commit eadf59d82fa406acbdd557e8b1bed19bf6ecfde4 Author: ModelHub XC Date: Fri Aug 28 08:48:18 2026 +0800 初始化项目，由ModelHub XC社区提供模型 Model: Neelectric/Llama-3.1-8B-Instruct_SFT_sciencev00.15 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..e9c4501 --- /dev/null +++ b/README.md @@ -0,0 +1,59 @@ +--- +base_model: meta-llama/Llama-3.1-8B-Instruct +datasets: Neelectric/wildguardmix_Llama-3.1-8B-Instruct_4096toks_refusals_only +library_name: transformers +model_name: Llama-3.1-8B-Instruct_SFT_sciencev00.15 +tags: +- generated_from_trainer +- open-r1 +- sft +- trl +licence: license +--- + +# Model Card for Llama-3.1-8B-Instruct_SFT_sciencev00.15 + +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/wildguardmix_Llama-3.1-8B-Instruct_4096toks_refusals_only](https://huggingface.co/datasets/Neelectric/wildguardmix_Llama-3.1-8B-Instruct_4096toks_refusals_only) 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_sciencev00.15", 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_science/runs/hb1r0hih) + + +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 +@software{vonwerra2020trl, + title = {{TRL: Transformers Reinforcement Learning}}, + author = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin}, + license = {Apache-2.0}, + url = {https://github.com/huggingface/trl}, + year = {2020} +} +``` \ No newline at end of file diff --git a/all_results.json b/all_results.json new file mode 100644 index 0000000..82d89dd --- /dev/null +++ b/all_results.json @@ -0,0 +1,8 @@ +{ + "total_flos": 5.664122751544197e+17, + "train_loss": 0.30678034723186126, + "train_runtime": 2812.6177, + "train_samples": 31266, + "train_samples_per_second": 22.233, + "train_steps_per_second": 1.39 +} \ No newline at end of file diff --git a/chat_template.jinja b/chat_template.jinja new file mode 100644 index 0000000..0ab931a --- /dev/null +++ b/chat_template.jinja @@ -0,0 +1,121 @@ +{{- 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) %} + {%- if message['role'] == 'assistant' %} + {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' }} + {% generation %} + {{- message['content'] | trim + '<|eot_id|>' }} + {% endgeneration %} + {%- else %} + {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' }} + {%- endif %} + {%- 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 %} + {{- '<|start_header_id|>assistant<|end_header_id|>\n\n' -}} + {% generation %} + {%- if builtin_tools is defined and tool_call.name in builtin_tools %} + {{- "<|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 %} + {{- '{"name": "' + tool_call.name + '", ' }} + {{- '"parameters": ' }} + {{- tool_call.arguments | tojson }} + {{- "}" }} + {%- endif %} + {%- if builtin_tools is defined %} + {{- "<|eom_id|>" }} + {%- else %} + {{- "<|eot_id|>" }} + {%- endif %} + {% endgeneration %} + {%- 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 %} \ No newline at end of file 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..ed3e96d --- /dev/null +++ b/model-00001-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:516b0dd9630d1417193499be66fbe2136bb0788bfad4391e1a281fe63c731fad +size 4976698672 diff --git a/model-00002-of-00004.safetensors b/model-00002-of-00004.safetensors new file mode 100644 index 0000000..da5cb54 --- /dev/null +++ b/model-00002-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32bce45e95fbbfbd25a62529bd0cdce2062586a6d085532beba7cefc4d37b698 +size 4999802720 diff --git a/model-00003-of-00004.safetensors b/model-00003-of-00004.safetensors new file mode 100644 index 0000000..e382e27 --- /dev/null +++ b/model-00003-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bd23d76392e0db60a106aea13605d32eb71fa885b7cd11966fdaecfebe61c53 +size 4915916176 diff --git a/model-00004-of-00004.safetensors b/model-00004-of-00004.safetensors new file mode 100644 index 0000000..711b682 --- /dev/null +++ b/model-00004-of-00004.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b16ad0a67c51348424b03417cac4dd52bd86449a4840b18f8d8703d08b7bf67 +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..82d89dd --- /dev/null +++ b/train_results.json @@ -0,0 +1,8 @@ +{ + "total_flos": 5.664122751544197e+17, + "train_loss": 0.30678034723186126, + "train_runtime": 2812.6177, + "train_samples": 31266, + "train_samples_per_second": 22.233, + "train_steps_per_second": 1.39 +} \ No newline at end of file diff --git a/trainer_state.json b/trainer_state.json new file mode 100644 index 0000000..af31a40 --- /dev/null +++ b/trainer_state.json @@ -0,0 +1,35233 @@ +{ + "best_global_step": null, + "best_metric": null, + "best_model_checkpoint": null, + "epoch": 2.0, + "eval_steps": 500, + "global_step": 3910, + "is_hyper_param_search": false, + "is_local_process_zero": true, + "is_world_process_zero": true, + "log_history": [ + { + "epoch": 0.0005115089514066496, + "grad_norm": 178.24435424804688, + "learning_rate": 0.0, + "loss": 2.3229, + "mean_token_accuracy": 0.7896848320960999, + "num_tokens": 2797.0, + "step": 1 + }, + { + "epoch": 0.0010230179028132991, + "grad_norm": 269.0757751464844, + "learning_rate": 2.557544757033248e-09, + "loss": 3.5846, + "mean_token_accuracy": 0.7011135816574097, + "num_tokens": 5778.0, + "step": 2 + }, + { + "epoch": 0.001534526854219949, + "grad_norm": 143.24859619140625, + "learning_rate": 5.115089514066496e-09, + "loss": 2.1878, + "mean_token_accuracy": 0.7541686296463013, + "num_tokens": 8922.0, + "step": 3 + }, + { + "epoch": 0.0020460358056265983, + "grad_norm": 141.7488555908203, + "learning_rate": 7.672634271099745e-09, + "loss": 2.3032, + "mean_token_accuracy": 0.7885234355926514, + "num_tokens": 11828.0, + "step": 4 + }, + { + "epoch": 0.0025575447570332483, + "grad_norm": 52.665504455566406, + "learning_rate": 1.0230179028132992e-08, + "loss": 0.9437, + "mean_token_accuracy": 0.827370822429657, + "num_tokens": 16163.0, + "step": 5 + }, + { + "epoch": 0.003069053708439898, + "grad_norm": 64.31586456298828, + "learning_rate": 1.278772378516624e-08, + "loss": 1.1972, + "mean_token_accuracy": 0.8139883875846863, + "num_tokens": 19641.0, + "step": 6 + }, + { + "epoch": 0.0035805626598465474, + "grad_norm": 143.20518493652344, + "learning_rate": 1.534526854219949e-08, + "loss": 2.1651, + "mean_token_accuracy": 0.8090519905090332, + "num_tokens": 22219.0, + "step": 7 + }, + { + "epoch": 0.004092071611253197, + "grad_norm": 139.1504669189453, + "learning_rate": 1.7902813299232734e-08, + "loss": 2.0921, + "mean_token_accuracy": 0.8322995901107788, + "num_tokens": 26106.0, + "step": 8 + }, + { + "epoch": 0.004603580562659846, + "grad_norm": 158.3728790283203, + "learning_rate": 2.0460358056265984e-08, + "loss": 2.28, + "mean_token_accuracy": 0.7669402360916138, + "num_tokens": 28781.0, + "step": 9 + }, + { + "epoch": 0.005115089514066497, + "grad_norm": 157.36215209960938, + "learning_rate": 