, + "grad_norm": 1.5730769634246826, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8817446231842041, + "num_tokens": 512521683.0, + "step": 14597 + }, + { + "epoch": 2.7108635097493035, + "grad_norm": 1.5450390577316284, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8982473611831665, + "num_tokens": 512561727.0, + "step": 14598 + }, + { + "epoch": 2.711049210770659, + "grad_norm": 1.5104140043258667, + "learning_rate": 1e-06, + "loss": 0.301, + "mean_token_accuracy": 0.8937753438949585, + "num_tokens": 512602624.0, + "step": 14599 + }, + { + "epoch": 2.711234911792015, + "grad_norm": 1.6791006326675415, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8789048194885254, + "num_tokens": 512635327.0, + "step": 14600 + }, + { + "epoch": 2.71142061281337, + "grad_norm": 1.5277162790298462, + "learning_rate": 1e-06, + "loss": 0.2805, + "mean_token_accuracy": 0.9020577669143677, + "num_tokens": 512667059.0, + "step": 14601 + }, + { + "epoch": 2.711606313834726, + "grad_norm": 1.4564907550811768, + "learning_rate": 1e-06, + "loss": 0.2891, + "mean_token_accuracy": 0.8997132182121277, + "num_tokens": 512704820.0, + "step": 14602 + }, + { + "epoch": 2.7117920148560817, + "grad_norm": 1.491767168045044, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.885563850402832, + "num_tokens": 512746389.0, + "step": 14603 + }, + { + "epoch": 2.7119777158774374, + "grad_norm": 1.6290754079818726, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8915597796440125, + "num_tokens": 512777008.0, + "step": 14604 + }, + { + "epoch": 2.712163416898793, + "grad_norm": 1.5577327013015747, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8774479031562805, + "num_tokens": 512813264.0, + "step": 14605 + }, + { + "epoch": 2.7123491179201484, + "grad_norm": 1.6499332189559937, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8884790539741516, + "num_tokens": 512846968.0, + "step": 14606 + }, + { + "epoch": 2.712534818941504, + "grad_norm": 1.5990521907806396, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8883341550827026, + "num_tokens": 512881425.0, + "step": 14607 + }, + { + "epoch": 2.71272051996286, + "grad_norm": 1.5860718488693237, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8879039287567139, + "num_tokens": 512912386.0, + "step": 14608 + }, + { + "epoch": 2.712906220984215, + "grad_norm": 1.5969926118850708, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8891854882240295, + "num_tokens": 512949015.0, + "step": 14609 + }, + { + "epoch": 2.713091922005571, + "grad_norm": 1.5435689687728882, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8891034126281738, + "num_tokens": 512984455.0, + "step": 14610 + }, + { + "epoch": 2.7132776230269267, + "grad_norm": 1.5477041006088257, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8887872695922852, + "num_tokens": 513019079.0, + "step": 14611 + }, + { + "epoch": 2.7134633240482824, + "grad_norm": 1.6867314577102661, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8798616528511047, + "num_tokens": 513052607.0, + "step": 14612 + }, + { + "epoch": 2.713649025069638, + "grad_norm": 1.4627184867858887, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8900653719902039, + "num_tokens": 513092810.0, + "step": 14613 + }, + { + "epoch": 2.7138347260909934, + "grad_norm": 1.647257685661316, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8895361423492432, + "num_tokens": 513122530.0, + "step": 14614 + }, + { + "epoch": 2.714020427112349, + "grad_norm": 1.5508660078048706, + "learning_rate": 1e-06, + "loss": 0.2664, + "mean_token_accuracy": 0.9035733938217163, + "num_tokens": 513154672.0, + "step": 14615 + }, + { + "epoch": 2.714206128133705, + "grad_norm": 1.6061654090881348, + "learning_rate": 1e-06, + "loss": 0.3836, + "mean_token_accuracy": 0.8665414452552795, + "num_tokens": 513190882.0, + "step": 14616 + }, + { + "epoch": 2.71439182915506, + "grad_norm": 1.5587985515594482, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8890300989151001, + "num_tokens": 513225777.0, + "step": 14617 + }, + { + "epoch": 2.714577530176416, + "grad_norm": 1.6196030378341675, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.881550669670105, + "num_tokens": 513261169.0, + "step": 14618 + }, + { + "epoch": 2.7147632311977716, + "grad_norm": 1.6080290079116821, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8930084705352783, + "num_tokens": 513298906.0, + "step": 14619 + }, + { + "epoch": 2.7149489322191274, + "grad_norm": 1.499518871307373, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8856001496315002, + "num_tokens": 513335339.0, + "step": 14620 + }, + { + "epoch": 2.715134633240483, + "grad_norm": 1.6674504280090332, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8872467279434204, + "num_tokens": 513364864.0, + "step": 14621 + }, + { + "epoch": 2.7153203342618384, + "grad_norm": 1.6262412071228027, + "learning_rate": 1e-06, + "loss": 0.368, + "mean_token_accuracy": 0.8714851140975952, + "num_tokens": 513400843.0, + "step": 14622 + }, + { + "epoch": 2.715506035283194, + "grad_norm": 1.505881428718567, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.8963009119033813, + "num_tokens": 513433463.0, + "step": 14623 + }, + { + "epoch": 2.7156917363045494, + "grad_norm": 1.5250391960144043, + "learning_rate": 1e-06, + "loss": 0.3064, + "mean_token_accuracy": 0.8914825320243835, + "num_tokens": 513471434.0, + "step": 14624 + }, + { + "epoch": 2.715877437325905, + "grad_norm": 1.5348658561706543, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8870817422866821, + "num_tokens": 513510689.0, + "step": 14625 + }, + { + "epoch": 2.716063138347261, + "grad_norm": 1.6313881874084473, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8741233348846436, + "num_tokens": 513547126.0, + "step": 14626 + }, + { + "epoch": 2.7162488393686166, + "grad_norm": 1.4845627546310425, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8969076871871948, + "num_tokens": 513583825.0, + "step": 14627 + }, + { + "epoch": 2.7164345403899723, + "grad_norm": 1.539844274520874, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.878311038017273, + "num_tokens": 513621380.0, + "step": 14628 + }, + { + "epoch": 2.7166202414113276, + "grad_norm": 1.6562482118606567, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.880551278591156, + "num_tokens": 513651720.0, + "step": 14629 + }, + { + "epoch": 2.7168059424326834, + "grad_norm": 1.5475353002548218, + "learning_rate": 1e-06, + "loss": 0.2758, + "mean_token_accuracy": 0.9024418592453003, + "num_tokens": 513683733.0, + "step": 14630 + }, + { + "epoch": 2.716991643454039, + "grad_norm": 1.4651304483413696, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8915331959724426, + "num_tokens": 513722491.0, + "step": 14631 + }, + { + "epoch": 2.7171773444753944, + "grad_norm": 1.7686703205108643, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8798701763153076, + "num_tokens": 513755865.0, + "step": 14632 + }, + { + "epoch": 2.71736304549675, + "grad_norm": 1.644527554512024, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8776575326919556, + "num_tokens": 513788026.0, + "step": 14633 + }, + { + "epoch": 2.717548746518106, + "grad_norm": 1.544547438621521, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.878643274307251, + "num_tokens": 513827283.0, + "step": 14634 + }, + { + "epoch": 2.7177344475394616, + "grad_norm": 1.6702386140823364, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8781145215034485, + "num_tokens": 513859992.0, + "step": 14635 + }, + { + "epoch": 2.7179201485608173, + "grad_norm": 1.49202299118042, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8801965713500977, + "num_tokens": 513900970.0, + "step": 14636 + }, + { + "epoch": 2.7181058495821726, + "grad_norm": 1.6644577980041504, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8863569498062134, + "num_tokens": 513935044.0, + "step": 14637 + }, + { + "epoch": 2.7182915506035283, + "grad_norm": 1.6295995712280273, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8843733072280884, + "num_tokens": 513968085.0, + "step": 14638 + }, + { + "epoch": 2.718477251624884, + "grad_norm": 1.5268480777740479, + "learning_rate": 1e-06, + "loss": 0.2711, + "mean_token_accuracy": 0.9011284112930298, + "num_tokens": 513999736.0, + "step": 14639 + }, + { + "epoch": 2.7186629526462394, + "grad_norm": 1.596634864807129, + "learning_rate": 1e-06, + "loss": 0.3831, + "mean_token_accuracy": 0.8655548095703125, + "num_tokens": 514041956.0, + "step": 14640 + }, + { + "epoch": 2.718848653667595, + "grad_norm": 1.623558521270752, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.884404718875885, + "num_tokens": 514076253.0, + "step": 14641 + }, + { + "epoch": 2.719034354688951, + "grad_norm": 1.5775439739227295, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8736786842346191, + "num_tokens": 514113531.0, + "step": 14642 + }, + { + "epoch": 2.7192200557103066, + "grad_norm": 1.56496262550354, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8891007900238037, + "num_tokens": 514148678.0, + "step": 14643 + }, + { + "epoch": 2.7194057567316623, + "grad_norm": 1.610558271408081, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8925944566726685, + "num_tokens": 514178636.0, + "step": 14644 + }, + { + "epoch": 2.7195914577530176, + "grad_norm": 1.4859628677368164, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8736324310302734, + "num_tokens": 514217988.0, + "step": 14645 + }, + { + "epoch": 2.7197771587743733, + "grad_norm": 1.4698208570480347, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8846625089645386, + "num_tokens": 514263300.0, + "step": 14646 + }, + { + "epoch": 2.7199628597957286, + "grad_norm": 1.6142346858978271, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8859831094741821, + "num_tokens": 514296669.0, + "step": 14647 + }, + { + "epoch": 2.7201485608170843, + "grad_norm": 1.7374529838562012, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8844849467277527, + "num_tokens": 514325116.0, + "step": 14648 + }, + { + "epoch": 2.72033426183844, + "grad_norm": 1.4763169288635254, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8883254528045654, + "num_tokens": 514360866.0, + "step": 14649 + }, + { + "epoch": 2.720519962859796, + "grad_norm": 1.5732065439224243, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8914618492126465, + "num_tokens": 514393106.0, + "step": 14650 + }, + { + "epoch": 2.7207056638811515, + "grad_norm": 1.6094655990600586, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8891329169273376, + "num_tokens": 514429574.0, + "step": 14651 + }, + { + "epoch": 2.720891364902507, + "grad_norm": 1.4251726865768433, + "learning_rate": 1e-06, + "loss": 0.2887, + "mean_token_accuracy": 0.895958662033081, + "num_tokens": 514467837.0, + "step": 14652 + }, + { + "epoch": 2.7210770659238626, + "grad_norm": 1.5149723291397095, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8906965255737305, + "num_tokens": 514505960.0, + "step": 14653 + }, + { + "epoch": 2.7212627669452183, + "grad_norm": 1.3971902132034302, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8944871425628662, + "num_tokens": 514546992.0, + "step": 14654 + }, + { + "epoch": 2.7214484679665736, + "grad_norm": 1.4641655683517456, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8980060815811157, + "num_tokens": 514585644.0, + "step": 14655 + }, + { + "epoch": 2.7216341689879293, + "grad_norm": 1.5215036869049072, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8815840482711792, + "num_tokens": 514623384.0, + "step": 14656 + }, + { + "epoch": 2.721819870009285, + "grad_norm": 1.633692741394043, + "learning_rate": 1e-06, + "loss": 0.3572, + "mean_token_accuracy": 0.87957364320755, + "num_tokens": 514657521.0, + "step": 14657 + }, + { + "epoch": 2.722005571030641, + "grad_norm": 1.5450119972229004, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8797460794448853, + "num_tokens": 514692107.0, + "step": 14658 + }, + { + "epoch": 2.7221912720519965, + "grad_norm": 1.5421191453933716, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8824152946472168, + "num_tokens": 514730254.0, + "step": 14659 + }, + { + "epoch": 2.722376973073352, + "grad_norm": 1.740478754043579, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8883723020553589, + "num_tokens": 514762009.0, + "step": 14660 + }, + { + "epoch": 2.7225626740947075, + "grad_norm": 1.5788122415542603, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8835804462432861, + "num_tokens": 514795581.0, + "step": 14661 + }, + { + "epoch": 2.7227483751160633, + "grad_norm": 1.7586407661437988, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8914468288421631, + "num_tokens": 514820336.0, + "step": 14662 + }, + { + "epoch": 2.7229340761374186, + "grad_norm": 1.683169960975647, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8845199942588806, + "num_tokens": 514852239.0, + "step": 14663 + }, + { + "epoch": 2.7231197771587743, + "grad_norm": 1.4665247201919556, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8913235068321228, + "num_tokens": 514889840.0, + "step": 14664 + }, + { + "epoch": 2.72330547818013, + "grad_norm": 1.631488561630249, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8838800191879272, + "num_tokens": 514923133.0, + "step": 14665 + }, + { + "epoch": 2.7234911792014858, + "grad_norm": 1.54737389087677, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8834092617034912, + "num_tokens": 514959412.0, + "step": 14666 + }, + { + "epoch": 2.7236768802228415, + "grad_norm": 1.5960917472839355, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8740084171295166, + "num_tokens": 514997553.0, + "step": 14667 + }, + { + "epoch": 2.723862581244197, + "grad_norm": 1.513113260269165, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8914217352867126, + "num_tokens": 515036507.0, + "step": 14668 + }, + { + "epoch": 2.7240482822655525, + "grad_norm": 1.7905168533325195, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.9022221565246582, + "num_tokens": 515060721.0, + "step": 14669 + }, + { + "epoch": 2.724233983286908, + "grad_norm": 1.711898684501648, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8970743417739868, + "num_tokens": 515088653.0, + "step": 14670 + }, + { + "epoch": 2.7244196843082635, + "grad_norm": 1.96273672580719, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8702077865600586, + "num_tokens": 515114564.0, + "step": 14671 + }, + { + "epoch": 2.7246053853296193, + "grad_norm": 1.3821077346801758, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.900049090385437, + "num_tokens": 515155048.0, + "step": 14672 + }, + { + "epoch": 2.724791086350975, + "grad_norm": 1.6067488193511963, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8705161809921265, + "num_tokens": 515189975.0, + "step": 14673 + }, + { + "epoch": 2.7249767873723307, + "grad_norm": 1.5775572061538696, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8885234594345093, + "num_tokens": 515224478.0, + "step": 14674 + }, + { + "epoch": 2.725162488393686, + "grad_norm": 1.7311515808105469, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8977652788162231, + "num_tokens": 515249709.0, + "step": 14675 + }, + { + "epoch": 2.7253481894150418, + "grad_norm": 1.5948885679244995, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8920835256576538, + "num_tokens": 515284864.0, + "step": 14676 + }, + { + "epoch": 2.7255338904363975, + "grad_norm": 1.6564853191375732, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8875964283943176, + "num_tokens": 515317838.0, + "step": 14677 + }, + { + "epoch": 2.7257195914577528, + "grad_norm": 1.5965123176574707, + "learning_rate": 1e-06, + "loss": 0.3016, + "mean_token_accuracy": 0.8961958289146423, + "num_tokens": 515349217.0, + "step": 14678 + }, + { + "epoch": 2.7259052924791085, + "grad_norm": 1.624640703201294, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.8921254873275757, + "num_tokens": 515380892.0, + "step": 14679 + }, + { + "epoch": 2.7260909935004642, + "grad_norm": 1.589521050453186, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8866831064224243, + "num_tokens": 515416876.0, + "step": 14680 + }, + { + "epoch": 2.72627669452182, + "grad_norm": 1.635165810585022, + "learning_rate": 1e-06, + "loss": 0.3398, + "mean_token_accuracy": 0.8824239373207092, + "num_tokens": 515450173.0, + "step": 14681 + }, + { + "epoch": 2.7264623955431757, + "grad_norm": 1.5754458904266357, + "learning_rate": 1e-06, + "loss": 0.2729, + "mean_token_accuracy": 0.9039684534072876, + "num_tokens": 515479804.0, + "step": 14682 + }, + { + "epoch": 2.726648096564531, + "grad_norm": 1.6552973985671997, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8792816400527954, + "num_tokens": 515512599.0, + "step": 14683 + }, + { + "epoch": 2.7268337975858867, + "grad_norm": 1.7310081720352173, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8902835845947266, + "num_tokens": 515537112.0, + "step": 14684 + }, + { + "epoch": 2.7270194986072425, + "grad_norm": 1.5967795848846436, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8845028877258301, + "num_tokens": 515572032.0, + "step": 14685 + }, + { + "epoch": 2.7272051996285978, + "grad_norm": 1.5569671392440796, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8918381929397583, + "num_tokens": 515603717.0, + "step": 14686 + }, + { + "epoch": 2.7273909006499535, + "grad_norm": 1.6933561563491821, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.875217080116272, + "num_tokens": 515636652.0, + "step": 14687 + }, + { + "epoch": 2.727576601671309, + "grad_norm": 1.5178645849227905, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8754673600196838, + "num_tokens": 515675600.0, + "step": 14688 + }, + { + "epoch": 2.727762302692665, + "grad_norm": 1.587388038635254, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8923389315605164, + "num_tokens": 515709516.0, + "step": 14689 + }, + { + "epoch": 2.7279480037140207, + "grad_norm": 1.7330894470214844, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.884964108467102, + "num_tokens": 515743999.0, + "step": 14690 + }, + { + "epoch": 2.728133704735376, + "grad_norm": 1.4798680543899536, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8821827173233032, + "num_tokens": 515784833.0, + "step": 14691 + }, + { + "epoch": 2.7283194057567317, + "grad_norm": 1.3777942657470703, + "learning_rate": 1e-06, + "loss": 0.2735, + "mean_token_accuracy": 0.9027823805809021, + "num_tokens": 515819521.0, + "step": 14692 + }, + { + "epoch": 2.7285051067780874, + "grad_norm": 1.647141695022583, + "learning_rate": 1e-06, + "loss": 0.2929, + "mean_token_accuracy": 0.8960818648338318, + "num_tokens": 515848474.0, + "step": 14693 + }, + { + "epoch": 2.7286908077994427, + "grad_norm": 1.4741939306259155, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.8976336717605591, + "num_tokens": 515885698.0, + "step": 14694 + }, + { + "epoch": 2.7288765088207985, + "grad_norm": 1.534452199935913, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.881662130355835, + "num_tokens": 515922929.0, + "step": 14695 + }, + { + "epoch": 2.729062209842154, + "grad_norm": 1.5996665954589844, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.879231333732605, + "num_tokens": 515958618.0, + "step": 14696 + }, + { + "epoch": 2.72924791086351, + "grad_norm": 1.5852234363555908, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8852683901786804, + "num_tokens": 515992442.0, + "step": 14697 + }, + { + "epoch": 2.729433611884865, + "grad_norm": 1.7240501642227173, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.8969964385032654, + "num_tokens": 516021647.0, + "step": 14698 + }, + { + "epoch": 2.729619312906221, + "grad_norm": 1.6105260848999023, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8874279260635376, + "num_tokens": 516057626.0, + "step": 14699 + }, + { + "epoch": 2.7298050139275767, + "grad_norm": 1.7009516954421997, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8906888365745544, + "num_tokens": 516086399.0, + "step": 14700 + }, + { + "epoch": 2.729990714948932, + "grad_norm": 1.584366798400879, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8820135593414307, + "num_tokens": 516120655.0, + "step": 14701 + }, + { + "epoch": 2.7301764159702877, + "grad_norm": 1.6087456941604614, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.889477014541626, + "num_tokens": 516151940.0, + "step": 14702 + }, + { + "epoch": 2.7303621169916434, + "grad_norm": 1.538055419921875, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8886117935180664, + "num_tokens": 516191877.0, + "step": 14703 + }, + { + "epoch": 2.730547818012999, + "grad_norm": 1.5773067474365234, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8881357908248901, + "num_tokens": 516228141.0, + "step": 14704 + }, + { + "epoch": 2.730733519034355, + "grad_norm": 1.6606465578079224, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8930337429046631, + "num_tokens": 516260310.0, + "step": 14705 + }, + { + "epoch": 2.73091922005571, + "grad_norm": 1.5584911108016968, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8881243467330933, + "num_tokens": 516294876.0, + "step": 14706 + }, + { + "epoch": 2.731104921077066, + "grad_norm": 1.5133165121078491, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.884678840637207, + "num_tokens": 516334463.0, + "step": 14707 + }, + { + "epoch": 2.7312906220984217, + "grad_norm": 1.5418494939804077, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8776556253433228, + "num_tokens": 516371197.0, + "step": 14708 + }, + { + "epoch": 2.731476323119777, + "grad_norm": 1.6001149415969849, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.880937397480011, + "num_tokens": 516405668.0, + "step": 14709 + }, + { + "epoch": 2.7316620241411327, + "grad_norm": 1.5407862663269043, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8811373710632324, + "num_tokens": 516440985.0, + "step": 14710 + }, + { + "epoch": 2.7318477251624884, + "grad_norm": 1.5641108751296997, + "learning_rate": 1e-06, + "loss": 0.3558, + "mean_token_accuracy": 0.8804866075515747, + "num_tokens": 516479031.0, + "step": 14711 + }, + { + "epoch": 2.732033426183844, + "grad_norm": 1.5727053880691528, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8833154439926147, + "num_tokens": 516520711.0, + "step": 14712 + }, + { + "epoch": 2.7322191272052, + "grad_norm": 1.5289630889892578, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8806232213973999, + "num_tokens": 516558975.0, + "step": 14713 + }, + { + "epoch": 2.732404828226555, + "grad_norm": 1.395652174949646, + "learning_rate": 1e-06, + "loss": 0.2905, + "mean_token_accuracy": 0.8980283141136169, + "num_tokens": 516600983.0, + "step": 14714 + }, + { + "epoch": 2.732590529247911, + "grad_norm": 1.6033291816711426, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8846232891082764, + "num_tokens": 516636729.0, + "step": 14715 + }, + { + "epoch": 2.7327762302692666, + "grad_norm": 1.4441827535629272, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.8943172693252563, + "num_tokens": 516670410.0, + "step": 14716 + }, + { + "epoch": 2.732961931290622, + "grad_norm": 1.6876801252365112, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8724752068519592, + "num_tokens": 516703227.0, + "step": 14717 + }, + { + "epoch": 2.7331476323119777, + "grad_norm": 1.6679179668426514, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8579146862030029, + "num_tokens": 516738107.0, + "step": 14718 + }, + { + "epoch": 2.7333333333333334, + "grad_norm": 1.3670989274978638, + "learning_rate": 1e-06, + "loss": 0.2643, + "mean_token_accuracy": 0.9034085273742676, + "num_tokens": 516777527.0, + "step": 14719 + }, + { + "epoch": 2.733519034354689, + "grad_norm": 1.5575299263000488, + "learning_rate": 1e-06, + "loss": 0.32, + "mean_token_accuracy": 0.8890373110771179, + "num_tokens": 516816007.0, + "step": 14720 + }, + { + "epoch": 2.7337047353760444, + "grad_norm": 1.5778717994689941, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8834476470947266, + "num_tokens": 516852732.0, + "step": 14721 + }, + { + "epoch": 2.7338904363974, + "grad_norm": 1.5980174541473389, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.8913437128067017, + "num_tokens": 516886702.0, + "step": 14722 + }, + { + "epoch": 2.734076137418756, + "grad_norm": 1.3636565208435059, + "learning_rate": 1e-06, + "loss": 0.2869, + "mean_token_accuracy": 0.8959130048751831, + "num_tokens": 516926717.0, + "step": 14723 + }, + { + "epoch": 2.734261838440111, + "grad_norm": 1.5777318477630615, + "learning_rate": 1e-06, + "loss": 0.2773, + "mean_token_accuracy": 0.9023332595825195, + "num_tokens": 516959428.0, + "step": 14724 + }, + { + "epoch": 2.734447539461467, + "grad_norm": 1.4595003128051758, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8855923414230347, + "num_tokens": 517001317.0, + "step": 14725 + }, + { + "epoch": 2.7346332404828226, + "grad_norm": 1.748645305633545, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8721415996551514, + "num_tokens": 517031970.0, + "step": 14726 + }, + { + "epoch": 2.7348189415041784, + "grad_norm": 1.4797492027282715, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8900954723358154, + "num_tokens": 517070609.0, + "step": 14727 + }, + { + "epoch": 2.735004642525534, + "grad_norm": 1.5820294618606567, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8900170922279358, + "num_tokens": 517104457.0, + "step": 14728 + }, + { + "epoch": 2.7351903435468894, + "grad_norm": 1.6915773153305054, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8918348550796509, + "num_tokens": 517136151.0, + "step": 14729 + }, + { + "epoch": 2.735376044568245, + "grad_norm": 1.555031180381775, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8870089054107666, + "num_tokens": 517171037.0, + "step": 14730 + }, + { + "epoch": 2.735561745589601, + "grad_norm": 1.5506101846694946, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8869727849960327, + "num_tokens": 517208149.0, + "step": 14731 + }, + { + "epoch": 2.735747446610956, + "grad_norm": 1.3718855381011963, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8920696973800659, + "num_tokens": 517249179.0, + "step": 14732 + }, + { + "epoch": 2.735933147632312, + "grad_norm": 1.484337329864502, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8921435475349426, + "num_tokens": 517286184.0, + "step": 14733 + }, + { + "epoch": 2.7361188486536676, + "grad_norm": 1.6030950546264648, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.8868817090988159, + "num_tokens": 517323803.0, + "step": 14734 + }, + { + "epoch": 2.7363045496750233, + "grad_norm": 1.4786971807479858, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8942716121673584, + "num_tokens": 517362551.0, + "step": 14735 + }, + { + "epoch": 2.736490250696379, + "grad_norm": 1.4661320447921753, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8885445594787598, + "num_tokens": 517400816.0, + "step": 14736 + }, + { + "epoch": 2.7366759517177344, + "grad_norm": 1.5338009595870972, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8879692554473877, + "num_tokens": 517440314.0, + "step": 14737 + }, + { + "epoch": 2.73686165273909, + "grad_norm": 1.6619466543197632, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8767206072807312, + "num_tokens": 517472608.0, + "step": 14738 + }, + { + "epoch": 2.737047353760446, + "grad_norm": 1.848788857460022, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8830995559692383, + "num_tokens": 517500035.0, + "step": 14739 + }, + { + "epoch": 2.737233054781801, + "grad_norm": 1.6468791961669922, + "learning_rate": 1e-06, + "loss": 0.3807, + "mean_token_accuracy": 0.8740314245223999, + "num_tokens": 517533447.0, + "step": 14740 + }, + { + "epoch": 2.737418755803157, + "grad_norm": 1.4473878145217896, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.896764874458313, + "num_tokens": 517571544.0, + "step": 14741 + }, + { + "epoch": 2.7376044568245126, + "grad_norm": 1.4980376958847046, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.9053906202316284, + "num_tokens": 517604772.0, + "step": 14742 + }, + { + "epoch": 2.7377901578458683, + "grad_norm": 1.5314562320709229, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8783053159713745, + "num_tokens": 517642761.0, + "step": 14743 + }, + { + "epoch": 2.7379758588672236, + "grad_norm": 1.385489583015442, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8917124271392822, + "num_tokens": 517684345.0, + "step": 14744 + }, + { + "epoch": 2.7381615598885793, + "grad_norm": 1.4832898378372192, + "learning_rate": 1e-06, + "loss": 0.2708, + "mean_token_accuracy": 0.9037202000617981, + "num_tokens": 517720457.0, + "step": 14745 + }, + { + "epoch": 2.738347260909935, + "grad_norm": 1.5075266361236572, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8846097588539124, + "num_tokens": 517756891.0, + "step": 14746 + }, + { + "epoch": 2.7385329619312904, + "grad_norm": 1.4903053045272827, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8926380276679993, + "num_tokens": 517790878.0, + "step": 14747 + }, + { + "epoch": 2.738718662952646, + "grad_norm": 1.6521638631820679, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8789108395576477, + "num_tokens": 517823542.0, + "step": 14748 + }, + { + "epoch": 2.738904363974002, + "grad_norm": 1.548112154006958, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.877476692199707, + "num_tokens": 517859064.0, + "step": 14749 + }, + { + "epoch": 2.7390900649953576, + "grad_norm": 1.500915765762329, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8720371723175049, + "num_tokens": 517902714.0, + "step": 14750 + }, + { + "epoch": 2.7392757660167133, + "grad_norm": 1.6204673051834106, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8927894830703735, + "num_tokens": 517932356.0, + "step": 14751 + }, + { + "epoch": 2.7394614670380686, + "grad_norm": 1.623360514640808, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.8968510627746582, + "num_tokens": 517964147.0, + "step": 14752 + }, + { + "epoch": 2.7396471680594243, + "grad_norm": 1.5407963991165161, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.893771231174469, + "num_tokens": 517998097.0, + "step": 14753 + }, + { + "epoch": 2.73983286908078, + "grad_norm": 1.597009539604187, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8817960619926453, + "num_tokens": 518033623.0, + "step": 14754 + }, + { + "epoch": 2.7400185701021353, + "grad_norm": 1.658639907836914, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8774171471595764, + "num_tokens": 518069036.0, + "step": 14755 + }, + { + "epoch": 2.740204271123491, + "grad_norm": 1.5543664693832397, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8823258280754089, + "num_tokens": 518106375.0, + "step": 14756 + }, + { + "epoch": 2.740389972144847, + "grad_norm": 1.4293831586837769, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8822853565216064, + "num_tokens": 518146264.0, + "step": 14757 + }, + { + "epoch": 2.7405756731662025, + "grad_norm": 1.6426773071289062, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8807190656661987, + "num_tokens": 518178217.0, + "step": 14758 + }, + { + "epoch": 2.7407613741875583, + "grad_norm": 1.6605552434921265, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8876194953918457, + "num_tokens": 518208395.0, + "step": 14759 + }, + { + "epoch": 2.7409470752089136, + "grad_norm": 1.4703004360198975, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8822239637374878, + "num_tokens": 518249690.0, + "step": 14760 + }, + { + "epoch": 2.7411327762302693, + "grad_norm": 1.6542508602142334, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8978923559188843, + "num_tokens": 518277265.0, + "step": 14761 + }, + { + "epoch": 2.741318477251625, + "grad_norm": 1.513542652130127, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8748154640197754, + "num_tokens": 518314271.0, + "step": 14762 + }, + { + "epoch": 2.7415041782729803, + "grad_norm": 1.3850455284118652, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.9000136256217957, + "num_tokens": 518353339.0, + "step": 14763 + }, + { + "epoch": 2.741689879294336, + "grad_norm": 1.4709149599075317, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8792168498039246, + "num_tokens": 518397075.0, + "step": 14764 + }, + { + "epoch": 2.741875580315692, + "grad_norm": 1.5165774822235107, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8854029178619385, + "num_tokens": 518433787.0, + "step": 14765 + }, + { + "epoch": 2.7420612813370475, + "grad_norm": 1.6240625381469727, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8791686296463013, + "num_tokens": 518467658.0, + "step": 14766 + }, + { + "epoch": 2.742246982358403, + "grad_norm": 1.4914098978042603, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8927934169769287, + "num_tokens": 518504986.0, + "step": 14767 + }, + { + "epoch": 2.7424326833797585, + "grad_norm": 1.6301990747451782, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8952312469482422, + "num_tokens": 518537177.0, + "step": 14768 + }, + { + "epoch": 2.7426183844011143, + "grad_norm": 1.721712589263916, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8748394250869751, + "num_tokens": 518570655.0, + "step": 14769 + }, + { + "epoch": 2.7428040854224696, + "grad_norm": 1.479884147644043, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8878434300422668, + "num_tokens": 518614044.0, + "step": 14770 + }, + { + "epoch": 2.7429897864438253, + "grad_norm": 1.518027663230896, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8869292736053467, + "num_tokens": 518651464.0, + "step": 14771 + }, + { + "epoch": 2.743175487465181, + "grad_norm": 1.6286944150924683, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8756333589553833, + "num_tokens": 518686351.0, + "step": 14772 + }, + { + "epoch": 2.7433611884865368, + "grad_norm": 1.503456950187683, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8872895240783691, + "num_tokens": 518722783.0, + "step": 14773 + }, + { + "epoch": 2.7435468895078925, + "grad_norm": 1.6651901006698608, + "learning_rate": 1e-06, + "loss": 0.3359, + "mean_token_accuracy": 0.8819942474365234, + "num_tokens": 518753858.0, + "step": 14774 + }, + { + "epoch": 2.743732590529248, + "grad_norm": 1.42386794090271, + "learning_rate": 1e-06, + "loss": 0.2659, + "mean_token_accuracy": 0.9029724597930908, + "num_tokens": 518789004.0, + "step": 14775 + }, + { + "epoch": 2.7439182915506035, + "grad_norm": 1.5299227237701416, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8898768424987793, + "num_tokens": 518823727.0, + "step": 14776 + }, + { + "epoch": 2.7441039925719592, + "grad_norm": 1.6652247905731201, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8999688625335693, + "num_tokens": 518857053.0, + "step": 14777 + }, + { + "epoch": 2.7442896935933145, + "grad_norm": 1.4888291358947754, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8874490857124329, + "num_tokens": 518895308.0, + "step": 14778 + }, + { + "epoch": 2.7444753946146703, + "grad_norm": 1.4551390409469604, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8909229040145874, + "num_tokens": 518932115.0, + "step": 14779 + }, + { + "epoch": 2.744661095636026, + "grad_norm": 1.6388776302337646, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8882127404212952, + "num_tokens": 518961081.0, + "step": 14780 + }, + { + "epoch": 2.7448467966573817, + "grad_norm": 1.5583635568618774, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8870788812637329, + "num_tokens": 518995625.0, + "step": 14781 + }, + { + "epoch": 2.7450324976787375, + "grad_norm": 1.6580743789672852, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8870298266410828, + "num_tokens": 519024801.0, + "step": 14782 + }, + { + "epoch": 2.7452181987000928, + "grad_norm": 1.476949691772461, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.886626124382019, + "num_tokens": 519068126.0, + "step": 14783 + }, + { + "epoch": 2.7454038997214485, + "grad_norm": 1.4332187175750732, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8960763216018677, + "num_tokens": 519105980.0, + "step": 14784 + }, + { + "epoch": 2.7455896007428042, + "grad_norm": 1.5325568914413452, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8828269839286804, + "num_tokens": 519141394.0, + "step": 14785 + }, + { + "epoch": 2.7457753017641595, + "grad_norm": 1.48885977268219, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8743919134140015, + "num_tokens": 519182846.0, + "step": 14786 + }, + { + "epoch": 2.7459610027855152, + "grad_norm": 1.604906439781189, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8733682036399841, + "num_tokens": 519221057.0, + "step": 14787 + }, + { + "epoch": 2.746146703806871, + "grad_norm": 1.5675585269927979, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8861249685287476, + "num_tokens": 519255393.0, + "step": 14788 + }, + { + "epoch": 2.7463324048282267, + "grad_norm": 1.528232216835022, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8945348262786865, + "num_tokens": 519293657.0, + "step": 14789 + }, + { + "epoch": 2.7465181058495824, + "grad_norm": 1.5474891662597656, + "learning_rate": 1e-06, + "loss": 0.3319, + "mean_token_accuracy": 0.8843808770179749, + "num_tokens": 519331818.0, + "step": 14790 + }, + { + "epoch": 2.7467038068709377, + "grad_norm": 1.5455535650253296, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8720163702964783, + "num_tokens": 519372714.0, + "step": 14791 + }, + { + "epoch": 2.7468895078922935, + "grad_norm": 1.620489239692688, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8890380859375, + "num_tokens": 519405039.0, + "step": 14792 + }, + { + "epoch": 2.7470752089136488, + "grad_norm": 1.5286787748336792, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8927069902420044, + "num_tokens": 519440981.0, + "step": 14793 + }, + { + "epoch": 2.7472609099350045, + "grad_norm": 1.4624100923538208, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8876251578330994, + "num_tokens": 519480942.0, + "step": 14794 + }, + { + "epoch": 2.7474466109563602, + "grad_norm": 1.601048231124878, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8842096328735352, + "num_tokens": 519515833.0, + "step": 14795 + }, + { + "epoch": 2.747632311977716, + "grad_norm": 1.5316640138626099, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8835639953613281, + "num_tokens": 519551335.0, + "step": 14796 + }, + { + "epoch": 2.7478180129990717, + "grad_norm": 1.5818642377853394, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8883827924728394, + "num_tokens": 519584864.0, + "step": 14797 + }, + { + "epoch": 2.748003714020427, + "grad_norm": 1.5245552062988281, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.888187050819397, + "num_tokens": 519620203.0, + "step": 14798 + }, + { + "epoch": 2.7481894150417827, + "grad_norm": 1.5657031536102295, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8906601667404175, + "num_tokens": 519653447.0, + "step": 14799 + }, + { + "epoch": 2.7483751160631384, + "grad_norm": 1.4748754501342773, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8939171433448792, + "num_tokens": 519689736.0, + "step": 14800 + }, + { + "epoch": 2.7485608170844937, + "grad_norm": 1.4654697179794312, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8947889804840088, + "num_tokens": 519723423.0, + "step": 14801 + }, + { + "epoch": 2.7487465181058495, + "grad_norm": 1.7365896701812744, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8773597478866577, + "num_tokens": 519756217.0, + "step": 14802 + }, + { + "epoch": 2.748932219127205, + "grad_norm": 1.5938889980316162, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8829928636550903, + "num_tokens": 519791536.0, + "step": 14803 + }, + { + "epoch": 2.749117920148561, + "grad_norm": 1.6214731931686401, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8760175704956055, + "num_tokens": 519827431.0, + "step": 14804 + }, + { + "epoch": 2.7493036211699167, + "grad_norm": 1.521369218826294, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8952547907829285, + "num_tokens": 519861485.0, + "step": 14805 + }, + { + "epoch": 2.749489322191272, + "grad_norm": 1.563491702079773, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.902161717414856, + "num_tokens": 519896910.0, + "step": 14806 + }, + { + "epoch": 2.7496750232126277, + "grad_norm": 1.641022801399231, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.8786609172821045, + "num_tokens": 519931940.0, + "step": 14807 + }, + { + "epoch": 2.7498607242339834, + "grad_norm": 1.556462287902832, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8842746019363403, + "num_tokens": 519969479.0, + "step": 14808 + }, + { + "epoch": 2.7500464252553387, + "grad_norm": 1.5179353952407837, + "learning_rate": 1e-06, + "loss": 0.2659, + "mean_token_accuracy": 0.9043882489204407, + "num_tokens": 520001966.0, + "step": 14809 + }, + { + "epoch": 2.7502321262766944, + "grad_norm": 1.4988811016082764, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8938689231872559, + "num_tokens": 520037321.0, + "step": 14810 + }, + { + "epoch": 2.75041782729805, + "grad_norm": 1.6084553003311157, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8864532709121704, + "num_tokens": 520070308.0, + "step": 14811 + }, + { + "epoch": 2.750603528319406, + "grad_norm": 1.6752972602844238, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8781956434249878, + "num_tokens": 520104442.0, + "step": 14812 + }, + { + "epoch": 2.7507892293407616, + "grad_norm": 1.5771770477294922, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8879760503768921, + "num_tokens": 520139715.0, + "step": 14813 + }, + { + "epoch": 2.750974930362117, + "grad_norm": 1.5417066812515259, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8879406452178955, + "num_tokens": 520173910.0, + "step": 14814 + }, + { + "epoch": 2.7511606313834727, + "grad_norm": 1.4997427463531494, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8817800879478455, + "num_tokens": 520214627.0, + "step": 14815 + }, + { + "epoch": 2.751346332404828, + "grad_norm": 1.6296777725219727, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8899799585342407, + "num_tokens": 520247195.0, + "step": 14816 + }, + { + "epoch": 2.7515320334261837, + "grad_norm": 1.639888048171997, + "learning_rate": 1e-06, + "loss": 0.3556, + "mean_token_accuracy": 0.8794679641723633, + "num_tokens": 520284488.0, + "step": 14817 + }, + { + "epoch": 2.7517177344475394, + "grad_norm": 1.4124035835266113, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8904545903205872, + "num_tokens": 520325886.0, + "step": 14818 + }, + { + "epoch": 2.751903435468895, + "grad_norm": 1.4433598518371582, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8815563321113586, + "num_tokens": 520367499.0, + "step": 14819 + }, + { + "epoch": 2.752089136490251, + "grad_norm": 1.5295761823654175, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8964019417762756, + "num_tokens": 520403326.0, + "step": 14820 + }, + { + "epoch": 2.752274837511606, + "grad_norm": 1.453517198562622, + "learning_rate": 1e-06, + "loss": 0.2966, + "mean_token_accuracy": 0.893783450126648, + "num_tokens": 520440802.0, + "step": 14821 + }, + { + "epoch": 2.752460538532962, + "grad_norm": 1.4515234231948853, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8954992294311523, + "num_tokens": 520478282.0, + "step": 14822 + }, + { + "epoch": 2.7526462395543176, + "grad_norm": 1.5539830923080444, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8814867734909058, + "num_tokens": 520516312.0, + "step": 14823 + }, + { + "epoch": 2.752831940575673, + "grad_norm": 1.4264265298843384, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8913207650184631, + "num_tokens": 520555831.0, + "step": 14824 + }, + { + "epoch": 2.7530176415970287, + "grad_norm": 1.5713834762573242, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8878349661827087, + "num_tokens": 520596118.0, + "step": 14825 + }, + { + "epoch": 2.7532033426183844, + "grad_norm": 1.684701681137085, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8800156116485596, + "num_tokens": 520628288.0, + "step": 14826 + }, + { + "epoch": 2.75338904363974, + "grad_norm": 1.5169737339019775, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8841919898986816, + "num_tokens": 520662693.0, + "step": 14827 + }, + { + "epoch": 2.753574744661096, + "grad_norm": 1.467090129852295, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.8828566670417786, + "num_tokens": 520706062.0, + "step": 14828 + }, + { + "epoch": 2.753760445682451, + "grad_norm": 1.6052972078323364, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.8957945704460144, + "num_tokens": 520737937.0, + "step": 14829 + }, + { + "epoch": 2.753946146703807, + "grad_norm": 1.4169334173202515, + "learning_rate": 1e-06, + "loss": 0.2763, + "mean_token_accuracy": 0.9043761491775513, + "num_tokens": 520773565.0, + "step": 14830 + }, + { + "epoch": 2.7541318477251626, + "grad_norm": 1.4895074367523193, + "learning_rate": 1e-06, + "loss": 0.3623, + "mean_token_accuracy": 0.8739545941352844, + "num_tokens": 520815313.0, + "step": 14831 + }, + { + "epoch": 2.754317548746518, + "grad_norm": 1.5410715341567993, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8847991824150085, + "num_tokens": 520851266.0, + "step": 14832 + }, + { + "epoch": 2.7545032497678736, + "grad_norm": 1.6038457155227661, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.890151858329773, + "num_tokens": 520885863.0, + "step": 14833 + }, + { + "epoch": 2.7546889507892294, + "grad_norm": 1.6905646324157715, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.876914918422699, + "num_tokens": 520916939.0, + "step": 14834 + }, + { + "epoch": 2.754874651810585, + "grad_norm": 1.467700719833374, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.896746814250946, + "num_tokens": 520956603.0, + "step": 14835 + }, + { + "epoch": 2.755060352831941, + "grad_norm": 1.5881013870239258, + "learning_rate": 1e-06, + "loss": 0.379, + "mean_token_accuracy": 0.8680496215820312, + "num_tokens": 520993072.0, + "step": 14836 + }, + { + "epoch": 2.755246053853296, + "grad_norm": 1.5664219856262207, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8846623301506042, + "num_tokens": 521027328.0, + "step": 14837 + }, + { + "epoch": 2.755431754874652, + "grad_norm": 1.734695315361023, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.8962200880050659, + "num_tokens": 521053661.0, + "step": 14838 + }, + { + "epoch": 2.755617455896007, + "grad_norm": 1.4870344400405884, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8790624141693115, + "num_tokens": 521097162.0, + "step": 14839 + }, + { + "epoch": 2.755803156917363, + "grad_norm": 1.574293613433838, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8927927017211914, + "num_tokens": 521130489.0, + "step": 14840 + }, + { + "epoch": 2.7559888579387186, + "grad_norm": 1.5132417678833008, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8765518665313721, + "num_tokens": 521175214.0, + "step": 14841 + }, + { + "epoch": 2.7561745589600744, + "grad_norm": 1.6625546216964722, + "learning_rate": 1e-06, + "loss": 0.3854, + "mean_token_accuracy": 0.8675087690353394, + "num_tokens": 521211684.0, + "step": 14842 + }, + { + "epoch": 2.75636025998143, + "grad_norm": 1.609938621520996, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8895779848098755, + "num_tokens": 521243517.0, + "step": 14843 + }, + { + "epoch": 2.7565459610027854, + "grad_norm": 1.4699243307113647, + "learning_rate": 1e-06, + "loss": 0.2942, + "mean_token_accuracy": 0.8986005783081055, + "num_tokens": 521276804.0, + "step": 14844 + }, + { + "epoch": 2.756731662024141, + "grad_norm": 1.5234588384628296, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8733233213424683, + "num_tokens": 521317702.0, + "step": 14845 + }, + { + "epoch": 2.756917363045497, + "grad_norm": 1.5011699199676514, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8861774206161499, + "num_tokens": 521356578.0, + "step": 14846 + }, + { + "epoch": 2.757103064066852, + "grad_norm": 1.3793630599975586, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8893039226531982, + "num_tokens": 521397663.0, + "step": 14847 + }, + { + "epoch": 2.757288765088208, + "grad_norm": 1.6369574069976807, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8804725408554077, + "num_tokens": 521430669.0, + "step": 14848 + }, + { + "epoch": 2.7574744661095636, + "grad_norm": 1.578177571296692, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8951976895332336, + "num_tokens": 521463395.0, + "step": 14849 + }, + { + "epoch": 2.7576601671309193, + "grad_norm": 1.681036353111267, + "learning_rate": 1e-06, + "loss": 0.2544, + "mean_token_accuracy": 0.9076712727546692, + "num_tokens": 521489883.0, + "step": 14850 + }, + { + "epoch": 2.757845868152275, + "grad_norm": 1.590864658355713, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8835903406143188, + "num_tokens": 521524772.0, + "step": 14851 + }, + { + "epoch": 2.7580315691736303, + "grad_norm": 1.6449353694915771, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.890769362449646, + "num_tokens": 521556215.0, + "step": 14852 + }, + { + "epoch": 2.758217270194986, + "grad_norm": 1.650355339050293, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.889568567276001, + "num_tokens": 521588708.0, + "step": 14853 + }, + { + "epoch": 2.758402971216342, + "grad_norm": 1.5628024339675903, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8969086408615112, + "num_tokens": 521623515.0, + "step": 14854 + }, + { + "epoch": 2.758588672237697, + "grad_norm": 1.7130680084228516, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8718481063842773, + "num_tokens": 521652588.0, + "step": 14855 + }, + { + "epoch": 2.758774373259053, + "grad_norm": 1.487487554550171, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.8991693258285522, + "num_tokens": 521690297.0, + "step": 14856 + }, + { + "epoch": 2.7589600742804086, + "grad_norm": 1.592742919921875, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8833655118942261, + "num_tokens": 521727194.0, + "step": 14857 + }, + { + "epoch": 2.7591457753017643, + "grad_norm": 1.531445026397705, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8945271968841553, + "num_tokens": 521761474.0, + "step": 14858 + }, + { + "epoch": 2.75933147632312, + "grad_norm": 1.4643019437789917, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8813366889953613, + "num_tokens": 521803453.0, + "step": 14859 + }, + { + "epoch": 2.7595171773444753, + "grad_norm": 1.5647984743118286, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8773468136787415, + "num_tokens": 521839429.0, + "step": 14860 + }, + { + "epoch": 2.759702878365831, + "grad_norm": 1.5578246116638184, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8902291059494019, + "num_tokens": 521875235.0, + "step": 14861 + }, + { + "epoch": 2.759888579387187, + "grad_norm": 1.6904460191726685, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8829957246780396, + "num_tokens": 521904863.0, + "step": 14862 + }, + { + "epoch": 2.760074280408542, + "grad_norm": 1.524982213973999, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8882235884666443, + "num_tokens": 521942462.0, + "step": 14863 + }, + { + "epoch": 2.760259981429898, + "grad_norm": 1.6205949783325195, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8980900049209595, + "num_tokens": 521974943.0, + "step": 14864 + }, + { + "epoch": 2.7604456824512535, + "grad_norm": 1.8027400970458984, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8728012442588806, + "num_tokens": 522006065.0, + "step": 14865 + }, + { + "epoch": 2.7606313834726093, + "grad_norm": 1.6394983530044556, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8932817578315735, + "num_tokens": 522038618.0, + "step": 14866 + }, + { + "epoch": 2.7608170844939646, + "grad_norm": 1.5761401653289795, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8867365121841431, + "num_tokens": 522071234.0, + "step": 14867 + }, + { + "epoch": 2.7610027855153203, + "grad_norm": 1.6185802221298218, + "learning_rate": 1e-06, + "loss": 0.3518, + "mean_token_accuracy": 0.8789790868759155, + "num_tokens": 522103289.0, + "step": 14868 + }, + { + "epoch": 2.761188486536676, + "grad_norm": 1.5011587142944336, + "learning_rate": 1e-06, + "loss": 0.3077, + "mean_token_accuracy": 0.8930535316467285, + "num_tokens": 522137631.0, + "step": 14869 + }, + { + "epoch": 2.7613741875580313, + "grad_norm": 1.6789498329162598, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8847990036010742, + "num_tokens": 522170515.0, + "step": 14870 + }, + { + "epoch": 2.761559888579387, + "grad_norm": 1.6295404434204102, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8863611221313477, + "num_tokens": 522201805.0, + "step": 14871 + }, + { + "epoch": 2.761745589600743, + "grad_norm": 1.4763977527618408, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8896019458770752, + "num_tokens": 522238809.0, + "step": 14872 + }, + { + "epoch": 2.7619312906220985, + "grad_norm": 1.61478853225708, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8796222805976868, + "num_tokens": 522272069.0, + "step": 14873 + }, + { + "epoch": 2.7621169916434543, + "grad_norm": 1.502624273300171, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8923808336257935, + "num_tokens": 522306512.0, + "step": 14874 + }, + { + "epoch": 2.7623026926648095, + "grad_norm": 1.680544137954712, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8860692381858826, + "num_tokens": 522340068.0, + "step": 14875 + }, + { + "epoch": 2.7624883936861653, + "grad_norm": 1.5494226217269897, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8907191157341003, + "num_tokens": 522372237.0, + "step": 14876 + }, + { + "epoch": 2.762674094707521, + "grad_norm": 1.6159182786941528, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8890820741653442, + "num_tokens": 522407393.0, + "step": 14877 + }, + { + "epoch": 2.7628597957288763, + "grad_norm": 1.5392160415649414, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8846175670623779, + "num_tokens": 522444843.0, + "step": 14878 + }, + { + "epoch": 2.763045496750232, + "grad_norm": 1.7082858085632324, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8771706819534302, + "num_tokens": 522477781.0, + "step": 14879 + }, + { + "epoch": 2.7632311977715878, + "grad_norm": 1.5711520910263062, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8863131999969482, + "num_tokens": 522510863.0, + "step": 14880 + }, + { + "epoch": 2.7634168987929435, + "grad_norm": 1.4104493856430054, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.897485613822937, + "num_tokens": 522550128.0, + "step": 14881 + }, + { + "epoch": 2.7636025998142992, + "grad_norm": 1.636823296546936, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8844822645187378, + "num_tokens": 522580940.0, + "step": 14882 + }, + { + "epoch": 2.7637883008356545, + "grad_norm": 1.574596881866455, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.886504054069519, + "num_tokens": 522615550.0, + "step": 14883 + }, + { + "epoch": 2.7639740018570103, + "grad_norm": 1.5239568948745728, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8959502577781677, + "num_tokens": 522651881.0, + "step": 14884 + }, + { + "epoch": 2.764159702878366, + "grad_norm": 1.5530755519866943, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8735684752464294, + "num_tokens": 522689241.0, + "step": 14885 + }, + { + "epoch": 2.7643454038997213, + "grad_norm": 1.4910591840744019, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8861281275749207, + "num_tokens": 522726728.0, + "step": 14886 + }, + { + "epoch": 2.764531104921077, + "grad_norm": 1.465144395828247, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8818469047546387, + "num_tokens": 522765073.0, + "step": 14887 + }, + { + "epoch": 2.7647168059424327, + "grad_norm": 1.555387020111084, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8869969248771667, + "num_tokens": 522800075.0, + "step": 14888 + }, + { + "epoch": 2.7649025069637885, + "grad_norm": 1.5639485120773315, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8925354480743408, + "num_tokens": 522835502.0, + "step": 14889 + }, + { + "epoch": 2.7650882079851438, + "grad_norm": 1.635114312171936, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8916242122650146, + "num_tokens": 522864283.0, + "step": 14890 + }, + { + "epoch": 2.7652739090064995, + "grad_norm": 1.4591847658157349, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8904350996017456, + "num_tokens": 522901971.0, + "step": 14891 + }, + { + "epoch": 2.7654596100278552, + "grad_norm": 1.6595243215560913, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8752015233039856, + "num_tokens": 522937411.0, + "step": 14892 + }, + { + "epoch": 2.7656453110492105, + "grad_norm": 1.667429804801941, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8782554268836975, + "num_tokens": 522973844.0, + "step": 14893 + }, + { + "epoch": 2.7658310120705663, + "grad_norm": 1.4507511854171753, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8913945555686951, + "num_tokens": 523010675.0, + "step": 14894 + }, + { + "epoch": 2.766016713091922, + "grad_norm": 1.5419743061065674, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8878816366195679, + "num_tokens": 523044640.0, + "step": 14895 + }, + { + "epoch": 2.7662024141132777, + "grad_norm": 1.4801503419876099, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8901695609092712, + "num_tokens": 523084342.0, + "step": 14896 + }, + { + "epoch": 2.7663881151346335, + "grad_norm": 1.5547642707824707, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8817211389541626, + "num_tokens": 523121181.0, + "step": 14897 + }, + { + "epoch": 2.7665738161559887, + "grad_norm": 1.4380658864974976, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8743370175361633, + "num_tokens": 523164804.0, + "step": 14898 + }, + { + "epoch": 2.7667595171773445, + "grad_norm": 1.4878549575805664, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8829548954963684, + "num_tokens": 523208484.0, + "step": 14899 + }, + { + "epoch": 2.7669452181987, + "grad_norm": 1.4830182790756226, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8886973857879639, + "num_tokens": 523246213.0, + "step": 14900 + }, + { + "epoch": 2.7671309192200555, + "grad_norm": 1.4287055730819702, + "learning_rate": 1e-06, + "loss": 0.2753, + "mean_token_accuracy": 0.9007370471954346, + "num_tokens": 523282802.0, + "step": 14901 + }, + { + "epoch": 2.7673166202414112, + "grad_norm": 1.564502239227295, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.888629138469696, + "num_tokens": 523316128.0, + "step": 14902 + }, + { + "epoch": 2.767502321262767, + "grad_norm": 1.6349629163742065, + "learning_rate": 1e-06, + "loss": 0.2833, + "mean_token_accuracy": 0.900202751159668, + "num_tokens": 523347659.0, + "step": 14903 + }, + { + "epoch": 2.7676880222841227, + "grad_norm": 1.6006927490234375, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8896169066429138, + "num_tokens": 523378509.0, + "step": 14904 + }, + { + "epoch": 2.7678737233054784, + "grad_norm": 1.5312529802322388, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8915626406669617, + "num_tokens": 523413491.0, + "step": 14905 + }, + { + "epoch": 2.7680594243268337, + "grad_norm": 1.556657075881958, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.890739917755127, + "num_tokens": 523449606.0, + "step": 14906 + }, + { + "epoch": 2.7682451253481895, + "grad_norm": 1.5151193141937256, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8935688734054565, + "num_tokens": 523486557.0, + "step": 14907 + }, + { + "epoch": 2.768430826369545, + "grad_norm": 1.4701294898986816, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8891592621803284, + "num_tokens": 523523459.0, + "step": 14908 + }, + { + "epoch": 2.7686165273909005, + "grad_norm": 1.473796010017395, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8819435834884644, + "num_tokens": 523564782.0, + "step": 14909 + }, + { + "epoch": 2.768802228412256, + "grad_norm": 1.7368053197860718, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8901669383049011, + "num_tokens": 523592707.0, + "step": 14910 + }, + { + "epoch": 2.768987929433612, + "grad_norm": 1.455881953239441, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8930439949035645, + "num_tokens": 523629243.0, + "step": 14911 + }, + { + "epoch": 2.7691736304549677, + "grad_norm": 1.5448161363601685, + "learning_rate": 1e-06, + "loss": 0.2929, + "mean_token_accuracy": 0.896710991859436, + "num_tokens": 523661672.0, + "step": 14912 + }, + { + "epoch": 2.769359331476323, + "grad_norm": 1.5013550519943237, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8874295353889465, + "num_tokens": 523701206.0, + "step": 14913 + }, + { + "epoch": 2.7695450324976787, + "grad_norm": 1.5098987817764282, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8946364521980286, + "num_tokens": 523739509.0, + "step": 14914 + }, + { + "epoch": 2.7697307335190344, + "grad_norm": 1.4847317934036255, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8906915187835693, + "num_tokens": 523776153.0, + "step": 14915 + }, + { + "epoch": 2.7699164345403897, + "grad_norm": 1.5223147869110107, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.8843628168106079, + "num_tokens": 523810493.0, + "step": 14916 + }, + { + "epoch": 2.7701021355617454, + "grad_norm": 1.4204574823379517, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8962776064872742, + "num_tokens": 523848030.0, + "step": 14917 + }, + { + "epoch": 2.770287836583101, + "grad_norm": 1.3653335571289062, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8883790969848633, + "num_tokens": 523892372.0, + "step": 14918 + }, + { + "epoch": 2.770473537604457, + "grad_norm": 1.5764390230178833, + "learning_rate": 1e-06, + "loss": 0.4045, + "mean_token_accuracy": 0.8602076768875122, + "num_tokens": 523931599.0, + "step": 14919 + }, + { + "epoch": 2.7706592386258126, + "grad_norm": 1.4893449544906616, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.9001538157463074, + "num_tokens": 523966948.0, + "step": 14920 + }, + { + "epoch": 2.770844939647168, + "grad_norm": 1.5917807817459106, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8921142816543579, + "num_tokens": 523997962.0, + "step": 14921 + }, + { + "epoch": 2.7710306406685237, + "grad_norm": 1.5313189029693604, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8808504343032837, + "num_tokens": 524032784.0, + "step": 14922 + }, + { + "epoch": 2.7712163416898794, + "grad_norm": 1.5326348543167114, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8808797597885132, + "num_tokens": 524069796.0, + "step": 14923 + }, + { + "epoch": 2.7714020427112347, + "grad_norm": 1.6709892749786377, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8714642524719238, + "num_tokens": 524106876.0, + "step": 14924 + }, + { + "epoch": 2.7715877437325904, + "grad_norm": 1.64032781124115, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8874849081039429, + "num_tokens": 524141640.0, + "step": 14925 + }, + { + "epoch": 2.771773444753946, + "grad_norm": 1.687668800354004, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8638665676116943, + "num_tokens": 524176877.0, + "step": 14926 + }, + { + "epoch": 2.771959145775302, + "grad_norm": 1.5606611967086792, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8932715058326721, + "num_tokens": 524209560.0, + "step": 14927 + }, + { + "epoch": 2.7721448467966576, + "grad_norm": 1.480139970779419, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8960338830947876, + "num_tokens": 524246733.0, + "step": 14928 + }, + { + "epoch": 2.772330547818013, + "grad_norm": 1.6005630493164062, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.9016473293304443, + "num_tokens": 524278080.0, + "step": 14929 + }, + { + "epoch": 2.7725162488393686, + "grad_norm": 1.4289547204971313, + "learning_rate": 1e-06, + "loss": 0.2765, + "mean_token_accuracy": 0.9008061289787292, + "num_tokens": 524316040.0, + "step": 14930 + }, + { + "epoch": 2.7727019498607244, + "grad_norm": 1.5986558198928833, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8758344054222107, + "num_tokens": 524350956.0, + "step": 14931 + }, + { + "epoch": 2.7728876508820797, + "grad_norm": 1.6614207029342651, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8848849534988403, + "num_tokens": 524383493.0, + "step": 14932 + }, + { + "epoch": 2.7730733519034354, + "grad_norm": 1.4982738494873047, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.895746111869812, + "num_tokens": 524422197.0, + "step": 14933 + }, + { + "epoch": 2.773259052924791, + "grad_norm": 1.4354240894317627, + "learning_rate": 1e-06, + "loss": 0.2754, + "mean_token_accuracy": 0.8998342156410217, + "num_tokens": 524457561.0, + "step": 14934 + }, + { + "epoch": 2.773444753946147, + "grad_norm": 1.6391880512237549, + "learning_rate": 1e-06, + "loss": 0.3451, + "mean_token_accuracy": 0.8802710771560669, + "num_tokens": 524489737.0, + "step": 14935 + }, + { + "epoch": 2.773630454967502, + "grad_norm": 1.578316330909729, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.8926585912704468, + "num_tokens": 524521146.0, + "step": 14936 + }, + { + "epoch": 2.773816155988858, + "grad_norm": 1.7434964179992676, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8854147791862488, + "num_tokens": 524552166.0, + "step": 14937 + }, + { + "epoch": 2.7740018570102136, + "grad_norm": 1.56466805934906, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8863906860351562, + "num_tokens": 524588957.0, + "step": 14938 + }, + { + "epoch": 2.774187558031569, + "grad_norm": 1.5252039432525635, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.8795856833457947, + "num_tokens": 524627113.0, + "step": 14939 + }, + { + "epoch": 2.7743732590529246, + "grad_norm": 1.4497284889221191, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8759028911590576, + "num_tokens": 524671129.0, + "step": 14940 + }, + { + "epoch": 2.7745589600742804, + "grad_norm": 1.6539846658706665, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8779476284980774, + "num_tokens": 524706472.0, + "step": 14941 + }, + { + "epoch": 2.774744661095636, + "grad_norm": 1.5174146890640259, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8873413801193237, + "num_tokens": 524743510.0, + "step": 14942 + }, + { + "epoch": 2.774930362116992, + "grad_norm": 1.4715415239334106, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8953592777252197, + "num_tokens": 524780910.0, + "step": 14943 + }, + { + "epoch": 2.775116063138347, + "grad_norm": 1.3699951171875, + "learning_rate": 1e-06, + "loss": 0.2735, + "mean_token_accuracy": 0.9015425443649292, + "num_tokens": 524819318.0, + "step": 14944 + }, + { + "epoch": 2.775301764159703, + "grad_norm": 1.4592176675796509, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8874648213386536, + "num_tokens": 524859265.0, + "step": 14945 + }, + { + "epoch": 2.7754874651810586, + "grad_norm": 1.5366168022155762, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8843003511428833, + "num_tokens": 524894333.0, + "step": 14946 + }, + { + "epoch": 2.775673166202414, + "grad_norm": 1.452487587928772, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8940203785896301, + "num_tokens": 524933730.0, + "step": 14947 + }, + { + "epoch": 2.7758588672237696, + "grad_norm": 1.665397047996521, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8695027828216553, + "num_tokens": 524968075.0, + "step": 14948 + }, + { + "epoch": 2.7760445682451254, + "grad_norm": 1.6283299922943115, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8844529390335083, + "num_tokens": 524999556.0, + "step": 14949 + }, + { + "epoch": 2.776230269266481, + "grad_norm": 1.5289448499679565, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8877788782119751, + "num_tokens": 525036033.0, + "step": 14950 + }, + { + "epoch": 2.776415970287837, + "grad_norm": 1.6680996417999268, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8729937672615051, + "num_tokens": 525071136.0, + "step": 14951 + }, + { + "epoch": 2.776601671309192, + "grad_norm": 1.4786711931228638, + "learning_rate": 1e-06, + "loss": 0.2793, + "mean_token_accuracy": 0.8977707624435425, + "num_tokens": 525109259.0, + "step": 14952 + }, + { + "epoch": 2.776787372330548, + "grad_norm": 1.5377013683319092, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8823589086532593, + "num_tokens": 525147100.0, + "step": 14953 + }, + { + "epoch": 2.7769730733519036, + "grad_norm": 1.5914733409881592, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8730373382568359, + "num_tokens": 525188171.0, + "step": 14954 + }, + { + "epoch": 2.777158774373259, + "grad_norm": 1.615474820137024, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8830251693725586, + "num_tokens": 525220420.0, + "step": 14955 + }, + { + "epoch": 2.7773444753946146, + "grad_norm": 1.5616254806518555, + "learning_rate": 1e-06, + "loss": 0.2898, + "mean_token_accuracy": 0.8989173769950867, + "num_tokens": 525255827.0, + "step": 14956 + }, + { + "epoch": 2.7775301764159703, + "grad_norm": 1.4293299913406372, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8776023387908936, + "num_tokens": 525298045.0, + "step": 14957 + }, + { + "epoch": 2.777715877437326, + "grad_norm": 1.6703284978866577, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8880143165588379, + "num_tokens": 525330391.0, + "step": 14958 + }, + { + "epoch": 2.777901578458682, + "grad_norm": 1.496181845664978, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8961342573165894, + "num_tokens": 525364844.0, + "step": 14959 + }, + { + "epoch": 2.778087279480037, + "grad_norm": 1.885591983795166, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8763458728790283, + "num_tokens": 525390824.0, + "step": 14960 + }, + { + "epoch": 2.778272980501393, + "grad_norm": 1.6675325632095337, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.871974766254425, + "num_tokens": 525424290.0, + "step": 14961 + }, + { + "epoch": 2.778458681522748, + "grad_norm": 1.6226245164871216, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8815842866897583, + "num_tokens": 525458500.0, + "step": 14962 + }, + { + "epoch": 2.778644382544104, + "grad_norm": 1.6197062730789185, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8891329765319824, + "num_tokens": 525491075.0, + "step": 14963 + }, + { + "epoch": 2.7788300835654596, + "grad_norm": 1.772572636604309, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8800030946731567, + "num_tokens": 525521961.0, + "step": 14964 + }, + { + "epoch": 2.7790157845868153, + "grad_norm": 1.6733026504516602, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8862057328224182, + "num_tokens": 525551346.0, + "step": 14965 + }, + { + "epoch": 2.779201485608171, + "grad_norm": 1.5794785022735596, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8868992924690247, + "num_tokens": 525592593.0, + "step": 14966 + }, + { + "epoch": 2.7793871866295263, + "grad_norm": 1.6826529502868652, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8801262378692627, + "num_tokens": 525622762.0, + "step": 14967 + }, + { + "epoch": 2.779572887650882, + "grad_norm": 1.5249472856521606, + "learning_rate": 1e-06, + "loss": 0.2977, + "mean_token_accuracy": 0.8921715617179871, + "num_tokens": 525658090.0, + "step": 14968 + }, + { + "epoch": 2.779758588672238, + "grad_norm": 1.5615921020507812, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8925543427467346, + "num_tokens": 525693065.0, + "step": 14969 + }, + { + "epoch": 2.779944289693593, + "grad_norm": 1.4380202293395996, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.888514518737793, + "num_tokens": 525733437.0, + "step": 14970 + }, + { + "epoch": 2.780129990714949, + "grad_norm": 1.5025016069412231, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8881049752235413, + "num_tokens": 525770622.0, + "step": 14971 + }, + { + "epoch": 2.7803156917363046, + "grad_norm": 1.756229043006897, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8749188184738159, + "num_tokens": 525799496.0, + "step": 14972 + }, + { + "epoch": 2.7805013927576603, + "grad_norm": 1.5536484718322754, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8775247931480408, + "num_tokens": 525837328.0, + "step": 14973 + }, + { + "epoch": 2.780687093779016, + "grad_norm": 1.682777762413025, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8823492527008057, + "num_tokens": 525869358.0, + "step": 14974 + }, + { + "epoch": 2.7808727948003713, + "grad_norm": 1.7119802236557007, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8850889205932617, + "num_tokens": 525901561.0, + "step": 14975 + }, + { + "epoch": 2.781058495821727, + "grad_norm": 1.6324421167373657, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8862537145614624, + "num_tokens": 525934752.0, + "step": 14976 + }, + { + "epoch": 2.7812441968430828, + "grad_norm": 1.6342744827270508, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8803117275238037, + "num_tokens": 525973728.0, + "step": 14977 + }, + { + "epoch": 2.781429897864438, + "grad_norm": 1.64366614818573, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8756250143051147, + "num_tokens": 526007482.0, + "step": 14978 + }, + { + "epoch": 2.781615598885794, + "grad_norm": 1.6497257947921753, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8916361331939697, + "num_tokens": 526038499.0, + "step": 14979 + }, + { + "epoch": 2.7818012999071495, + "grad_norm": 1.5841985940933228, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8868023157119751, + "num_tokens": 526074094.0, + "step": 14980 + }, + { + "epoch": 2.7819870009285053, + "grad_norm": 1.5216090679168701, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8839856386184692, + "num_tokens": 526111862.0, + "step": 14981 + }, + { + "epoch": 2.782172701949861, + "grad_norm": 1.604937195777893, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8727120161056519, + "num_tokens": 526145523.0, + "step": 14982 + }, + { + "epoch": 2.7823584029712163, + "grad_norm": 1.5799742937088013, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8800942301750183, + "num_tokens": 526181900.0, + "step": 14983 + }, + { + "epoch": 2.782544103992572, + "grad_norm": 1.4767379760742188, + "learning_rate": 1e-06, + "loss": 0.3064, + "mean_token_accuracy": 0.8886622786521912, + "num_tokens": 526218582.0, + "step": 14984 + }, + { + "epoch": 2.7827298050139273, + "grad_norm": 1.608668327331543, + "learning_rate": 1e-06, + "loss": 0.2893, + "mean_token_accuracy": 0.894463837146759, + "num_tokens": 526253468.0, + "step": 14985 + }, + { + "epoch": 2.782915506035283, + "grad_norm": 1.6098923683166504, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8845281004905701, + "num_tokens": 526288674.0, + "step": 14986 + }, + { + "epoch": 2.7831012070566388, + "grad_norm": 1.471893548965454, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8889131546020508, + "num_tokens": 526326199.0, + "step": 14987 + }, + { + "epoch": 2.7832869080779945, + "grad_norm": 1.650829792022705, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.879890501499176, + "num_tokens": 526364142.0, + "step": 14988 + }, + { + "epoch": 2.7834726090993502, + "grad_norm": 1.5148630142211914, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8912357091903687, + "num_tokens": 526398721.0, + "step": 