2.301790281329923e-08, + "loss": 2.4264, + "mean_token_accuracy": 0.7941558361053467, + "num_tokens": 31415.0, + "step": 10 + }, + { + "epoch": 0.005626598465473146, + "grad_norm": 137.9846954345703, + "learning_rate": 2.557544757033248e-08, + "loss": 1.8915, + "mean_token_accuracy": 0.8200650215148926, + "num_tokens": 33985.0, + "step": 11 + }, + { + "epoch": 0.006138107416879796, + "grad_norm": 237.4044952392578, + "learning_rate": 2.8132992327365726e-08, + "loss": 3.2348, + "mean_token_accuracy": 0.7327378988265991, + "num_tokens": 36248.0, + "step": 12 + }, + { + "epoch": 0.006649616368286445, + "grad_norm": 192.54115295410156, + "learning_rate": 3.069053708439898e-08, + "loss": 2.7739, + "mean_token_accuracy": 0.7563351988792419, + "num_tokens": 38735.0, + "step": 13 + }, + { + "epoch": 0.007161125319693095, + "grad_norm": 102.98716735839844, + "learning_rate": 3.3248081841432223e-08, + "loss": 1.7731, + "mean_token_accuracy": 0.7895292639732361, + "num_tokens": 41763.0, + "step": 14 + }, + { + "epoch": 0.0076726342710997444, + "grad_norm": 149.59825134277344, + "learning_rate": 3.580562659846547e-08, + "loss": 2.2015, + "mean_token_accuracy": 0.7935357093811035, + "num_tokens": 44436.0, + "step": 15 + }, + { + "epoch": 0.008184143222506393, + "grad_norm": 179.1138458251953, + "learning_rate": 3.836317135549872e-08, + "loss": 2.4866, + "mean_token_accuracy": 0.7900997996330261, + "num_tokens": 47058.0, + "step": 16 + }, + { + "epoch": 0.008695652173913044, + "grad_norm": 198.84425354003906, + "learning_rate": 4.092071611253197e-08, + "loss": 2.803, + "mean_token_accuracy": 0.7637479305267334, + "num_tokens": 50408.0, + "step": 17 + }, + { + "epoch": 0.009207161125319692, + "grad_norm": 75.51876831054688, + "learning_rate": 4.347826086956521e-08, + "loss": 1.2034, + "mean_token_accuracy": 0.8161088228225708, + "num_tokens": 53773.0, + "step": 18 + }, + { + "epoch": 0.009718670076726343, + "grad_norm": 38.22001647949219, + "learning_rate": 4.603580562659846e-08, + "loss": 0.7701, + "mean_token_accuracy": 0.8549056053161621, + "num_tokens": 57558.0, + "step": 19 + }, + { + "epoch": 0.010230179028132993, + "grad_norm": 239.12725830078125, + "learning_rate": 4.8593350383631707e-08, + "loss": 3.0961, + "mean_token_accuracy": 0.7830601930618286, + "num_tokens": 60486.0, + "step": 20 + }, + { + "epoch": 0.010741687979539642, + "grad_norm": 85.97297668457031, + "learning_rate": 5.115089514066496e-08, + "loss": 1.4409, + "mean_token_accuracy": 0.8411673307418823, + "num_tokens": 63765.0, + "step": 21 + }, + { + "epoch": 0.011253196930946292, + "grad_norm": 144.9137420654297, + "learning_rate": 5.370843989769821e-08, + "loss": 2.1206, + "mean_token_accuracy": 0.8174169063568115, + "num_tokens": 66752.0, + "step": 22 + }, + { + "epoch": 0.011764705882352941, + "grad_norm": 207.23130798339844, + "learning_rate": 5.626598465473145e-08, + "loss": 2.6898, + "mean_token_accuracy": 0.7919356822967529, + "num_tokens": 69830.0, + "step": 23 + }, + { + "epoch": 0.012276214833759591, + "grad_norm": 187.57318115234375, + "learning_rate": 5.88235294117647e-08, + "loss": 2.5319, + "mean_token_accuracy": 0.7950634956359863, + "num_tokens": 72996.0, + "step": 24 + }, + { + "epoch": 0.01278772378516624, + "grad_norm": 71.59149932861328, + "learning_rate": 6.138107416879796e-08, + "loss": 1.0823, + "mean_token_accuracy": 0.8276670575141907, + "num_tokens": 76344.0, + "step": 25 + }, + { + "epoch": 0.01329923273657289, + "grad_norm": 90.45720672607422, + "learning_rate": 6.393861892583121e-08, + "loss": 1.3127, + "mean_token_accuracy": 0.80290287733078, + "num_tokens": 79200.0, + "step": 26 + }, + { + "epoch": 0.01381074168797954, + "grad_norm": 161.12864685058594, + "learning_rate": 6.649616368286445e-08, + "loss": 2.2243, + "mean_token_accuracy": 0.7905029058456421, + "num_tokens": 82304.0, + "step": 27 + }, + { + "epoch": 0.01432225063938619, + "grad_norm": 118.32920837402344, + "learning_rate": 6.905370843989768e-08, + "loss": 1.6398, + "mean_token_accuracy": 0.7871684432029724, + "num_tokens": 85478.0, + "step": 28 + }, + { + "epoch": 0.014833759590792838, + "grad_norm": 104.62891387939453, + "learning_rate": 7.161125319693093e-08, + "loss": 1.5101, + "mean_token_accuracy": 0.8198884129524231, + "num_tokens": 88804.0, + "step": 29 + }, + { + "epoch": 0.015345268542199489, + "grad_norm": 85.83583068847656, + "learning_rate": 7.416879795396419e-08, + "loss": 1.222, + "mean_token_accuracy": 0.8161230683326721, + "num_tokens": 91877.0, + "step": 30 + }, + { + "epoch": 0.015856777493606138, + "grad_norm": 90.76548767089844, + "learning_rate": 7.672634271099744e-08, + "loss": 1.2639, + "mean_token_accuracy": 0.82970130443573, + "num_tokens": 95274.0, + "step": 31 + }, + { + "epoch": 0.016368286445012786, + "grad_norm": 75.00279998779297, + "learning_rate": 7.928388746803069e-08, + "loss": 1.0885, + "mean_token_accuracy": 0.8382601737976074, + "num_tokens": 98659.0, + "step": 32 + }, + { + "epoch": 0.01687979539641944, + "grad_norm": 174.1860809326172, + "learning_rate": 8.184143222506394e-08, + "loss": 2.2152, + "mean_token_accuracy": 0.7910646200180054, + "num_tokens": 101369.0, + "step": 33 + }, + { + "epoch": 0.017391304347826087, + "grad_norm": 102.26188659667969, + "learning_rate": 8.439897698209717e-08, + "loss": 1.3898, + "mean_token_accuracy": 0.7661615014076233, + "num_tokens": 104944.0, + "step": 34 + }, + { + "epoch": 0.017902813299232736, + "grad_norm": 61.695377349853516, + "learning_rate": 8.695652173913042e-08, + "loss": 0.8288, + "mean_token_accuracy": 0.8378233909606934, + "num_tokens": 108615.0, + "step": 35 + }, + { + "epoch": 0.018414322250639385, + "grad_norm": 193.62203979492188, + "learning_rate": 8.951406649616368e-08, + "loss": 2.0522, + "mean_token_accuracy": 0.8460657596588135, + "num_tokens": 111551.0, + "step": 36 + }, + { + "epoch": 0.018925831202046037, + "grad_norm": 213.60400390625, + "learning_rate": 9.207161125319693e-08, + "loss": 2.2981, + "mean_token_accuracy": 0.7845360040664673, + "num_tokens": 113995.0, + "step": 37 + }, + { + "epoch": 0.019437340153452685, + "grad_norm": 210.23880004882812, + "learning_rate": 9.462915601023018e-08, + "loss": 2.3782, + "mean_token_accuracy": 0.768383264541626, + "num_tokens": 116912.0, + "step": 38 + }, + { + "epoch": 0.019948849104859334, + "grad_norm": 107.99103546142578, + "learning_rate": 9.718670076726341e-08, + "loss": 1.3588, + "mean_token_accuracy": 0.8315550088882446, + "num_tokens": 120076.0, + "step": 39 + }, + { + "epoch": 0.020460358056265986, + "grad_norm": 116.34110260009766, + "learning_rate": 9.974424552429666e-08, + "loss": 1.3651, + "mean_token_accuracy": 0.8258199691772461, + "num_tokens": 123640.0, + "step": 40 + }, + { + "epoch": 0.020971867007672635, + "grad_norm": 