14989 + }, + { + "epoch": 2.7836583101207055, + "grad_norm": 1.6825677156448364, + "learning_rate": 1e-06, + "loss": 0.3899, + "mean_token_accuracy": 0.8634934425354004, + "num_tokens": 526433165.0, + "step": 14990 + }, + { + "epoch": 2.7838440111420613, + "grad_norm": 1.6965124607086182, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8981461524963379, + "num_tokens": 526466330.0, + "step": 14991 + }, + { + "epoch": 2.784029712163417, + "grad_norm": 1.5439422130584717, + "learning_rate": 1e-06, + "loss": 0.2613, + "mean_token_accuracy": 0.902665376663208, + "num_tokens": 526497923.0, + "step": 14992 + }, + { + "epoch": 2.7842154131847723, + "grad_norm": 1.5377336740493774, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8941574096679688, + "num_tokens": 526529283.0, + "step": 14993 + }, + { + "epoch": 2.784401114206128, + "grad_norm": 1.6081480979919434, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.884028434753418, + "num_tokens": 526561355.0, + "step": 14994 + }, + { + "epoch": 2.7845868152274837, + "grad_norm": 1.5633642673492432, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.8961132764816284, + "num_tokens": 526594843.0, + "step": 14995 + }, + { + "epoch": 2.7847725162488395, + "grad_norm": 1.4596807956695557, + "learning_rate": 1e-06, + "loss": 0.3268, + "mean_token_accuracy": 0.8858314752578735, + "num_tokens": 526635199.0, + "step": 14996 + }, + { + "epoch": 2.784958217270195, + "grad_norm": 1.4523873329162598, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.893794596195221, + "num_tokens": 526670266.0, + "step": 14997 + }, + { + "epoch": 2.7851439182915505, + "grad_norm": 1.5066395998001099, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.89463871717453, + "num_tokens": 526705605.0, + "step": 14998 + }, + { + "epoch": 2.7853296193129062, + "grad_norm": 1.4694116115570068, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8843746185302734, + "num_tokens": 526747299.0, + "step": 14999 + }, + { + "epoch": 2.785515320334262, + "grad_norm": 1.7651808261871338, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8905906677246094, + "num_tokens": 526772595.0, + "step": 15000 + }, + { + "epoch": 2.7857010213556173, + "grad_norm": 1.7798652648925781, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8878175020217896, + "num_tokens": 526800740.0, + "step": 15001 + }, + { + "epoch": 2.785886722376973, + "grad_norm": 1.4840434789657593, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8854771852493286, + "num_tokens": 526836455.0, + "step": 15002 + }, + { + "epoch": 2.7860724233983287, + "grad_norm": 1.520506501197815, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8932340741157532, + "num_tokens": 526870366.0, + "step": 15003 + }, + { + "epoch": 2.7862581244196845, + "grad_norm": 1.7036848068237305, + "learning_rate": 1e-06, + "loss": 0.3834, + "mean_token_accuracy": 0.870469868183136, + "num_tokens": 526905439.0, + "step": 15004 + }, + { + "epoch": 2.78644382544104, + "grad_norm": 1.7329106330871582, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.887641429901123, + "num_tokens": 526932705.0, + "step": 15005 + }, + { + "epoch": 2.7866295264623955, + "grad_norm": 1.5570805072784424, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8745276927947998, + "num_tokens": 526971787.0, + "step": 15006 + }, + { + "epoch": 2.786815227483751, + "grad_norm": 1.4502756595611572, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8767527341842651, + "num_tokens": 527011828.0, + "step": 15007 + }, + { + "epoch": 2.7870009285051065, + "grad_norm": 1.4880913496017456, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8819260597229004, + "num_tokens": 527049123.0, + "step": 15008 + }, + { + "epoch": 2.7871866295264622, + "grad_norm": 1.6158367395401, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8682796359062195, + "num_tokens": 527085031.0, + "step": 15009 + }, + { + "epoch": 2.787372330547818, + "grad_norm": 1.4863402843475342, + "learning_rate": 1e-06, + "loss": 0.295, + "mean_token_accuracy": 0.8939378261566162, + "num_tokens": 527122183.0, + "step": 15010 + }, + { + "epoch": 2.7875580315691737, + "grad_norm": 1.4619431495666504, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8853017091751099, + "num_tokens": 527158800.0, + "step": 15011 + }, + { + "epoch": 2.7877437325905294, + "grad_norm": 1.575941801071167, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8901530504226685, + "num_tokens": 527191609.0, + "step": 15012 + }, + { + "epoch": 2.7879294336118847, + "grad_norm": 1.4260995388031006, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8976713418960571, + "num_tokens": 527230842.0, + "step": 15013 + }, + { + "epoch": 2.7881151346332405, + "grad_norm": 1.5021109580993652, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8937057852745056, + "num_tokens": 527264995.0, + "step": 15014 + }, + { + "epoch": 2.788300835654596, + "grad_norm": 1.678743600845337, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.862421989440918, + "num_tokens": 527298271.0, + "step": 15015 + }, + { + "epoch": 2.7884865366759515, + "grad_norm": 1.5305854082107544, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8954648375511169, + "num_tokens": 527332531.0, + "step": 15016 + }, + { + "epoch": 2.788672237697307, + "grad_norm": 1.5596098899841309, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8813405632972717, + "num_tokens": 527369177.0, + "step": 15017 + }, + { + "epoch": 2.788857938718663, + "grad_norm": 1.639009714126587, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8723436594009399, + "num_tokens": 527401306.0, + "step": 15018 + }, + { + "epoch": 2.7890436397400187, + "grad_norm": 1.5882056951522827, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8904917240142822, + "num_tokens": 527434637.0, + "step": 15019 + }, + { + "epoch": 2.7892293407613744, + "grad_norm": 1.412426233291626, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8884966969490051, + "num_tokens": 527474576.0, + "step": 15020 + }, + { + "epoch": 2.7894150417827297, + "grad_norm": 1.5324804782867432, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8958032131195068, + "num_tokens": 527509128.0, + "step": 15021 + }, + { + "epoch": 2.7896007428040854, + "grad_norm": 1.5088692903518677, + "learning_rate": 1e-06, + "loss": 0.2701, + "mean_token_accuracy": 0.9009733200073242, + "num_tokens": 527545466.0, + "step": 15022 + }, + { + "epoch": 2.789786443825441, + "grad_norm": 1.5193005800247192, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8799420595169067, + "num_tokens": 527582741.0, + "step": 15023 + }, + { + "epoch": 2.7899721448467965, + "grad_norm": 1.7029461860656738, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8812900185585022, + "num_tokens": 527614199.0, + "step": 15024 + }, + { + "epoch": 2.790157845868152, + "grad_norm": 1.5722743272781372, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.8826887011528015, + "num_tokens": 527655472.0, + "step": 15025 + }, + { + "epoch": 2.790343546889508, + "grad_norm": 1.593066930770874, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8928961157798767, + "num_tokens": 527689961.0, + "step": 15026 + }, + { + "epoch": 2.7905292479108637, + "grad_norm": 1.4055029153823853, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.8940111398696899, + "num_tokens": 527730939.0, + "step": 15027 + }, + { + "epoch": 2.7907149489322194, + "grad_norm": 1.447611689567566, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8956528306007385, + "num_tokens": 527768022.0, + "step": 15028 + }, + { + "epoch": 2.7909006499535747, + "grad_norm": 1.70442795753479, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8858785033226013, + "num_tokens": 527798268.0, + "step": 15029 + }, + { + "epoch": 2.7910863509749304, + "grad_norm": 1.6175940036773682, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8860608339309692, + "num_tokens": 527833402.0, + "step": 15030 + }, + { + "epoch": 2.791272051996286, + "grad_norm": 1.4979498386383057, + "learning_rate": 1e-06, + "loss": 0.3606, + "mean_token_accuracy": 0.8767659068107605, + "num_tokens": 527875203.0, + "step": 15031 + }, + { + "epoch": 2.7914577530176414, + "grad_norm": 1.61014986038208, + "learning_rate": 1e-06, + "loss": 0.268, + "mean_token_accuracy": 0.9045947790145874, + "num_tokens": 527904554.0, + "step": 15032 + }, + { + "epoch": 2.791643454038997, + "grad_norm": 1.4644694328308105, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8805789947509766, + "num_tokens": 527947440.0, + "step": 15033 + }, + { + "epoch": 2.791829155060353, + "grad_norm": 1.5348751544952393, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8820228576660156, + "num_tokens": 527983754.0, + "step": 15034 + }, + { + "epoch": 2.7920148560817086, + "grad_norm": 1.405165195465088, + "learning_rate": 1e-06, + "loss": 0.2781, + "mean_token_accuracy": 0.901686429977417, + "num_tokens": 528025222.0, + "step": 15035 + }, + { + "epoch": 2.792200557103064, + "grad_norm": 1.4578168392181396, + "learning_rate": 1e-06, + "loss": 0.2664, + "mean_token_accuracy": 0.9045600891113281, + "num_tokens": 528060368.0, + "step": 15036 + }, + { + "epoch": 2.7923862581244197, + "grad_norm": 1.6076806783676147, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8814125657081604, + "num_tokens": 528093019.0, + "step": 15037 + }, + { + "epoch": 2.7925719591457754, + "grad_norm": 1.584206461906433, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8928344249725342, + "num_tokens": 528130785.0, + "step": 15038 + }, + { + "epoch": 2.7927576601671307, + "grad_norm": 1.6690593957901, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8846027851104736, + "num_tokens": 528162381.0, + "step": 15039 + }, + { + "epoch": 2.7929433611884864, + "grad_norm": 1.6269158124923706, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8819037079811096, + "num_tokens": 528195985.0, + "step": 15040 + }, + { + "epoch": 2.793129062209842, + "grad_norm": 1.634491205215454, + "learning_rate": 1e-06, + "loss": 0.3553, + "mean_token_accuracy": 0.8780908584594727, + "num_tokens": 528234040.0, + "step": 15041 + }, + { + "epoch": 2.793314763231198, + "grad_norm": 1.6668469905853271, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8766182065010071, + "num_tokens": 528264225.0, + "step": 15042 + }, + { + "epoch": 2.7935004642525536, + "grad_norm": 1.5107260942459106, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8880590200424194, + "num_tokens": 528300523.0, + "step": 15043 + }, + { + "epoch": 2.793686165273909, + "grad_norm": 1.4568716287612915, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8936002254486084, + "num_tokens": 528337159.0, + "step": 15044 + }, + { + "epoch": 2.7938718662952646, + "grad_norm": 1.446999192237854, + "learning_rate": 1e-06, + "loss": 0.2688, + "mean_token_accuracy": 0.9048009514808655, + "num_tokens": 528371561.0, + "step": 15045 + }, + { + "epoch": 2.7940575673166204, + "grad_norm": 1.691608190536499, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8743480443954468, + "num_tokens": 528403685.0, + "step": 15046 + }, + { + "epoch": 2.7942432683379756, + "grad_norm": 1.5946078300476074, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8927146196365356, + "num_tokens": 528434202.0, + "step": 15047 + }, + { + "epoch": 2.7944289693593314, + "grad_norm": 1.7078754901885986, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8897309303283691, + "num_tokens": 528463224.0, + "step": 15048 + }, + { + "epoch": 2.794614670380687, + "grad_norm": 1.4976820945739746, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8885447382926941, + "num_tokens": 528499148.0, + "step": 15049 + }, + { + "epoch": 2.794800371402043, + "grad_norm": 1.5783865451812744, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8796712160110474, + "num_tokens": 528533681.0, + "step": 15050 + }, + { + "epoch": 2.7949860724233986, + "grad_norm": 1.6970868110656738, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8855156898498535, + "num_tokens": 528569879.0, + "step": 15051 + }, + { + "epoch": 2.795171773444754, + "grad_norm": 1.5966355800628662, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8808759450912476, + "num_tokens": 528605385.0, + "step": 15052 + }, + { + "epoch": 2.7953574744661096, + "grad_norm": 1.6390281915664673, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8874807357788086, + "num_tokens": 528635263.0, + "step": 15053 + }, + { + "epoch": 2.7955431754874653, + "grad_norm": 1.687929630279541, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8836368322372437, + "num_tokens": 528665147.0, + "step": 15054 + }, + { + "epoch": 2.7957288765088206, + "grad_norm": 1.5476986169815063, + "learning_rate": 1e-06, + "loss": 0.3491, + "mean_token_accuracy": 0.8772024512290955, + "num_tokens": 528700744.0, + "step": 15055 + }, + { + "epoch": 2.7959145775301764, + "grad_norm": 1.5724658966064453, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.8924453258514404, + "num_tokens": 528735779.0, + "step": 15056 + }, + { + "epoch": 2.796100278551532, + "grad_norm": 1.6208692789077759, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8788297176361084, + "num_tokens": 528770879.0, + "step": 15057 + }, + { + "epoch": 2.796285979572888, + "grad_norm": 1.41495943069458, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.8953148126602173, + "num_tokens": 528807526.0, + "step": 15058 + }, + { + "epoch": 2.796471680594243, + "grad_norm": 1.6817152500152588, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.889633059501648, + "num_tokens": 528837768.0, + "step": 15059 + }, + { + "epoch": 2.796657381615599, + "grad_norm": 1.6716398000717163, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8961591720581055, + "num_tokens": 528867586.0, + "step": 15060 + }, + { + "epoch": 2.7968430826369546, + "grad_norm": 1.5914703607559204, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.879098117351532, + "num_tokens": 528902975.0, + "step": 15061 + }, + { + "epoch": 2.79702878365831, + "grad_norm": 1.8158036470413208, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8939621448516846, + "num_tokens": 528927550.0, + "step": 15062 + }, + { + "epoch": 2.7972144846796656, + "grad_norm": 1.591173768043518, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.9009667038917542, + "num_tokens": 528956710.0, + "step": 15063 + }, + { + "epoch": 2.7974001857010213, + "grad_norm": 1.5278897285461426, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8902772665023804, + "num_tokens": 528991796.0, + "step": 15064 + }, + { + "epoch": 2.797585886722377, + "grad_norm": 1.4523411989212036, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8968873023986816, + "num_tokens": 529029427.0, + "step": 15065 + }, + { + "epoch": 2.797771587743733, + "grad_norm": 1.4829683303833008, + "learning_rate": 1e-06, + "loss": 0.2819, + "mean_token_accuracy": 0.902672290802002, + "num_tokens": 529068775.0, + "step": 15066 + }, + { + "epoch": 2.797957288765088, + "grad_norm": 1.4893321990966797, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.8888953328132629, + "num_tokens": 529109471.0, + "step": 15067 + }, + { + "epoch": 2.798142989786444, + "grad_norm": 1.6581346988677979, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8972738981246948, + "num_tokens": 529136339.0, + "step": 15068 + }, + { + "epoch": 2.7983286908077996, + "grad_norm": 1.4159197807312012, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8913595080375671, + "num_tokens": 529177072.0, + "step": 15069 + }, + { + "epoch": 2.798514391829155, + "grad_norm": 1.5769233703613281, + "learning_rate": 1e-06, + "loss": 0.3629, + "mean_token_accuracy": 0.8734759092330933, + "num_tokens": 529213740.0, + "step": 15070 + }, + { + "epoch": 2.7987000928505106, + "grad_norm": 1.4412230253219604, + "learning_rate": 1e-06, + "loss": 0.2825, + "mean_token_accuracy": 0.8982864618301392, + "num_tokens": 529252580.0, + "step": 15071 + }, + { + "epoch": 2.7988857938718663, + "grad_norm": 1.402488350868225, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8888930082321167, + "num_tokens": 529293679.0, + "step": 15072 + }, + { + "epoch": 2.799071494893222, + "grad_norm": 1.5929322242736816, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8880258798599243, + "num_tokens": 529326521.0, + "step": 15073 + }, + { + "epoch": 2.799257195914578, + "grad_norm": 1.5501625537872314, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8859197497367859, + "num_tokens": 529363445.0, + "step": 15074 + }, + { + "epoch": 2.799442896935933, + "grad_norm": 1.518441081047058, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8873406648635864, + "num_tokens": 529397829.0, + "step": 15075 + }, + { + "epoch": 2.799628597957289, + "grad_norm": 1.5437812805175781, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8917406797409058, + "num_tokens": 529431574.0, + "step": 15076 + }, + { + "epoch": 2.7998142989786445, + "grad_norm": 1.5952450037002563, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8905547857284546, + "num_tokens": 529467948.0, + "step": 15077 + }, + { + "epoch": 2.8, + "grad_norm": 1.4736822843551636, + "learning_rate": 1e-06, + "loss": 0.2936, + "mean_token_accuracy": 0.8936251997947693, + "num_tokens": 529505157.0, + "step": 15078 + }, + { + "epoch": 2.8001857010213556, + "grad_norm": 1.578203558921814, + "learning_rate": 1e-06, + "loss": 0.2856, + "mean_token_accuracy": 0.9020594358444214, + "num_tokens": 529533056.0, + "step": 15079 + }, + { + "epoch": 2.8003714020427113, + "grad_norm": 1.4798029661178589, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8860657215118408, + "num_tokens": 529570560.0, + "step": 15080 + }, + { + "epoch": 2.800557103064067, + "grad_norm": 1.5649113655090332, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8693000078201294, + "num_tokens": 529610080.0, + "step": 15081 + }, + { + "epoch": 2.8007428040854223, + "grad_norm": 1.4430210590362549, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8882274031639099, + "num_tokens": 529649267.0, + "step": 15082 + }, + { + "epoch": 2.800928505106778, + "grad_norm": 1.5403778553009033, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8858526349067688, + "num_tokens": 529687450.0, + "step": 15083 + }, + { + "epoch": 2.8011142061281338, + "grad_norm": 1.5632281303405762, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8725183010101318, + "num_tokens": 529724542.0, + "step": 15084 + }, + { + "epoch": 2.801299907149489, + "grad_norm": 1.488147497177124, + "learning_rate": 1e-06, + "loss": 0.2877, + "mean_token_accuracy": 0.8977952599525452, + "num_tokens": 529761693.0, + "step": 15085 + }, + { + "epoch": 2.801485608170845, + "grad_norm": 1.5521987676620483, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8737939596176147, + "num_tokens": 529799530.0, + "step": 15086 + }, + { + "epoch": 2.8016713091922005, + "grad_norm": 1.613684058189392, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8911618590354919, + "num_tokens": 529833792.0, + "step": 15087 + }, + { + "epoch": 2.8018570102135563, + "grad_norm": 1.6026830673217773, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.884377658367157, + "num_tokens": 529869994.0, + "step": 15088 + }, + { + "epoch": 2.802042711234912, + "grad_norm": 1.613194465637207, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8871967196464539, + "num_tokens": 529901836.0, + "step": 15089 + }, + { + "epoch": 2.8022284122562673, + "grad_norm": 1.734894037246704, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8862038254737854, + "num_tokens": 529929598.0, + "step": 15090 + }, + { + "epoch": 2.802414113277623, + "grad_norm": 1.5921095609664917, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8959668874740601, + "num_tokens": 529959742.0, + "step": 15091 + }, + { + "epoch": 2.8025998142989788, + "grad_norm": 1.5609543323516846, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8827316761016846, + "num_tokens": 529996385.0, + "step": 15092 + }, + { + "epoch": 2.802785515320334, + "grad_norm": 1.4097168445587158, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.8978261947631836, + "num_tokens": 530034010.0, + "step": 15093 + }, + { + "epoch": 2.8029712163416898, + "grad_norm": 1.490796446800232, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8868070244789124, + "num_tokens": 530071553.0, + "step": 15094 + }, + { + "epoch": 2.8031569173630455, + "grad_norm": 1.6726033687591553, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8868670463562012, + "num_tokens": 530102836.0, + "step": 15095 + }, + { + "epoch": 2.8033426183844012, + "grad_norm": 1.6608326435089111, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8855876922607422, + "num_tokens": 530136327.0, + "step": 15096 + }, + { + "epoch": 2.803528319405757, + "grad_norm": 1.6391352415084839, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8861956000328064, + "num_tokens": 530166865.0, + "step": 15097 + }, + { + "epoch": 2.8037140204271123, + "grad_norm": 1.5791770219802856, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.881760835647583, + "num_tokens": 530205318.0, + "step": 15098 + }, + { + "epoch": 2.803899721448468, + "grad_norm": 1.5473322868347168, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8923583030700684, + "num_tokens": 530243234.0, + "step": 15099 + }, + { + "epoch": 2.8040854224698237, + "grad_norm": 1.5879334211349487, + "learning_rate": 1e-06, + "loss": 0.2899, + "mean_token_accuracy": 0.8995383977890015, + "num_tokens": 530276695.0, + "step": 15100 + }, + { + "epoch": 2.804271123491179, + "grad_norm": 1.6625173091888428, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8939253687858582, + "num_tokens": 530308194.0, + "step": 15101 + }, + { + "epoch": 2.8044568245125348, + "grad_norm": 1.4796593189239502, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8951247334480286, + "num_tokens": 530345715.0, + "step": 15102 + }, + { + "epoch": 2.8046425255338905, + "grad_norm": 1.570162057876587, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8857548236846924, + "num_tokens": 530380819.0, + "step": 15103 + }, + { + "epoch": 2.804828226555246, + "grad_norm": 1.6089338064193726, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8630387783050537, + "num_tokens": 530419144.0, + "step": 15104 + }, + { + "epoch": 2.8050139275766015, + "grad_norm": 1.6728392839431763, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8837152719497681, + "num_tokens": 530450081.0, + "step": 15105 + }, + { + "epoch": 2.8051996285979572, + "grad_norm": 1.5169668197631836, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8944706916809082, + "num_tokens": 530487022.0, + "step": 15106 + }, + { + "epoch": 2.805385329619313, + "grad_norm": 1.5493721961975098, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8921499252319336, + "num_tokens": 530521659.0, + "step": 15107 + }, + { + "epoch": 2.8055710306406683, + "grad_norm": 1.713999629020691, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8717993497848511, + "num_tokens": 530553234.0, + "step": 15108 + }, + { + "epoch": 2.805756731662024, + "grad_norm": 1.6589152812957764, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8895503282546997, + "num_tokens": 530581141.0, + "step": 15109 + }, + { + "epoch": 2.8059424326833797, + "grad_norm": 1.5724893808364868, + "learning_rate": 1e-06, + "loss": 0.3002, + "mean_token_accuracy": 0.8941761255264282, + "num_tokens": 530612510.0, + "step": 15110 + }, + { + "epoch": 2.8061281337047355, + "grad_norm": 1.455668568611145, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8954293727874756, + "num_tokens": 530653932.0, + "step": 15111 + }, + { + "epoch": 2.806313834726091, + "grad_norm": 1.5090508460998535, + "learning_rate": 1e-06, + "loss": 0.3034, + "mean_token_accuracy": 0.8886942863464355, + "num_tokens": 530691756.0, + "step": 15112 + }, + { + "epoch": 2.8064995357474465, + "grad_norm": 1.6054260730743408, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8668493032455444, + "num_tokens": 530730399.0, + "step": 15113 + }, + { + "epoch": 2.806685236768802, + "grad_norm": 1.5425100326538086, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8926228880882263, + "num_tokens": 530764521.0, + "step": 15114 + }, + { + "epoch": 2.806870937790158, + "grad_norm": 1.5096641778945923, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.8774257898330688, + "num_tokens": 530801729.0, + "step": 15115 + }, + { + "epoch": 2.8070566388115132, + "grad_norm": 1.4529603719711304, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.8910110592842102, + "num_tokens": 530838504.0, + "step": 15116 + }, + { + "epoch": 2.807242339832869, + "grad_norm": 1.656787395477295, + "learning_rate": 1e-06, + "loss": 0.3557, + "mean_token_accuracy": 0.8824785947799683, + "num_tokens": 530873001.0, + "step": 15117 + }, + { + "epoch": 2.8074280408542247, + "grad_norm": 1.5171689987182617, + "learning_rate": 1e-06, + "loss": 0.3653, + "mean_token_accuracy": 0.8742028474807739, + "num_tokens": 530914003.0, + "step": 15118 + }, + { + "epoch": 2.8076137418755804, + "grad_norm": 1.7145397663116455, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8835540413856506, + "num_tokens": 530941906.0, + "step": 15119 + }, + { + "epoch": 2.807799442896936, + "grad_norm": 1.535975456237793, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8781801462173462, + "num_tokens": 530980400.0, + "step": 15120 + }, + { + "epoch": 2.8079851439182915, + "grad_norm": 1.5139232873916626, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8865129947662354, + "num_tokens": 531018779.0, + "step": 15121 + }, + { + "epoch": 2.808170844939647, + "grad_norm": 1.413167119026184, + "learning_rate": 1e-06, + "loss": 0.2792, + "mean_token_accuracy": 0.8985055685043335, + "num_tokens": 531058532.0, + "step": 15122 + }, + { + "epoch": 2.808356545961003, + "grad_norm": 1.4481828212738037, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8962135314941406, + "num_tokens": 531097718.0, + "step": 15123 + }, + { + "epoch": 2.808542246982358, + "grad_norm": 1.5818729400634766, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8896222710609436, + "num_tokens": 531128711.0, + "step": 15124 + }, + { + "epoch": 2.808727948003714, + "grad_norm": 1.6090381145477295, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.9011350870132446, + "num_tokens": 531158918.0, + "step": 15125 + }, + { + "epoch": 2.8089136490250697, + "grad_norm": 1.4628493785858154, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8789267539978027, + "num_tokens": 531199362.0, + "step": 15126 + }, + { + "epoch": 2.8090993500464254, + "grad_norm": 1.4762763977050781, + "learning_rate": 1e-06, + "loss": 0.2625, + "mean_token_accuracy": 0.903459906578064, + "num_tokens": 531235573.0, + "step": 15127 + }, + { + "epoch": 2.809285051067781, + "grad_norm": 1.6374948024749756, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8840779662132263, + "num_tokens": 531270266.0, + "step": 15128 + }, + { + "epoch": 2.8094707520891364, + "grad_norm": 1.5635528564453125, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.8967078328132629, + "num_tokens": 531304239.0, + "step": 15129 + }, + { + "epoch": 2.809656453110492, + "grad_norm": 1.6147348880767822, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8821132779121399, + "num_tokens": 531340370.0, + "step": 15130 + }, + { + "epoch": 2.8098421541318475, + "grad_norm": 1.4883933067321777, + "learning_rate": 1e-06, + "loss": 0.2799, + "mean_token_accuracy": 0.9015893936157227, + "num_tokens": 531374740.0, + "step": 15131 + }, + { + "epoch": 2.810027855153203, + "grad_norm": 1.6092060804367065, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8756016492843628, + "num_tokens": 531413478.0, + "step": 15132 + }, + { + "epoch": 2.810213556174559, + "grad_norm": 1.6461310386657715, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8830327391624451, + "num_tokens": 531450158.0, + "step": 15133 + }, + { + "epoch": 2.8103992571959147, + "grad_norm": 1.53116774559021, + "learning_rate": 1e-06, + "loss": 0.2884, + "mean_token_accuracy": 0.8987824320793152, + "num_tokens": 531481509.0, + "step": 15134 + }, + { + "epoch": 2.8105849582172704, + "grad_norm": 1.7224255800247192, + "learning_rate": 1e-06, + "loss": 0.3012, + "mean_token_accuracy": 0.894506573677063, + "num_tokens": 531508509.0, + "step": 15135 + }, + { + "epoch": 2.8107706592386257, + "grad_norm": 1.5659120082855225, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8805248141288757, + "num_tokens": 531544467.0, + "step": 15136 + }, + { + "epoch": 2.8109563602599814, + "grad_norm": 1.5113263130187988, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8899115920066833, + "num_tokens": 531584945.0, + "step": 15137 + }, + { + "epoch": 2.811142061281337, + "grad_norm": 1.3905763626098633, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.894977867603302, + "num_tokens": 531626050.0, + "step": 15138 + }, + { + "epoch": 2.8113277623026924, + "grad_norm": 1.4858556985855103, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8891816139221191, + "num_tokens": 531662231.0, + "step": 15139 + }, + { + "epoch": 2.811513463324048, + "grad_norm": 1.5127192735671997, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8893826007843018, + "num_tokens": 531699538.0, + "step": 15140 + }, + { + "epoch": 2.811699164345404, + "grad_norm": 1.6023740768432617, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8666591048240662, + "num_tokens": 531738611.0, + "step": 15141 + }, + { + "epoch": 2.8118848653667596, + "grad_norm": 1.565976619720459, + "learning_rate": 1e-06, + "loss": 0.294, + "mean_token_accuracy": 0.896926999092102, + "num_tokens": 531768003.0, + "step": 15142 + }, + { + "epoch": 2.8120705663881154, + "grad_norm": 1.5287460088729858, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8858889937400818, + "num_tokens": 531801070.0, + "step": 15143 + }, + { + "epoch": 2.8122562674094707, + "grad_norm": 1.4634994268417358, + "learning_rate": 1e-06, + "loss": 0.2887, + "mean_token_accuracy": 0.8962425589561462, + "num_tokens": 531835626.0, + "step": 15144 + }, + { + "epoch": 2.8124419684308264, + "grad_norm": 1.4106640815734863, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8905864357948303, + "num_tokens": 531875636.0, + "step": 15145 + }, + { + "epoch": 2.812627669452182, + "grad_norm": 1.5427886247634888, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8871592283248901, + "num_tokens": 531909794.0, + "step": 15146 + }, + { + "epoch": 2.8128133704735374, + "grad_norm": 1.421731948852539, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8835549354553223, + "num_tokens": 531954372.0, + "step": 15147 + }, + { + "epoch": 2.812999071494893, + "grad_norm": 1.5815881490707397, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8777693510055542, + "num_tokens": 531989716.0, + "step": 15148 + }, + { + "epoch": 2.813184772516249, + "grad_norm": 1.5326545238494873, + "learning_rate": 1e-06, + "loss": 0.2796, + "mean_token_accuracy": 0.9021717309951782, + "num_tokens": 532024611.0, + "step": 15149 + }, + { + "epoch": 2.8133704735376046, + "grad_norm": 1.8658989667892456, + "learning_rate": 1e-06, + "loss": 0.2739, + "mean_token_accuracy": 0.9001821279525757, + "num_tokens": 532052417.0, + "step": 15150 + }, + { + "epoch": 2.8135561745589603, + "grad_norm": 1.4780490398406982, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8811424374580383, + "num_tokens": 532093897.0, + "step": 15151 + }, + { + "epoch": 2.8137418755803156, + "grad_norm": 1.6097803115844727, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8805394172668457, + "num_tokens": 532130718.0, + "step": 15152 + }, + { + "epoch": 2.8139275766016714, + "grad_norm": 1.4524980783462524, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8911787867546082, + "num_tokens": 532168356.0, + "step": 15153 + }, + { + "epoch": 2.8141132776230267, + "grad_norm": 1.502591609954834, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.8927305936813354, + "num_tokens": 532201672.0, + "step": 15154 + }, + { + "epoch": 2.8142989786443824, + "grad_norm": 1.404098629951477, + "learning_rate": 1e-06, + "loss": 0.2593, + "mean_token_accuracy": 0.9089870452880859, + "num_tokens": 532241523.0, + "step": 15155 + }, + { + "epoch": 2.814484679665738, + "grad_norm": 1.6976693868637085, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8791359663009644, + "num_tokens": 532271129.0, + "step": 15156 + }, + { + "epoch": 2.814670380687094, + "grad_norm": 1.6621501445770264, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.874870777130127, + "num_tokens": 532308172.0, + "step": 15157 + }, + { + "epoch": 2.8148560817084496, + "grad_norm": 1.4160010814666748, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.891549289226532, + "num_tokens": 532345999.0, + "step": 15158 + }, + { + "epoch": 2.815041782729805, + "grad_norm": 1.6413021087646484, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8767309188842773, + "num_tokens": 532379514.0, + "step": 15159 + }, + { + "epoch": 2.8152274837511606, + "grad_norm": 1.5425753593444824, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.889372706413269, + "num_tokens": 532417951.0, + "step": 15160 + }, + { + "epoch": 2.8154131847725163, + "grad_norm": 1.4566776752471924, + "learning_rate": 1e-06, + "loss": 0.2849, + "mean_token_accuracy": 0.899124801158905, + "num_tokens": 532453760.0, + "step": 15161 + }, + { + "epoch": 2.8155988857938716, + "grad_norm": 1.579890251159668, + "learning_rate": 1e-06, + "loss": 0.2796, + "mean_token_accuracy": 0.9008240103721619, + "num_tokens": 532487406.0, + "step": 15162 + }, + { + "epoch": 2.8157845868152274, + "grad_norm": 1.604164958000183, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8842859268188477, + "num_tokens": 532520344.0, + "step": 15163 + }, + { + "epoch": 2.815970287836583, + "grad_norm": 1.8795393705368042, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8812737464904785, + "num_tokens": 532548462.0, + "step": 15164 + }, + { + "epoch": 2.816155988857939, + "grad_norm": 1.5728440284729004, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8890977501869202, + "num_tokens": 532580672.0, + "step": 15165 + }, + { + "epoch": 2.8163416898792946, + "grad_norm": 1.6291992664337158, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.896163284778595, + "num_tokens": 532613083.0, + "step": 15166 + }, + { + "epoch": 2.81652739090065, + "grad_norm": 1.6069144010543823, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8969588279724121, + "num_tokens": 532645921.0, + "step": 15167 + }, + { + "epoch": 2.8167130919220056, + "grad_norm": 1.526419758796692, + "learning_rate": 1e-06, + "loss": 0.2894, + "mean_token_accuracy": 0.898493230342865, + "num_tokens": 532680947.0, + "step": 15168 + }, + { + "epoch": 2.8168987929433613, + "grad_norm": 1.6085458993911743, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8847488164901733, + "num_tokens": 532711607.0, + "step": 15169 + }, + { + "epoch": 2.8170844939647166, + "grad_norm": 1.4209092855453491, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8845675587654114, + "num_tokens": 532753759.0, + "step": 15170 + }, + { + "epoch": 2.8172701949860723, + "grad_norm": 1.5747088193893433, + "learning_rate": 1e-06, + "loss": 0.3731, + "mean_token_accuracy": 0.8716115951538086, + "num_tokens": 532793428.0, + "step": 15171 + }, + { + "epoch": 2.817455896007428, + "grad_norm": 1.4598157405853271, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8921594023704529, + "num_tokens": 532830691.0, + "step": 15172 + }, + { + "epoch": 2.817641597028784, + "grad_norm": 1.627130389213562, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.89460289478302, + "num_tokens": 532863416.0, + "step": 15173 + }, + { + "epoch": 2.8178272980501395, + "grad_norm": 1.4643943309783936, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8820513486862183, + "num_tokens": 532903878.0, + "step": 15174 + }, + { + "epoch": 2.818012999071495, + "grad_norm": 1.5220293998718262, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8920514583587646, + "num_tokens": 532938600.0, + "step": 15175 + }, + { + "epoch": 2.8181987000928506, + "grad_norm": 1.5345299243927002, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8758074045181274, + "num_tokens": 532976693.0, + "step": 15176 + }, + { + "epoch": 2.8183844011142063, + "grad_norm": 1.6630561351776123, + "learning_rate": 1e-06, + "loss": 0.3141, + "mean_token_accuracy": 0.8918068408966064, + "num_tokens": 533007327.0, + "step": 15177 + }, + { + "epoch": 2.8185701021355616, + "grad_norm": 1.5179208517074585, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8983308672904968, + "num_tokens": 533041161.0, + "step": 15178 + }, + { + "epoch": 2.8187558031569173, + "grad_norm": 1.645881175994873, + "learning_rate": 1e-06, + "loss": 0.3848, + "mean_token_accuracy": 0.8680375218391418, + "num_tokens": 533076097.0, + "step": 15179 + }, + { + "epoch": 2.818941504178273, + "grad_norm": 1.3913134336471558, + "learning_rate": 1e-06, + "loss": 0.2789, + "mean_token_accuracy": 0.902165412902832, + "num_tokens": 533115414.0, + "step": 15180 + }, + { + "epoch": 2.819127205199629, + "grad_norm": 1.6114720106124878, + "learning_rate": 1e-06, + "loss": 0.3673, + "mean_token_accuracy": 0.8727317452430725, + "num_tokens": 533148540.0, + "step": 15181 + }, + { + "epoch": 2.819312906220984, + "grad_norm": 1.5081151723861694, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.883592426776886, + "num_tokens": 533186087.0, + "step": 15182 + }, + { + "epoch": 2.81949860724234, + "grad_norm": 1.522792935371399, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8894532918930054, + "num_tokens": 533224437.0, + "step": 15183 + }, + { + "epoch": 2.8196843082636955, + "grad_norm": 1.5315065383911133, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8926784992218018, + "num_tokens": 