53.283172607421875, + "learning_rate": 1.0230179028132991e-07, + "loss": 0.6687, + "mean_token_accuracy": 0.8232872486114502, + "num_tokens": 127598.0, + "step": 41 + }, + { + "epoch": 0.021483375959079284, + "grad_norm": 146.0146026611328, + "learning_rate": 1.0485933503836316e-07, + "loss": 1.1581, + "mean_token_accuracy": 0.7739042043685913, + "num_tokens": 130700.0, + "step": 42 + }, + { + "epoch": 0.021994884910485932, + "grad_norm": 197.12623596191406, + "learning_rate": 1.0741687979539642e-07, + "loss": 1.4347, + "mean_token_accuracy": 0.8275335431098938, + "num_tokens": 134031.0, + "step": 43 + }, + { + "epoch": 0.022506393861892585, + "grad_norm": 257.8228759765625, + "learning_rate": 1.0997442455242967e-07, + "loss": 1.6859, + "mean_token_accuracy": 0.7454214692115784, + "num_tokens": 137720.0, + "step": 44 + }, + { + "epoch": 0.023017902813299233, + "grad_norm": 220.6326446533203, + "learning_rate": 1.125319693094629e-07, + "loss": 1.3573, + "mean_token_accuracy": 0.8486558794975281, + "num_tokens": 141038.0, + "step": 45 + }, + { + "epoch": 0.023529411764705882, + "grad_norm": 200.6830291748047, + "learning_rate": 1.1508951406649615e-07, + "loss": 1.2739, + "mean_token_accuracy": 0.8022428750991821, + "num_tokens": 144841.0, + "step": 46 + }, + { + "epoch": 0.02404092071611253, + "grad_norm": 54.78773498535156, + "learning_rate": 1.176470588235294e-07, + "loss": 0.6242, + "mean_token_accuracy": 0.8289681673049927, + "num_tokens": 148810.0, + "step": 47 + }, + { + "epoch": 0.024552429667519183, + "grad_norm": 113.60277557373047, + "learning_rate": 1.2020460358056265e-07, + "loss": 0.8507, + "mean_token_accuracy": 0.804018497467041, + "num_tokens": 151734.0, + "step": 48 + }, + { + "epoch": 0.02506393861892583, + "grad_norm": 127.46772766113281, + "learning_rate": 1.2276214833759592e-07, + "loss": 0.8609, + "mean_token_accuracy": 0.7839692234992981, + "num_tokens": 154349.0, + "step": 49 + }, + { + "epoch": 0.02557544757033248, + "grad_norm": 133.61952209472656, + "learning_rate": 1.2531969309462913e-07, + "loss": 1.1099, + "mean_token_accuracy": 0.763914942741394, + "num_tokens": 156874.0, + "step": 50 + }, + { + "epoch": 0.02608695652173913, + "grad_norm": 96.55607604980469, + "learning_rate": 1.2787723785166242e-07, + "loss": 0.7179, + "mean_token_accuracy": 0.8386679887771606, + "num_tokens": 159826.0, + "step": 51 + }, + { + "epoch": 0.02659846547314578, + "grad_norm": 109.2497329711914, + "learning_rate": 1.3043478260869563e-07, + "loss": 0.8025, + "mean_token_accuracy": 0.8415220975875854, + "num_tokens": 162874.0, + "step": 52 + }, + { + "epoch": 0.02710997442455243, + "grad_norm": 92.47949981689453, + "learning_rate": 1.329923273657289e-07, + "loss": 0.61, + "mean_token_accuracy": 0.8539420366287231, + "num_tokens": 165572.0, + "step": 53 + }, + { + "epoch": 0.02762148337595908, + "grad_norm": 82.9727783203125, + "learning_rate": 1.3554987212276213e-07, + "loss": 0.7146, + "mean_token_accuracy": 0.8121727705001831, + "num_tokens": 168259.0, + "step": 54 + }, + { + "epoch": 0.028132992327365727, + "grad_norm": 57.08842468261719, + "learning_rate": 1.3810741687979537e-07, + "loss": 0.6255, + "mean_token_accuracy": 0.8097832202911377, + "num_tokens": 171327.0, + "step": 55 + }, + { + "epoch": 0.02864450127877238, + "grad_norm": 59.159141540527344, + "learning_rate": 1.4066496163682863e-07, + "loss": 0.7626, + "mean_token_accuracy": 0.7698606252670288, + "num_tokens": 173918.0, + "step": 56 + }, + { + "epoch": 0.029156010230179028, + "grad_norm": 36.09717559814453, + "learning_rate": 1.4322250639386187e-07, + "loss": 0.4397, + "mean_token_accuracy": 0.8506565093994141, + "num_tokens": 176699.0, + "step": 57 + }, + { + "epoch": 0.029667519181585677, + "grad_norm": 19.960472106933594, + "learning_rate": 1.4578005115089513e-07, + "loss": 0.4493, + "mean_token_accuracy": 0.82008957862854, + "num_tokens": 180516.0, + "step": 58 + }, + { + "epoch": 0.03017902813299233, + "grad_norm": 14.012887954711914, + "learning_rate": 1.4833759590792837e-07, + "loss": 0.3231, + "mean_token_accuracy": 0.8378205895423889, + "num_tokens": 184095.0, + "step": 59 + }, + { + "epoch": 0.030690537084398978, + "grad_norm": 54.16360092163086, + "learning_rate": 1.508951406649616e-07, + "loss": 0.6894, + "mean_token_accuracy": 0.7745775580406189, + "num_tokens": 186894.0, + "step": 60 + }, + { + "epoch": 0.031202046035805626, + "grad_norm": 24.884572982788086, + "learning_rate": 1.5345268542199487e-07, + "loss": 0.4442, + "mean_token_accuracy": 0.8268241882324219, + "num_tokens": 191215.0, + "step": 61 + }, + { + "epoch": 0.031713554987212275, + "grad_norm": 35.67756652832031, + "learning_rate": 1.560102301790281e-07, + "loss": 0.5132, + "mean_token_accuracy": 0.810157299041748, + "num_tokens": 193951.0, + "step": 62 + }, + { + "epoch": 0.032225063938618924, + "grad_norm": 25.885570526123047, + "learning_rate": 1.5856777493606137e-07, + "loss": 0.3944, + "mean_token_accuracy": 0.8766423463821411, + "num_tokens": 197234.0, + "step": 63 + }, + { + "epoch": 0.03273657289002557, + "grad_norm": 14.613554954528809, + "learning_rate": 1.611253196930946e-07, + "loss": 0.3161, + "mean_token_accuracy": 0.8670029640197754, + "num_tokens": 201651.0, + "step": 64 + }, + { + "epoch": 0.03324808184143223, + "grad_norm": 22.87023162841797, + "learning_rate": 1.6368286445012787e-07, + "loss": 0.4132, + "mean_token_accuracy": 0.8424674868583679, + "num_tokens": 204513.0, + "step": 65 + }, + { + "epoch": 0.03375959079283888, + "grad_norm": 39.33483123779297, + "learning_rate": 1.662404092071611e-07, + "loss": 0.5283, + "mean_token_accuracy": 0.8234544992446899, + "num_tokens": 207765.0, + "step": 66 + }, + { + "epoch": 0.034271099744245526, + "grad_norm": 34.351993560791016, + "learning_rate": 1.6879795396419435e-07, + "loss": 0.4652, + "mean_token_accuracy": 0.8392084836959839, + "num_tokens": 210265.0, + "step": 67 + }, + { + "epoch": 0.034782608695652174, + "grad_norm": 13.335749626159668, + "learning_rate": 1.713554987212276e-07, + "loss": 0.3255, + "mean_token_accuracy": 0.8738463521003723, + "num_tokens": 214485.0, + "step": 68 + }, + { + "epoch": 0.03529411764705882, + "grad_norm": 29.632858276367188, + "learning_rate": 1.7391304347826085e-07, + "loss": 0.4615, + "mean_token_accuracy": 0.8507890701293945, + "num_tokens": 217367.0, + "step": 69 + }, + { + "epoch": 0.03580562659846547, + "grad_norm": 6.474399566650391, + "learning_rate": 1.764705882352941e-07, + "loss": 0.1064, + "mean_token_accuracy": 0.8944461941719055, + "num_tokens": 222051.0, + "step": 70 + }, + { + "epoch": 0.03631713554987212, + "grad_norm": 20.859630584716797, + "learning_rate": 1.7902813299232735e-07, + "loss": 0.3877, + "mean_token_accuracy": 0.834736704826355, + "num_tokens": 225492.0, + "step": 71 + }, + { + "epoch": 