533257492.0, + "step": 15184 + }, + { + "epoch": 2.819870009285051, + "grad_norm": 1.5138086080551147, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8922433257102966, + "num_tokens": 533293347.0, + "step": 15185 + }, + { + "epoch": 2.8200557103064066, + "grad_norm": 1.5896931886672974, + "learning_rate": 1e-06, + "loss": 0.2986, + "mean_token_accuracy": 0.8943508863449097, + "num_tokens": 533324840.0, + "step": 15186 + }, + { + "epoch": 2.8202414113277623, + "grad_norm": 1.5274043083190918, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8836104869842529, + "num_tokens": 533362441.0, + "step": 15187 + }, + { + "epoch": 2.820427112349118, + "grad_norm": 1.5418660640716553, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8776018619537354, + "num_tokens": 533399214.0, + "step": 15188 + }, + { + "epoch": 2.8206128133704738, + "grad_norm": 1.5850231647491455, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8783238530158997, + "num_tokens": 533433795.0, + "step": 15189 + }, + { + "epoch": 2.820798514391829, + "grad_norm": 1.7294509410858154, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8809826374053955, + "num_tokens": 533467367.0, + "step": 15190 + }, + { + "epoch": 2.820984215413185, + "grad_norm": 1.5713061094284058, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8808057308197021, + "num_tokens": 533506059.0, + "step": 15191 + }, + { + "epoch": 2.8211699164345405, + "grad_norm": 1.6216826438903809, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8880178928375244, + "num_tokens": 533538261.0, + "step": 15192 + }, + { + "epoch": 2.821355617455896, + "grad_norm": 1.5175220966339111, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8961871266365051, + "num_tokens": 533574934.0, + "step": 15193 + }, + { + "epoch": 2.8215413184772515, + "grad_norm": 1.586726427078247, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8851559162139893, + "num_tokens": 533609698.0, + "step": 15194 + }, + { + "epoch": 2.8217270194986073, + "grad_norm": 1.5799589157104492, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.897969126701355, + "num_tokens": 533638587.0, + "step": 15195 + }, + { + "epoch": 2.821912720519963, + "grad_norm": 1.6072126626968384, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8887583613395691, + "num_tokens": 533671960.0, + "step": 15196 + }, + { + "epoch": 2.8220984215413187, + "grad_norm": 1.5992335081100464, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8797301650047302, + "num_tokens": 533705782.0, + "step": 15197 + }, + { + "epoch": 2.822284122562674, + "grad_norm": 1.575985312461853, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.8883742094039917, + "num_tokens": 533741803.0, + "step": 15198 + }, + { + "epoch": 2.8224698235840298, + "grad_norm": 1.7056427001953125, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8903434872627258, + "num_tokens": 533768741.0, + "step": 15199 + }, + { + "epoch": 2.8226555246053855, + "grad_norm": 1.4620721340179443, + "learning_rate": 1e-06, + "loss": 0.3308, + "mean_token_accuracy": 0.8813638687133789, + "num_tokens": 533812081.0, + "step": 15200 + }, + { + "epoch": 2.822841225626741, + "grad_norm": 1.4949735403060913, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.8932475447654724, + "num_tokens": 533848598.0, + "step": 15201 + }, + { + "epoch": 2.8230269266480965, + "grad_norm": 1.5525766611099243, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8836358785629272, + "num_tokens": 533886027.0, + "step": 15202 + }, + { + "epoch": 2.8232126276694522, + "grad_norm": 1.5515111684799194, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.878948986530304, + "num_tokens": 533922360.0, + "step": 15203 + }, + { + "epoch": 2.823398328690808, + "grad_norm": 1.5118646621704102, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8912296295166016, + "num_tokens": 533956200.0, + "step": 15204 + }, + { + "epoch": 2.8235840297121633, + "grad_norm": 1.5003650188446045, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8849491477012634, + "num_tokens": 533994977.0, + "step": 15205 + }, + { + "epoch": 2.823769730733519, + "grad_norm": 1.511275053024292, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8959130048751831, + "num_tokens": 534027518.0, + "step": 15206 + }, + { + "epoch": 2.8239554317548747, + "grad_norm": 1.4607621431350708, + "learning_rate": 1e-06, + "loss": 0.2769, + "mean_token_accuracy": 0.9012973308563232, + "num_tokens": 534062982.0, + "step": 15207 + }, + { + "epoch": 2.82414113277623, + "grad_norm": 1.5678786039352417, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8931714296340942, + "num_tokens": 534093813.0, + "step": 15208 + }, + { + "epoch": 2.8243268337975858, + "grad_norm": 1.5173678398132324, + "learning_rate": 1e-06, + "loss": 0.3681, + "mean_token_accuracy": 0.8720557689666748, + "num_tokens": 534137068.0, + "step": 15209 + }, + { + "epoch": 2.8245125348189415, + "grad_norm": 1.351143479347229, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8950693011283875, + "num_tokens": 534181544.0, + "step": 15210 + }, + { + "epoch": 2.8246982358402972, + "grad_norm": 1.65339195728302, + "learning_rate": 1e-06, + "loss": 0.3145, + "mean_token_accuracy": 0.8917263746261597, + "num_tokens": 534209703.0, + "step": 15211 + }, + { + "epoch": 2.824883936861653, + "grad_norm": 1.5456463098526, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8865703344345093, + "num_tokens": 534248398.0, + "step": 15212 + }, + { + "epoch": 2.8250696378830082, + "grad_norm": 1.6079914569854736, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8885540962219238, + "num_tokens": 534280752.0, + "step": 15213 + }, + { + "epoch": 2.825255338904364, + "grad_norm": 1.5888608694076538, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8873206377029419, + "num_tokens": 534315106.0, + "step": 15214 + }, + { + "epoch": 2.8254410399257197, + "grad_norm": 1.4934240579605103, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8985750675201416, + "num_tokens": 534348525.0, + "step": 15215 + }, + { + "epoch": 2.825626740947075, + "grad_norm": 1.5431137084960938, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8672460317611694, + "num_tokens": 534390116.0, + "step": 15216 + }, + { + "epoch": 2.8258124419684307, + "grad_norm": 1.6705242395401, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8763083219528198, + "num_tokens": 534424209.0, + "step": 15217 + }, + { + "epoch": 2.8259981429897865, + "grad_norm": 1.768025279045105, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8871091604232788, + "num_tokens": 534452933.0, + "step": 15218 + }, + { + "epoch": 2.826183844011142, + "grad_norm": 1.5507341623306274, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8793007135391235, + "num_tokens": 534486724.0, + "step": 15219 + }, + { + "epoch": 2.826369545032498, + "grad_norm": 1.5690579414367676, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8888017535209656, + "num_tokens": 534521969.0, + "step": 15220 + }, + { + "epoch": 2.826555246053853, + "grad_norm": 1.5813838243484497, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8829801082611084, + "num_tokens": 534555899.0, + "step": 15221 + }, + { + "epoch": 2.826740947075209, + "grad_norm": 1.6206943988800049, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8952586650848389, + "num_tokens": 534591295.0, + "step": 15222 + }, + { + "epoch": 2.8269266480965647, + "grad_norm": 1.4833803176879883, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.888058602809906, + "num_tokens": 534630879.0, + "step": 15223 + }, + { + "epoch": 2.82711234911792, + "grad_norm": 1.5494160652160645, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8858339786529541, + "num_tokens": 534668265.0, + "step": 15224 + }, + { + "epoch": 2.8272980501392757, + "grad_norm": 1.5482553243637085, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8864631652832031, + "num_tokens": 534704279.0, + "step": 15225 + }, + { + "epoch": 2.8274837511606314, + "grad_norm": 1.6015199422836304, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8732398748397827, + "num_tokens": 534741024.0, + "step": 15226 + }, + { + "epoch": 2.827669452181987, + "grad_norm": 1.8274425268173218, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8874751925468445, + "num_tokens": 534765385.0, + "step": 15227 + }, + { + "epoch": 2.8278551532033425, + "grad_norm": 1.5972718000411987, + "learning_rate": 1e-06, + "loss": 0.3512, + "mean_token_accuracy": 0.8783770799636841, + "num_tokens": 534807227.0, + "step": 15228 + }, + { + "epoch": 2.828040854224698, + "grad_norm": 1.6459764242172241, + "learning_rate": 1e-06, + "loss": 0.2731, + "mean_token_accuracy": 0.9019776582717896, + "num_tokens": 534834777.0, + "step": 15229 + }, + { + "epoch": 2.828226555246054, + "grad_norm": 1.5525007247924805, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8878660202026367, + "num_tokens": 534867710.0, + "step": 15230 + }, + { + "epoch": 2.828412256267409, + "grad_norm": 1.6989272832870483, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8787839412689209, + "num_tokens": 534898766.0, + "step": 15231 + }, + { + "epoch": 2.828597957288765, + "grad_norm": 1.595685601234436, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8818540573120117, + "num_tokens": 534933022.0, + "step": 15232 + }, + { + "epoch": 2.8287836583101207, + "grad_norm": 1.3957518339157104, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8993403911590576, + "num_tokens": 534972949.0, + "step": 15233 + }, + { + "epoch": 2.8289693593314764, + "grad_norm": 1.647530436515808, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8801977634429932, + "num_tokens": 535007070.0, + "step": 15234 + }, + { + "epoch": 2.829155060352832, + "grad_norm": 1.4733525514602661, + "learning_rate": 1e-06, + "loss": 0.2823, + "mean_token_accuracy": 0.9009565114974976, + "num_tokens": 535044848.0, + "step": 15235 + }, + { + "epoch": 2.8293407613741874, + "grad_norm": 1.434157371520996, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8913429379463196, + "num_tokens": 535085969.0, + "step": 15236 + }, + { + "epoch": 2.829526462395543, + "grad_norm": 1.4458723068237305, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8883562088012695, + "num_tokens": 535122631.0, + "step": 15237 + }, + { + "epoch": 2.829712163416899, + "grad_norm": 1.601338267326355, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8861938714981079, + "num_tokens": 535156546.0, + "step": 15238 + }, + { + "epoch": 2.829897864438254, + "grad_norm": 1.5290030241012573, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8744397163391113, + "num_tokens": 535195443.0, + "step": 15239 + }, + { + "epoch": 2.83008356545961, + "grad_norm": 1.646196722984314, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8818517327308655, + "num_tokens": 535236007.0, + "step": 15240 + }, + { + "epoch": 2.8302692664809657, + "grad_norm": 1.6819653511047363, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8887513875961304, + "num_tokens": 535266833.0, + "step": 15241 + }, + { + "epoch": 2.8304549675023214, + "grad_norm": 1.5342435836791992, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8815621733665466, + "num_tokens": 535304516.0, + "step": 15242 + }, + { + "epoch": 2.830640668523677, + "grad_norm": 1.7993652820587158, + "learning_rate": 1e-06, + "loss": 0.3802, + "mean_token_accuracy": 0.8713164329528809, + "num_tokens": 535334563.0, + "step": 15243 + }, + { + "epoch": 2.8308263695450324, + "grad_norm": 1.5507891178131104, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8896411657333374, + "num_tokens": 535366785.0, + "step": 15244 + }, + { + "epoch": 2.831012070566388, + "grad_norm": 1.850452184677124, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.8763242959976196, + "num_tokens": 535397798.0, + "step": 15245 + }, + { + "epoch": 2.831197771587744, + "grad_norm": 1.4057830572128296, + "learning_rate": 1e-06, + "loss": 0.2591, + "mean_token_accuracy": 0.9054582118988037, + "num_tokens": 535435780.0, + "step": 15246 + }, + { + "epoch": 2.831383472609099, + "grad_norm": 1.5183407068252563, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8973121643066406, + "num_tokens": 535468597.0, + "step": 15247 + }, + { + "epoch": 2.831569173630455, + "grad_norm": 1.543974757194519, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8973188400268555, + "num_tokens": 535504089.0, + "step": 15248 + }, + { + "epoch": 2.8317548746518106, + "grad_norm": 1.4449186325073242, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.8989837169647217, + "num_tokens": 535539482.0, + "step": 15249 + }, + { + "epoch": 2.8319405756731664, + "grad_norm": 1.6260837316513062, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8827292919158936, + "num_tokens": 535571421.0, + "step": 15250 + }, + { + "epoch": 2.8321262766945217, + "grad_norm": 1.652607798576355, + "learning_rate": 1e-06, + "loss": 0.2976, + "mean_token_accuracy": 0.8917180299758911, + "num_tokens": 535601704.0, + "step": 15251 + }, + { + "epoch": 2.8323119777158774, + "grad_norm": 1.7050929069519043, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8774231672286987, + "num_tokens": 535632960.0, + "step": 15252 + }, + { + "epoch": 2.832497678737233, + "grad_norm": 1.6144849061965942, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8692107200622559, + "num_tokens": 535670037.0, + "step": 15253 + }, + { + "epoch": 2.8326833797585884, + "grad_norm": 1.4789340496063232, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8939759731292725, + "num_tokens": 535705081.0, + "step": 15254 + }, + { + "epoch": 2.832869080779944, + "grad_norm": 1.48002028465271, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.8927570581436157, + "num_tokens": 535743221.0, + "step": 15255 + }, + { + "epoch": 2.8330547818013, + "grad_norm": 1.6403379440307617, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8902660608291626, + "num_tokens": 535777542.0, + "step": 15256 + }, + { + "epoch": 2.8332404828226556, + "grad_norm": 1.4037479162216187, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8848106265068054, + "num_tokens": 535817240.0, + "step": 15257 + }, + { + "epoch": 2.8334261838440113, + "grad_norm": 1.5812610387802124, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8766480684280396, + "num_tokens": 535857713.0, + "step": 15258 + }, + { + "epoch": 2.8336118848653666, + "grad_norm": 1.6398167610168457, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8810170888900757, + "num_tokens": 535886763.0, + "step": 15259 + }, + { + "epoch": 2.8337975858867224, + "grad_norm": 1.411787509918213, + "learning_rate": 1e-06, + "loss": 0.2423, + "mean_token_accuracy": 0.91378253698349, + "num_tokens": 535919690.0, + "step": 15260 + }, + { + "epoch": 2.833983286908078, + "grad_norm": 1.6044899225234985, + "learning_rate": 1e-06, + "loss": 0.2932, + "mean_token_accuracy": 0.8966041803359985, + "num_tokens": 535948863.0, + "step": 15261 + }, + { + "epoch": 2.8341689879294334, + "grad_norm": 1.5225605964660645, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8882026672363281, + "num_tokens": 535988295.0, + "step": 15262 + }, + { + "epoch": 2.834354688950789, + "grad_norm": 1.4244123697280884, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8912612795829773, + "num_tokens": 536026858.0, + "step": 15263 + }, + { + "epoch": 2.834540389972145, + "grad_norm": 1.4851454496383667, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8849482536315918, + "num_tokens": 536063423.0, + "step": 15264 + }, + { + "epoch": 2.8347260909935006, + "grad_norm": 1.4593223333358765, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8849378824234009, + "num_tokens": 536105436.0, + "step": 15265 + }, + { + "epoch": 2.8349117920148563, + "grad_norm": 1.481630802154541, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8951430320739746, + "num_tokens": 536139889.0, + "step": 15266 + }, + { + "epoch": 2.8350974930362116, + "grad_norm": 1.7115788459777832, + "learning_rate": 1e-06, + "loss": 0.3355, + "mean_token_accuracy": 0.8843315839767456, + "num_tokens": 536171939.0, + "step": 15267 + }, + { + "epoch": 2.8352831940575673, + "grad_norm": 1.783565878868103, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8764504194259644, + "num_tokens": 536202994.0, + "step": 15268 + }, + { + "epoch": 2.835468895078923, + "grad_norm": 1.5263546705245972, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8772987723350525, + "num_tokens": 536241802.0, + "step": 15269 + }, + { + "epoch": 2.8356545961002784, + "grad_norm": 1.4441267251968384, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8913803100585938, + "num_tokens": 536282656.0, + "step": 15270 + }, + { + "epoch": 2.835840297121634, + "grad_norm": 1.5993871688842773, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8831932544708252, + "num_tokens": 536316949.0, + "step": 15271 + }, + { + "epoch": 2.83602599814299, + "grad_norm": 1.531515121459961, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8816008567810059, + "num_tokens": 536351623.0, + "step": 15272 + }, + { + "epoch": 2.8362116991643456, + "grad_norm": 1.8580368757247925, + "learning_rate": 1e-06, + "loss": 0.3468, + "mean_token_accuracy": 0.881177544593811, + "num_tokens": 536374585.0, + "step": 15273 + }, + { + "epoch": 2.836397400185701, + "grad_norm": 1.416491985321045, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.890813946723938, + "num_tokens": 536414311.0, + "step": 15274 + }, + { + "epoch": 2.8365831012070566, + "grad_norm": 1.4652749300003052, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.87913978099823, + "num_tokens": 536456785.0, + "step": 15275 + }, + { + "epoch": 2.8367688022284123, + "grad_norm": 1.6105904579162598, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8802609443664551, + "num_tokens": 536492071.0, + "step": 15276 + }, + { + "epoch": 2.8369545032497676, + "grad_norm": 1.6384308338165283, + "learning_rate": 1e-06, + "loss": 0.3684, + "mean_token_accuracy": 0.8737309575080872, + "num_tokens": 536528519.0, + "step": 15277 + }, + { + "epoch": 2.8371402042711233, + "grad_norm": 1.6780253648757935, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8953492641448975, + "num_tokens": 536556865.0, + "step": 15278 + }, + { + "epoch": 2.837325905292479, + "grad_norm": 1.5869516134262085, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8956214785575867, + "num_tokens": 536590738.0, + "step": 15279 + }, + { + "epoch": 2.837511606313835, + "grad_norm": 1.4235024452209473, + "learning_rate": 1e-06, + "loss": 0.3019, + "mean_token_accuracy": 0.8911015391349792, + "num_tokens": 536630108.0, + "step": 15280 + }, + { + "epoch": 2.8376973073351905, + "grad_norm": 1.4432768821716309, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.883647084236145, + "num_tokens": 536670676.0, + "step": 15281 + }, + { + "epoch": 2.837883008356546, + "grad_norm": 1.5017433166503906, + "learning_rate": 1e-06, + "loss": 0.3554, + "mean_token_accuracy": 0.8747934103012085, + "num_tokens": 536710174.0, + "step": 15282 + }, + { + "epoch": 2.8380687093779016, + "grad_norm": 1.5968389511108398, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8866502046585083, + "num_tokens": 536744401.0, + "step": 15283 + }, + { + "epoch": 2.8382544103992573, + "grad_norm": 1.5551941394805908, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8867169618606567, + "num_tokens": 536779575.0, + "step": 15284 + }, + { + "epoch": 2.8384401114206126, + "grad_norm": 1.7653062343597412, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8872648477554321, + "num_tokens": 536806666.0, + "step": 15285 + }, + { + "epoch": 2.8386258124419683, + "grad_norm": 1.4708843231201172, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8808426856994629, + "num_tokens": 536846011.0, + "step": 15286 + }, + { + "epoch": 2.838811513463324, + "grad_norm": 1.5575830936431885, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8919447660446167, + "num_tokens": 536877959.0, + "step": 15287 + }, + { + "epoch": 2.83899721448468, + "grad_norm": 1.4323711395263672, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.884899914264679, + "num_tokens": 536918464.0, + "step": 15288 + }, + { + "epoch": 2.8391829155060355, + "grad_norm": 1.6216883659362793, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8931856751441956, + "num_tokens": 536952893.0, + "step": 15289 + }, + { + "epoch": 2.839368616527391, + "grad_norm": 1.583044171333313, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8955783247947693, + "num_tokens": 536984893.0, + "step": 15290 + }, + { + "epoch": 2.8395543175487465, + "grad_norm": 1.5709269046783447, + "learning_rate": 1e-06, + "loss": 0.3597, + "mean_token_accuracy": 0.8756564855575562, + "num_tokens": 537023490.0, + "step": 15291 + }, + { + "epoch": 2.8397400185701023, + "grad_norm": 1.5823726654052734, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8760203719139099, + "num_tokens": 537061522.0, + "step": 15292 + }, + { + "epoch": 2.8399257195914576, + "grad_norm": 1.5550646781921387, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8815111517906189, + "num_tokens": 537096705.0, + "step": 15293 + }, + { + "epoch": 2.8401114206128133, + "grad_norm": 1.5657622814178467, + "learning_rate": 1e-06, + "loss": 0.2973, + "mean_token_accuracy": 0.8942772150039673, + "num_tokens": 537131633.0, + "step": 15294 + }, + { + "epoch": 2.840297121634169, + "grad_norm": 1.5486394166946411, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8892554044723511, + "num_tokens": 537168062.0, + "step": 15295 + }, + { + "epoch": 2.8404828226555248, + "grad_norm": 1.5709645748138428, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8839083909988403, + "num_tokens": 537204104.0, + "step": 15296 + }, + { + "epoch": 2.8406685236768805, + "grad_norm": 1.549116611480713, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8798477649688721, + "num_tokens": 537241414.0, + "step": 15297 + }, + { + "epoch": 2.840854224698236, + "grad_norm": 1.5930628776550293, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.882422685623169, + "num_tokens": 537279189.0, + "step": 15298 + }, + { + "epoch": 2.8410399257195915, + "grad_norm": 1.6543974876403809, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8707964420318604, + "num_tokens": 537313283.0, + "step": 15299 + }, + { + "epoch": 2.841225626740947, + "grad_norm": 1.4948434829711914, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8873204588890076, + "num_tokens": 537350844.0, + "step": 15300 + }, + { + "epoch": 2.8414113277623025, + "grad_norm": 1.6183313131332397, + "learning_rate": 1e-06, + "loss": 0.3534, + "mean_token_accuracy": 0.8776058554649353, + "num_tokens": 537383162.0, + "step": 15301 + }, + { + "epoch": 2.8415970287836583, + "grad_norm": 1.5685752630233765, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8685731291770935, + "num_tokens": 537423632.0, + "step": 15302 + }, + { + "epoch": 2.841782729805014, + "grad_norm": 1.435930609703064, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.891247034072876, + "num_tokens": 537462876.0, + "step": 15303 + }, + { + "epoch": 2.8419684308263697, + "grad_norm": 1.5523135662078857, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8842135071754456, + "num_tokens": 537498380.0, + "step": 15304 + }, + { + "epoch": 2.842154131847725, + "grad_norm": 1.5796113014221191, + "learning_rate": 1e-06, + "loss": 0.2909, + "mean_token_accuracy": 0.8978848457336426, + "num_tokens": 537529775.0, + "step": 15305 + }, + { + "epoch": 2.8423398328690808, + "grad_norm": 1.4194703102111816, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8965470194816589, + "num_tokens": 537573682.0, + "step": 15306 + }, + { + "epoch": 2.8425255338904365, + "grad_norm": 1.6753343343734741, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8697646856307983, + "num_tokens": 537604492.0, + "step": 15307 + }, + { + "epoch": 2.842711234911792, + "grad_norm": 1.4147024154663086, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.8974397778511047, + "num_tokens": 537639865.0, + "step": 15308 + }, + { + "epoch": 2.8428969359331475, + "grad_norm": 1.5584626197814941, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8758906722068787, + "num_tokens": 537676783.0, + "step": 15309 + }, + { + "epoch": 2.8430826369545033, + "grad_norm": 1.522947072982788, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8916208744049072, + "num_tokens": 537710423.0, + "step": 15310 + }, + { + "epoch": 2.843268337975859, + "grad_norm": 1.624640703201294, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8926060199737549, + "num_tokens": 537741698.0, + "step": 15311 + }, + { + "epoch": 2.8434540389972147, + "grad_norm": 1.8195232152938843, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8750113248825073, + "num_tokens": 537772562.0, + "step": 15312 + }, + { + "epoch": 2.84363974001857, + "grad_norm": 1.4569191932678223, + "learning_rate": 1e-06, + "loss": 0.3371, + "mean_token_accuracy": 0.883310854434967, + "num_tokens": 537811920.0, + "step": 15313 + }, + { + "epoch": 2.8438254410399257, + "grad_norm": 1.6406033039093018, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8939600586891174, + "num_tokens": 537844371.0, + "step": 15314 + }, + { + "epoch": 2.8440111420612815, + "grad_norm": 1.6786518096923828, + "learning_rate": 1e-06, + "loss": 0.2898, + "mean_token_accuracy": 0.8931785821914673, + "num_tokens": 537874541.0, + "step": 15315 + }, + { + "epoch": 2.8441968430826368, + "grad_norm": 1.4500808715820312, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8932971954345703, + "num_tokens": 537914695.0, + "step": 15316 + }, + { + "epoch": 2.8443825441039925, + "grad_norm": 1.5300358533859253, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8874485492706299, + "num_tokens": 537952242.0, + "step": 15317 + }, + { + "epoch": 2.8445682451253482, + "grad_norm": 1.849280834197998, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8771038055419922, + "num_tokens": 537980322.0, + "step": 15318 + }, + { + "epoch": 2.844753946146704, + "grad_norm": 1.7698944807052612, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8755226731300354, + "num_tokens": 538012461.0, + "step": 15319 + }, + { + "epoch": 2.8449396471680597, + "grad_norm": 1.5265616178512573, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.8963755369186401, + "num_tokens": 538045128.0, + "step": 15320 + }, + { + "epoch": 2.845125348189415, + "grad_norm": 1.5103511810302734, + "learning_rate": 1e-06, + "loss": 0.2682, + "mean_token_accuracy": 0.9017454385757446, + "num_tokens": 538075820.0, + "step": 15321 + }, + { + "epoch": 2.8453110492107707, + "grad_norm": 1.6028401851654053, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.891674280166626, + "num_tokens": 538111038.0, + "step": 15322 + }, + { + "epoch": 2.845496750232126, + "grad_norm": 1.5260266065597534, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8884472846984863, + "num_tokens": 538149281.0, + "step": 15323 + }, + { + "epoch": 2.8456824512534817, + "grad_norm": 1.553092122077942, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8919614553451538, + "num_tokens": 538183916.0, + "step": 15324 + }, + { + "epoch": 2.8458681522748375, + "grad_norm": 1.4977699518203735, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8998360633850098, + "num_tokens": 538219313.0, + "step": 15325 + }, + { + "epoch": 2.846053853296193, + "grad_norm": 1.588323950767517, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8930083513259888, + "num_tokens": 538251971.0, + "step": 15326 + }, + { + "epoch": 2.846239554317549, + "grad_norm": 1.5738980770111084, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8921254873275757, + "num_tokens": 538283468.0, + "step": 15327 + }, + { + "epoch": 2.8464252553389042, + "grad_norm": 1.6796236038208008, + "learning_rate": 1e-06, + "loss": 0.2843, + "mean_token_accuracy": 0.8996006846427917, + "num_tokens": 538312940.0, + "step": 15328 + }, + { + "epoch": 2.84661095636026, + "grad_norm": 1.4274359941482544, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.9010107517242432, + "num_tokens": 538350502.0, + "step": 15329 + }, + { + "epoch": 2.8467966573816157, + "grad_norm": 1.389144778251648, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8899227976799011, + "num_tokens": 538392860.0, + "step": 15330 + }, + { + "epoch": 2.846982358402971, + "grad_norm": 1.5615602731704712, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8815891742706299, + "num_tokens": 538429704.0, + "step": 15331 + }, + { + "epoch": 2.8471680594243267, + "grad_norm": 1.617570400238037, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8822670578956604, + "num_tokens": 538463636.0, + "step": 15332 + }, + { + "epoch": 2.8473537604456824, + "grad_norm": 1.5148082971572876, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.8902124166488647, + "num_tokens": 538499864.0, + "step": 15333 + }, + { + "epoch": 2.847539461467038, + "grad_norm": 1.6094967126846313, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8782728314399719, + "num_tokens": 538539006.0, + "step": 15334 + }, + { + "epoch": 2.847725162488394, + "grad_norm": 1.4898537397384644, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8907879590988159, + "num_tokens": 538576042.0, + "step": 15335 + }, + { + "epoch": 2.847910863509749, + "grad_norm": 1.8187501430511475, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8847770094871521, + "num_tokens": 538602230.0, + "step": 15336 + }, + { + "epoch": 2.848096564531105, + "grad_norm": 1.3992201089859009, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8985872268676758, + "num_tokens": 538639304.0, + "step": 15337 + }, + { + "epoch": 2.8482822655524607, + "grad_norm": 1.6146897077560425, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8786367774009705, + "num_tokens": 538674509.0, + "step": 15338 + }, + { + "epoch": 2.848467966573816, + "grad_norm": 1.560337781906128, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8972474336624146, + "num_tokens": 538708343.0, + "step": 15339 + }, + { + "epoch": 2.8486536675951717, + "grad_norm": 1.6950736045837402, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8855985403060913, + "num_tokens": 538739107.0, + "step": 15340 + }, + { + "epoch": 2.8488393686165274, + "grad_norm": 1.6643216609954834, + "learning_rate": 1e-06, + "loss": 0.342, + "mean_token_accuracy": 0.8816797137260437, + "num_tokens": 538771909.0, + "step": 15341 + }, + { + "epoch": 2.849025069637883, + "grad_norm": 1.5948559045791626, + "learning_rate": 1e-06, + "loss": 0.3635, + "mean_token_accuracy": 0.8735219836235046, + "num_tokens": 538810814.0, + "step": 15342 + }, + { + "epoch": 2.849210770659239, + "grad_norm": 1.6344540119171143, + "learning_rate": 1e-06, + "loss": 0.3781, + "mean_token_accuracy": 0.8745909929275513, + "num_tokens": 538844539.0, + "step": 15343 + }, + { + "epoch": 2.849396471680594, + "grad_norm": 1.6591235399246216, + "learning_rate": 1e-06, + "loss": 0.3174, + "mean_token_accuracy": 0.8855054974555969, + "num_tokens": 538878374.0, + "step": 15344 + }, + { + "epoch": 2.84958217270195, + "grad_norm": 1.6517356634140015, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8864614963531494, + "num_tokens": 538911790.0, + "step": 15345 + }, + { + "epoch": 2.8497678737233056, + "grad_norm": 1.444237232208252, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.889997124671936, + "num_tokens": 538949956.0, + "step": 15346 + }, + { + "epoch": 2.849953574744661, + "grad_norm": 1.7720195055007935, + "learning_rate": 1e-06, + "loss": 0.3627, + "mean_token_accuracy": 0.8746145963668823, + "num_tokens": 538980952.0, + "step": 15347 + }, + { + "epoch": 2.8501392757660167, + "grad_norm": 1.5963979959487915, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8874572515487671, + "num_tokens": 539013461.0, + "step": 15348 + }, + { + "epoch": 2.8503249767873724, + "grad_norm": 1.546872615814209, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8882569074630737, + "num_tokens": 539047234.0, + "step": 15349 + }, + { + "epoch": 2.850510677808728, + "grad_norm": 1.4881072044372559, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8912913799285889, + "num_tokens": 539083057.0, + "step": 15350 + }, + { + "epoch": 2.8506963788300834, + "grad_norm": 1.7433618307113647, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8970255851745605, + "num_tokens": 539110761.0, + "step": 15351 + }, + { + "epoch": 2.850882079851439, + "grad_norm": 1.6307621002197266, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8918135166168213, + "num_tokens": 539140660.0, + "step": 15352 + }, + { + "epoch": 2.851067780872795, + "grad_norm": 1.686683177947998, + "learning_rate": 1e-06, + "loss": 0.406, + "mean_token_accuracy": 0.8631582260131836, + "num_tokens": 539175103.0, + "step": 15353 + }, + { + "epoch": 2.85125348189415, + "grad_norm": 1.5723237991333008, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8926242589950562, + "num_tokens": 539211220.0, + "step": 15354 + }, + { + "epoch": 2.851439182915506, + "grad_norm": 1.6493269205093384, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8899902105331421, + "num_tokens": 539246879.0, + "step": 15355 + }, + { + "epoch": 2.8516248839368616, + "grad_norm": 1.5491200685501099, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8774237036705017, + "num_tokens": 539286381.0, + "step": 15356 + }, + { + "epoch": 2.8518105849582174, + "grad_norm": 1.5437593460083008, + "learning_rate": 1e-06, + "loss": 0.2911, + "mean_token_accuracy": 0.8965060114860535, + "num_tokens": 539319772.0, + "step": 15357 + }, + { + "epoch": 2.851996285979573, + "grad_norm": 1.4413325786590576, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8856428265571594, + "num_tokens": 539360131.0, + "step": 15358 + }, + { + "epoch": 2.8521819870009284, + "grad_norm": 1.587827444076538, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8877965211868286, + "num_tokens": 539398469.0, + "step": 15359 + }, + { + "epoch": 2.852367688022284, + "grad_norm": 1.414514183998108, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8811436295509338, + "num_tokens": 539442138.0, + "step": 15360 + }, + { + "epoch": 2.85255338904364, + "grad_norm": 1.709874153137207, + "learning_rate": 1e-06, + "loss": 0.3473, + "mean_token_accuracy": 0.8806343674659729, + "num_tokens": 539474749.0, + "step": 15361 + }, + { + "epoch": 2.852739090064995, + "grad_norm": 1.7194859981536865, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.8940308094024658, + "num_tokens": 539501714.0, + "step": 15362 + }, + { + "epoch": 2.852924791086351, + "grad_norm": 1.4834487438201904, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8814789056777954, + "num_tokens": 539540883.0, + "step": 15363 + }, + { + "epoch": 2.8531104921077066, + "grad_norm": 1.607682704925537, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8792147040367126, + "num_tokens": 539577653.0, + "step": 15364 + }, + { + "epoch": 2.8532961931290624, + "grad_norm": 1.534737467765808, + "learning_rate": 1e-06, + "loss": 0.3566, + "mean_token_accuracy": 0.8795568346977234, + "num_tokens": 539617700.0, + "step": 15365 + }, + { + "epoch": 2.853481894150418, + "grad_norm": 1.646919846534729, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8966255187988281, + "num_tokens": 539648076.0, + "step": 15366 + }, + { + "epoch": 2.8536675951717734, + "grad_norm": 1.5636982917785645, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8815414905548096, + "num_tokens": 539688539.0, + "step": 15367 + }, + { + "epoch": 2.853853296193129, + "grad_norm": 1.638538122177124, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8833625912666321, + "num_tokens": 539723778.0, + "step": 15368 + }, + { + "epoch": 2.854038997214485, + "grad_norm": 1.6977927684783936, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8841170072555542, + "num_tokens": 539755015.0, + "step": 15369 + }, + { + "epoch": 2.85422469823584, + "grad_norm": 1.6211187839508057, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8965282440185547, + "num_tokens": 539786428.0, + "step": 15370 + }, + { + "epoch": 2.854410399257196, + "grad_norm": 1.5447769165039062, + "learning_rate": 1e-06, + "loss": 0.3331, + "mean_token_accuracy": 0.8825505375862122, + "num_tokens": 539824623.0, + "step": 15371 + }, + { + "epoch": 2.8545961002785516, + "grad_norm": 1.886927843093872, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8812558650970459, + "num_tokens": 539853862.0, + "step": 15372 + }, + { + "epoch": 2.8547818012999073, + "grad_norm": 1.670175313949585, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8934398293495178, + "num_tokens": 539885500.0, + "step": 15373 + }, + { + "epoch": 2.8549675023212626, + "grad_norm": 1.5805696249008179, + "learning_rate": 1e-06, + "loss": 0.366, + "mean_token_accuracy": 0.8744813203811646, + "num_tokens": 539924624.0, + "step": 15374 + }, + { + "epoch": 2.8551532033426184, + "grad_norm": 1.5143728256225586, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.874190092086792, + "num_tokens": 539964858.0, + "step": 15375 + }, + { + "epoch": 2.855338904363974, + "grad_norm": 1.4802477359771729, + "learning_rate": 1e-06, + "loss": 0.3604, + "mean_token_accuracy": 0.8766322135925293, + "num_tokens": 540004709.0, + "step": 15376 + }, + { + "epoch": 2.8555246053853294, + "grad_norm": 1.5308678150177002, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8778514862060547, + "num_tokens": 540043997.0, + "step": 15377 + }, + { + "epoch": 2.855710306406685, + "grad_norm": 1.5500649213790894, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8906527757644653, + "num_tokens": 540082006.0, + "step": 15378 + }, + { + "epoch": 2.855896007428041, + "grad_norm": 1.6461162567138672, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8739769458770752, + "num_tokens": 