0.03682864450127877, + "grad_norm": 9.837139129638672, + "learning_rate": 1.815856777493606e-07, + "loss": 0.2869, + "mean_token_accuracy": 0.9089845418930054, + "num_tokens": 230461.0, + "step": 72 + }, + { + "epoch": 0.037340153452685425, + "grad_norm": 31.206445693969727, + "learning_rate": 1.8414322250639385e-07, + "loss": 0.3678, + "mean_token_accuracy": 0.8622485399246216, + "num_tokens": 233104.0, + "step": 73 + }, + { + "epoch": 0.037851662404092073, + "grad_norm": 11.22103214263916, + "learning_rate": 1.867007672634271e-07, + "loss": 0.3885, + "mean_token_accuracy": 0.8613430857658386, + "num_tokens": 238509.0, + "step": 74 + }, + { + "epoch": 0.03836317135549872, + "grad_norm": 10.894756317138672, + "learning_rate": 1.8925831202046035e-07, + "loss": 0.2176, + "mean_token_accuracy": 0.9081374406814575, + "num_tokens": 243231.0, + "step": 75 + }, + { + "epoch": 0.03887468030690537, + "grad_norm": 29.264318466186523, + "learning_rate": 1.918158567774936e-07, + "loss": 0.4445, + "mean_token_accuracy": 0.8471688628196716, + "num_tokens": 245727.0, + "step": 76 + }, + { + "epoch": 0.03938618925831202, + "grad_norm": 32.82176208496094, + "learning_rate": 1.9437340153452683e-07, + "loss": 0.4695, + "mean_token_accuracy": 0.8451350927352905, + "num_tokens": 248981.0, + "step": 77 + }, + { + "epoch": 0.03989769820971867, + "grad_norm": 38.709495544433594, + "learning_rate": 1.969309462915601e-07, + "loss": 0.5413, + "mean_token_accuracy": 0.824944257736206, + "num_tokens": 252431.0, + "step": 78 + }, + { + "epoch": 0.04040920716112532, + "grad_norm": 27.581270217895508, + "learning_rate": 1.9948849104859333e-07, + "loss": 0.4184, + "mean_token_accuracy": 0.8414168357849121, + "num_tokens": 255447.0, + "step": 79 + }, + { + "epoch": 0.04092071611253197, + "grad_norm": 15.641094207763672, + "learning_rate": 2.020460358056266e-07, + "loss": 0.1611, + "mean_token_accuracy": 0.9130299091339111, + "num_tokens": 259419.0, + "step": 80 + }, + { + "epoch": 0.04143222506393862, + "grad_norm": 16.915897369384766, + "learning_rate": 2.0460358056265983e-07, + "loss": 0.3851, + "mean_token_accuracy": 0.8672155141830444, + "num_tokens": 262986.0, + "step": 81 + }, + { + "epoch": 0.04194373401534527, + "grad_norm": 23.797605514526367, + "learning_rate": 2.071611253196931e-07, + "loss": 0.3275, + "mean_token_accuracy": 0.9015141725540161, + "num_tokens": 266265.0, + "step": 82 + }, + { + "epoch": 0.04245524296675192, + "grad_norm": 21.605716705322266, + "learning_rate": 2.0971867007672633e-07, + "loss": 0.3852, + "mean_token_accuracy": 0.8909030556678772, + "num_tokens": 269049.0, + "step": 83 + }, + { + "epoch": 0.04296675191815857, + "grad_norm": 29.510509490966797, + "learning_rate": 2.1227621483375957e-07, + "loss": 0.461, + "mean_token_accuracy": 0.8185452818870544, + "num_tokens": 272479.0, + "step": 84 + }, + { + "epoch": 0.043478260869565216, + "grad_norm": 15.19318962097168, + "learning_rate": 2.1483375959079283e-07, + "loss": 0.2377, + "mean_token_accuracy": 0.9147399663925171, + "num_tokens": 275080.0, + "step": 85 + }, + { + "epoch": 0.043989769820971865, + "grad_norm": 17.34721565246582, + "learning_rate": 2.1739130434782607e-07, + "loss": 0.2867, + "mean_token_accuracy": 0.8800281286239624, + "num_tokens": 278674.0, + "step": 86 + }, + { + "epoch": 0.044501278772378514, + "grad_norm": 20.198760986328125, + "learning_rate": 2.1994884910485933e-07, + "loss": 0.2938, + "mean_token_accuracy": 0.8668062686920166, + "num_tokens": 282269.0, + "step": 87 + }, + { + "epoch": 0.04501278772378517, + "grad_norm": 29.35466194152832, + "learning_rate": 2.2250639386189257e-07, + "loss": 0.3016, + "mean_token_accuracy": 0.8933764100074768, + "num_tokens": 284758.0, + "step": 88 + }, + { + "epoch": 0.04552429667519182, + "grad_norm": 25.371225357055664, + "learning_rate": 2.250639386189258e-07, + "loss": 0.2975, + "mean_token_accuracy": 0.8665649890899658, + "num_tokens": 288420.0, + "step": 89 + }, + { + "epoch": 0.04603580562659847, + "grad_norm": 42.065406799316406, + "learning_rate": 2.2762148337595907e-07, + "loss": 0.4182, + "mean_token_accuracy": 0.8326417207717896, + "num_tokens": 290972.0, + "step": 90 + }, + { + "epoch": 0.046547314578005115, + "grad_norm": 20.71299934387207, + "learning_rate": 2.301790281329923e-07, + "loss": 0.3144, + "mean_token_accuracy": 0.8859845399856567, + "num_tokens": 293959.0, + "step": 91 + }, + { + "epoch": 0.047058823529411764, + "grad_norm": 17.148942947387695, + "learning_rate": 2.3273657289002557e-07, + "loss": 0.2859, + "mean_token_accuracy": 0.9123847484588623, + "num_tokens": 296999.0, + "step": 92 + }, + { + "epoch": 0.04757033248081841, + "grad_norm": 13.404889106750488, + "learning_rate": 2.352941176470588e-07, + "loss": 0.3883, + "mean_token_accuracy": 0.8552314043045044, + "num_tokens": 300355.0, + "step": 93 + }, + { + "epoch": 0.04808184143222506, + "grad_norm": 20.698440551757812, + "learning_rate": 2.3785166240409207e-07, + "loss": 0.2778, + "mean_token_accuracy": 0.8984112739562988, + "num_tokens": 303259.0, + "step": 94 + }, + { + "epoch": 0.04859335038363171, + "grad_norm": 27.2398681640625, + "learning_rate": 2.404092071611253e-07, + "loss": 0.2763, + "mean_token_accuracy": 0.9022860527038574, + "num_tokens": 306024.0, + "step": 95 + }, + { + "epoch": 0.049104859335038366, + "grad_norm": 22.13515281677246, + "learning_rate": 2.4296675191815855e-07, + "loss": 0.278, + "mean_token_accuracy": 0.9062796235084534, + "num_tokens": 308518.0, + "step": 96 + }, + { + "epoch": 0.049616368286445015, + "grad_norm": 20.62944793701172, + "learning_rate": 2.4552429667519184e-07, + "loss": 0.4701, + "mean_token_accuracy": 0.8488901853561401, + "num_tokens": 311595.0, + "step": 97 + }, + { + "epoch": 0.05012787723785166, + "grad_norm": 20.998424530029297, + "learning_rate": 2.48081841432225e-07, + "loss": 0.3137, + "mean_token_accuracy": 0.8738723993301392, + "num_tokens": 314934.0, + "step": 98 + }, + { + "epoch": 0.05063938618925831, + "grad_norm": 15.533212661743164, + "learning_rate": 2.5063938618925826e-07, + "loss": 0.2746, + "mean_token_accuracy": 0.8795336484909058, + "num_tokens": 318236.0, + "step": 99 + }, + { + "epoch": 0.05115089514066496, + "grad_norm": 27.825117111206055, + "learning_rate": 2.531969309462916e-07, + "loss": 0.4071, + "mean_token_accuracy": 0.8719491362571716, + "num_tokens": 321122.0, + "step": 100 + }, + { + "epoch": 0.05166240409207161, + "grad_norm": 18.763154983520508, + "learning_rate": 2.5575447570332484e-07, + "loss": 0.4031, + "mean_token_accuracy": 0.8682801723480225, + "num_tokens": 324787.0, + "step": 101 + }, + { + "epoch": 0.05217391304347826, + "grad_norm": 22.935169219970703, + "learning_rate": 2.58312020460358e-07, + "loss": 0.4241, + "mean_token_accuracy": 0.843114972114563, + "num_tokens": 327808.0, + "step": 102 + }, + { + "epoch": 0.052685421994884914, + "grad_norm": 17.16703987121582, + "learning_rate": 2.6086956521739126e-07, + "loss": 0.3693, + "mean_token_accuracy": 0.8559282422065735, + "num_tokens": 331219.0, + "step": 103 + }, + { + "epoch": 0.05319693094629156, + "grad_norm": 18.996002197265625, + "learning_rate": 2.634271099744245e-07, + "loss": 0.2799, + "mean_token_accuracy": 0.8920080661773682, + "num_tokens": 334076.0, + "step": 104 + }, + { + "epoch": 0.05370843989769821, + "grad_norm": 23.53285789489746, + "learning_rate": 2.659846547314578e-07, + "loss": 0.3198, + "mean_token_accuracy": 0.8700375556945801, + "num_tokens": 337037.0, + "step": 105 + }, + { + "epoch": 0.05421994884910486, + "grad_norm": 24.903053283691406, + "learning_rate": 2.68542199488491e-07, + "loss": 0.3947, + "mean_token_accuracy": 0.8315317630767822, + "num_tokens": 340498.0, + "step": 106 + }, + { + "epoch": 0.05473145780051151, + "grad_norm": 27.963050842285156, + "learning_rate": 2.7109974424552426e-07, + "loss": 0.3842, + "mean_token_accuracy": 0.8637053966522217, + "num_tokens": 343153.0, + "step": 107 + }, + { + "epoch": 0.05524296675191816, + "grad_norm": 18.79988670349121, + "learning_rate": 2.736572890025575e-07, + "loss": 0.2263, + "mean_token_accuracy": 0.9064067602157593, + "num_tokens": 346312.0, + "step": 108 + }, + { + "epoch": 0.055754475703324806, + "grad_norm": 10.160648345947266, + "learning_rate": 2.7621483375959074e-07, + "loss": 0.2318, + "mean_token_accuracy": 0.8982808589935303, + "num_tokens": 350229.0, + "step": 109 + }, + { + "epoch": 0.056265984654731455, + "grad_norm": 12.265483856201172, + "learning_rate": 2.7877237851662403e-07, + "loss": 0.4316, + "mean_token_accuracy": 0.8697227239608765, + "num_tokens": 353731.0, + "step": 110 + }, + { + "epoch": 0.05677749360613811, + "grad_norm": 18.477947235107422, + "learning_rate": 2.8132992327365726e-07, + "loss": 0.3292, + "mean_token_accuracy": 0.8818848729133606, + "num_tokens": 356287.0, + "step": 111 + }, + { + "epoch": 0.05728900255754476, + "grad_norm": 24.21441650390625, + "learning_rate": 2.838874680306905e-07, + "loss": 0.3576, + "mean_token_accuracy": 0.8753139972686768, + "num_tokens": 358905.0, + "step": 112 + }, + { + "epoch": 0.05780051150895141, + "grad_norm": 11.783536911010742, + "learning_rate": 2.8644501278772374e-07, + "loss": 0.3047, + "mean_token_accuracy": 0.8412572741508484, + "num_tokens": 362488.0, + "step": 113 + }, + { + "epoch": 0.058312020460358056, + "grad_norm": 7.969364643096924, + "learning_rate": 2.89002557544757e-07, + "loss": 0.19, + "mean_token_accuracy": 0.9087700247764587, + "num_tokens": 366370.0, + "step": 114 + }, + { + "epoch": 0.058823529411764705, + "grad_norm": 25.384963989257812, + "learning_rate": 2.9156010230179027e-07, + "loss": 0.3159, + "mean_token_accuracy": 0.890160322189331, + "num_tokens": 369931.0, + "step": 115 + }, + { + "epoch": 0.059335038363171354, + "grad_norm": 17.172136306762695, + "learning_rate": 2.941176470588235e-07, + "loss": 0.2646, + "mean_token_accuracy": 0.9070903062820435, + "num_tokens": 372793.0, + "step": 116 + }, + { + "epoch": 0.059846547314578, + "grad_norm": 10.245344161987305, + "learning_rate": 2.9667519181585674e-07, + "loss": 0.2944, + "mean_token_accuracy": 0.892909049987793, + "num_tokens": 377647.0, + "step": 117 + }, + { + "epoch": 0.06035805626598466, + "grad_norm": 28.328933715820312, + "learning_rate": 2.9923273657289e-07, + "loss": 0.4005, + "mean_token_accuracy": 0.8677277565002441, + "num_tokens": 380201.0, + "step": 118 + }, + { + "epoch": 0.06086956521739131, + "grad_norm": 15.422980308532715, + "learning_rate": 3.017902813299232e-07, + "loss": 0.3552, + "mean_token_accuracy": 0.8636158108711243, + "num_tokens": 383716.0, + "step": 119 + }, + { + "epoch": 0.061381074168797956, + "grad_norm": 29.055530548095703, + "learning_rate": 3.043478260869565e-07, + "loss": 0.3237, + "mean_token_accuracy": 0.877373993396759, + "num_tokens": 387421.0, + "step": 120 + }, + { + "epoch": 0.061892583120204604, + "grad_norm": 23.368236541748047, + "learning_rate": 3.0690537084398974e-07, + "loss": 0.2469, + "mean_token_accuracy": 0.915671706199646, + "num_tokens": 390114.0, + "step": 121 + }, + { + "epoch": 0.06240409207161125, + "grad_norm": 18.77473258972168, + "learning_rate": 3.09462915601023e-07, + "loss": 0.2407, + "mean_token_accuracy": 0.8971856832504272, + "num_tokens": 392631.0, + "step": 122 + }, + { + "epoch": 0.06291560102301791, + "grad_norm": 22.172239303588867, + "learning_rate": 3.120204603580562e-07, + "loss": 0.3878, + "mean_token_accuracy": 0.8388662934303284, + "num_tokens": 395859.0, + "step": 123 + }, + { + "epoch": 0.06342710997442455, + "grad_norm": 24.349529266357422, + "learning_rate": 3.1457800511508945e-07, + "loss": 0.3578, + "mean_token_accuracy": 0.8794122338294983, + "num_tokens": 398951.0, + "step": 124 + }, + { + "epoch": 0.0639386189258312, + "grad_norm": 10.460365295410156, + "learning_rate": 3.1713554987212274e-07, + "loss": 0.1832, + "mean_token_accuracy": 0.9220331907272339, + "num_tokens": 402235.0, + "step": 125 + }, + { + "epoch": 0.06445012787723785, + "grad_norm": 23.418947219848633, + "learning_rate": 3.19693094629156e-07, + "loss": 0.3947, + "mean_token_accuracy": 0.8443590402603149, + "num_tokens": 404773.0, + "step": 126 + }, + { + "epoch": 0.0649616368286445, + "grad_norm": 14.503310203552246, + "learning_rate": 3.222506393861892e-07, + "loss": 0.3694, + "mean_token_accuracy": 0.887751579284668, + "num_tokens": 408060.0, + "step": 127 + }, + { + "epoch": 0.06547314578005115, + "grad_norm": 24.164749145507812, + "learning_rate": 3.2480818414322246e-07, + "loss": 0.3077, + "mean_token_accuracy": 0.8854978680610657, + "num_tokens": 411168.0, + "step": 128 + }, + { + "epoch": 0.0659846547314578, + "grad_norm": 14.65926742553711, + "learning_rate": 3.2736572890025575e-07, + "loss": 0.2095, + "mean_token_accuracy": 0.9001630544662476, + "num_tokens": 414080.0, + "step": 129 + }, + { + "epoch": 0.06649616368286446, + "grad_norm": 27.39354705810547, + "learning_rate": 3.29923273657289e-07, + "loss": 0.2808, + "mean_token_accuracy": 0.8919434547424316, + "num_tokens": 416709.0, + "step": 130 + }, + { + "epoch": 0.0670076726342711, + "grad_norm": 27.415010452270508, + "learning_rate": 3.324808184143222e-07, + "loss": 0.3459, + "mean_token_accuracy": 0.8762794137001038, + "num_tokens": 419502.0, + "step": 131 + }, + { + "epoch": 0.06751918158567775, + "grad_norm": 31.00101089477539, + "learning_rate": 3.3503836317135546e-07, + "loss": 0.4114, + "mean_token_accuracy": 0.851691722869873, + "num_tokens": 422637.0, + "step": 132 + }, + { + "epoch": 0.0680306905370844, + "grad_norm": 40.224571228027344, + "learning_rate": 3.375959079283887e-07, + "loss": 