540116528.0, + "step": 15379 + }, + { + "epoch": 2.8560817084493966, + "grad_norm": 1.507714033126831, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8861196637153625, + "num_tokens": 540156116.0, + "step": 15380 + }, + { + "epoch": 2.8562674094707523, + "grad_norm": 1.4959845542907715, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8831610679626465, + "num_tokens": 540197580.0, + "step": 15381 + }, + { + "epoch": 2.8564531104921076, + "grad_norm": 1.4834020137786865, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8902086019515991, + "num_tokens": 540235166.0, + "step": 15382 + }, + { + "epoch": 2.8566388115134633, + "grad_norm": 1.6659563779830933, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8861056566238403, + "num_tokens": 540267542.0, + "step": 15383 + }, + { + "epoch": 2.856824512534819, + "grad_norm": 1.5179290771484375, + "learning_rate": 1e-06, + "loss": 0.3092, + "mean_token_accuracy": 0.8915674686431885, + "num_tokens": 540302898.0, + "step": 15384 + }, + { + "epoch": 2.8570102135561743, + "grad_norm": 1.753786563873291, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.886231005191803, + "num_tokens": 540330226.0, + "step": 15385 + }, + { + "epoch": 2.85719591457753, + "grad_norm": 1.563171148300171, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8859274387359619, + "num_tokens": 540367769.0, + "step": 15386 + }, + { + "epoch": 2.857381615598886, + "grad_norm": 1.4899309873580933, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8886748552322388, + "num_tokens": 540406877.0, + "step": 15387 + }, + { + "epoch": 2.8575673166202415, + "grad_norm": 1.5477818250656128, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.886378288269043, + "num_tokens": 540443473.0, + "step": 15388 + }, + { + "epoch": 2.8577530176415973, + "grad_norm": 1.5440571308135986, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8923628330230713, + "num_tokens": 540477487.0, + "step": 15389 + }, + { + "epoch": 2.8579387186629526, + "grad_norm": 1.6621742248535156, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8758180141448975, + "num_tokens": 540509524.0, + "step": 15390 + }, + { + "epoch": 2.8581244196843083, + "grad_norm": 1.7492647171020508, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8838769197463989, + "num_tokens": 540541293.0, + "step": 15391 + }, + { + "epoch": 2.858310120705664, + "grad_norm": 1.689069390296936, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8964924812316895, + "num_tokens": 540570609.0, + "step": 15392 + }, + { + "epoch": 2.8584958217270193, + "grad_norm": 1.5888054370880127, + "learning_rate": 1e-06, + "loss": 0.2936, + "mean_token_accuracy": 0.8935102820396423, + "num_tokens": 540604716.0, + "step": 15393 + }, + { + "epoch": 2.858681522748375, + "grad_norm": 1.546449899673462, + "learning_rate": 1e-06, + "loss": 0.2643, + "mean_token_accuracy": 0.9060050845146179, + "num_tokens": 540636634.0, + "step": 15394 + }, + { + "epoch": 2.858867223769731, + "grad_norm": 1.5386271476745605, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.9017839431762695, + "num_tokens": 540668797.0, + "step": 15395 + }, + { + "epoch": 2.8590529247910865, + "grad_norm": 1.6197454929351807, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8838742971420288, + "num_tokens": 540702467.0, + "step": 15396 + }, + { + "epoch": 2.859238625812442, + "grad_norm": 1.6826715469360352, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8757641911506653, + "num_tokens": 540737088.0, + "step": 15397 + }, + { + "epoch": 2.8594243268337975, + "grad_norm": 1.4404851198196411, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8765590190887451, + "num_tokens": 540777468.0, + "step": 15398 + }, + { + "epoch": 2.8596100278551533, + "grad_norm": 1.4845975637435913, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8917337656021118, + "num_tokens": 540814470.0, + "step": 15399 + }, + { + "epoch": 2.8597957288765086, + "grad_norm": 1.5971267223358154, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8755355477333069, + "num_tokens": 540851501.0, + "step": 15400 + }, + { + "epoch": 2.8599814298978643, + "grad_norm": 1.6036344766616821, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.8868923187255859, + "num_tokens": 540882156.0, + "step": 15401 + }, + { + "epoch": 2.86016713091922, + "grad_norm": 1.4445804357528687, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8843780755996704, + "num_tokens": 540921486.0, + "step": 15402 + }, + { + "epoch": 2.8603528319405758, + "grad_norm": 1.5506446361541748, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.892471969127655, + "num_tokens": 540956728.0, + "step": 15403 + }, + { + "epoch": 2.8605385329619315, + "grad_norm": 1.5318106412887573, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8852220177650452, + "num_tokens": 540994663.0, + "step": 15404 + }, + { + "epoch": 2.860724233983287, + "grad_norm": 1.4457244873046875, + "learning_rate": 1e-06, + "loss": 0.2878, + "mean_token_accuracy": 0.8984226584434509, + "num_tokens": 541032212.0, + "step": 15405 + }, + { + "epoch": 2.8609099350046425, + "grad_norm": 1.6573631763458252, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8717522621154785, + "num_tokens": 541067813.0, + "step": 15406 + }, + { + "epoch": 2.8610956360259983, + "grad_norm": 1.7699296474456787, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8759945034980774, + "num_tokens": 541106929.0, + "step": 15407 + }, + { + "epoch": 2.8612813370473535, + "grad_norm": 1.4388052225112915, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.888918399810791, + "num_tokens": 541150639.0, + "step": 15408 + }, + { + "epoch": 2.8614670380687093, + "grad_norm": 1.5767793655395508, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.8959258794784546, + "num_tokens": 541183702.0, + "step": 15409 + }, + { + "epoch": 2.861652739090065, + "grad_norm": 1.4769337177276611, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8855369091033936, + "num_tokens": 541222695.0, + "step": 15410 + }, + { + "epoch": 2.8618384401114207, + "grad_norm": 1.585991382598877, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.8917738199234009, + "num_tokens": 541252425.0, + "step": 15411 + }, + { + "epoch": 2.8620241411327765, + "grad_norm": 1.5033146142959595, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8843562602996826, + "num_tokens": 541286687.0, + "step": 15412 + }, + { + "epoch": 2.8622098421541318, + "grad_norm": 1.4696708917617798, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8833247423171997, + "num_tokens": 541324170.0, + "step": 15413 + }, + { + "epoch": 2.8623955431754875, + "grad_norm": 1.8053058385849, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8657352328300476, + "num_tokens": 541354543.0, + "step": 15414 + }, + { + "epoch": 2.8625812441968432, + "grad_norm": 1.548991084098816, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8908895254135132, + "num_tokens": 541388157.0, + "step": 15415 + }, + { + "epoch": 2.8627669452181985, + "grad_norm": 1.590842366218567, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8836528062820435, + "num_tokens": 541421578.0, + "step": 15416 + }, + { + "epoch": 2.8629526462395543, + "grad_norm": 1.5194430351257324, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8936135768890381, + "num_tokens": 541458143.0, + "step": 15417 + }, + { + "epoch": 2.86313834726091, + "grad_norm": 1.486796498298645, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8928006887435913, + "num_tokens": 541492928.0, + "step": 15418 + }, + { + "epoch": 2.8633240482822657, + "grad_norm": 1.7881606817245483, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8817240595817566, + "num_tokens": 541520343.0, + "step": 15419 + }, + { + "epoch": 2.863509749303621, + "grad_norm": 1.5528610944747925, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8792637586593628, + "num_tokens": 541555171.0, + "step": 15420 + }, + { + "epoch": 2.8636954503249767, + "grad_norm": 1.659935474395752, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8757323026657104, + "num_tokens": 541588889.0, + "step": 15421 + }, + { + "epoch": 2.8638811513463325, + "grad_norm": 1.649024486541748, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.876266360282898, + "num_tokens": 541620731.0, + "step": 15422 + }, + { + "epoch": 2.8640668523676878, + "grad_norm": 1.4953875541687012, + "learning_rate": 1e-06, + "loss": 0.2761, + "mean_token_accuracy": 0.9027701020240784, + "num_tokens": 541656154.0, + "step": 15423 + }, + { + "epoch": 2.8642525533890435, + "grad_norm": 1.5880310535430908, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.8838560581207275, + "num_tokens": 541688958.0, + "step": 15424 + }, + { + "epoch": 2.8644382544103992, + "grad_norm": 1.7160844802856445, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8834143877029419, + "num_tokens": 541720390.0, + "step": 15425 + }, + { + "epoch": 2.864623955431755, + "grad_norm": 1.7895197868347168, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8811051845550537, + "num_tokens": 541754476.0, + "step": 15426 + }, + { + "epoch": 2.8648096564531107, + "grad_norm": 1.4561599493026733, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.9000747203826904, + "num_tokens": 541790016.0, + "step": 15427 + }, + { + "epoch": 2.864995357474466, + "grad_norm": 1.5733267068862915, + "learning_rate": 1e-06, + "loss": 0.2824, + "mean_token_accuracy": 0.8976745009422302, + "num_tokens": 541822171.0, + "step": 15428 + }, + { + "epoch": 2.8651810584958217, + "grad_norm": 1.4844528436660767, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8865300416946411, + "num_tokens": 541860801.0, + "step": 15429 + }, + { + "epoch": 2.8653667595171775, + "grad_norm": 1.4877924919128418, + "learning_rate": 1e-06, + "loss": 0.2721, + "mean_token_accuracy": 0.9006614685058594, + "num_tokens": 541892958.0, + "step": 15430 + }, + { + "epoch": 2.8655524605385327, + "grad_norm": 1.6156971454620361, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8842579126358032, + "num_tokens": 541929225.0, + "step": 15431 + }, + { + "epoch": 2.8657381615598885, + "grad_norm": 1.4630827903747559, + "learning_rate": 1e-06, + "loss": 0.2893, + "mean_token_accuracy": 0.8964030742645264, + "num_tokens": 541967461.0, + "step": 15432 + }, + { + "epoch": 2.865923862581244, + "grad_norm": 1.6912448406219482, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8812886476516724, + "num_tokens": 541999220.0, + "step": 15433 + }, + { + "epoch": 2.8661095636026, + "grad_norm": 1.6498349905014038, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.8846322298049927, + "num_tokens": 542031562.0, + "step": 15434 + }, + { + "epoch": 2.8662952646239557, + "grad_norm": 1.5671343803405762, + "learning_rate": 1e-06, + "loss": 0.2728, + "mean_token_accuracy": 0.9016956090927124, + "num_tokens": 542063746.0, + "step": 15435 + }, + { + "epoch": 2.866480965645311, + "grad_norm": 1.6176773309707642, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8771561980247498, + "num_tokens": 542095610.0, + "step": 15436 + }, + { + "epoch": 2.8666666666666667, + "grad_norm": 1.5395243167877197, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8849745988845825, + "num_tokens": 542129793.0, + "step": 15437 + }, + { + "epoch": 2.8668523676880224, + "grad_norm": 1.5313915014266968, + "learning_rate": 1e-06, + "loss": 0.2933, + "mean_token_accuracy": 0.8959054946899414, + "num_tokens": 542165617.0, + "step": 15438 + }, + { + "epoch": 2.8670380687093777, + "grad_norm": 1.6403775215148926, + "learning_rate": 1e-06, + "loss": 0.2845, + "mean_token_accuracy": 0.8965779542922974, + "num_tokens": 542199609.0, + "step": 15439 + }, + { + "epoch": 2.8672237697307335, + "grad_norm": 1.673353910446167, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8831149339675903, + "num_tokens": 542231354.0, + "step": 15440 + }, + { + "epoch": 2.867409470752089, + "grad_norm": 1.41581130027771, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8886804580688477, + "num_tokens": 542274056.0, + "step": 15441 + }, + { + "epoch": 2.867595171773445, + "grad_norm": 1.6265524625778198, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8844470381736755, + "num_tokens": 542306231.0, + "step": 15442 + }, + { + "epoch": 2.8677808727948, + "grad_norm": 1.5971444845199585, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8968468904495239, + "num_tokens": 542337867.0, + "step": 15443 + }, + { + "epoch": 2.867966573816156, + "grad_norm": 1.5871118307113647, + "learning_rate": 1e-06, + "loss": 0.2873, + "mean_token_accuracy": 0.8979135751724243, + "num_tokens": 542367737.0, + "step": 15444 + }, + { + "epoch": 2.8681522748375117, + "grad_norm": 1.625137209892273, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8908610343933105, + "num_tokens": 542398106.0, + "step": 15445 + }, + { + "epoch": 2.868337975858867, + "grad_norm": 1.7165489196777344, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8814303874969482, + "num_tokens": 542430139.0, + "step": 15446 + }, + { + "epoch": 2.8685236768802227, + "grad_norm": 1.4712111949920654, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8816652297973633, + "num_tokens": 542470121.0, + "step": 15447 + }, + { + "epoch": 2.8687093779015784, + "grad_norm": 1.6003875732421875, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8996900320053101, + "num_tokens": 542501592.0, + "step": 15448 + }, + { + "epoch": 2.868895078922934, + "grad_norm": 1.4483625888824463, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8772685527801514, + "num_tokens": 542540040.0, + "step": 15449 + }, + { + "epoch": 2.86908077994429, + "grad_norm": 1.6773574352264404, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8742306232452393, + "num_tokens": 542571277.0, + "step": 15450 + }, + { + "epoch": 2.869266480965645, + "grad_norm": 1.8628227710723877, + "learning_rate": 1e-06, + "loss": 0.388, + "mean_token_accuracy": 0.866258442401886, + "num_tokens": 542602750.0, + "step": 15451 + }, + { + "epoch": 2.869452181987001, + "grad_norm": 1.3807083368301392, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8886260986328125, + "num_tokens": 542642252.0, + "step": 15452 + }, + { + "epoch": 2.8696378830083566, + "grad_norm": 1.4675461053848267, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8956750631332397, + "num_tokens": 542675276.0, + "step": 15453 + }, + { + "epoch": 2.869823584029712, + "grad_norm": 1.5571401119232178, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.892867922782898, + "num_tokens": 542706391.0, + "step": 15454 + }, + { + "epoch": 2.8700092850510677, + "grad_norm": 1.5622739791870117, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8893178105354309, + "num_tokens": 542738305.0, + "step": 15455 + }, + { + "epoch": 2.8701949860724234, + "grad_norm": 1.3901700973510742, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8963507413864136, + "num_tokens": 542777384.0, + "step": 15456 + }, + { + "epoch": 2.870380687093779, + "grad_norm": 1.4088237285614014, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8873955607414246, + "num_tokens": 542814870.0, + "step": 15457 + }, + { + "epoch": 2.870566388115135, + "grad_norm": 1.5587435960769653, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.881924033164978, + "num_tokens": 542854626.0, + "step": 15458 + }, + { + "epoch": 2.87075208913649, + "grad_norm": 1.385337233543396, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8885195255279541, + "num_tokens": 542900528.0, + "step": 15459 + }, + { + "epoch": 2.870937790157846, + "grad_norm": 1.6115875244140625, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8801536560058594, + "num_tokens": 542936264.0, + "step": 15460 + }, + { + "epoch": 2.8711234911792016, + "grad_norm": 1.7033480405807495, + "learning_rate": 1e-06, + "loss": 0.3631, + "mean_token_accuracy": 0.8752111196517944, + "num_tokens": 542969586.0, + "step": 15461 + }, + { + "epoch": 2.871309192200557, + "grad_norm": 1.7345969676971436, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8765430450439453, + "num_tokens": 543001030.0, + "step": 15462 + }, + { + "epoch": 2.8714948932219126, + "grad_norm": 1.3996162414550781, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8953724503517151, + "num_tokens": 543040146.0, + "step": 15463 + }, + { + "epoch": 2.8716805942432684, + "grad_norm": 1.545029878616333, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8868799209594727, + "num_tokens": 543073778.0, + "step": 15464 + }, + { + "epoch": 2.871866295264624, + "grad_norm": 1.5909459590911865, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8827757835388184, + "num_tokens": 543106683.0, + "step": 15465 + }, + { + "epoch": 2.87205199628598, + "grad_norm": 1.5069750547409058, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8818407654762268, + "num_tokens": 543147846.0, + "step": 15466 + }, + { + "epoch": 2.872237697307335, + "grad_norm": 1.5909522771835327, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8863964676856995, + "num_tokens": 543181165.0, + "step": 15467 + }, + { + "epoch": 2.872423398328691, + "grad_norm": 1.6042481660842896, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8922539949417114, + "num_tokens": 543217701.0, + "step": 15468 + }, + { + "epoch": 2.872609099350046, + "grad_norm": 1.5551373958587646, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8989046216011047, + "num_tokens": 543250453.0, + "step": 15469 + }, + { + "epoch": 2.872794800371402, + "grad_norm": 1.7223875522613525, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8825814723968506, + "num_tokens": 543279640.0, + "step": 15470 + }, + { + "epoch": 2.8729805013927576, + "grad_norm": 1.5121296644210815, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8946370482444763, + "num_tokens": 543313023.0, + "step": 15471 + }, + { + "epoch": 2.8731662024141134, + "grad_norm": 1.5221729278564453, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8878879547119141, + "num_tokens": 543349061.0, + "step": 15472 + }, + { + "epoch": 2.873351903435469, + "grad_norm": 1.7698734998703003, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8818994760513306, + "num_tokens": 543379680.0, + "step": 15473 + }, + { + "epoch": 2.8735376044568244, + "grad_norm": 1.506782054901123, + "learning_rate": 1e-06, + "loss": 0.3609, + "mean_token_accuracy": 0.8746446371078491, + "num_tokens": 543421719.0, + "step": 15474 + }, + { + "epoch": 2.87372330547818, + "grad_norm": 1.627435564994812, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8879635334014893, + "num_tokens": 543454371.0, + "step": 15475 + }, + { + "epoch": 2.873909006499536, + "grad_norm": 1.5149810314178467, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8874998092651367, + "num_tokens": 543490184.0, + "step": 15476 + }, + { + "epoch": 2.874094707520891, + "grad_norm": 1.4503040313720703, + "learning_rate": 1e-06, + "loss": 0.3203, + "mean_token_accuracy": 0.886425256729126, + "num_tokens": 543527446.0, + "step": 15477 + }, + { + "epoch": 2.874280408542247, + "grad_norm": 1.6119258403778076, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8792301416397095, + "num_tokens": 543562245.0, + "step": 15478 + }, + { + "epoch": 2.8744661095636026, + "grad_norm": 1.5660218000411987, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.8925438523292542, + "num_tokens": 543595262.0, + "step": 15479 + }, + { + "epoch": 2.8746518105849583, + "grad_norm": 1.5058547258377075, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8815730810165405, + "num_tokens": 543634876.0, + "step": 15480 + }, + { + "epoch": 2.874837511606314, + "grad_norm": 1.6290873289108276, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8817927837371826, + "num_tokens": 543668019.0, + "step": 15481 + }, + { + "epoch": 2.8750232126276694, + "grad_norm": 1.6223433017730713, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.875821590423584, + "num_tokens": 543701314.0, + "step": 15482 + }, + { + "epoch": 2.875208913649025, + "grad_norm": 1.4231258630752563, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8899462223052979, + "num_tokens": 543743027.0, + "step": 15483 + }, + { + "epoch": 2.875394614670381, + "grad_norm": 1.6587814092636108, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8806982040405273, + "num_tokens": 543776532.0, + "step": 15484 + }, + { + "epoch": 2.875580315691736, + "grad_norm": 1.4942682981491089, + "learning_rate": 1e-06, + "loss": 0.2777, + "mean_token_accuracy": 0.9015202522277832, + "num_tokens": 543810922.0, + "step": 15485 + }, + { + "epoch": 2.875766016713092, + "grad_norm": 1.5676164627075195, + "learning_rate": 1e-06, + "loss": 0.3234, + "mean_token_accuracy": 0.8899105191230774, + "num_tokens": 543845659.0, + "step": 15486 + }, + { + "epoch": 2.8759517177344476, + "grad_norm": 1.6320557594299316, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8807990550994873, + "num_tokens": 543879401.0, + "step": 15487 + }, + { + "epoch": 2.8761374187558033, + "grad_norm": 1.472164511680603, + "learning_rate": 1e-06, + "loss": 0.3096, + "mean_token_accuracy": 0.8909656405448914, + "num_tokens": 543917679.0, + "step": 15488 + }, + { + "epoch": 2.876323119777159, + "grad_norm": 1.4853167533874512, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8963724374771118, + "num_tokens": 543954849.0, + "step": 15489 + }, + { + "epoch": 2.8765088207985143, + "grad_norm": 1.645592451095581, + "learning_rate": 1e-06, + "loss": 0.36, + "mean_token_accuracy": 0.8787510395050049, + "num_tokens": 543992499.0, + "step": 15490 + }, + { + "epoch": 2.87669452181987, + "grad_norm": 1.5319052934646606, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.892187237739563, + "num_tokens": 544027551.0, + "step": 15491 + }, + { + "epoch": 2.8768802228412254, + "grad_norm": 1.4392969608306885, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8946936130523682, + "num_tokens": 544065014.0, + "step": 15492 + }, + { + "epoch": 2.877065923862581, + "grad_norm": 1.4611191749572754, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8806573152542114, + "num_tokens": 544105210.0, + "step": 15493 + }, + { + "epoch": 2.877251624883937, + "grad_norm": 1.563744068145752, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8844490647315979, + "num_tokens": 544142331.0, + "step": 15494 + }, + { + "epoch": 2.8774373259052926, + "grad_norm": 1.560884714126587, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8922178745269775, + "num_tokens": 544176379.0, + "step": 15495 + }, + { + "epoch": 2.8776230269266483, + "grad_norm": 1.5614697933197021, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8917917609214783, + "num_tokens": 544217871.0, + "step": 15496 + }, + { + "epoch": 2.8778087279480036, + "grad_norm": 1.4921246767044067, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8952544927597046, + "num_tokens": 544254051.0, + "step": 15497 + }, + { + "epoch": 2.8779944289693593, + "grad_norm": 1.6176856756210327, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.891207218170166, + "num_tokens": 544285151.0, + "step": 15498 + }, + { + "epoch": 2.878180129990715, + "grad_norm": 1.5602185726165771, + "learning_rate": 1e-06, + "loss": 0.3615, + "mean_token_accuracy": 0.8722456693649292, + "num_tokens": 544321851.0, + "step": 15499 + }, + { + "epoch": 2.8783658310120703, + "grad_norm": 1.6677660942077637, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8885185718536377, + "num_tokens": 544353549.0, + "step": 15500 + }, + { + "epoch": 2.878551532033426, + "grad_norm": 1.4529902935028076, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8758496642112732, + "num_tokens": 544395029.0, + "step": 15501 + }, + { + "epoch": 2.878737233054782, + "grad_norm": 1.6133326292037964, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8929301500320435, + "num_tokens": 544426226.0, + "step": 15502 + }, + { + "epoch": 2.8789229340761375, + "grad_norm": 1.6087470054626465, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8907737135887146, + "num_tokens": 544455903.0, + "step": 15503 + }, + { + "epoch": 2.8791086350974933, + "grad_norm": 1.461133599281311, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.8993473052978516, + "num_tokens": 544492234.0, + "step": 15504 + }, + { + "epoch": 2.8792943361188486, + "grad_norm": 1.5502009391784668, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8806365728378296, + "num_tokens": 544532138.0, + "step": 15505 + }, + { + "epoch": 2.8794800371402043, + "grad_norm": 1.566127896308899, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8974697589874268, + "num_tokens": 544566399.0, + "step": 15506 + }, + { + "epoch": 2.87966573816156, + "grad_norm": 1.4940215349197388, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8832492232322693, + "num_tokens": 544607226.0, + "step": 15507 + }, + { + "epoch": 2.8798514391829153, + "grad_norm": 1.4457465410232544, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8876636624336243, + "num_tokens": 544645955.0, + "step": 15508 + }, + { + "epoch": 2.880037140204271, + "grad_norm": 1.5851202011108398, + "learning_rate": 1e-06, + "loss": 0.3068, + "mean_token_accuracy": 0.8916918635368347, + "num_tokens": 544677636.0, + "step": 15509 + }, + { + "epoch": 2.8802228412256268, + "grad_norm": 1.512509822845459, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8676782846450806, + "num_tokens": 544720764.0, + "step": 15510 + }, + { + "epoch": 2.8804085422469825, + "grad_norm": 1.7107081413269043, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8800438642501831, + "num_tokens": 544753685.0, + "step": 15511 + }, + { + "epoch": 2.8805942432683382, + "grad_norm": 1.56722891330719, + "learning_rate": 1e-06, + "loss": 0.2948, + "mean_token_accuracy": 0.8940038681030273, + "num_tokens": 544786009.0, + "step": 15512 + }, + { + "epoch": 2.8807799442896935, + "grad_norm": 1.4912526607513428, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.8992162346839905, + "num_tokens": 544824848.0, + "step": 15513 + }, + { + "epoch": 2.8809656453110493, + "grad_norm": 1.6040838956832886, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8829286098480225, + "num_tokens": 544856509.0, + "step": 15514 + }, + { + "epoch": 2.881151346332405, + "grad_norm": 1.5182335376739502, + "learning_rate": 1e-06, + "loss": 0.3021, + "mean_token_accuracy": 0.8931734561920166, + "num_tokens": 544890168.0, + "step": 15515 + }, + { + "epoch": 2.8813370473537603, + "grad_norm": 1.5672928094863892, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8875114917755127, + "num_tokens": 544921846.0, + "step": 15516 + }, + { + "epoch": 2.881522748375116, + "grad_norm": 1.4212682247161865, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8960375785827637, + "num_tokens": 544961583.0, + "step": 15517 + }, + { + "epoch": 2.8817084493964717, + "grad_norm": 1.7312246561050415, + "learning_rate": 1e-06, + "loss": 0.3541, + "mean_token_accuracy": 0.8781378269195557, + "num_tokens": 544989679.0, + "step": 15518 + }, + { + "epoch": 2.8818941504178275, + "grad_norm": 1.5122424364089966, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.9091899394989014, + "num_tokens": 545027514.0, + "step": 15519 + }, + { + "epoch": 2.8820798514391828, + "grad_norm": 1.53788423538208, + "learning_rate": 1e-06, + "loss": 0.3358, + "mean_token_accuracy": 0.8849002122879028, + "num_tokens": 545063051.0, + "step": 15520 + }, + { + "epoch": 2.8822655524605385, + "grad_norm": 1.4525222778320312, + "learning_rate": 1e-06, + "loss": 0.2757, + "mean_token_accuracy": 0.9025788307189941, + "num_tokens": 545096542.0, + "step": 15521 + }, + { + "epoch": 2.8824512534818942, + "grad_norm": 1.677952527999878, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.8931599855422974, + "num_tokens": 545124951.0, + "step": 15522 + }, + { + "epoch": 2.8826369545032495, + "grad_norm": 1.4739540815353394, + "learning_rate": 1e-06, + "loss": 0.2656, + "mean_token_accuracy": 0.903593897819519, + "num_tokens": 545161472.0, + "step": 15523 + }, + { + "epoch": 2.8828226555246053, + "grad_norm": 1.4217084646224976, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8855117559432983, + "num_tokens": 545204815.0, + "step": 15524 + }, + { + "epoch": 2.883008356545961, + "grad_norm": 1.5663378238677979, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.883568286895752, + "num_tokens": 545240267.0, + "step": 15525 + }, + { + "epoch": 2.8831940575673167, + "grad_norm": 1.5220211744308472, + "learning_rate": 1e-06, + "loss": 0.3044, + "mean_token_accuracy": 0.8961249589920044, + "num_tokens": 545278634.0, + "step": 15526 + }, + { + "epoch": 2.8833797585886725, + "grad_norm": 1.4796489477157593, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8854402899742126, + "num_tokens": 545313725.0, + "step": 15527 + }, + { + "epoch": 2.8835654596100277, + "grad_norm": 1.6012580394744873, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8741446733474731, + "num_tokens": 545347885.0, + "step": 15528 + }, + { + "epoch": 2.8837511606313835, + "grad_norm": 1.5816200971603394, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8864790201187134, + "num_tokens": 545381427.0, + "step": 15529 + }, + { + "epoch": 2.883936861652739, + "grad_norm": 1.5168393850326538, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8883799910545349, + "num_tokens": 545418282.0, + "step": 15530 + }, + { + "epoch": 2.8841225626740945, + "grad_norm": 1.6152344942092896, + "learning_rate": 1e-06, + "loss": 0.4011, + "mean_token_accuracy": 0.8656163811683655, + "num_tokens": 545456570.0, + "step": 15531 + }, + { + "epoch": 2.8843082636954502, + "grad_norm": 1.6062543392181396, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8928962349891663, + "num_tokens": 545487593.0, + "step": 15532 + }, + { + "epoch": 2.884493964716806, + "grad_norm": 1.5169347524642944, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8796401619911194, + "num_tokens": 545527451.0, + "step": 15533 + }, + { + "epoch": 2.8846796657381617, + "grad_norm": 1.724158525466919, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8861143589019775, + "num_tokens": 545554323.0, + "step": 15534 + }, + { + "epoch": 2.8848653667595174, + "grad_norm": 1.8261481523513794, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.885087251663208, + "num_tokens": 545583426.0, + "step": 15535 + }, + { + "epoch": 2.8850510677808727, + "grad_norm": 1.6361820697784424, + "learning_rate": 1e-06, + "loss": 0.3288, + "mean_token_accuracy": 0.8841325044631958, + "num_tokens": 545617118.0, + "step": 15536 + }, + { + "epoch": 2.8852367688022285, + "grad_norm": 1.6394009590148926, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8819512724876404, + "num_tokens": 545648680.0, + "step": 15537 + }, + { + "epoch": 2.885422469823584, + "grad_norm": 1.4425984621047974, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.89115971326828, + "num_tokens": 545689581.0, + "step": 15538 + }, + { + "epoch": 2.8856081708449395, + "grad_norm": 1.3764433860778809, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8983579874038696, + "num_tokens": 545729237.0, + "step": 15539 + }, + { + "epoch": 2.885793871866295, + "grad_norm": 1.559454083442688, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8862453699111938, + "num_tokens": 545764397.0, + "step": 15540 + }, + { + "epoch": 2.885979572887651, + "grad_norm": 1.5282272100448608, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8857093453407288, + "num_tokens": 545800831.0, + "step": 15541 + }, + { + "epoch": 2.8861652739090067, + "grad_norm": 1.4568122625350952, + "learning_rate": 1e-06, + "loss": 0.3186, + "mean_token_accuracy": 0.8865252137184143, + "num_tokens": 545841558.0, + "step": 15542 + }, + { + "epoch": 2.886350974930362, + "grad_norm": 1.6397581100463867, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8759129047393799, + "num_tokens": 545874310.0, + "step": 15543 + }, + { + "epoch": 2.8865366759517177, + "grad_norm": 1.4617265462875366, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8853211998939514, + "num_tokens": 545916669.0, + "step": 15544 + }, + { + "epoch": 2.8867223769730734, + "grad_norm": 1.483289361000061, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8808708190917969, + "num_tokens": 545953748.0, + "step": 15545 + }, + { + "epoch": 2.8869080779944287, + "grad_norm": 1.7857606410980225, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.89207923412323, + "num_tokens": 545981986.0, + "step": 15546 + }, + { + "epoch": 2.8870937790157845, + "grad_norm": 1.6863011121749878, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.876521646976471, + "num_tokens": 546013173.0, + "step": 15547 + }, + { + "epoch": 2.88727948003714, + "grad_norm": 1.4990636110305786, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8877670764923096, + "num_tokens": 546048601.0, + "step": 15548 + }, + { + "epoch": 2.887465181058496, + "grad_norm": 1.6917222738265991, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.877427875995636, + "num_tokens": 546082476.0, + "step": 15549 + }, + { + "epoch": 2.8876508820798517, + "grad_norm": 1.6325056552886963, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8830548524856567, + "num_tokens": 546113659.0, + "step": 15550 + }, + { + "epoch": 2.887836583101207, + "grad_norm": 1.6840606927871704, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8834232091903687, + "num_tokens": 546144870.0, + "step": 15551 + }, + { + "epoch": 2.8880222841225627, + "grad_norm": 1.526106834411621, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8950672149658203, + "num_tokens": 546181670.0, + "step": 15552 + }, + { + "epoch": 2.8882079851439184, + "grad_norm": 1.5614705085754395, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8907029032707214, + "num_tokens": 546216869.0, + "step": 15553 + }, + { + "epoch": 2.8883936861652737, + "grad_norm": 1.5572789907455444, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.895946204662323, + "num_tokens": 546246732.0, + "step": 15554 + }, + { + "epoch": 2.8885793871866294, + "grad_norm": 1.5837985277175903, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8884062170982361, + "num_tokens": 546281406.0, + "step": 15555 + }, + { + "epoch": 2.888765088207985, + "grad_norm": 1.5977979898452759, + "learning_rate": 1e-06, + "loss": 0.2791, + "mean_token_accuracy": 0.8988848924636841, + "num_tokens": 546312117.0, + "step": 15556 + }, + { + "epoch": 2.888950789229341, + "grad_norm": 1.5415335893630981, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8905304074287415, + "num_tokens": 546348096.0, + "step": 15557 + }, + { + "epoch": 2.8891364902506966, + "grad_norm": 1.5322922468185425, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8874861001968384, + "num_tokens": 546382185.0, + "step": 15558 + }, + { + "epoch": 2.889322191272052, + "grad_norm": 1.4790170192718506, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8748884201049805, + "num_tokens": 546425037.0, + "step": 15559 + }, + { + "epoch": 2.8895078922934077, + "grad_norm": 1.7269866466522217, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8864378929138184, + "num_tokens": 546456267.0, + "step": 15560 + }, + { + "epoch": 2.8896935933147634, + "grad_norm": 1.5522410869598389, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8864291310310364, + "num_tokens": 546493715.0, + "step": 15561 + }, + { + "epoch": 2.8898792943361187, + "grad_norm": 1.6165817975997925, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8714263439178467, + "num_tokens": 546527353.0, + "step": 15562 + }, + { + "epoch": 2.8900649953574744, + "grad_norm": 1.527980089187622, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8942037224769592, + "num_tokens": 546560471.0, + "step": 15563 + }, + { + "epoch": 2.89025069637883, + "grad_norm": 1.7057260274887085, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8834125995635986, + "num_tokens": 546593648.0, + "step": 15564 + }, + { + "epoch": 2.890436397400186, + "grad_norm": 1.583856463432312, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8850557804107666, + "num_tokens": 546628440.0, + "step": 15565 + }, + { + "epoch": 2.890622098421541, + "grad_norm": 1.5095685720443726, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8809250593185425, + "num_tokens": 546669618.0, + "step": 15566 + }, + { + "epoch": 2.890807799442897, + "grad_norm": 1.4748470783233643, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8921406269073486, + "num_tokens": 546705041.0, + "step": 15567 + }, + { + "epoch": 2.8909935004642526, + "grad_norm": 1.6256539821624756, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8808462619781494, + "num_tokens": 546739032.0, + "step": 15568 + }, + { + "epoch": 2.891179201485608, + "grad_norm": 1.7019840478897095, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8809194564819336, + "num_tokens": 546766316.0, + "step": 15569 + }, + { + "epoch": 2.8913649025069637, + "grad_norm": 1.5584161281585693, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8953002691268921, + "num_tokens": 546801106.0, + "step": 15570 + }, + { + "epoch": 2.8915506035283194, + "grad_norm": 1.4817719459533691, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8903896808624268, + "num_tokens": 546836761.0, + "step": 15571 + }, + { + "epoch": 2.891736304549675, + "grad_norm": 1.5658694505691528, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8812795877456665, + "num_tokens": 546871574.0, + "step": 15572 + }, + { + "epoch": 2.891922005571031, + "grad_norm": 1.4620994329452515, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.8964465260505676, + "num_tokens": 546914605.0, + "step": 15573 + }, + { + "epoch": 2.892107706592386, + "grad_norm": 1.444860816001892, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8928921818733215, + "num_tokens": 546952680.0, + "step": 15574 + }, + { + "epoch": 2.892293407613742, + "grad_norm": 1.5637552738189697, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8796302080154419, + "num_tokens": 546987133.0, + "step": 15575 + }, + { + "epoch": 2.8924791086350976, + "grad_norm": 1.8223392963409424, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8841198682785034, + "num_tokens": 547018485.0, + "step": 15576 + }, + { + "epoch": 2.892664809656453, + "grad_norm": 1.7873417139053345, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8886545300483704, + "num_tokens": 547051306.0, + "step": 15577 + }, + { + "epoch": 2.8928505106778086, + "grad_norm": 1.495959758758545, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8901227712631226, + "num_tokens": 547086844.0, + "step": 15578 + }, + { + "epoch": 2.8930362116991644, + "grad_norm": 1.656017541885376, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.896140456199646, + "num_tokens": 547120092.0, + "step": 15579 + }, + { + "epoch": 2.89322191272052, + "grad_norm": 1.5885674953460693, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8742613792419434, + "num_tokens": 547156023.0, + "step": 15580 + }, + { + "epoch": 2.893407613741876, + "grad_norm": 1.574705719947815, + "learning_rate": 1e-06, + "loss": 0.3415, + "mean_token_accuracy": 0.8803911209106445, + "num_tokens": 547191872.0, + "step": 15581 + }, + { + "epoch": 2.893593314763231, + "grad_norm": 1.5122900009155273, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8970436453819275, + "num_tokens": 547231223.0, + "step": 15582 + }, + { + "epoch": 2.893779015784587, + "grad_norm": 1.689549446105957, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8844202160835266, + "num_tokens": 547269234.0, + "step": 15583 + }, + { + "epoch": 2.8939647168059426, + "grad_norm": 1.6690499782562256, + "learning_rate": 1e-06, + "loss": 0.315, + "mean_token_accuracy": 0.8894325494766235, + "num_tokens": 547305257.0, + "step": 15584 + }, + { + "epoch": 2.894150417827298, + "grad_norm": 1.5210806131362915, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8818512558937073, + "num_tokens": 547340288.0, + "step": 15585 + }, + { + "epoch": 2.8943361188486536, + "grad_norm": 1.4818605184555054, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8916565179824829, + "num_tokens": 547377479.0, + "step": 15586 + }, + { + "epoch": 2.8945218198700093, + "grad_norm": 1.6088789701461792, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.8950687646865845, + "num_tokens": 547411077.0, + "step": 15587 + }, + { + "epoch": 2.894707520891365, + "grad_norm": 1.4928278923034668, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8798376321792603, + "num_tokens": 547454524.0, + "step": 15588 + }, + { + "epoch": 2.8948932219127204, + "grad_norm": 1.8571090698242188, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8806466460227966, + "num_tokens": 547483375.0, + "step": 15589 + }, + { + "epoch": 2.895078922934076, + "grad_norm": 1.6727441549301147, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8928413391113281, + "num_tokens": 547517005.0, + "step": 15590 + }, + { + "epoch": 2.895264623955432, + "grad_norm": 1.6630170345306396, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8833609819412231, + "num_tokens": 547548515.0, + "step": 15591 + }, + { + "epoch": 2.895450324976787, + "grad_norm": 1.6094504594802856, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8927686214447021, + "num_tokens": 547581669.0, + "step": 15592 + }, + { + "epoch": 2.895636025998143, + "grad_norm": 1.6205745935440063, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8906805515289307, + "num_tokens": 547612948.0, + "step": 15593 + }, + { + "epoch": 2.8958217270194986, + "grad_norm": 1.519274353981018, + "learning_rate": 1e-06, + "loss": 0.3029, + "mean_token_accuracy": 0.8914754390716553, + "num_tokens": 547647209.0, + "step": 15594 + }, + { + "epoch": 2.8960074280408543, + "grad_norm": 1.7370907068252563, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8805623650550842, + "num_tokens": 547680286.0, + "step": 15595 + }, + { + "epoch": 2.89619312906221, + "grad_norm": 1.417799711227417, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.906251072883606, + "num_tokens": 547713647.0, + "step": 15596 + }, + { + "epoch": 2.8963788300835653, + "grad_norm": 1.314775824546814, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.8956204652786255, + "num_tokens": 547754042.0, + "step": 15597 + }, + { + "epoch": 2.896564531104921, + "grad_norm": 1.3935712575912476, + "learning_rate": 1e-06, + "loss": 0.2745, + "mean_token_accuracy": 0.9012285470962524, + "num_tokens": 547793979.0, + "step": 15598 + }, + { + "epoch": 2.896750232126277, + "grad_norm": 1.4076536893844604, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.892379105091095, + "num_tokens": 547834829.0, + "step": 15599 + }, + { + "epoch": 2.896935933147632, + "grad_norm": 1.7042925357818604, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8807861804962158, + "num_tokens": 547865090.0, + "step": 15600 + }, + { + "epoch": 2.897121634168988, + "grad_norm": 1.4328784942626953, + "learning_rate": 1e-06, + "loss": 0.2758, + "mean_token_accuracy": 0.9015981554985046, + "num_tokens": 547901485.0, + "step": 15601 + }, + { + "epoch": 2.8973073351903436, + "grad_norm": 1.5324851274490356, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8806764483451843, + "num_tokens": 547940578.0, + "step": 15602 + }, + { + "epoch": 2.8974930362116993, + "grad_norm": 1.5971592664718628, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8845754265785217, + "num_tokens": 547973595.0, + "step": 15603 + }, + { + "epoch": 2.897678737233055, + "grad_norm": 1.5177021026611328, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8859935998916626, + "num_tokens": 548009623.0, + "step": 15604 + }, + { + "epoch": 2.8978644382544103, + "grad_norm": 1.3820346593856812, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.886069118976593, + "num_tokens": 548054487.0, + "step": 15605 + }, + { + "epoch": 2.898050139275766, + "grad_norm": 1.5429115295410156, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8927063345909119, + "num_tokens": 548087543.0, + "step": 15606 + }, + { + "epoch": 2.898235840297122, + "grad_norm": 1.5231493711471558, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8768024444580078, + "num_tokens": 548128220.0, + "step": 15607 + }, + { + "epoch": 2.898421541318477, + "grad_norm": 1.413083791732788, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8897964358329773, + "num_tokens": 548170778.0, + "step": 15608 + }, + { + "epoch": 2.898607242339833, + "grad_norm": 1.5598362684249878, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8806824088096619, + "num_tokens": 548212939.0, + "step": 15609 + }, + { + "epoch": 2.8987929433611885, + "grad_norm": 1.5693830251693726, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8770979046821594, + "num_tokens": 548251002.0, + "step": 15610 + }, + { + "epoch": 2.8989786443825443, + "grad_norm": 1.6945216655731201, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8809075355529785, + "num_tokens": 548280775.0, + "step": 15611 + }, + { + "epoch": 2.8991643454038996, + "grad_norm": 1.455462098121643, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8878349661827087, + "num_tokens": 548318705.0, + "step": 15612 + }, + { + "epoch": 2.8993500464252553, + "grad_norm": 1.4235682487487793, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8921826481819153, + "num_tokens": 548358221.0, + "step": 15613 + }, + { + "epoch": 2.899535747446611, + "grad_norm": 1.719942569732666, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8751726150512695, + "num_tokens": 548387623.0, + "step": 15614 + }, + { + "epoch": 2.8997214484679663, + "grad_norm": 1.508721947669983, + "learning_rate": 1e-06, + "loss": 0.2855, + "mean_token_accuracy": 0.8993006944656372, + "num_tokens": 548419120.0, + "step": 15615 + }, + { + "epoch": 2.899907149489322, + "grad_norm": 1.4762182235717773, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8809291124343872, + "num_tokens": 548457304.0, + "step": 15616 + }, + { + "epoch": 2.9000928505106778, + "grad_norm": 1.5224918127059937, + "learning_rate": 1e-06, + "loss": 0.2691, + "mean_token_accuracy": 0.9040068984031677, + "num_tokens": 548486890.0, + "step": 15617 + }, + { + "epoch": 2.9002785515320335, + "grad_norm": 1.4841442108154297, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8831208944320679, + "num_tokens": 548525977.0, + "step": 15618 + }, + { + "epoch": 2.9004642525533892, + "grad_norm": 1.4725594520568848, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8918263912200928, + "num_tokens": 548563170.0, + "step": 15619 + }, + { + "epoch": 2.9006499535747445, + "grad_norm": 1.4937286376953125, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8991814255714417, + "num_tokens": 548599582.0, + "step": 15620 + }, + { + "epoch": 2.9008356545961003, + "grad_norm": 1.5200265645980835, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8844360113143921, + "num_tokens": 548635021.0, + "step": 15621 + }, + { + "epoch": 2.901021355617456, + "grad_norm": 1.5036046504974365, + "learning_rate": 1e-06, + "loss": 0.3516, + "mean_token_accuracy": 0.8760032653808594, + "num_tokens": 548676269.0, + "step": 15622 + }, + { + "epoch": 2.9012070566388113, + "grad_norm": 1.7029541730880737, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8699378371238708, + "num_tokens": 548707634.0, + "step": 15623 + }, + { + "epoch": 2.901392757660167, + "grad_norm": 1.533563494682312, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8977037668228149, + "num_tokens": 548740718.0, + "step": 15624 + }, + { + "epoch": 2.9015784586815228, + "grad_norm": 1.6580214500427246, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8773897886276245, + "num_tokens": 548774125.0, + "step": 15625 + }, + { + "epoch": 2.9017641597028785, + "grad_norm": 1.3707448244094849, + "learning_rate": 1e-06, + "loss": 0.2618, + "mean_token_accuracy": 0.9058912396430969, + "num_tokens": 548813512.0, + "step": 15626 + }, + { + "epoch": 2.901949860724234, + "grad_norm": 1.6597217321395874, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8788532018661499, + "num_tokens": 548843847.0, + "step": 15627 + }, + { + "epoch": 2.9021355617455895, + "grad_norm": 1.5926918983459473, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8920848369598389, + "num_tokens": 548874275.0, + "step": 15628 + }, + { + "epoch": 2.9023212627669452, + "grad_norm": 1.6590057611465454, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.8856262564659119, + "num_tokens": 548906930.0, + "step": 15629 + }, + { + "epoch": 2.902506963788301, + "grad_norm": 1.474696397781372, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8951200246810913, + "num_tokens": 548947828.0, + "step": 15630 + }, + { + "epoch": 2.9026926648096563, + "grad_norm": 1.5290372371673584, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.8952361345291138, + "num_tokens": 548983324.0, + "step": 15631 + }, + { + "epoch": 2.902878365831012, + "grad_norm": 1.5631330013275146, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8944608569145203, + "num_tokens": 549015575.0, + "step": 15632 + }, + { + "epoch": 2.9030640668523677, + "grad_norm": 1.4588677883148193, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.894160270690918, + "num_tokens": 549053257.0, + "step": 15633 + }, + { + "epoch": 2.9032497678737235, + "grad_norm": 1.696186900138855, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8960645198822021, + "num_tokens": 549080338.0, + "step": 15634 + }, + { + "epoch": 2.903435468895079, + "grad_norm": 1.6677707433700562, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8973986506462097, + "num_tokens": 549109082.0, + "step": 15635 + }, + { + "epoch": 2.9036211699164345, + "grad_norm": 1.4880433082580566, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8896793127059937, + "num_tokens": 549144950.0, + "step": 15636 + }, + { + "epoch": 2.90380687093779, + "grad_norm": 1.534946084022522, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8866024017333984, + "num_tokens": 549179469.0, + "step": 15637 + }, + { + "epoch": 2.9039925719591455, + "grad_norm": 1.5579394102096558, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.889167308807373, + "num_tokens": 549214848.0, + "step": 15638 + }, + { + "epoch": 2.9041782729805012, + "grad_norm": 1.5624977350234985, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8879599571228027, + "num_tokens": 549248208.0, + "step": 15639 + }, + { + "epoch": 2.904363974001857, + "grad_norm": 1.463335394859314, + "learning_rate": 1e-06, + "loss": 0.3275, + "mean_token_accuracy": 0.8849267959594727, + "num_tokens": 549288655.0, + "step": 15640 + }, + { + "epoch": 2.9045496750232127, + "grad_norm": 1.47721529006958, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8746082186698914, + "num_tokens": 549333626.0, + "step": 15641 + }, + { + "epoch": 2.9047353760445684, + "grad_norm": 1.379586935043335, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8866758346557617, + "num_tokens": 549375165.0, + "step": 15642 + }, + { + "epoch": 2.9049210770659237, + "grad_norm": 1.471899390220642, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8901534080505371, + "num_tokens": 549410801.0, + "step": 15643 + }, + { + "epoch": 2.9051067780872795, + "grad_norm": 1.5518068075180054, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8851084113121033, + "num_tokens": 549450297.0, + "step": 15644 + }, + { + "epoch": 2.905292479108635, + "grad_norm": 1.827897310256958, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8806235790252686, + "num_tokens": 549478324.0, + "step": 15645 + }, + { + "epoch": 2.9054781801299905, + "grad_norm": 1.7169429063796997, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8749669194221497, + "num_tokens": 549510495.0, + "step": 15646 + }, + { + "epoch": 2.905663881151346, + "grad_norm": 1.4695466756820679, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8827037811279297, + "num_tokens": 549552231.0, + "step": 15647 + }, + { + "epoch": 2.905849582172702, + "grad_norm": 1.6402727365493774, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8923681974411011, + "num_tokens": 549584284.0, + "step": 15648 + }, + { + "epoch": 2.9060352831940577, + "grad_norm": 1.7973579168319702, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8761550188064575, + "num_tokens": 549616788.0, + "step": 15649 + }, + { + "epoch": 2.9062209842154134, + "grad_norm": 1.6310371160507202, + "learning_rate": 1e-06, + "loss": 0.3219, + "mean_token_accuracy": 0.8882924914360046, + "num_tokens": 549650687.0, + "step": 15650 + }, + { + "epoch": 2.9064066852367687, + "grad_norm": 1.4087281227111816, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.88456130027771, + "num_tokens": 549692583.0, + "step": 15651 + }, + { + "epoch": 2.9065923862581244, + "grad_norm": 1.361153244972229, + "learning_rate": 1e-06, + "loss": 0.3049, + "mean_token_accuracy": 0.8926081657409668, + "num_tokens": 549733840.0, + "step": 15652 + }, + { + "epoch": 2.90677808727948, + "grad_norm": 1.6065467596054077, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8893779516220093, + "num_tokens": 549768048.0, + "step": 15653 + }, + { + "epoch": 2.9069637883008355, + "grad_norm": 1.4880021810531616, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8908768892288208, + "num_tokens": 549802569.0, + "step": 15654 + }, + { + "epoch": 2.907149489322191, + "grad_norm": 1.4077876806259155, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8901086449623108, + "num_tokens": 549842480.0, + "step": 15655 + }, + { + "epoch": 2.907335190343547, + "grad_norm": 1.5177925825119019, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8908343315124512, + "num_tokens": 549878352.0, + "step": 15656 + }, + { + "epoch": 2.9075208913649027, + "grad_norm": 1.4938431978225708, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8802171945571899, + "num_tokens": 549914868.0, + "step": 15657 + }, + { + "epoch": 2.9077065923862584, + "grad_norm": 1.767123818397522, + "learning_rate": 1e-06, + "loss": 0.3828, + "mean_token_accuracy": 0.8687561750411987, + "num_tokens": 549951111.0, + "step": 15658 + }, + { + "epoch": 2.9078922934076137, + "grad_norm": 1.4184784889221191, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.9001399278640747, + "num_tokens": 549991230.0, + "step": 15659 + }, + { + "epoch": 2.9080779944289694, + "grad_norm": 1.4656167030334473, + "learning_rate": 1e-06, + "loss": 0.2767, + "mean_token_accuracy": 0.8985396027565002, + "num_tokens": 550025942.0, + "step": 15660 + }, + { + "epoch": 2.9082636954503247, + "grad_norm": 1.7323447465896606, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8874136209487915, + "num_tokens": 550054891.0, + "step": 15661 + }, + { + "epoch": 2.9084493964716804, + "grad_norm": 1.4289048910140991, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.8937127590179443, + "num_tokens": 550097434.0, + "step": 15662 + }, + { + "epoch": 2.908635097493036, + "grad_norm": 1.6471282243728638, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8757637739181519, + "num_tokens": 550130656.0, + "step": 15663 + }, + { + "epoch": 2.908820798514392, + "grad_norm": 1.5530507564544678, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8829874992370605, + "num_tokens": 550166970.0, + "step": 15664 + }, + { + "epoch": 2.9090064995357476, + "grad_norm": 1.589161992073059, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8760193586349487, + "num_tokens": 550204256.0, + "step": 15665 + }, + { + "epoch": 2.909192200557103, + "grad_norm": 1.532942295074463, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.882897138595581, + "num_tokens": 550242481.0, + "step": 15666 + }, + { + "epoch": 2.9093779015784587, + "grad_norm": 1.5573160648345947, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.8970412015914917, + "num_tokens": 550273973.0, + "step": 15667 + }, + { + "epoch": 2.9095636025998144, + "grad_norm": 1.5447584390640259, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.8959090709686279, + "num_tokens": 550306009.0, + "step": 15668 + }, + { + "epoch": 2.9097493036211697, + "grad_norm": 1.6258901357650757, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8799178004264832, + "num_tokens": 550342683.0, + "step": 15669 + }, + { + "epoch": 2.9099350046425254, + "grad_norm": 1.4705308675765991, + "learning_rate": 1e-06, + "loss": 0.3049, + "mean_token_accuracy": 0.8895076513290405, + "num_tokens": 550380445.0, + "step": 15670 + }, + { + "epoch": 2.910120705663881, + "grad_norm": 1.5936744213104248, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8767057061195374, + "num_tokens": 550414717.0, + "step": 15671 + }, + { + "epoch": 2.910306406685237, + "grad_norm": 1.6118590831756592, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8875868320465088, + "num_tokens": 550447620.0, + "step": 15672 + }, + { + "epoch": 2.9104921077065926, + "grad_norm": 1.4741002321243286, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.8910930156707764, + "num_tokens": 550486708.0, + "step": 15673 + }, + { + "epoch": 2.910677808727948, + "grad_norm": 1.4215514659881592, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8947328329086304, + "num_tokens": 550524998.0, + "step": 15674 + }, + { + "epoch": 2.9108635097493036, + "grad_norm": 1.5180904865264893, + "learning_rate": 1e-06, + "loss": 0.2796, + "mean_token_accuracy": 0.8992739915847778, + "num_tokens": 550560946.0, + "step": 15675 + }, + { + "epoch": 2.9110492107706594, + "grad_norm": 1.4676194190979004, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8862553834915161, + "num_tokens": 550599599.0, + "step": 15676 + }, + { + "epoch": 2.9112349117920147, + "grad_norm": 1.4723454713821411, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.896337628364563, + "num_tokens": 550637854.0, + "step": 15677 + }, + { + "epoch": 2.9114206128133704, + "grad_norm": 1.4976329803466797, + "learning_rate": 1e-06, + "loss": 0.2666, + "mean_token_accuracy": 0.905443012714386, + "num_tokens": 550671481.0, + "step": 15678 + }, + { + "epoch": 2.911606313834726, + "grad_norm": 1.5579665899276733, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8906564116477966, + "num_tokens": 550703125.0, + "step": 15679 + }, + { + "epoch": 2.911792014856082, + "grad_norm": 1.723336100578308, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8801009058952332, + "num_tokens": 550732438.0, + "step": 15680 + }, + { + "epoch": 2.9119777158774376, + "grad_norm": 1.4195231199264526, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8947075009346008, + "num_tokens": 550774194.0, + "step": 15681 + }, + { + "epoch": 2.912163416898793, + "grad_norm": 1.5375003814697266, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8865504264831543, + "num_tokens": 550810544.0, + "step": 15682 + }, + { + "epoch": 2.9123491179201486, + "grad_norm": 1.6165125370025635, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8887752890586853, + "num_tokens": 550842229.0, + "step": 15683 + }, + { + "epoch": 2.9125348189415043, + "grad_norm": 1.4943667650222778, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8941808938980103, + "num_tokens": 550877703.0, + "step": 15684 + }, + { + "epoch": 2.9127205199628596, + "grad_norm": 1.5307859182357788, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8819072842597961, + "num_tokens": 550919216.0, + "step": 15685 + }, + { + "epoch": 2.9129062209842154, + "grad_norm": 1.56471848487854, + "learning_rate": 1e-06, + "loss": 0.3148, + "mean_token_accuracy": 0.8904649019241333, + "num_tokens": 550951996.0, + "step": 15686 + }, + { + "epoch": 2.913091922005571, + "grad_norm": 1.526210069656372, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8839190006256104, + "num_tokens": 550988949.0, + "step": 15687 + }, + { + "epoch": 2.913277623026927, + "grad_norm": 1.5865458250045776, + "learning_rate": 1e-06, + "loss": 0.3762, + "mean_token_accuracy": 0.8715319633483887, + "num_tokens": 551028744.0, + "step": 15688 + }, + { + "epoch": 2.913463324048282, + "grad_norm": 1.5872734785079956, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8893672227859497, + "num_tokens": 551064129.0, + "step": 15689 + }, + { + "epoch": 2.913649025069638, + "grad_norm": 1.7050011157989502, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8901427984237671, + "num_tokens": 551092005.0, + "step": 15690 + }, + { + "epoch": 2.9138347260909936, + "grad_norm": 1.5152393579483032, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8928738236427307, + "num_tokens": 551125439.0, + "step": 15691 + }, + { + "epoch": 2.914020427112349, + "grad_norm": 1.4779291152954102, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8805237412452698, + "num_tokens": 551167144.0, + "step": 15692 + }, + { + "epoch": 2.9142061281337046, + "grad_norm": 1.5355043411254883, + "learning_rate": 1e-06, + "loss": 0.2768, + "mean_token_accuracy": 0.9026175737380981, + "num_tokens": 551199330.0, + "step": 15693 + }, + { + "epoch": 2.9143918291550603, + "grad_norm": 1.7557586431503296, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8875165581703186, + "num_tokens": 551225955.0, + "step": 15694 + }, + { + "epoch": 2.914577530176416, + "grad_norm": 1.470118522644043, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8904620409011841, + "num_tokens": 551265164.0, + "step": 15695 + }, + { + "epoch": 2.914763231197772, + "grad_norm": 1.520577311515808, + "learning_rate": 1e-06, + "loss": 0.3689, + "mean_token_accuracy": 0.8702137470245361, + "num_tokens": 551305264.0, + "step": 15696 + }, + { + "epoch": 2.914948932219127, + "grad_norm": 1.5512229204177856, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8912316560745239, + "num_tokens": 551344827.0, + "step": 15697 + }, + { + "epoch": 2.915134633240483, + "grad_norm": 1.4723819494247437, + "learning_rate": 1e-06, + "loss": 0.2984, + "mean_token_accuracy": 0.8948881030082703, + "num_tokens": 551384019.0, + "step": 15698 + }, + { + "epoch": 2.9153203342618386, + "grad_norm": 1.6224254369735718, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8773905634880066, + "num_tokens": 551418719.0, + "step": 15699 + }, + { + "epoch": 2.915506035283194, + "grad_norm": 1.5429154634475708, + "learning_rate": 1e-06, + "loss": 0.3115, + "mean_token_accuracy": 0.8926956653594971, + "num_tokens": 551453974.0, + "step": 15700 + }, + { + "epoch": 2.9156917363045496, + "grad_norm": 1.4959720373153687, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.8858150839805603, + "num_tokens": 551494217.0, + "step": 15701 + }, + { + "epoch": 2.9158774373259053, + "grad_norm": 1.5170488357543945, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8838143348693848, + "num_tokens": 551531378.0, + "step": 15702 + }, + { + "epoch": 2.916063138347261, + "grad_norm": 1.4831852912902832, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8823642730712891, + "num_tokens": 551569469.0, + "step": 15703 + }, + { + "epoch": 2.916248839368617, + "grad_norm": 1.4440891742706299, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8962334990501404, + "num_tokens": 551606609.0, + "step": 15704 + }, + { + "epoch": 2.916434540389972, + "grad_norm": 1.5774630308151245, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8747842311859131, + "num_tokens": 551640400.0, + "step": 15705 + }, + { + "epoch": 2.916620241411328, + "grad_norm": 1.973997950553894, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8710427284240723, + "num_tokens": 551666739.0, + "step": 15706 + }, + { + "epoch": 2.9168059424326835, + "grad_norm": 1.601977825164795, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8926796317100525, + "num_tokens": 551698204.0, + "step": 15707 + }, + { + "epoch": 2.916991643454039, + "grad_norm": 1.5499231815338135, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8806902766227722, + "num_tokens": 551735870.0, + "step": 15708 + }, + { + "epoch": 2.9171773444753946, + "grad_norm": 1.4185664653778076, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8907894492149353, + "num_tokens": 551774665.0, + "step": 15709 + }, + { + "epoch": 2.9173630454967503, + "grad_norm": 1.7926799058914185, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8918575048446655, + "num_tokens": 551802721.0, + "step": 15710 + }, + { + "epoch": 2.917548746518106, + "grad_norm": 1.5639954805374146, + "learning_rate": 1e-06, + "loss": 0.2911, + "mean_token_accuracy": 0.8963301777839661, + "num_tokens": 551835551.0, + "step": 15711 + }, + { + "epoch": 2.9177344475394613, + "grad_norm": 1.513629674911499, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8742691874504089, + "num_tokens": 551875170.0, + "step": 15712 + }, + { + "epoch": 2.917920148560817, + "grad_norm": 1.4332276582717896, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.8950780034065247, + "num_tokens": 551912808.0, + "step": 15713 + }, + { + "epoch": 2.918105849582173, + "grad_norm": 1.5845009088516235, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8769292831420898, + "num_tokens": 551949256.0, + "step": 15714 + }, + { + "epoch": 2.918291550603528, + "grad_norm": 1.5655548572540283, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8791942596435547, + "num_tokens": 551983910.0, + "step": 15715 + }, + { + "epoch": 2.918477251624884, + "grad_norm": 1.5687050819396973, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8789128661155701, + "num_tokens": 552022913.0, + "step": 15716 + }, + { + "epoch": 2.9186629526462395, + "grad_norm": 1.5259853601455688, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8914245367050171, + "num_tokens": 552056953.0, + "step": 15717 + }, + { + "epoch": 2.9188486536675953, + "grad_norm": 1.5511746406555176, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8798435926437378, + "num_tokens": 552095033.0, + "step": 15718 + }, + { + "epoch": 2.919034354688951, + "grad_norm": 1.5129703283309937, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8886557817459106, + "num_tokens": 552130960.0, + "step": 15719 + }, + { + "epoch": 2.9192200557103063, + "grad_norm": 1.5248061418533325, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8803914785385132, + "num_tokens": 552167960.0, + "step": 15720 + }, + { + "epoch": 2.919405756731662, + "grad_norm": 1.6309800148010254, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8779007792472839, + "num_tokens": 552199003.0, + "step": 15721 + }, + { + "epoch": 2.9195914577530178, + "grad_norm": 1.4169328212738037, + "learning_rate": 1e-06, + "loss": 0.255, + "mean_token_accuracy": 0.9072555303573608, + "num_tokens": 552234457.0, + "step": 15722 + }, + { + "epoch": 2.919777158774373, + "grad_norm": 1.550968885421753, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8909223675727844, + "num_tokens": 552267073.0, + "step": 15723 + }, + { + "epoch": 2.919962859795729, + "grad_norm": 1.5346693992614746, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8836547136306763, + "num_tokens": 552303570.0, + "step": 15724 + }, + { + "epoch": 2.9201485608170845, + "grad_norm": 1.6204288005828857, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8792111873626709, + "num_tokens": 552338324.0, + "step": 15725 + }, + { + "epoch": 2.9203342618384402, + "grad_norm": 1.652767539024353, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8862481117248535, + "num_tokens": 552371873.0, + "step": 15726 + }, + { + "epoch": 2.920519962859796, + "grad_norm": 1.535893201828003, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8900350332260132, + "num_tokens": 552404728.0, + "step": 15727 + }, + { + "epoch": 2.9207056638811513, + "grad_norm": 1.7926334142684937, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.882326602935791, + "num_tokens": 552440772.0, + "step": 15728 + }, + { + "epoch": 2.920891364902507, + "grad_norm": 1.5327552556991577, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8761101961135864, + "num_tokens": 552479489.0, + "step": 15729 + }, + { + "epoch": 2.9210770659238627, + "grad_norm": 1.7188196182250977, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8854155540466309, + "num_tokens": 552510136.0, + "step": 15730 + }, + { + "epoch": 2.921262766945218, + "grad_norm": 1.5997889041900635, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8906813263893127, + "num_tokens": 552547584.0, + "step": 15731 + }, + { + "epoch": 2.9214484679665738, + "grad_norm": 1.4796913862228394, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8832703828811646, + "num_tokens": 552584225.0, + "step": 15732 + }, + { + "epoch": 2.9216341689879295, + "grad_norm": 1.5251227617263794, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8860316276550293, + "num_tokens": 552621767.0, + "step": 15733 + }, + { + "epoch": 2.9218198700092852, + "grad_norm": 1.4416182041168213, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8863060474395752, + "num_tokens": 552664719.0, + "step": 15734 + }, + { + "epoch": 2.9220055710306405, + "grad_norm": 1.4994884729385376, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8955386281013489, + "num_tokens": 552702903.0, + "step": 15735 + }, + { + "epoch": 2.9221912720519962, + "grad_norm": 1.7026795148849487, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8794432878494263, + "num_tokens": 552733603.0, + "step": 15736 + }, + { + "epoch": 2.922376973073352, + "grad_norm": 1.4840716123580933, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8844947814941406, + "num_tokens": 552772209.0, + "step": 15737 + }, + { + "epoch": 2.9225626740947073, + "grad_norm": 1.5705510377883911, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8887895345687866, + "num_tokens": 552804153.0, + "step": 15738 + }, + { + "epoch": 2.922748375116063, + "grad_norm": 1.4524933099746704, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8755137920379639, + "num_tokens": 552844390.0, + "step": 15739 + }, + { + "epoch": 2.9229340761374187, + "grad_norm": 1.4796088933944702, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8946027159690857, + "num_tokens": 552882389.0, + "step": 15740 + }, + { + "epoch": 2.9231197771587745, + "grad_norm": 1.5293611288070679, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8960894346237183, + "num_tokens": 552913827.0, + "step": 15741 + }, + { + "epoch": 2.92330547818013, + "grad_norm": 1.5163058042526245, + "learning_rate": 1e-06, + "loss": 0.3504, + "mean_token_accuracy": 0.874518632888794, + "num_tokens": 552955615.0, + "step": 15742 + }, + { + "epoch": 2.9234911792014855, + "grad_norm": 1.4296098947525024, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8791937828063965, + "num_tokens": 552999251.0, + "step": 15743 + }, + { + "epoch": 2.9236768802228412, + "grad_norm": 1.793546199798584, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8944685459136963, + "num_tokens": 553022979.0, + "step": 15744 + }, + { + "epoch": 2.923862581244197, + "grad_norm": 1.6419976949691772, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8839073181152344, + "num_tokens": 553058117.0, + "step": 15745 + }, + { + "epoch": 2.9240482822655522, + "grad_norm": 1.5765327215194702, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8953272104263306, + "num_tokens": 553090963.0, + "step": 15746 + }, + { + "epoch": 2.924233983286908, + "grad_norm": 1.4891291856765747, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8911547660827637, + "num_tokens": 553129154.0, + "step": 15747 + }, + { + "epoch": 2.9244196843082637, + "grad_norm": 1.6769788265228271, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8897050023078918, + "num_tokens": 553156399.0, + "step": 15748 + }, + { + "epoch": 2.9246053853296194, + "grad_norm": 1.5666935443878174, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8845469951629639, + "num_tokens": 553190880.0, + "step": 15749 + }, + { + "epoch": 2.924791086350975, + "grad_norm": 1.5523688793182373, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8860447406768799, + "num_tokens": 553232904.0, + "step": 15750 + }, + { + "epoch": 2.9249767873723305, + "grad_norm": 1.5324503183364868, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8926385641098022, + "num_tokens": 553265390.0, + "step": 15751 + }, + { + "epoch": 2.925162488393686, + "grad_norm": 1.4165937900543213, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8915483951568604, + "num_tokens": 553304180.0, + "step": 15752 + }, + { + "epoch": 2.925348189415042, + "grad_norm": 1.732370376586914, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8903528451919556, + "num_tokens": 553332823.0, + "step": 15753 + }, + { + "epoch": 2.925533890436397, + "grad_norm": 1.4519842863082886, + "learning_rate": 1e-06, + "loss": 0.275, + "mean_token_accuracy": 0.9027951955795288, + "num_tokens": 553369641.0, + "step": 15754 + }, + { + "epoch": 2.925719591457753, + "grad_norm": 1.5505515336990356, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8947715163230896, + "num_tokens": 553402581.0, + "step": 15755 + }, + { + "epoch": 2.9259052924791087, + "grad_norm": 1.5518136024475098, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8865119814872742, + "num_tokens": 553438121.0, + "step": 15756 + }, + { + "epoch": 2.9260909935004644, + "grad_norm": 1.5376884937286377, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8929367065429688, + "num_tokens": 553470636.0, + "step": 15757 + }, + { + "epoch": 2.9262766945218197, + "grad_norm": 1.5203478336334229, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8900389075279236, + "num_tokens": 553509443.0, + "step": 15758 + }, + { + "epoch": 2.9264623955431754, + "grad_norm": 1.6106535196304321, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8761440515518188, + "num_tokens": 553547016.0, + "step": 15759 + }, + { + "epoch": 2.926648096564531, + "grad_norm": 1.475448489189148, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8918391466140747, + "num_tokens": 553584533.0, + "step": 15760 + }, + { + "epoch": 2.9268337975858865, + "grad_norm": 1.435903549194336, + "learning_rate": 1e-06, + "loss": 0.2848, + "mean_token_accuracy": 0.8983156085014343, + "num_tokens": 553620529.0, + "step": 15761 + }, + { + "epoch": 2.927019498607242, + "grad_norm": 1.4722241163253784, + "learning_rate": 1e-06, + "loss": 0.2964, + "mean_token_accuracy": 0.8936355710029602, + "num_tokens": 553656335.0, + "step": 15762 + }, + { + "epoch": 2.927205199628598, + "grad_norm": 1.4653290510177612, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8933284282684326, + "num_tokens": 553693635.0, + "step": 15763 + }, + { + "epoch": 2.9273909006499537, + "grad_norm": 1.527280330657959, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8797542452812195, + "num_tokens": 553730116.0, + "step": 15764 + }, + { + "epoch": 2.9275766016713094, + "grad_norm": 1.65292489528656, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8854187726974487, + "num_tokens": 553765072.0, + "step": 15765 + }, + { + "epoch": 2.9277623026926647, + "grad_norm": 1.7243068218231201, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8893032670021057, + "num_tokens": 553794106.0, + "step": 15766 + }, + { + "epoch": 2.9279480037140204, + "grad_norm": 1.547127604484558, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.8817430138587952, + "num_tokens": 553831026.0, + "step": 15767 + }, + { + "epoch": 2.928133704735376, + "grad_norm": 1.4471968412399292, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8819919228553772, + "num_tokens": 553871503.0, + "step": 15768 + }, + { + "epoch": 2.9283194057567314, + "grad_norm": 1.5962979793548584, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8940024971961975, + "num_tokens": 553903288.0, + "step": 15769 + }, + { + "epoch": 2.928505106778087, + "grad_norm": 1.8554401397705078, + "learning_rate": 1e-06, + "loss": 0.3905, + "mean_token_accuracy": 0.8636438250541687, + "num_tokens": 553933430.0, + "step": 15770 + }, + { + "epoch": 2.928690807799443, + "grad_norm": 