0.429, + "mean_token_accuracy": 0.8526999354362488, + "num_tokens": 425273.0, + "step": 133 + }, + { + "epoch": 0.06854219948849105, + "grad_norm": 18.503984451293945, + "learning_rate": 3.40153452685422e-07, + "loss": 0.3066, + "mean_token_accuracy": 0.8810898065567017, + "num_tokens": 428080.0, + "step": 134 + }, + { + "epoch": 0.06905370843989769, + "grad_norm": 28.158891677856445, + "learning_rate": 3.427109974424552e-07, + "loss": 0.3711, + "mean_token_accuracy": 0.8567726016044617, + "num_tokens": 430891.0, + "step": 135 + }, + { + "epoch": 0.06956521739130435, + "grad_norm": 12.90997314453125, + "learning_rate": 3.4526854219948846e-07, + "loss": 0.2611, + "mean_token_accuracy": 0.9025558233261108, + "num_tokens": 433971.0, + "step": 136 + }, + { + "epoch": 0.070076726342711, + "grad_norm": 32.700984954833984, + "learning_rate": 3.478260869565217e-07, + "loss": 0.3428, + "mean_token_accuracy": 0.8640915155410767, + "num_tokens": 437515.0, + "step": 137 + }, + { + "epoch": 0.07058823529411765, + "grad_norm": 21.443288803100586, + "learning_rate": 3.5038363171355494e-07, + "loss": 0.3546, + "mean_token_accuracy": 0.8884392976760864, + "num_tokens": 440379.0, + "step": 138 + }, + { + "epoch": 0.0710997442455243, + "grad_norm": 14.941019058227539, + "learning_rate": 3.529411764705882e-07, + "loss": 0.3342, + "mean_token_accuracy": 0.8596026301383972, + "num_tokens": 443645.0, + "step": 139 + }, + { + "epoch": 0.07161125319693094, + "grad_norm": 22.549388885498047, + "learning_rate": 3.5549872122762146e-07, + "loss": 0.2918, + "mean_token_accuracy": 0.8888661861419678, + "num_tokens": 446554.0, + "step": 140 + }, + { + "epoch": 0.0721227621483376, + "grad_norm": 22.392559051513672, + "learning_rate": 3.580562659846547e-07, + "loss": 0.31, + "mean_token_accuracy": 0.871440052986145, + "num_tokens": 449447.0, + "step": 141 + }, + { + "epoch": 0.07263427109974424, + "grad_norm": 25.39676284790039, + "learning_rate": 3.6061381074168794e-07, + "loss": 0.3417, + "mean_token_accuracy": 0.8671771287918091, + "num_tokens": 452218.0, + "step": 142 + }, + { + "epoch": 0.0731457800511509, + "grad_norm": 16.479408264160156, + "learning_rate": 3.631713554987212e-07, + "loss": 0.3882, + "mean_token_accuracy": 0.8517661690711975, + "num_tokens": 455101.0, + "step": 143 + }, + { + "epoch": 0.07365728900255754, + "grad_norm": 11.995569229125977, + "learning_rate": 3.6572890025575446e-07, + "loss": 0.2058, + "mean_token_accuracy": 0.8966461420059204, + "num_tokens": 458694.0, + "step": 144 + }, + { + "epoch": 0.0741687979539642, + "grad_norm": 11.710745811462402, + "learning_rate": 3.682864450127877e-07, + "loss": 0.1657, + "mean_token_accuracy": 0.9056930541992188, + "num_tokens": 461264.0, + "step": 145 + }, + { + "epoch": 0.07468030690537085, + "grad_norm": 15.025140762329102, + "learning_rate": 3.7084398976982094e-07, + "loss": 0.2735, + "mean_token_accuracy": 0.8678882122039795, + "num_tokens": 464337.0, + "step": 146 + }, + { + "epoch": 0.07519181585677749, + "grad_norm": 17.179922103881836, + "learning_rate": 3.734015345268542e-07, + "loss": 0.4533, + "mean_token_accuracy": 0.8483802080154419, + "num_tokens": 467684.0, + "step": 147 + }, + { + "epoch": 0.07570332480818415, + "grad_norm": 24.835493087768555, + "learning_rate": 3.759590792838874e-07, + "loss": 0.4429, + "mean_token_accuracy": 0.84512859582901, + "num_tokens": 470653.0, + "step": 148 + }, + { + "epoch": 0.07621483375959079, + "grad_norm": 23.85975456237793, + "learning_rate": 3.785166240409207e-07, + "loss": 0.3868, + "mean_token_accuracy": 0.8446352481842041, + "num_tokens": 473295.0, + "step": 149 + }, + { + "epoch": 0.07672634271099744, + "grad_norm": 21.695446014404297, + "learning_rate": 3.8107416879795394e-07, + "loss": 0.2861, + "mean_token_accuracy": 0.8709262013435364, + "num_tokens": 476064.0, + "step": 150 + }, + { + "epoch": 0.07723785166240409, + "grad_norm": 28.86606216430664, + "learning_rate": 3.836317135549872e-07, + "loss": 0.3882, + "mean_token_accuracy": 0.8589136600494385, + "num_tokens": 479045.0, + "step": 151 + }, + { + "epoch": 0.07774936061381074, + "grad_norm": 14.61961555480957, + "learning_rate": 3.861892583120204e-07, + "loss": 0.3263, + "mean_token_accuracy": 0.8818495869636536, + "num_tokens": 482128.0, + "step": 152 + }, + { + "epoch": 0.0782608695652174, + "grad_norm": 17.72173309326172, + "learning_rate": 3.8874680306905365e-07, + "loss": 0.3335, + "mean_token_accuracy": 0.8658561110496521, + "num_tokens": 484994.0, + "step": 153 + }, + { + "epoch": 0.07877237851662404, + "grad_norm": 14.253996849060059, + "learning_rate": 3.9130434782608694e-07, + "loss": 0.3105, + "mean_token_accuracy": 0.8617784380912781, + "num_tokens": 488584.0, + "step": 154 + }, + { + "epoch": 0.0792838874680307, + "grad_norm": 26.471559524536133, + "learning_rate": 3.938618925831202e-07, + "loss": 0.2531, + "mean_token_accuracy": 0.9083852767944336, + "num_tokens": 490906.0, + "step": 155 + }, + { + "epoch": 0.07979539641943734, + "grad_norm": 17.500732421875, + "learning_rate": 3.964194373401534e-07, + "loss": 0.3442, + "mean_token_accuracy": 0.84794020652771, + "num_tokens": 493817.0, + "step": 156 + }, + { + "epoch": 0.08030690537084399, + "grad_norm": 27.501888275146484, + "learning_rate": 3.9897698209718666e-07, + "loss": 0.3358, + "mean_token_accuracy": 0.8779575824737549, + "num_tokens": 497009.0, + "step": 157 + }, + { + "epoch": 0.08081841432225063, + "grad_norm": 17.622556686401367, + "learning_rate": 4.0153452685421995e-07, + "loss": 0.2627, + "mean_token_accuracy": 0.892830491065979, + "num_tokens": 499895.0, + "step": 158 + }, + { + "epoch": 0.08132992327365729, + "grad_norm": 22.043292999267578, + "learning_rate": 4.040920716112532e-07, + "loss": 0.4305, + "mean_token_accuracy": 0.8164557218551636, + "num_tokens": 502689.0, + "step": 159 + }, + { + "epoch": 0.08184143222506395, + "grad_norm": 17.908912658691406, + "learning_rate": 4.066496163682864e-07, + "loss": 0.301, + "mean_token_accuracy": 0.8504549264907837, + "num_tokens": 506050.0, + "step": 160 + }, + { + "epoch": 0.08235294117647059, + "grad_norm": 23.02467918395996, + "learning_rate": 4.0920716112531966e-07, + "loss": 0.338, + "mean_token_accuracy": 0.8616349697113037, + "num_tokens": 508483.0, + "step": 161 + }, + { + "epoch": 0.08286445012787724, + "grad_norm": 20.75882911682129, + "learning_rate": 4.117647058823529e-07, + "loss": 0.2847, + "mean_token_accuracy": 0.8853994607925415, + "num_tokens": 510862.0, + "step": 162 + }, + { + "epoch": 0.08337595907928388, + "grad_norm": 17.08759117126465, + "learning_rate": 4.143222506393862e-07, + "loss": 0.2794, + "mean_token_accuracy": 0.9035633206367493, + "num_tokens": 513759.0, + "step": 163 + }, + { + "epoch": 0.08388746803069054, + "grad_norm": 11.089282035827637, + "learning_rate": 4.168797953964194e-07, + "loss": 0.2782, + "mean_token_accuracy": 0.8845057487487793, + "num_tokens": 517372.0, + "step": 164 + }, + { + "epoch": 0.08439897698209718, + "grad_norm": 15.210725784301758, + "learning_rate": 4.1943734015345266e-07, + "loss": 0.3187, + "mean_token_accuracy": 0.8725647926330566, + "num_tokens": 521186.0, + "step": 165 + }, + { + "epoch": 0.08491048593350384, + "grad_norm": 9.48412799835205, + "learning_rate": 4.219948849104859e-07, + "loss": 0.2839, + "mean_token_accuracy": 0.8890608549118042, + "num_tokens": 524824.0, + "step": 166 + }, + { + "epoch": 0.08542199488491048, + "grad_norm": 18.62152099609375, + "learning_rate": 4.2455242966751913e-07, + "loss": 0.3233, + "mean_token_accuracy": 0.8461313843727112, + "num_tokens": 529024.0, + "step": 167 + }, + { + "epoch": 0.08593350383631713, + "grad_norm": 9.747681617736816, + "learning_rate": 4.271099744245524e-07, + "loss": 0.2359, + "mean_token_accuracy": 0.9112660884857178, + "num_tokens": 533075.0, + "step": 168 + }, + { + "epoch": 0.08644501278772379, + "grad_norm": 27.892515182495117, + "learning_rate": 4.2966751918158566e-07, + "loss": 0.2925, + "mean_token_accuracy": 0.8798779845237732, + "num_tokens": 536210.0, + "step": 169 + }, + { + "epoch": 0.08695652173913043, + "grad_norm": 18.331941604614258, + "learning_rate": 4.322250639386189e-07, + "loss": 0.2778, + "mean_token_accuracy": 0.8987711071968079, + "num_tokens": 539011.0, + "step": 170 + }, + { + "epoch": 0.08746803069053709, + "grad_norm": 13.895400047302246, + "learning_rate": 4.3478260869565214e-07, + "loss": 0.3266, + "mean_token_accuracy": 0.8624075651168823, + "num_tokens": 542266.0, + "step": 171 + }, + { + "epoch": 0.08797953964194373, + "grad_norm": 21.15460205078125, + "learning_rate": 4.3734015345268537e-07, + "loss": 0.2906, + "mean_token_accuracy": 0.8881990313529968, + "num_tokens": 545290.0, + "step": 172 + }, + { + "epoch": 0.08849104859335039, + "grad_norm": 18.483774185180664, + "learning_rate": 4.3989769820971866e-07, + "loss": 0.336, + "mean_token_accuracy": 0.8840640187263489, + "num_tokens": 548126.0, + "step": 173 + }, + { + "epoch": 0.08900255754475703, + "grad_norm": 16.760847091674805, + "learning_rate": 4.424552429667519e-07, + "loss": 0.2804, + "mean_token_accuracy": 0.9025435447692871, + "num_tokens": 551183.0, + "step": 174 + }, + { + "epoch": 0.08951406649616368, + "grad_norm": 7.33767032623291, + "learning_rate": 4.4501278772378514e-07, + "loss": 0.2232, + "mean_token_accuracy": 0.918347954750061, + "num_tokens": 555610.0, + "step": 175 + }, + { + "epoch": 0.09002557544757034, + "grad_norm": 28.174320220947266, + "learning_rate": 4.475703324808184e-07, + "loss": 0.3581, + "mean_token_accuracy": 0.8751071095466614, + "num_tokens": 559686.0, + "step": 176 + }, + { + "epoch": 0.09053708439897698, + "grad_norm": 20.724994659423828, + "learning_rate": 4.501278772378516e-07, + "loss": 0.3143, + "mean_token_accuracy": 0.8852145671844482, + "num_tokens": 563016.0, + "step": 177 + }, + { + "epoch": 0.09104859335038364, + "grad_norm": 22.395971298217773, + "learning_rate": 4.526854219948849e-07, + "loss": 0.2902, + "mean_token_accuracy": 0.8888736963272095, + "num_tokens": 566355.0, + "step": 178 + }, + { + "epoch": 0.09156010230179028, + "grad_norm": 15.162639617919922, + "learning_rate": 4.5524296675191814e-07, + "loss": 0.2297, + "mean_token_accuracy": 0.9090532064437866, + "num_tokens": 569542.0, + "step": 179 + }, + { + "epoch": 0.09207161125319693, + "grad_norm": 30.4066104888916, + "learning_rate": 4.578005115089514e-07, + "loss": 0.4305, + "mean_token_accuracy": 0.8547916412353516, + "num_tokens": 572377.0, + "step": 180 + }, + { + "epoch": 0.09258312020460358, + "grad_norm": 10.455379486083984, + "learning_rate": 4.603580562659846e-07, + "loss": 0.2876, + "mean_token_accuracy": 0.877539873123169, + "num_tokens": 575929.0, + "step": 181 + }, + { + "epoch": 0.09309462915601023, + "grad_norm": 18.545860290527344, + "learning_rate": 4.6291560102301785e-07, + "loss": 0.2302, + "mean_token_accuracy": 0.8899485468864441, + "num_tokens": 578864.0, + "step": 182 + }, + { + "epoch": 0.09360613810741689, + "grad_norm": 17.00993537902832, + "learning_rate": 4.6547314578005114e-07, + "loss": 0.3878, + "mean_token_accuracy": 0.8544268608093262, + "num_tokens": 581681.0, + "step": 183 + }, + { + "epoch": 0.09411764705882353, + "grad_norm": 12.781731605529785, + "learning_rate": 4.680306905370844e-07, + "loss": 0.2897, + "mean_token_accuracy": 0.9243752360343933, + "num_tokens": 585164.0, + "step": 184 + }, + { + "epoch": 0.09462915601023018, + "grad_norm": 17.036840438842773, + "learning_rate": 4.705882352941176e-07, + "loss": 0.4011, + "mean_token_accuracy": 0.8517594337463379, + "num_tokens": 588490.0, + "step": 185 + }, + { + "epoch": 0.09514066496163683, + "grad_norm": 19.410757064819336, + "learning_rate": 4.7314578005115085e-07, + "loss": 0.2857, + "mean_token_accuracy": 0.8885026574134827, + "num_tokens": 591076.0, + "step": 186 + }, + { + "epoch": 0.09565217391304348, + "grad_norm": 9.726699829101562, + "learning_rate": 4.7570332480818414e-07, + "loss": 0.2499, + "mean_token_accuracy": 0.8528520464897156, + "num_tokens": 594940.0, + "step": 187 + }, + { + "epoch": 0.09616368286445012, + "grad_norm": 25.838125228881836, + "learning_rate": 4.782608695652174e-07, + "loss": 0.4053, + "mean_token_accuracy": 0.8495023250579834, + "num_tokens": 597574.0, + "step": 188 + }, + { + "epoch": 0.09667519181585678, + "grad_norm": 12.847740173339844, + "learning_rate": 4.808184143222506e-07, + "loss": 0.3506, + "mean_token_accuracy": 0.8257908821105957, + "num_tokens": 600403.0, + "step": 189 + }, + { + "epoch": 0.09718670076726342, + "grad_norm": 16.734188079833984, + "learning_rate": 4.833759590792839e-07, + "loss": 0.2243, + "mean_token_accuracy": 0.9069640636444092, + "num_tokens": 603327.0, + "step": 190 + }, + { + "epoch": 0.09769820971867008, + "grad_norm": 18.02349090576172, + "learning_rate": 4.859335038363171e-07, + "loss": 0.3534, + "mean_token_accuracy": 0.8543546199798584, + "num_tokens": 606878.0, + "step": 191 + }, + { + "epoch": 0.09820971867007673, + "grad_norm": 10.642441749572754, + "learning_rate": 4.884910485933504e-07, + "loss": 0.2849, + "mean_token_accuracy": 0.8836093544960022, + "num_tokens": 610748.0, + "step": 192 + }, + { + "epoch": 0.09872122762148337, + "grad_norm": 17.641307830810547, + "learning_rate": 4.910485933503837e-07, + "loss": 0.3204, + "mean_token_accuracy": 0.8876261711120605, + "num_tokens": 613541.0, + "step": 193 + }, + { + "epoch": 0.09923273657289003, + "grad_norm": 21.679807662963867, + "learning_rate": 4.936061381074169e-07, + "loss": 0.3009, + "mean_token_accuracy": 0.8832826614379883, + "num_tokens": 616620.0, + "step": 194 + }, + { + "epoch": 