1.6362498998641968, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8775021433830261, + "num_tokens": 553965775.0, + "step": 15771 + }, + { + "epoch": 2.9288765088207986, + "grad_norm": 1.6029605865478516, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8706430196762085, + "num_tokens": 554000632.0, + "step": 15772 + }, + { + "epoch": 2.9290622098421544, + "grad_norm": 1.6962167024612427, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8817975521087646, + "num_tokens": 554029595.0, + "step": 15773 + }, + { + "epoch": 2.9292479108635097, + "grad_norm": 1.5733445882797241, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8903883099555969, + "num_tokens": 554061647.0, + "step": 15774 + }, + { + "epoch": 2.9294336118848654, + "grad_norm": 1.6082894802093506, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8799433708190918, + "num_tokens": 554095786.0, + "step": 15775 + }, + { + "epoch": 2.929619312906221, + "grad_norm": 1.46778404712677, + "learning_rate": 1e-06, + "loss": 0.2877, + "mean_token_accuracy": 0.8982908725738525, + "num_tokens": 554131286.0, + "step": 15776 + }, + { + "epoch": 2.9298050139275764, + "grad_norm": 1.529786467552185, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.8987934589385986, + "num_tokens": 554161574.0, + "step": 15777 + }, + { + "epoch": 2.929990714948932, + "grad_norm": 1.581178069114685, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8683902025222778, + "num_tokens": 554196114.0, + "step": 15778 + }, + { + "epoch": 2.930176415970288, + "grad_norm": 1.4069496393203735, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.890803873538971, + "num_tokens": 554235936.0, + "step": 15779 + }, + { + "epoch": 2.9303621169916436, + "grad_norm": 1.4235261678695679, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8869208693504333, + "num_tokens": 554278920.0, + "step": 15780 + }, + { + "epoch": 2.930547818012999, + "grad_norm": 1.4742642641067505, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8893525004386902, + "num_tokens": 554317381.0, + "step": 15781 + }, + { + "epoch": 2.9307335190343546, + "grad_norm": 1.713835597038269, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8848792314529419, + "num_tokens": 554352710.0, + "step": 15782 + }, + { + "epoch": 2.9309192200557104, + "grad_norm": 1.6551318168640137, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8820798397064209, + "num_tokens": 554386290.0, + "step": 15783 + }, + { + "epoch": 2.9311049210770657, + "grad_norm": 1.660983920097351, + "learning_rate": 1e-06, + "loss": 0.3116, + "mean_token_accuracy": 0.8929467797279358, + "num_tokens": 554414234.0, + "step": 15784 + }, + { + "epoch": 2.9312906220984214, + "grad_norm": 1.6935521364212036, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8822548389434814, + "num_tokens": 554443869.0, + "step": 15785 + }, + { + "epoch": 2.931476323119777, + "grad_norm": 1.661402940750122, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8805174827575684, + "num_tokens": 554476501.0, + "step": 15786 + }, + { + "epoch": 2.931662024141133, + "grad_norm": 1.513392686843872, + "learning_rate": 1e-06, + "loss": 0.3341, + "mean_token_accuracy": 0.8843785524368286, + "num_tokens": 554513391.0, + "step": 15787 + }, + { + "epoch": 2.9318477251624886, + "grad_norm": 1.6303110122680664, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8792600631713867, + "num_tokens": 554546914.0, + "step": 15788 + }, + { + "epoch": 2.932033426183844, + "grad_norm": 1.611006736755371, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8806872367858887, + "num_tokens": 554579452.0, + "step": 15789 + }, + { + "epoch": 2.9322191272051996, + "grad_norm": 1.613861083984375, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8898409605026245, + "num_tokens": 554615315.0, + "step": 15790 + }, + { + "epoch": 2.9324048282265553, + "grad_norm": 1.6189619302749634, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8768514394760132, + "num_tokens": 554651222.0, + "step": 15791 + }, + { + "epoch": 2.9325905292479106, + "grad_norm": 1.494315266609192, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8943297266960144, + "num_tokens": 554687498.0, + "step": 15792 + }, + { + "epoch": 2.9327762302692664, + "grad_norm": 1.5715446472167969, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8887057304382324, + "num_tokens": 554722787.0, + "step": 15793 + }, + { + "epoch": 2.932961931290622, + "grad_norm": 1.467041015625, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8928430676460266, + "num_tokens": 554761201.0, + "step": 15794 + }, + { + "epoch": 2.933147632311978, + "grad_norm": 1.5007164478302002, + "learning_rate": 1e-06, + "loss": 0.2991, + "mean_token_accuracy": 0.896018922328949, + "num_tokens": 554795426.0, + "step": 15795 + }, + { + "epoch": 2.9333333333333336, + "grad_norm": 1.5686030387878418, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8859913945198059, + "num_tokens": 554830100.0, + "step": 15796 + }, + { + "epoch": 2.933519034354689, + "grad_norm": 1.6685243844985962, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8850736021995544, + "num_tokens": 554858305.0, + "step": 15797 + }, + { + "epoch": 2.9337047353760446, + "grad_norm": 1.5018714666366577, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8946326971054077, + "num_tokens": 554893219.0, + "step": 15798 + }, + { + "epoch": 2.9338904363974003, + "grad_norm": 1.3734052181243896, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.9022501707077026, + "num_tokens": 554932554.0, + "step": 15799 + }, + { + "epoch": 2.9340761374187556, + "grad_norm": 1.581676959991455, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.889683723449707, + "num_tokens": 554965489.0, + "step": 15800 + }, + { + "epoch": 2.9342618384401113, + "grad_norm": 1.6581857204437256, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.876925528049469, + "num_tokens": 554996382.0, + "step": 15801 + }, + { + "epoch": 2.934447539461467, + "grad_norm": 1.466787576675415, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8944578170776367, + "num_tokens": 555033015.0, + "step": 15802 + }, + { + "epoch": 2.934633240482823, + "grad_norm": 1.7914355993270874, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8818027973175049, + "num_tokens": 555061079.0, + "step": 15803 + }, + { + "epoch": 2.9348189415041785, + "grad_norm": 1.652321457862854, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8887124061584473, + "num_tokens": 555093390.0, + "step": 15804 + }, + { + "epoch": 2.935004642525534, + "grad_norm": 1.5579938888549805, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8965637683868408, + "num_tokens": 555128990.0, + "step": 15805 + }, + { + "epoch": 2.9351903435468896, + "grad_norm": 1.6779377460479736, + "learning_rate": 1e-06, + "loss": 0.3475, + "mean_token_accuracy": 0.8803422451019287, + "num_tokens": 555160859.0, + "step": 15806 + }, + { + "epoch": 2.935376044568245, + "grad_norm": 1.5691379308700562, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8909392356872559, + "num_tokens": 555194744.0, + "step": 15807 + }, + { + "epoch": 2.9355617455896006, + "grad_norm": 1.6660677194595337, + "learning_rate": 1e-06, + "loss": 0.2739, + "mean_token_accuracy": 0.9070485234260559, + "num_tokens": 555221955.0, + "step": 15808 + }, + { + "epoch": 2.9357474466109563, + "grad_norm": 1.6260992288589478, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8975071310997009, + "num_tokens": 555254110.0, + "step": 15809 + }, + { + "epoch": 2.935933147632312, + "grad_norm": 1.6934633255004883, + "learning_rate": 1e-06, + "loss": 0.352, + "mean_token_accuracy": 0.8830324411392212, + "num_tokens": 555283499.0, + "step": 15810 + }, + { + "epoch": 2.936118848653668, + "grad_norm": 1.4800258874893188, + "learning_rate": 1e-06, + "loss": 0.2965, + "mean_token_accuracy": 0.8940074443817139, + "num_tokens": 555317967.0, + "step": 15811 + }, + { + "epoch": 2.936304549675023, + "grad_norm": 1.6454848051071167, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8812969923019409, + "num_tokens": 555354367.0, + "step": 15812 + }, + { + "epoch": 2.936490250696379, + "grad_norm": 1.6604464054107666, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8858973979949951, + "num_tokens": 555387089.0, + "step": 15813 + }, + { + "epoch": 2.9366759517177345, + "grad_norm": 1.3487508296966553, + "learning_rate": 1e-06, + "loss": 0.2735, + "mean_token_accuracy": 0.9020441770553589, + "num_tokens": 555427264.0, + "step": 15814 + }, + { + "epoch": 2.93686165273909, + "grad_norm": 1.5331013202667236, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8931593894958496, + "num_tokens": 555462237.0, + "step": 15815 + }, + { + "epoch": 2.9370473537604456, + "grad_norm": 1.5626710653305054, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8764532804489136, + "num_tokens": 555497085.0, + "step": 15816 + }, + { + "epoch": 2.9372330547818013, + "grad_norm": 1.4163795709609985, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8871347904205322, + "num_tokens": 555537681.0, + "step": 15817 + }, + { + "epoch": 2.937418755803157, + "grad_norm": 1.5569441318511963, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8807815313339233, + "num_tokens": 555572722.0, + "step": 15818 + }, + { + "epoch": 2.9376044568245128, + "grad_norm": 1.5108715295791626, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8795920610427856, + "num_tokens": 555606475.0, + "step": 15819 + }, + { + "epoch": 2.937790157845868, + "grad_norm": 1.5118718147277832, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8937761783599854, + "num_tokens": 555640565.0, + "step": 15820 + }, + { + "epoch": 2.937975858867224, + "grad_norm": 1.5393179655075073, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8861647248268127, + "num_tokens": 555675678.0, + "step": 15821 + }, + { + "epoch": 2.9381615598885795, + "grad_norm": 1.5462266206741333, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8876620531082153, + "num_tokens": 555710122.0, + "step": 15822 + }, + { + "epoch": 2.938347260909935, + "grad_norm": 1.5624580383300781, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8774527311325073, + "num_tokens": 555749424.0, + "step": 15823 + }, + { + "epoch": 2.9385329619312905, + "grad_norm": 1.4450044631958008, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8893846869468689, + "num_tokens": 555788745.0, + "step": 15824 + }, + { + "epoch": 2.9387186629526463, + "grad_norm": 1.5722800493240356, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8858861923217773, + "num_tokens": 555819770.0, + "step": 15825 + }, + { + "epoch": 2.938904363974002, + "grad_norm": 1.6838459968566895, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8763456344604492, + "num_tokens": 555852055.0, + "step": 15826 + }, + { + "epoch": 2.9390900649953577, + "grad_norm": 1.5594156980514526, + "learning_rate": 1e-06, + "loss": 0.3662, + "mean_token_accuracy": 0.8730947971343994, + "num_tokens": 555889745.0, + "step": 15827 + }, + { + "epoch": 2.939275766016713, + "grad_norm": 1.4634404182434082, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.8980353474617004, + "num_tokens": 555925646.0, + "step": 15828 + }, + { + "epoch": 2.9394614670380688, + "grad_norm": 1.5828641653060913, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.897011399269104, + "num_tokens": 555955411.0, + "step": 15829 + }, + { + "epoch": 2.939647168059424, + "grad_norm": 1.487593173980713, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8962792158126831, + "num_tokens": 555990130.0, + "step": 15830 + }, + { + "epoch": 2.93983286908078, + "grad_norm": 1.4001387357711792, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8949605822563171, + "num_tokens": 556029924.0, + "step": 15831 + }, + { + "epoch": 2.9400185701021355, + "grad_norm": 1.7220063209533691, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8798108100891113, + "num_tokens": 556058405.0, + "step": 15832 + }, + { + "epoch": 2.9402042711234913, + "grad_norm": 1.5848966836929321, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8894137740135193, + "num_tokens": 556090418.0, + "step": 15833 + }, + { + "epoch": 2.940389972144847, + "grad_norm": 1.5432348251342773, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8805069327354431, + "num_tokens": 556128967.0, + "step": 15834 + }, + { + "epoch": 2.9405756731662023, + "grad_norm": 1.6519135236740112, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8920389413833618, + "num_tokens": 556160017.0, + "step": 15835 + }, + { + "epoch": 2.940761374187558, + "grad_norm": 1.4840959310531616, + "learning_rate": 1e-06, + "loss": 0.2774, + "mean_token_accuracy": 0.9001460075378418, + "num_tokens": 556192905.0, + "step": 15836 + }, + { + "epoch": 2.9409470752089137, + "grad_norm": 1.6367064714431763, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8944424986839294, + "num_tokens": 556219619.0, + "step": 15837 + }, + { + "epoch": 2.941132776230269, + "grad_norm": 1.7410880327224731, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8873271942138672, + "num_tokens": 556251745.0, + "step": 15838 + }, + { + "epoch": 2.9413184772516248, + "grad_norm": 1.5823588371276855, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8784385919570923, + "num_tokens": 556290877.0, + "step": 15839 + }, + { + "epoch": 2.9415041782729805, + "grad_norm": 1.669848084449768, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8703628778457642, + "num_tokens": 556325261.0, + "step": 15840 + }, + { + "epoch": 2.9416898792943362, + "grad_norm": 1.4615392684936523, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8883726596832275, + "num_tokens": 556365148.0, + "step": 15841 + }, + { + "epoch": 2.941875580315692, + "grad_norm": 1.5356117486953735, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8847218751907349, + "num_tokens": 556401450.0, + "step": 15842 + }, + { + "epoch": 2.9420612813370473, + "grad_norm": 1.6184331178665161, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8938974142074585, + "num_tokens": 556432749.0, + "step": 15843 + }, + { + "epoch": 2.942246982358403, + "grad_norm": 1.5490809679031372, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8776063919067383, + "num_tokens": 556472508.0, + "step": 15844 + }, + { + "epoch": 2.9424326833797587, + "grad_norm": 1.5564719438552856, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8807591199874878, + "num_tokens": 556508363.0, + "step": 15845 + }, + { + "epoch": 2.942618384401114, + "grad_norm": 1.6164222955703735, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.9018605947494507, + "num_tokens": 556539489.0, + "step": 15846 + }, + { + "epoch": 2.9428040854224697, + "grad_norm": 1.5103641748428345, + "learning_rate": 1e-06, + "loss": 0.3522, + "mean_token_accuracy": 0.8793507814407349, + "num_tokens": 556581374.0, + "step": 15847 + }, + { + "epoch": 2.9429897864438255, + "grad_norm": 1.6821138858795166, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8827779293060303, + "num_tokens": 556611161.0, + "step": 15848 + }, + { + "epoch": 2.943175487465181, + "grad_norm": 1.3603825569152832, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.895552396774292, + "num_tokens": 556653363.0, + "step": 15849 + }, + { + "epoch": 2.943361188486537, + "grad_norm": 1.5439926385879517, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8916202187538147, + "num_tokens": 556684484.0, + "step": 15850 + }, + { + "epoch": 2.9435468895078922, + "grad_norm": 1.480973482131958, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8949395418167114, + "num_tokens": 556721843.0, + "step": 15851 + }, + { + "epoch": 2.943732590529248, + "grad_norm": 1.4111239910125732, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.8967173099517822, + "num_tokens": 556757138.0, + "step": 15852 + }, + { + "epoch": 2.9439182915506037, + "grad_norm": 1.5606025457382202, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.8968145847320557, + "num_tokens": 556786647.0, + "step": 15853 + }, + { + "epoch": 2.944103992571959, + "grad_norm": 1.4413948059082031, + "learning_rate": 1e-06, + "loss": 0.3166, + "mean_token_accuracy": 0.88893723487854, + "num_tokens": 556825344.0, + "step": 15854 + }, + { + "epoch": 2.9442896935933147, + "grad_norm": 1.5904654264450073, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8904008865356445, + "num_tokens": 556859117.0, + "step": 15855 + }, + { + "epoch": 2.9444753946146704, + "grad_norm": 1.4658327102661133, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.88555908203125, + "num_tokens": 556897367.0, + "step": 15856 + }, + { + "epoch": 2.944661095636026, + "grad_norm": 1.5159567594528198, + "learning_rate": 1e-06, + "loss": 0.3023, + "mean_token_accuracy": 0.8943669199943542, + "num_tokens": 556931531.0, + "step": 15857 + }, + { + "epoch": 2.9448467966573815, + "grad_norm": 1.531640648841858, + "learning_rate": 1e-06, + "loss": 0.3424, + "mean_token_accuracy": 0.8794885873794556, + "num_tokens": 556968267.0, + "step": 15858 + }, + { + "epoch": 2.945032497678737, + "grad_norm": 1.5114723443984985, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.89169842004776, + "num_tokens": 557003134.0, + "step": 15859 + }, + { + "epoch": 2.945218198700093, + "grad_norm": 1.5328552722930908, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8897890448570251, + "num_tokens": 557035997.0, + "step": 15860 + }, + { + "epoch": 2.9454038997214482, + "grad_norm": 1.576757788658142, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8942404985427856, + "num_tokens": 557068681.0, + "step": 15861 + }, + { + "epoch": 2.945589600742804, + "grad_norm": 1.650802731513977, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.897070050239563, + "num_tokens": 557098832.0, + "step": 15862 + }, + { + "epoch": 2.9457753017641597, + "grad_norm": 1.594131588935852, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8849156498908997, + "num_tokens": 557132268.0, + "step": 15863 + }, + { + "epoch": 2.9459610027855154, + "grad_norm": 1.74611234664917, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8712371587753296, + "num_tokens": 557162025.0, + "step": 15864 + }, + { + "epoch": 2.946146703806871, + "grad_norm": 1.618880033493042, + "learning_rate": 1e-06, + "loss": 0.3385, + "mean_token_accuracy": 0.8793114423751831, + "num_tokens": 557197446.0, + "step": 15865 + }, + { + "epoch": 2.9463324048282264, + "grad_norm": 1.474624752998352, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8844981789588928, + "num_tokens": 557235626.0, + "step": 15866 + }, + { + "epoch": 2.946518105849582, + "grad_norm": 1.477615237236023, + "learning_rate": 1e-06, + "loss": 0.2833, + "mean_token_accuracy": 0.8999829292297363, + "num_tokens": 557271438.0, + "step": 15867 + }, + { + "epoch": 2.946703806870938, + "grad_norm": 1.5432443618774414, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8903135061264038, + "num_tokens": 557303212.0, + "step": 15868 + }, + { + "epoch": 2.946889507892293, + "grad_norm": 1.6503467559814453, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.9030647873878479, + "num_tokens": 557330903.0, + "step": 15869 + }, + { + "epoch": 2.947075208913649, + "grad_norm": 1.7706658840179443, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8716306090354919, + "num_tokens": 557362817.0, + "step": 15870 + }, + { + "epoch": 2.9472609099350047, + "grad_norm": 1.5642058849334717, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8866859674453735, + "num_tokens": 557395802.0, + "step": 15871 + }, + { + "epoch": 2.9474466109563604, + "grad_norm": 1.5823478698730469, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.882972240447998, + "num_tokens": 557429964.0, + "step": 15872 + }, + { + "epoch": 2.947632311977716, + "grad_norm": 1.3945114612579346, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.896406888961792, + "num_tokens": 557468185.0, + "step": 15873 + }, + { + "epoch": 2.9478180129990714, + "grad_norm": 1.4935474395751953, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.8801896572113037, + "num_tokens": 557506066.0, + "step": 15874 + }, + { + "epoch": 2.948003714020427, + "grad_norm": 1.5320658683776855, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8925924301147461, + "num_tokens": 557540503.0, + "step": 15875 + }, + { + "epoch": 2.948189415041783, + "grad_norm": 1.5475273132324219, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8890680074691772, + "num_tokens": 557573452.0, + "step": 15876 + }, + { + "epoch": 2.948375116063138, + "grad_norm": 1.5880672931671143, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8750166893005371, + "num_tokens": 557609919.0, + "step": 15877 + }, + { + "epoch": 2.948560817084494, + "grad_norm": 1.5142998695373535, + "learning_rate": 1e-06, + "loss": 0.2814, + "mean_token_accuracy": 0.8988255262374878, + "num_tokens": 557646617.0, + "step": 15878 + }, + { + "epoch": 2.9487465181058496, + "grad_norm": 1.4784202575683594, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8984776735305786, + "num_tokens": 557682989.0, + "step": 15879 + }, + { + "epoch": 2.9489322191272054, + "grad_norm": 1.5272246599197388, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.884408712387085, + "num_tokens": 557719212.0, + "step": 15880 + }, + { + "epoch": 2.9491179201485607, + "grad_norm": 1.5826468467712402, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8792199492454529, + "num_tokens": 557755185.0, + "step": 15881 + }, + { + "epoch": 2.9493036211699164, + "grad_norm": 1.623892068862915, + "learning_rate": 1e-06, + "loss": 0.3281, + "mean_token_accuracy": 0.8845670223236084, + "num_tokens": 557788175.0, + "step": 15882 + }, + { + "epoch": 2.949489322191272, + "grad_norm": 1.6824733018875122, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8777450323104858, + "num_tokens": 557821071.0, + "step": 15883 + }, + { + "epoch": 2.9496750232126274, + "grad_norm": 1.5696905851364136, + "learning_rate": 1e-06, + "loss": 0.3414, + "mean_token_accuracy": 0.8794467449188232, + "num_tokens": 557858497.0, + "step": 15884 + }, + { + "epoch": 2.949860724233983, + "grad_norm": 1.5042595863342285, + "learning_rate": 1e-06, + "loss": 0.3103, + "mean_token_accuracy": 0.8897210359573364, + "num_tokens": 557893598.0, + "step": 15885 + }, + { + "epoch": 2.950046425255339, + "grad_norm": 1.4449243545532227, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8848199844360352, + "num_tokens": 557932506.0, + "step": 15886 + }, + { + "epoch": 2.9502321262766946, + "grad_norm": 1.5574195384979248, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8920867443084717, + "num_tokens": 557968159.0, + "step": 15887 + }, + { + "epoch": 2.9504178272980504, + "grad_norm": 1.578984260559082, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8890388011932373, + "num_tokens": 558000061.0, + "step": 15888 + }, + { + "epoch": 2.9506035283194056, + "grad_norm": 1.616981029510498, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8869398832321167, + "num_tokens": 558032882.0, + "step": 15889 + }, + { + "epoch": 2.9507892293407614, + "grad_norm": 1.4823098182678223, + "learning_rate": 1e-06, + "loss": 0.3084, + "mean_token_accuracy": 0.8889164924621582, + "num_tokens": 558069988.0, + "step": 15890 + }, + { + "epoch": 2.950974930362117, + "grad_norm": 1.5413224697113037, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8865848779678345, + "num_tokens": 558108015.0, + "step": 15891 + }, + { + "epoch": 2.9511606313834724, + "grad_norm": 1.617348551750183, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8905352354049683, + "num_tokens": 558143674.0, + "step": 15892 + }, + { + "epoch": 2.951346332404828, + "grad_norm": 1.5876588821411133, + "learning_rate": 1e-06, + "loss": 0.3481, + "mean_token_accuracy": 0.8782244324684143, + "num_tokens": 558181456.0, + "step": 15893 + }, + { + "epoch": 2.951532033426184, + "grad_norm": 1.4748555421829224, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8809565901756287, + "num_tokens": 558219176.0, + "step": 15894 + }, + { + "epoch": 2.9517177344475396, + "grad_norm": 1.5407530069351196, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.888938844203949, + "num_tokens": 558254329.0, + "step": 15895 + }, + { + "epoch": 2.9519034354688953, + "grad_norm": 1.8331953287124634, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8780203461647034, + "num_tokens": 558282041.0, + "step": 15896 + }, + { + "epoch": 2.9520891364902506, + "grad_norm": 1.7151800394058228, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8728326559066772, + "num_tokens": 558313869.0, + "step": 15897 + }, + { + "epoch": 2.9522748375116064, + "grad_norm": 1.410598635673523, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8912419080734253, + "num_tokens": 558354281.0, + "step": 15898 + }, + { + "epoch": 2.952460538532962, + "grad_norm": 1.8746750354766846, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8786208033561707, + "num_tokens": 558380104.0, + "step": 15899 + }, + { + "epoch": 2.9526462395543174, + "grad_norm": 1.5796594619750977, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.891281247138977, + "num_tokens": 558411736.0, + "step": 15900 + }, + { + "epoch": 2.952831940575673, + "grad_norm": 1.5902074575424194, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8931382894515991, + "num_tokens": 558443756.0, + "step": 15901 + }, + { + "epoch": 2.953017641597029, + "grad_norm": 1.6085785627365112, + "learning_rate": 1e-06, + "loss": 0.3537, + "mean_token_accuracy": 0.8771145343780518, + "num_tokens": 558479330.0, + "step": 15902 + }, + { + "epoch": 2.9532033426183846, + "grad_norm": 1.4760260581970215, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.8845696449279785, + "num_tokens": 558516554.0, + "step": 15903 + }, + { + "epoch": 2.95338904363974, + "grad_norm": 1.6063005924224854, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8714079856872559, + "num_tokens": 558556790.0, + "step": 15904 + }, + { + "epoch": 2.9535747446610956, + "grad_norm": 1.499850869178772, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8911923766136169, + "num_tokens": 558589254.0, + "step": 15905 + }, + { + "epoch": 2.9537604456824513, + "grad_norm": 1.6556026935577393, + "learning_rate": 1e-06, + "loss": 0.3591, + "mean_token_accuracy": 0.8750158548355103, + "num_tokens": 558623482.0, + "step": 15906 + }, + { + "epoch": 2.9539461467038066, + "grad_norm": 1.4782962799072266, + "learning_rate": 1e-06, + "loss": 0.2895, + "mean_token_accuracy": 0.8951877355575562, + "num_tokens": 558658787.0, + "step": 15907 + }, + { + "epoch": 2.9541318477251624, + "grad_norm": 1.5935473442077637, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8950988054275513, + "num_tokens": 558690335.0, + "step": 15908 + }, + { + "epoch": 2.954317548746518, + "grad_norm": 1.454345941543579, + "learning_rate": 1e-06, + "loss": 0.2746, + "mean_token_accuracy": 0.901285707950592, + "num_tokens": 558727349.0, + "step": 15909 + }, + { + "epoch": 2.954503249767874, + "grad_norm": 1.6194968223571777, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8839452266693115, + "num_tokens": 558761989.0, + "step": 15910 + }, + { + "epoch": 2.9546889507892296, + "grad_norm": 1.4332153797149658, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8867952823638916, + "num_tokens": 558803865.0, + "step": 15911 + }, + { + "epoch": 2.954874651810585, + "grad_norm": 1.6035223007202148, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8870035409927368, + "num_tokens": 558841260.0, + "step": 15912 + }, + { + "epoch": 2.9550603528319406, + "grad_norm": 1.5818052291870117, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.8809776306152344, + "num_tokens": 558875395.0, + "step": 15913 + }, + { + "epoch": 2.9552460538532963, + "grad_norm": 1.5767817497253418, + "learning_rate": 1e-06, + "loss": 0.3447, + "mean_token_accuracy": 0.8792965412139893, + "num_tokens": 558910450.0, + "step": 15914 + }, + { + "epoch": 2.9554317548746516, + "grad_norm": 1.4243460893630981, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8997576832771301, + "num_tokens": 558949438.0, + "step": 15915 + }, + { + "epoch": 2.9556174558960073, + "grad_norm": 1.5791436433792114, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8871282339096069, + "num_tokens": 558986607.0, + "step": 15916 + }, + { + "epoch": 2.955803156917363, + "grad_norm": 1.5568859577178955, + "learning_rate": 1e-06, + "loss": 0.3694, + "mean_token_accuracy": 0.8719267845153809, + "num_tokens": 559023074.0, + "step": 15917 + }, + { + "epoch": 2.955988857938719, + "grad_norm": 1.3606173992156982, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.898467481136322, + "num_tokens": 559065426.0, + "step": 15918 + }, + { + "epoch": 2.9561745589600745, + "grad_norm": 1.4257049560546875, + "learning_rate": 1e-06, + "loss": 0.2589, + "mean_token_accuracy": 0.907284140586853, + "num_tokens": 559101987.0, + "step": 15919 + }, + { + "epoch": 2.95636025998143, + "grad_norm": 1.6666855812072754, + "learning_rate": 1e-06, + "loss": 0.2682, + "mean_token_accuracy": 0.9026317000389099, + "num_tokens": 559131050.0, + "step": 15920 + }, + { + "epoch": 2.9565459610027855, + "grad_norm": 1.7032901048660278, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8791042566299438, + "num_tokens": 559161705.0, + "step": 15921 + }, + { + "epoch": 2.9567316620241413, + "grad_norm": 1.6687604188919067, + "learning_rate": 1e-06, + "loss": 0.3338, + "mean_token_accuracy": 0.8826720714569092, + "num_tokens": 559194686.0, + "step": 15922 + }, + { + "epoch": 2.9569173630454966, + "grad_norm": 1.5637140274047852, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8944774866104126, + "num_tokens": 559228663.0, + "step": 15923 + }, + { + "epoch": 2.9571030640668523, + "grad_norm": 1.5390273332595825, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8796634674072266, + "num_tokens": 559272190.0, + "step": 15924 + }, + { + "epoch": 2.957288765088208, + "grad_norm": 1.3343477249145508, + "learning_rate": 1e-06, + "loss": 0.3056, + "mean_token_accuracy": 0.8911982774734497, + "num_tokens": 559316557.0, + "step": 15925 + }, + { + "epoch": 2.9574744661095638, + "grad_norm": 1.6283067464828491, + "learning_rate": 1e-06, + "loss": 0.378, + "mean_token_accuracy": 0.8687893152236938, + "num_tokens": 559355508.0, + "step": 15926 + }, + { + "epoch": 2.957660167130919, + "grad_norm": 1.6129955053329468, + "learning_rate": 1e-06, + "loss": 0.2639, + "mean_token_accuracy": 0.9035179018974304, + "num_tokens": 559384343.0, + "step": 15927 + }, + { + "epoch": 2.957845868152275, + "grad_norm": 1.383646011352539, + "learning_rate": 1e-06, + "loss": 0.2757, + "mean_token_accuracy": 0.9007521867752075, + "num_tokens": 559424414.0, + "step": 15928 + }, + { + "epoch": 2.9580315691736305, + "grad_norm": 1.5075682401657104, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8837262392044067, + "num_tokens": 559459131.0, + "step": 15929 + }, + { + "epoch": 2.958217270194986, + "grad_norm": 1.4215444326400757, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8961591720581055, + "num_tokens": 559499427.0, + "step": 15930 + }, + { + "epoch": 2.9584029712163415, + "grad_norm": 1.588163137435913, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.8968892097473145, + "num_tokens": 559532825.0, + "step": 15931 + }, + { + "epoch": 2.9585886722376973, + "grad_norm": 1.5625420808792114, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8707000613212585, + "num_tokens": 559567951.0, + "step": 15932 + }, + { + "epoch": 2.958774373259053, + "grad_norm": 1.5160928964614868, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8904648423194885, + "num_tokens": 559605888.0, + "step": 15933 + }, + { + "epoch": 2.9589600742804087, + "grad_norm": 1.54266357421875, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8830858469009399, + "num_tokens": 559642958.0, + "step": 15934 + }, + { + "epoch": 2.959145775301764, + "grad_norm": 1.5768026113510132, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8846302032470703, + "num_tokens": 559677040.0, + "step": 15935 + }, + { + "epoch": 2.9593314763231198, + "grad_norm": 1.416856288909912, + "learning_rate": 1e-06, + "loss": 0.2623, + "mean_token_accuracy": 0.9057043790817261, + "num_tokens": 559716311.0, + "step": 15936 + }, + { + "epoch": 2.9595171773444755, + "grad_norm": 1.6129933595657349, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8832928538322449, + "num_tokens": 559754050.0, + "step": 15937 + }, + { + "epoch": 2.959702878365831, + "grad_norm": 1.6547822952270508, + "learning_rate": 1e-06, + "loss": 0.353, + "mean_token_accuracy": 0.8791537284851074, + "num_tokens": 559787256.0, + "step": 15938 + }, + { + "epoch": 2.9598885793871865, + "grad_norm": 1.6376893520355225, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.8854774236679077, + "num_tokens": 559820179.0, + "step": 15939 + }, + { + "epoch": 2.9600742804085423, + "grad_norm": 1.5106310844421387, + "learning_rate": 1e-06, + "loss": 0.3232, + "mean_token_accuracy": 0.8882150650024414, + "num_tokens": 559858382.0, + "step": 15940 + }, + { + "epoch": 2.960259981429898, + "grad_norm": 1.5398399829864502, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8964419364929199, + "num_tokens": 559891070.0, + "step": 15941 + }, + { + "epoch": 2.9604456824512537, + "grad_norm": 1.5287556648254395, + "learning_rate": 1e-06, + "loss": 0.3364, + "mean_token_accuracy": 0.8823797702789307, + "num_tokens": 559928999.0, + "step": 15942 + }, + { + "epoch": 2.960631383472609, + "grad_norm": 1.5517829656600952, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8832224607467651, + "num_tokens": 559965190.0, + "step": 15943 + }, + { + "epoch": 2.9608170844939647, + "grad_norm": 1.5610765218734741, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.8987516164779663, + "num_tokens": 559997200.0, + "step": 15944 + }, + { + "epoch": 2.9610027855153205, + "grad_norm": 1.570595383644104, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8846577405929565, + "num_tokens": 560033738.0, + "step": 15945 + }, + { + "epoch": 2.9611884865366758, + "grad_norm": 1.5253922939300537, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8848692178726196, + "num_tokens": 560065859.0, + "step": 15946 + }, + { + "epoch": 2.9613741875580315, + "grad_norm": 1.691565752029419, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8823424577713013, + "num_tokens": 560097362.0, + "step": 15947 + }, + { + "epoch": 2.9615598885793872, + "grad_norm": 1.3934099674224854, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8852784633636475, + "num_tokens": 560136957.0, + "step": 15948 + }, + { + "epoch": 2.961745589600743, + "grad_norm": 1.6067218780517578, + "learning_rate": 1e-06, + "loss": 0.3059, + "mean_token_accuracy": 0.8897914886474609, + "num_tokens": 560169185.0, + "step": 15949 + }, + { + "epoch": 2.9619312906220983, + "grad_norm": 1.6687065362930298, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8868046998977661, + "num_tokens": 560201290.0, + "step": 15950 + }, + { + "epoch": 2.962116991643454, + "grad_norm": 1.5097545385360718, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8935247659683228, + "num_tokens": 560240623.0, + "step": 15951 + }, + { + "epoch": 2.9623026926648097, + "grad_norm": 1.481905221939087, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8906232118606567, + "num_tokens": 560278161.0, + "step": 15952 + }, + { + "epoch": 2.962488393686165, + "grad_norm": 1.5515601634979248, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.9000169038772583, + "num_tokens": 560310032.0, + "step": 15953 + }, + { + "epoch": 2.9626740947075207, + "grad_norm": 1.6777114868164062, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8789812326431274, + "num_tokens": 560345229.0, + "step": 15954 + }, + { + "epoch": 2.9628597957288765, + "grad_norm": 1.6174508333206177, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8796854615211487, + "num_tokens": 560378554.0, + "step": 15955 + }, + { + "epoch": 2.963045496750232, + "grad_norm": 1.6028324365615845, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8634557723999023, + "num_tokens": 560417907.0, + "step": 15956 + }, + { + "epoch": 2.963231197771588, + "grad_norm": 1.6686939001083374, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8847227096557617, + "num_tokens": 560450122.0, + "step": 15957 + }, + { + "epoch": 2.9634168987929432, + "grad_norm": 1.6103768348693848, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8856726884841919, + "num_tokens": 560482577.0, + "step": 15958 + }, + { + "epoch": 2.963602599814299, + "grad_norm": 1.433422327041626, + "learning_rate": 1e-06, + "loss": 0.2743, + "mean_token_accuracy": 0.9010140895843506, + "num_tokens": 560518381.0, + "step": 15959 + }, + { + "epoch": 2.9637883008356547, + "grad_norm": 1.4826315641403198, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8887986540794373, + "num_tokens": 560557575.0, + "step": 15960 + }, + { + "epoch": 2.96397400185701, + "grad_norm": 1.508752465248108, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.9030517339706421, + "num_tokens": 560591877.0, + "step": 15961 + }, + { + "epoch": 2.9641597028783657, + "grad_norm": 1.5087019205093384, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8867948651313782, + "num_tokens": 560627777.0, + "step": 15962 + }, + { + "epoch": 2.9643454038997215, + "grad_norm": 1.6453765630722046, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8886479139328003, + "num_tokens": 560656003.0, + "step": 15963 + }, + { + "epoch": 2.964531104921077, + "grad_norm": 1.7397668361663818, + "learning_rate": 1e-06, + "loss": 0.4075, + "mean_token_accuracy": 0.8649715185165405, + "num_tokens": 560690206.0, + "step": 15964 + }, + { + "epoch": 2.964716805942433, + "grad_norm": 1.688234567642212, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8947128057479858, + "num_tokens": 560720347.0, + "step": 15965 + }, + { + "epoch": 2.964902506963788, + "grad_norm": 1.7169630527496338, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8879867196083069, + "num_tokens": 560748904.0, + "step": 15966 + }, + { + "epoch": 2.965088207985144, + "grad_norm": 1.639466643333435, + "learning_rate": 1e-06, + "loss": 0.3466, + "mean_token_accuracy": 0.8812649846076965, + "num_tokens": 560783442.0, + "step": 15967 + }, + { + "epoch": 2.9652739090064997, + "grad_norm": 1.5450060367584229, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8812395334243774, + "num_tokens": 560822382.0, + "step": 15968 + }, + { + "epoch": 2.965459610027855, + "grad_norm": 1.5413358211517334, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8684297204017639, + "num_tokens": 560861888.0, + "step": 15969 + }, + { + "epoch": 2.9656453110492107, + "grad_norm": 1.3999429941177368, + "learning_rate": 1e-06, + "loss": 0.2851, + "mean_token_accuracy": 0.8971693515777588, + "num_tokens": 560900956.0, + "step": 15970 + }, + { + "epoch": 2.9658310120705664, + "grad_norm": 1.5032870769500732, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8868823647499084, + "num_tokens": 560937776.0, + "step": 15971 + }, + { + "epoch": 2.966016713091922, + "grad_norm": 1.6382415294647217, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8745840787887573, + "num_tokens": 560969930.0, + "step": 15972 + }, + { + "epoch": 2.966202414113278, + "grad_norm": 1.5484060049057007, + "learning_rate": 1e-06, + "loss": 0.3243, + "mean_token_accuracy": 0.885188639163971, + "num_tokens": 561004183.0, + "step": 15973 + }, + { + "epoch": 2.966388115134633, + "grad_norm": 1.6144678592681885, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8822005987167358, + "num_tokens": 561035964.0, + "step": 15974 + }, + { + "epoch": 2.966573816155989, + "grad_norm": 1.4817216396331787, + "learning_rate": 1e-06, + "loss": 0.2894, + "mean_token_accuracy": 0.8938614726066589, + "num_tokens": 561074019.0, + "step": 15975 + }, + { + "epoch": 2.966759517177344, + "grad_norm": 1.5964653491973877, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.8899638056755066, + "num_tokens": 561105812.0, + "step": 15976 + }, + { + "epoch": 2.9669452181987, + "grad_norm": 1.5647412538528442, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.89335036277771, + "num_tokens": 561141485.0, + "step": 15977 + }, + { + "epoch": 2.9671309192200557, + "grad_norm": 1.5859894752502441, + "learning_rate": 1e-06, + "loss": 0.3382, + "mean_token_accuracy": 0.8802717924118042, + "num_tokens": 561177341.0, + "step": 15978 + }, + { + "epoch": 2.9673166202414114, + "grad_norm": 1.443729281425476, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8754561543464661, + "num_tokens": 561219587.0, + "step": 15979 + }, + { + "epoch": 2.967502321262767, + "grad_norm": 1.4695441722869873, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8847219944000244, + "num_tokens": 561259480.0, + "step": 15980 + }, + { + "epoch": 2.9676880222841224, + "grad_norm": 1.6589637994766235, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8696200847625732, + "num_tokens": 561292052.0, + "step": 15981 + }, + { + "epoch": 2.967873723305478, + "grad_norm": 1.4176567792892456, + "learning_rate": 1e-06, + "loss": 0.2824, + "mean_token_accuracy": 0.9007496237754822, + "num_tokens": 561331331.0, + "step": 15982 + }, + { + "epoch": 2.968059424326834, + "grad_norm": 1.6152715682983398, + "learning_rate": 1e-06, + "loss": 0.3261, + "mean_token_accuracy": 0.8866696357727051, + "num_tokens": 561365104.0, + "step": 15983 + }, + { + "epoch": 2.968245125348189, + "grad_norm": 1.6254998445510864, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8961087465286255, + "num_tokens": 561395089.0, + "step": 15984 + }, + { + "epoch": 2.968430826369545, + "grad_norm": 1.6040629148483276, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8782919645309448, + "num_tokens": 561432529.0, + "step": 15985 + }, + { + "epoch": 2.9686165273909007, + "grad_norm": 1.7179604768753052, + "learning_rate": 1e-06, + "loss": 0.3895, + "mean_token_accuracy": 0.8665825128555298, + "num_tokens": 561468979.0, + "step": 15986 + }, + { + "epoch": 2.9688022284122564, + "grad_norm": 1.5178941488265991, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8853208422660828, + "num_tokens": 561503712.0, + "step": 15987 + }, + { + "epoch": 2.968987929433612, + "grad_norm": 1.613228678703308, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.882784366607666, + "num_tokens": 561538871.0, + "step": 15988 + }, + { + "epoch": 2.9691736304549674, + "grad_norm": 1.614109992980957, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8826805949211121, + "num_tokens": 561572439.0, + "step": 15989 + }, + { + "epoch": 2.969359331476323, + "grad_norm": 1.5389697551727295, + "learning_rate": 1e-06, + "loss": 0.3397, + "mean_token_accuracy": 0.8788372278213501, + "num_tokens": 561610189.0, + "step": 15990 + }, + { + "epoch": 2.969545032497679, + "grad_norm": 1.6356042623519897, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8834247589111328, + "num_tokens": 561642668.0, + "step": 15991 + }, + { + "epoch": 2.969730733519034, + "grad_norm": 1.5468741655349731, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8920535445213318, + "num_tokens": 561673994.0, + "step": 15992 + }, + { + "epoch": 2.96991643454039, + "grad_norm": 1.589961290359497, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8928502202033997, + "num_tokens": 561705901.0, + "step": 15993 + }, + { + "epoch": 2.9701021355617456, + "grad_norm": 1.6889986991882324, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8853642344474792, + "num_tokens": 561736308.0, + "step": 15994 + }, + { + "epoch": 2.9702878365831014, + "grad_norm": 1.4532198905944824, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8894301056861877, + "num_tokens": 561775748.0, + "step": 15995 + }, + { + "epoch": 2.970473537604457, + "grad_norm": 1.4442139863967896, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8861860632896423, + "num_tokens": 561816262.0, + "step": 15996 + }, + { + "epoch": 2.9706592386258124, + "grad_norm": 1.7709962129592896, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8753273487091064, + "num_tokens": 561848256.0, + "step": 15997 + }, + { + "epoch": 2.970844939647168, + "grad_norm": 1.599699854850769, + "learning_rate": 1e-06, + "loss": 0.2838, + "mean_token_accuracy": 0.8975327014923096, + "num_tokens": 561881391.0, + "step": 15998 + }, + { + "epoch": 2.9710306406685234, + "grad_norm": 1.5217951536178589, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8974993824958801, + "num_tokens": 561917353.0, + "step": 15999 + }, + { + "epoch": 2.971216341689879, + "grad_norm": 1.5524816513061523, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8848051428794861, + "num_tokens": 561955629.0, + "step": 16000 + }, + { + "epoch": 2.971402042711235, + "grad_norm": 1.6127418279647827, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.885538637638092, + "num_tokens": 561990201.0, + "step": 16001 + }, + { + "epoch": 2.9715877437325906, + "grad_norm": 1.532429814338684, + "learning_rate": 1e-06, + "loss": 0.2879, + "mean_token_accuracy": 0.8979290723800659, + "num_tokens": 562019938.0, + "step": 16002 + }, + { + "epoch": 2.9717734447539463, + "grad_norm": 1.4983899593353271, + "learning_rate": 1e-06, + "loss": 0.3267, + "mean_token_accuracy": 0.8872909545898438, + "num_tokens": 562061255.0, + "step": 16003 + }, + { + "epoch": 2.9719591457753016, + "grad_norm": 1.7515225410461426, + "learning_rate": 1e-06, + "loss": 0.333, + "mean_token_accuracy": 0.8829030394554138, + "num_tokens": 562088015.0, + "step": 16004 + }, + { + "epoch": 2.9721448467966574, + "grad_norm": 1.4081294536590576, + "learning_rate": 1e-06, + "loss": 0.2797, + "mean_token_accuracy": 0.899859607219696, + "num_tokens": 562125060.0, + "step": 16005 + }, + { + "epoch": 2.972330547818013, + "grad_norm": 1.4215409755706787, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8946611881256104, + "num_tokens": 562162407.0, + "step": 16006 + }, + { + "epoch": 2.9725162488393684, + "grad_norm": 1.6112682819366455, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8878771066665649, + "num_tokens": 562196181.0, + "step": 16007 + }, + { + "epoch": 2.972701949860724, + "grad_norm": 1.4438117742538452, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8901551365852356, + "num_tokens": 562233958.0, + "step": 16008 + }, + { + "epoch": 2.97288765088208, + "grad_norm": 1.485498309135437, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8895214796066284, + "num_tokens": 562271641.0, + "step": 16009 + }, + { + "epoch": 2.9730733519034356, + "grad_norm": 1.4943623542785645, + "learning_rate": 1e-06, + "loss": 0.307, + "mean_token_accuracy": 0.8932698965072632, + "num_tokens": 562308449.0, + "step": 16010 + }, + { + "epoch": 2.9732590529247913, + "grad_norm": 1.607769250869751, + "learning_rate": 1e-06, + "loss": 0.3227, + "mean_token_accuracy": 0.8894559144973755, + "num_tokens": 562341353.0, + "step": 16011 + }, + { + "epoch": 2.9734447539461466, + "grad_norm": 1.6074920892715454, + "learning_rate": 1e-06, + "loss": 0.3051, + "mean_token_accuracy": 0.8921325206756592, + "num_tokens": 562378140.0, + "step": 16012 + }, + { + "epoch": 2.9736304549675023, + "grad_norm": 1.521510362625122, + "learning_rate": 1e-06, + "loss": 0.3144, + "mean_token_accuracy": 0.8892929553985596, + "num_tokens": 562414034.0, + "step": 16013 + }, + { + "epoch": 2.973816155988858, + "grad_norm": 1.6691968441009521, + "learning_rate": 1e-06, + "loss": 0.3494, + "mean_token_accuracy": 0.8783711194992065, + "num_tokens": 562444826.0, + "step": 16014 + }, + { + "epoch": 2.9740018570102134, + "grad_norm": 1.4837678670883179, + "learning_rate": 1e-06, + "loss": 0.2738, + "mean_token_accuracy": 0.9006146192550659, + "num_tokens": 562476875.0, + "step": 16015 + }, + { + "epoch": 2.974187558031569, + "grad_norm": 1.6584867238998413, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8758313655853271, + "num_tokens": 562511943.0, + "step": 16016 + }, + { + "epoch": 2.974373259052925, + "grad_norm": 1.4780610799789429, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8947170972824097, + "num_tokens": 562549406.0, + "step": 16017 + }, + { + "epoch": 2.9745589600742806, + "grad_norm": 1.5091207027435303, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8894568085670471, + "num_tokens": 562589082.0, + "step": 16018 + }, + { + "epoch": 2.9747446610956363, + "grad_norm": 1.5897217988967896, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8872966766357422, + "num_tokens": 562622870.0, + "step": 16019 + }, + { + "epoch": 2.9749303621169916, + "grad_norm": 1.5064787864685059, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8868326544761658, + "num_tokens": 562662026.0, + "step": 16020 + }, + { + "epoch": 2.9751160631383473, + "grad_norm": 1.5869300365447998, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.876944899559021, + "num_tokens": 562701382.0, + "step": 16021 + }, + { + "epoch": 2.975301764159703, + "grad_norm": 1.450927734375, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8941104412078857, + "num_tokens": 562739602.0, + "step": 16022 + }, + { + "epoch": 2.9754874651810583, + "grad_norm": 1.4794299602508545, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8859884142875671, + "num_tokens": 562779156.0, + "step": 16023 + }, + { + "epoch": 2.975673166202414, + "grad_norm": 1.4158860445022583, + "learning_rate": 1e-06, + "loss": 0.2653, + "mean_token_accuracy": 0.9066505432128906, + "num_tokens": 562812962.0, + "step": 16024 + }, + { + "epoch": 2.97585886722377, + "grad_norm": 1.4839119911193848, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8764793872833252, + "num_tokens": 562853736.0, + "step": 16025 + }, + { + "epoch": 2.9760445682451255, + "grad_norm": 1.6270631551742554, + "learning_rate": 1e-06, + "loss": 0.3244, + "mean_token_accuracy": 0.8881287574768066, + "num_tokens": 562886201.0, + "step": 16026 + }, + { + "epoch": 2.976230269266481, + "grad_norm": 1.536128044128418, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8951141834259033, + "num_tokens": 562919672.0, + "step": 16027 + }, + { + "epoch": 2.9764159702878366, + "grad_norm": 1.5647897720336914, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8899731040000916, + "num_tokens": 562952716.0, + "step": 16028 + }, + { + "epoch": 2.9766016713091923, + "grad_norm": 1.5456490516662598, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8927326798439026, + "num_tokens": 562989454.0, + "step": 16029 + }, + { + "epoch": 2.9767873723305476, + "grad_norm": 1.6146230697631836, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8941195011138916, + "num_tokens": 563021633.0, + "step": 16030 + }, + { + "epoch": 2.9769730733519033, + "grad_norm": 1.6153091192245483, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8887903690338135, + "num_tokens": 563052626.0, + "step": 16031 + }, + { + "epoch": 2.977158774373259, + "grad_norm": 1.6126192808151245, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8889634609222412, + "num_tokens": 563083347.0, + "step": 16032 + }, + { + "epoch": 2.9773444753946148, + "grad_norm": 1.6675506830215454, + "learning_rate": 1e-06, + "loss": 0.3402, + "mean_token_accuracy": 0.8795651793479919, + "num_tokens": 563121861.0, + "step": 16033 + }, + { + "epoch": 2.9775301764159705, + "grad_norm": 1.6474335193634033, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8938552141189575, + "num_tokens": 563154069.0, + "step": 16034 + }, + { + "epoch": 2.977715877437326, + "grad_norm": 1.517349123954773, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8952137231826782, + "num_tokens": 563188381.0, + "step": 16035 + }, + { + "epoch": 2.9779015784586815, + "grad_norm": 1.5423917770385742, + "learning_rate": 1e-06, + "loss": 0.2819, + "mean_token_accuracy": 0.897786021232605, + "num_tokens": 563220341.0, + "step": 16036 + }, + { + "epoch": 2.9780872794800373, + "grad_norm": 1.59151291847229, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8783513307571411, + "num_tokens": 563255669.0, + "step": 16037 + }, + { + "epoch": 2.9782729805013926, + "grad_norm": 1.5453743934631348, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8919240832328796, + "num_tokens": 563292856.0, + "step": 16038 + }, + { + "epoch": 2.9784586815227483, + "grad_norm": 1.559874176979065, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8926804065704346, + "num_tokens": 563324420.0, + "step": 16039 + }, + { + "epoch": 2.978644382544104, + "grad_norm": 1.5513354539871216, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8933494091033936, + "num_tokens": 563359404.0, + "step": 16040 + }, + { + "epoch": 2.9788300835654598, + "grad_norm": 1.623773455619812, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8808913230895996, + "num_tokens": 563391618.0, + "step": 16041 + }, + { + "epoch": 2.9790157845868155, + "grad_norm": 1.5292798280715942, + "learning_rate": 1e-06, + "loss": 0.2917, + "mean_token_accuracy": 0.8965203762054443, + "num_tokens": 563430112.0, + "step": 16042 + }, + { + "epoch": 2.9792014856081708, + "grad_norm": 1.7173932790756226, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8943468332290649, + "num_tokens": 563457692.0, + "step": 16043 + }, + { + "epoch": 2.9793871866295265, + "grad_norm": 1.693960189819336, + "learning_rate": 1e-06, + "loss": 0.341, + "mean_token_accuracy": 0.879150390625, + "num_tokens": 563490927.0, + "step": 16044 + }, + { + "epoch": 2.9795728876508822, + "grad_norm": 1.7748851776123047, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8793759942054749, + "num_tokens": 563520162.0, + "step": 16045 + }, + { + "epoch": 2.9797585886722375, + "grad_norm": 1.5949392318725586, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8825076818466187, + "num_tokens": 563558044.0, + "step": 16046 + }, + { + "epoch": 2.9799442896935933, + "grad_norm": 1.6331205368041992, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8841140270233154, + "num_tokens": 563593246.0, + "step": 16047 + }, + { + "epoch": 2.980129990714949, + "grad_norm": 1.6328850984573364, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8945884704589844, + "num_tokens": 563622680.0, + "step": 16048 + }, + { + "epoch": 2.9803156917363047, + "grad_norm": 1.644588828086853, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8773903846740723, + "num_tokens": 563655827.0, + "step": 16049 + }, + { + "epoch": 2.98050139275766, + "grad_norm": 1.731690526008606, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8818360567092896, + "num_tokens": 563686892.0, + "step": 16050 + }, + { + "epoch": 2.9806870937790158, + "grad_norm": 1.5441360473632812, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8802895545959473, + "num_tokens": 563725522.0, + "step": 16051 + }, + { + "epoch": 2.9808727948003715, + "grad_norm": 1.5824196338653564, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8801971077919006, + "num_tokens": 563758865.0, + "step": 16052 + }, + { + "epoch": 2.9810584958217268, + "grad_norm": 1.4490299224853516, + "learning_rate": 1e-06, + "loss": 0.2522, + "mean_token_accuracy": 0.907227098941803, + "num_tokens": 563789214.0, + "step": 16053 + }, + { + "epoch": 2.9812441968430825, + "grad_norm": 1.736991047859192, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8717691898345947, + "num_tokens": 563820647.0, + "step": 16054 + }, + { + "epoch": 2.9814298978644382, + "grad_norm": 1.4129502773284912, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8973028659820557, + "num_tokens": 563857917.0, + "step": 16055 + }, + { + "epoch": 2.981615598885794, + "grad_norm": 1.6274805068969727, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.886428952217102, + "num_tokens": 563893839.0, + "step": 16056 + }, + { + "epoch": 2.9818012999071497, + "grad_norm": 1.6960344314575195, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8844117522239685, + "num_tokens": 563925231.0, + "step": 16057 + }, + { + "epoch": 2.981987000928505, + "grad_norm": 1.4846450090408325, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8861500024795532, + "num_tokens": 563964847.0, + "step": 16058 + }, + { + "epoch": 2.9821727019498607, + "grad_norm": 1.5944550037384033, + "learning_rate": 1e-06, + "loss": 0.3146, + "mean_token_accuracy": 0.8915592432022095, + "num_tokens": 563998374.0, + "step": 16059 + }, + { + "epoch": 2.9823584029712165, + "grad_norm": 1.5585345029830933, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8860428333282471, + "num_tokens": 564034606.0, + "step": 16060 + }, + { + "epoch": 2.9825441039925717, + "grad_norm": 1.7011281251907349, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8781603574752808, + "num_tokens": 564066864.0, + "step": 16061 + }, + { + "epoch": 2.9827298050139275, + "grad_norm": 1.5162416696548462, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.890862226486206, + "num_tokens": 564099205.0, + "step": 16062 + }, + { + "epoch": 2.982915506035283, + "grad_norm": 1.5231558084487915, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8945889472961426, + "num_tokens": 564132833.0, + "step": 16063 + }, + { + "epoch": 2.983101207056639, + "grad_norm": 1.473829746246338, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8787760138511658, + "num_tokens": 564172919.0, + "step": 16064 + }, + { + "epoch": 2.9832869080779947, + "grad_norm": 1.601037621498108, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8704699277877808, + "num_tokens": 564207423.0, + "step": 16065 + }, + { + "epoch": 2.98347260909935, + "grad_norm": 1.5733872652053833, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8892396092414856, + "num_tokens": 564241827.0, + "step": 16066 + }, + { + "epoch": 2.9836583101207057, + "grad_norm": 1.5645911693572998, + "learning_rate": 1e-06, + "loss": 0.2888, + "mean_token_accuracy": 0.896457314491272, + "num_tokens": 564273014.0, + "step": 16067 + }, + { + "epoch": 2.9838440111420614, + "grad_norm": 1.7187726497650146, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8746976852416992, + "num_tokens": 564303615.0, + "step": 16068 + }, + { + "epoch": 2.9840297121634167, + "grad_norm": 1.5466586351394653, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8772366642951965, + "num_tokens": 564342476.0, + "step": 16069 + }, + { + "epoch": 2.9842154131847725, + "grad_norm": 1.4928133487701416, + "learning_rate": 1e-06, + "loss": 0.2739, + "mean_token_accuracy": 0.9029629826545715, + "num_tokens": 564374380.0, + "step": 16070 + }, + { + "epoch": 2.984401114206128, + "grad_norm": 1.590962290763855, + "learning_rate": 1e-06, + "loss": 0.3714, + "mean_token_accuracy": 0.8724557757377625, + "num_tokens": 564411588.0, + "step": 16071 + }, + { + "epoch": 2.984586815227484, + "grad_norm": 1.4043487310409546, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8860054016113281, + "num_tokens": 564455288.0, + "step": 16072 + }, + { + "epoch": 2.984772516248839, + "grad_norm": 1.6225837469100952, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8946916460990906, + "num_tokens": 564485932.0, + "step": 16073 + }, + { + "epoch": 2.984958217270195, + "grad_norm": 1.624912142753601, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.8821447491645813, + "num_tokens": 564517409.0, + "step": 16074 + }, + { + "epoch": 2.9851439182915507, + "grad_norm": 1.421244502067566, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8829295635223389, + "num_tokens": 564559114.0, + "step": 16075 + }, + { + "epoch": 2.985329619312906, + "grad_norm": 1.5966919660568237, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.885719895362854, + "num_tokens": 564593870.0, + "step": 16076 + }, + { + "epoch": 2.9855153203342617, + "grad_norm": 1.6893165111541748, + "learning_rate": 1e-06, + "loss": 0.3444, + "mean_token_accuracy": 0.8827134966850281, + "num_tokens": 564623003.0, + "step": 16077 + }, + { + "epoch": 2.9857010213556174, + "grad_norm": 1.5367063283920288, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8844877481460571, + "num_tokens": 564656694.0, + "step": 16078 + }, + { + "epoch": 2.985886722376973, + "grad_norm": 1.465192198753357, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8954708576202393, + "num_tokens": 564693604.0, + "step": 16079 + }, + { + "epoch": 2.986072423398329, + "grad_norm": 1.4908020496368408, + "learning_rate": 1e-06, + "loss": 0.3187, + "mean_token_accuracy": 0.8898749351501465, + "num_tokens": 564732256.0, + "step": 16080 + }, + { + "epoch": 2.986258124419684, + "grad_norm": 1.5354443788528442, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8955802917480469, + "num_tokens": 564766623.0, + "step": 16081 + }, + { + "epoch": 2.98644382544104, + "grad_norm": 1.5383391380310059, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.8825538158416748, + "num_tokens": 564805859.0, + "step": 16082 + }, + { + "epoch": 2.9866295264623957, + "grad_norm": 1.6425966024398804, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.888629674911499, + "num_tokens": 564840396.0, + "step": 16083 + }, + { + "epoch": 2.986815227483751, + "grad_norm": 1.331312656402588, + "learning_rate": 1e-06, + "loss": 0.2936, + "mean_token_accuracy": 0.8964943885803223, + "num_tokens": 564883552.0, + "step": 16084 + }, + { + "epoch": 2.9870009285051067, + "grad_norm": 1.4535661935806274, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8836034536361694, + "num_tokens": 564920429.0, + "step": 16085 + }, + { + "epoch": 2.9871866295264624, + "grad_norm": 1.552306890487671, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8848607540130615, + "num_tokens": 564954570.0, + "step": 16086 + }, + { + "epoch": 2.987372330547818, + "grad_norm": 1.5474125146865845, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8917425870895386, + "num_tokens": 564990622.0, + "step": 16087 + }, + { + "epoch": 2.987558031569174, + "grad_norm": 1.5539474487304688, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8735157251358032, + "num_tokens": 565028666.0, + "step": 16088 + }, + { + "epoch": 2.987743732590529, + "grad_norm": 1.5747023820877075, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.8930366039276123, + "num_tokens": 565061909.0, + "step": 16089 + }, + { + "epoch": 2.987929433611885, + "grad_norm": 1.4893232583999634, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.878911018371582, + "num_tokens": 565100386.0, + "step": 16090 + }, + { + "epoch": 2.9881151346332406, + "grad_norm": 1.5621272325515747, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8848896026611328, + "num_tokens": 565135927.0, + "step": 16091 + }, + { + "epoch": 2.988300835654596, + "grad_norm": 1.499593734741211, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8786421418190002, + "num_tokens": 565173557.0, + "step": 16092 + }, + { + "epoch": 2.9884865366759517, + "grad_norm": 1.4164305925369263, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8905646204948425, + "num_tokens": 565215877.0, + "step": 16093 + }, + { + "epoch": 2.9886722376973074, + "grad_norm": 1.605688214302063, + "learning_rate": 1e-06, + "loss": 0.3236, + "mean_token_accuracy": 0.880827009677887, + "num_tokens": 565248962.0, + "step": 16094 + }, + { + "epoch": 2.988857938718663, + "grad_norm": 1.4848248958587646, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8909800052642822, + "num_tokens": 565284000.0, + "step": 16095 + }, + { + "epoch": 2.9890436397400184, + "grad_norm": 1.6254132986068726, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8910051584243774, + "num_tokens": 565312279.0, + "step": 16096 + }, + { + "epoch": 2.989229340761374, + "grad_norm": 1.3974933624267578, + "learning_rate": 1e-06, + "loss": 0.3192, + "mean_token_accuracy": 0.8866189122200012, + "num_tokens": 565356781.0, + "step": 16097 + }, + { + "epoch": 2.98941504178273, + "grad_norm": 1.5740163326263428, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.896925687789917, + "num_tokens": 565389434.0, + "step": 16098 + }, + { + "epoch": 2.989600742804085, + "grad_norm": 1.6199053525924683, + "learning_rate": 1e-06, + "loss": 0.358, + "mean_token_accuracy": 0.8736671209335327, + "num_tokens": 565422825.0, + "step": 16099 + }, + { + "epoch": 2.989786443825441, + "grad_norm": 1.4613356590270996, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8958983421325684, + "num_tokens": 565460544.0, + "step": 16100 + }, + { + "epoch": 2.9899721448467966, + "grad_norm": 1.5411145687103271, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8836060762405396, + "num_tokens": 565497100.0, + "step": 16101 + }, + { + "epoch": 2.9901578458681524, + "grad_norm": 1.6809256076812744, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8687942028045654, + "num_tokens": 565538304.0, + "step": 16102 + }, + { + "epoch": 2.990343546889508, + "grad_norm": 1.7007125616073608, + "learning_rate": 1e-06, + "loss": 0.3403, + "mean_token_accuracy": 0.8813796043395996, + "num_tokens": 565568466.0, + "step": 16103 + }, + { + "epoch": 2.9905292479108634, + "grad_norm": 1.518175721168518, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.885547935962677, + "num_tokens": 565605440.0, + "step": 16104 + }, + { + "epoch": 2.990714948932219, + "grad_norm": 1.5414398908615112, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8947417736053467, + "num_tokens": 565640804.0, + "step": 16105 + }, + { + "epoch": 2.990900649953575, + "grad_norm": 1.5662652254104614, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.890267550945282, + "num_tokens": 565675664.0, + "step": 16106 + }, + { + "epoch": 2.99108635097493, + "grad_norm": 1.5166499614715576, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8816571235656738, + "num_tokens": 565712946.0, + "step": 16107 + }, + { + "epoch": 2.991272051996286, + "grad_norm": 1.6660047769546509, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8844906091690063, + "num_tokens": 565744332.0, + "step": 16108 + }, + { + "epoch": 2.9914577530176416, + "grad_norm": 1.5410858392715454, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8916735649108887, + "num_tokens": 565777985.0, + "step": 16109 + }, + { + "epoch": 2.9916434540389973, + "grad_norm": 1.6490960121154785, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8803672790527344, + "num_tokens": 565812142.0, + "step": 16110 + }, + { + "epoch": 2.991829155060353, + "grad_norm": 1.6446654796600342, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.8866031169891357, + "num_tokens": 565845956.0, + "step": 16111 + }, + { + "epoch": 2.9920148560817084, + "grad_norm": 1.5497796535491943, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8951704502105713, + "num_tokens": 565881692.0, + "step": 16112 + }, + { + "epoch": 2.992200557103064, + "grad_norm": 1.5597790479660034, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8863023519515991, + "num_tokens": 565920082.0, + "step": 16113 + }, + { + "epoch": 2.99238625812442, + "grad_norm": 1.4805784225463867, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8927222490310669, + "num_tokens": 565958154.0, + "step": 16114 + }, + { + "epoch": 2.992571959145775, + "grad_norm": 1.7064118385314941, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8939570188522339, + "num_tokens": 565988393.0, + "step": 16115 + }, + { + "epoch": 2.992757660167131, + "grad_norm": 1.4630515575408936, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8797780871391296, + "num_tokens": 566028639.0, + "step": 16116 + }, + { + "epoch": 2.9929433611884866, + "grad_norm": 1.6339747905731201, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.8850898146629333, + "num_tokens": 566059504.0, + "step": 16117 + }, + { + "epoch": 2.9931290622098423, + "grad_norm": 1.576966404914856, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.8937634229660034, + "num_tokens": 566088630.0, + "step": 16118 + }, + { + "epoch": 2.9933147632311976, + "grad_norm": 1.6886481046676636, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8923615217208862, + "num_tokens": 566116325.0, + "step": 16119 + }, + { + "epoch": 2.9935004642525533, + "grad_norm": 1.7711416482925415, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8750742673873901, + "num_tokens": 566145867.0, + "step": 16120 + }, + { + "epoch": 2.993686165273909, + "grad_norm": 1.5369298458099365, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8943254947662354, + "num_tokens": 566176832.0, + "step": 16121 + }, + { + "epoch": 2.9938718662952644, + "grad_norm": 1.6108957529067993, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.887239933013916, + "num_tokens": 566207889.0, + "step": 16122 + }, + { + "epoch": 2.99405756731662, + "grad_norm": 1.5677506923675537, + "learning_rate": 1e-06, + "loss": 0.3639, + "mean_token_accuracy": 0.8709226846694946, + "num_tokens": 566244785.0, + "step": 16123 + }, + { + "epoch": 2.994243268337976, + "grad_norm": 1.797215461730957, + "learning_rate": 1e-06, + "loss": 0.3532, + "mean_token_accuracy": 0.8764276504516602, + "num_tokens": 566272168.0, + "step": 16124 + }, + { + "epoch": 2.9944289693593316, + "grad_norm": 1.4784408807754517, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8839678764343262, + "num_tokens": 566308469.0, + "step": 16125 + }, + { + "epoch": 2.9946146703806873, + "grad_norm": 1.437311053276062, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8928772211074829, + "num_tokens": 566345567.0, + "step": 16126 + }, + { + "epoch": 2.9948003714020426, + "grad_norm": 1.5557451248168945, + "learning_rate": 1e-06, + "loss": 0.3307, + "mean_token_accuracy": 0.8862054347991943, + "num_tokens": 566379601.0, + "step": 16127 + }, + { + "epoch": 2.9949860724233983, + "grad_norm": 1.4461615085601807, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8934758305549622, + "num_tokens": 566417814.0, + "step": 16128 + }, + { + "epoch": 2.995171773444754, + "grad_norm": 1.4348880052566528, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8839563131332397, + "num_tokens": 566459747.0, + "step": 16129 + }, + { + "epoch": 2.9953574744661093, + "grad_norm": 1.476104497909546, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8912023305892944, + "num_tokens": 566492555.0, + "step": 16130 + }, + { + "epoch": 2.995543175487465, + "grad_norm": 1.6542456150054932, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8922313451766968, + "num_tokens": 566520873.0, + "step": 16131 + }, + { + "epoch": 2.995728876508821, + "grad_norm": 1.7034704685211182, + "learning_rate": 1e-06, + "loss": 0.294, + "mean_token_accuracy": 0.8968543410301208, + "num_tokens": 566550239.0, + "step": 16132 + }, + { + "epoch": 2.9959145775301765, + "grad_norm": 1.6231858730316162, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.887553334236145, + "num_tokens": 566582183.0, + "step": 16133 + }, + { + "epoch": 2.9961002785515323, + "grad_norm": 1.406538724899292, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8846012353897095, + "num_tokens": 566625948.0, + "step": 16134 + }, + { + "epoch": 2.9962859795728876, + "grad_norm": 1.46208655834198, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8947407007217407, + "num_tokens": 566662767.0, + "step": 16135 + }, + { + "epoch": 2.9964716805942433, + "grad_norm": 1.5466947555541992, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.880362868309021, + "num_tokens": 566701367.0, + "step": 16136 + }, + { + "epoch": 2.996657381615599, + "grad_norm": 1.5452909469604492, + "learning_rate": 1e-06, + "loss": 0.3469, + "mean_token_accuracy": 0.8814716935157776, + "num_tokens": 566738508.0, + "step": 16137 + }, + { + "epoch": 2.9968430826369543, + "grad_norm": 1.6813921928405762, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8837443590164185, + "num_tokens": 566771516.0, + "step": 16138 + }, + { + "epoch": 2.99702878365831, + "grad_norm": 1.5164364576339722, + "learning_rate": 1e-06, + "loss": 0.263, + "mean_token_accuracy": 0.9044084548950195, + "num_tokens": 566802920.0, + "step": 16139 + }, + { + "epoch": 2.997214484679666, + "grad_norm": 1.574737787246704, + "learning_rate": 1e-06, + "loss": 0.2995, + "mean_token_accuracy": 0.8936600089073181, + "num_tokens": 566834968.0, + "step": 16140 + }, + { + "epoch": 2.9974001857010215, + "grad_norm": 1.5576146841049194, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8969441056251526, + "num_tokens": 566865795.0, + "step": 16141 + }, + { + "epoch": 2.9975858867223772, + "grad_norm": 1.456183671951294, + "learning_rate": 1e-06, + "loss": 0.2666, + "mean_token_accuracy": 0.9036372900009155, + "num_tokens": 566900167.0, + "step": 16142 + }, + { + "epoch": 2.9977715877437325, + "grad_norm": 1.3907923698425293, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8924165368080139, + "num_tokens": 566940191.0, + "step": 16143 + }, + { + "epoch": 2.9979572887650883, + "grad_norm": 1.593958854675293, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8693256974220276, + "num_tokens": 566976214.0, + "step": 16144 + }, + { + "epoch": 2.9981429897864436, + "grad_norm": 1.536332607269287, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8812896013259888, + "num_tokens": 567013143.0, + "step": 16145 + }, + { + "epoch": 2.9983286908077993, + "grad_norm": 1.5881365537643433, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8841956853866577, + "num_tokens": 567048392.0, + "step": 16146 + }, + { + "epoch": 2.998514391829155, + "grad_norm": 1.472267746925354, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8743429183959961, + "num_tokens": 567091640.0, + "step": 16147 + }, + { + "epoch": 2.9987000928505108, + "grad_norm": 1.5699303150177002, + "learning_rate": 1e-06, + "loss": 0.3061, + "mean_token_accuracy": 0.8949753046035767, + "num_tokens": 567126756.0, + "step": 16148 + }, + { + "epoch": 2.9988857938718665, + "grad_norm": 1.5656375885009766, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8756282329559326, + "num_tokens": 567163687.0, + "step": 16149 + }, + { + "epoch": 2.999071494893222, + "grad_norm": 1.6744166612625122, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8849716782569885, + "num_tokens": 567191521.0, + "step": 16150 + }, + { + "epoch": 2.9992571959145775, + "grad_norm": 1.534080982208252, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8849948644638062, + "num_tokens": 567227644.0, + "step": 16151 + }, + { + "epoch": 2.9994428969359332, + "grad_norm": 1.597373604774475, + "learning_rate": 1e-06, + "loss": 0.2624, + "mean_token_accuracy": 0.9060046672821045, + "num_tokens": 567255836.0, + "step": 16152 + }, + { + "epoch": 2.9996285979572885, + "grad_norm": 1.7312955856323242, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8856887817382812, + "num_tokens": 567285246.0, + "step": 16153 + }, + { + "epoch": 2.9998142989786443, + "grad_norm": 1.559607982635498, + "learning_rate": 1e-06, + "loss": 0.3082, + "mean_token_accuracy": 0.887574315071106, + "num_tokens": 567317642.0, + "step": 16154 + }, + { + "epoch": 3.0, + "grad_norm": 1.3875731229782104, + "learning_rate": 1e-06, + "loss": 0.3066, + "mean_token_accuracy": 0.8899900913238525, + "num_tokens": 567357843.0, + "step": 16155 + }, + { + "epoch": 3.0, + "step": 16155, + "total_flos": 2.554786723790219e+19, + "train_loss": 0.3743892161239599, + "train_runtime": 22902.1621, + "train_samples_per_second": 11.286, + "train_steps_per_second": 0.705 + } + ], + "logging_steps": 1, + "max_steps": 16155, + "num_input_tokens_seen": 0, + "num_train_epochs": 3, + "save_steps": 8078, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 2.554786723790219e+19, + "train_batch_size": 4, + "trial_name": null, + "trial_params": null +} diff --git a/training_args.bin b/training_args.bin new file mode 100644 index 0000000..aa07c44 --- /dev/null +++ b/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2526a3879a34e541c6ae6f2c50153532428ce93408d4cdebd5123992fe5d237 +size 13073