0.09974424552429667, + "grad_norm": 11.517809867858887, + "learning_rate": 4.9616368286445e-07, + "loss": 0.3132, + "mean_token_accuracy": 0.8778504133224487, + "num_tokens": 620171.0, + "step": 195 + }, + { + "epoch": 0.10025575447570333, + "grad_norm": 11.921733856201172, + "learning_rate": 4.987212276214833e-07, + "loss": 0.2427, + "mean_token_accuracy": 0.8600891828536987, + "num_tokens": 624116.0, + "step": 196 + }, + { + "epoch": 0.10076726342710997, + "grad_norm": 12.040887832641602, + "learning_rate": 5.012787723785165e-07, + "loss": 0.3359, + "mean_token_accuracy": 0.8685547709465027, + "num_tokens": 627813.0, + "step": 197 + }, + { + "epoch": 0.10127877237851662, + "grad_norm": 20.811750411987305, + "learning_rate": 5.038363171355499e-07, + "loss": 0.3159, + "mean_token_accuracy": 0.8991944789886475, + "num_tokens": 630953.0, + "step": 198 + }, + { + "epoch": 0.10179028132992328, + "grad_norm": 12.80086898803711, + "learning_rate": 5.063938618925832e-07, + "loss": 0.3375, + "mean_token_accuracy": 0.8701936602592468, + "num_tokens": 634521.0, + "step": 199 + }, + { + "epoch": 0.10230179028132992, + "grad_norm": 35.81144714355469, + "learning_rate": 5.089514066496163e-07, + "loss": 0.4075, + "mean_token_accuracy": 0.8243008852005005, + "num_tokens": 637231.0, + "step": 200 + }, + { + "epoch": 0.10281329923273658, + "grad_norm": 23.38886833190918, + "learning_rate": 5.115089514066497e-07, + "loss": 0.3569, + "mean_token_accuracy": 0.8546783924102783, + "num_tokens": 639623.0, + "step": 201 + }, + { + "epoch": 0.10332480818414322, + "grad_norm": 15.044546127319336, + "learning_rate": 5.140664961636828e-07, + "loss": 0.2821, + "mean_token_accuracy": 0.9011663794517517, + "num_tokens": 643211.0, + "step": 202 + }, + { + "epoch": 0.10383631713554987, + "grad_norm": 16.520963668823242, + "learning_rate": 5.16624040920716e-07, + "loss": 0.2959, + "mean_token_accuracy": 0.8943748474121094, + "num_tokens": 646152.0, + "step": 203 + }, + { + "epoch": 0.10434782608695652, + "grad_norm": 22.29756736755371, + "learning_rate": 5.191815856777494e-07, + "loss": 0.3337, + "mean_token_accuracy": 0.8535147905349731, + "num_tokens": 648728.0, + "step": 204 + }, + { + "epoch": 0.10485933503836317, + "grad_norm": 18.05092430114746, + "learning_rate": 5.217391304347825e-07, + "loss": 0.3781, + "mean_token_accuracy": 0.8511215448379517, + "num_tokens": 651326.0, + "step": 205 + }, + { + "epoch": 0.10537084398976983, + "grad_norm": 18.265335083007812, + "learning_rate": 5.242966751918158e-07, + "loss": 0.3724, + "mean_token_accuracy": 0.8561387658119202, + "num_tokens": 654981.0, + "step": 206 + }, + { + "epoch": 0.10588235294117647, + "grad_norm": 9.182594299316406, + "learning_rate": 5.26854219948849e-07, + "loss": 0.169, + "mean_token_accuracy": 0.9266980886459351, + "num_tokens": 659280.0, + "step": 207 + }, + { + "epoch": 0.10639386189258312, + "grad_norm": 12.256073951721191, + "learning_rate": 5.294117647058823e-07, + "loss": 0.3343, + "mean_token_accuracy": 0.8641704320907593, + "num_tokens": 662348.0, + "step": 208 + }, + { + "epoch": 0.10690537084398977, + "grad_norm": 12.064291000366211, + "learning_rate": 5.319693094629156e-07, + "loss": 0.296, + "mean_token_accuracy": 0.859680712223053, + "num_tokens": 665649.0, + "step": 209 + }, + { + "epoch": 0.10741687979539642, + "grad_norm": 16.997743606567383, + "learning_rate": 5.345268542199488e-07, + "loss": 0.347, + "mean_token_accuracy": 0.8498463034629822, + "num_tokens": 668265.0, + "step": 210 + }, + { + "epoch": 0.10792838874680306, + "grad_norm": 20.833202362060547, + "learning_rate": 5.37084398976982e-07, + "loss": 0.3071, + "mean_token_accuracy": 0.8801729679107666, + "num_tokens": 671274.0, + "step": 211 + }, + { + "epoch": 0.10843989769820972, + "grad_norm": 21.287704467773438, + "learning_rate": 5.396419437340152e-07, + "loss": 0.2858, + "mean_token_accuracy": 0.8806939125061035, + "num_tokens": 673979.0, + "step": 212 + }, + { + "epoch": 0.10895140664961638, + "grad_norm": 20.602304458618164, + "learning_rate": 5.421994884910485e-07, + "loss": 0.3479, + "mean_token_accuracy": 0.857731819152832, + "num_tokens": 676847.0, + "step": 213 + }, + { + "epoch": 0.10946291560102302, + "grad_norm": 8.874015808105469, + "learning_rate": 5.447570332480818e-07, + "loss": 0.2036, + "mean_token_accuracy": 0.895555853843689, + "num_tokens": 681093.0, + "step": 214 + }, + { + "epoch": 0.10997442455242967, + "grad_norm": 22.511735916137695, + "learning_rate": 5.47314578005115e-07, + "loss": 0.3196, + "mean_token_accuracy": 0.8513257503509521, + "num_tokens": 683956.0, + "step": 215 + }, + { + "epoch": 0.11048593350383631, + "grad_norm": 20.287845611572266, + "learning_rate": 5.498721227621483e-07, + "loss": 0.338, + "mean_token_accuracy": 0.8582816123962402, + "num_tokens": 686708.0, + "step": 216 + }, + { + "epoch": 0.11099744245524297, + "grad_norm": 12.997525215148926, + "learning_rate": 5.524296675191815e-07, + "loss": 0.3016, + "mean_token_accuracy": 0.8804595470428467, + "num_tokens": 690554.0, + "step": 217 + }, + { + "epoch": 0.11150895140664961, + "grad_norm": 16.682626724243164, + "learning_rate": 5.549872122762148e-07, + "loss": 0.3441, + "mean_token_accuracy": 0.8873519897460938, + "num_tokens": 693381.0, + "step": 218 + }, + { + "epoch": 0.11202046035805627, + "grad_norm": 18.126689910888672, + "learning_rate": 5.575447570332481e-07, + "loss": 0.2697, + "mean_token_accuracy": 0.903224527835846, + "num_tokens": 696989.0, + "step": 219 + }, + { + "epoch": 0.11253196930946291, + "grad_norm": 16.248125076293945, + "learning_rate": 5.601023017902812e-07, + "loss": 0.3015, + "mean_token_accuracy": 0.883755087852478, + "num_tokens": 699966.0, + "step": 220 + }, + { + "epoch": 0.11304347826086956, + "grad_norm": 8.98157024383545, + "learning_rate": 5.626598465473145e-07, + "loss": 0.2006, + "mean_token_accuracy": 0.9146337509155273, + "num_tokens": 703841.0, + "step": 221 + }, + { + "epoch": 0.11355498721227622, + "grad_norm": 25.149261474609375, + "learning_rate": 5.652173913043477e-07, + "loss": 0.317, + "mean_token_accuracy": 0.8807467222213745, + "num_tokens": 706797.0, + "step": 222 + }, + { + "epoch": 0.11406649616368286, + "grad_norm": 18.816043853759766, + "learning_rate": 5.67774936061381e-07, + "loss": 0.2091, + "mean_token_accuracy": 0.9089944362640381, + "num_tokens": 709821.0, + "step": 223 + }, + { + "epoch": 0.11457800511508952, + "grad_norm": 20.906835556030273, + "learning_rate": 5.703324808184143e-07, + "loss": 0.325, + "mean_token_accuracy": 0.8844343423843384, + "num_tokens": 712326.0, + "step": 224 + }, + { + "epoch": 0.11508951406649616, + "grad_norm": 14.024846076965332, + "learning_rate": 5.728900255754475e-07, + "loss": 0.2443, + "mean_token_accuracy": 0.8806254863739014, + "num_tokens": 715462.0, + "step": 225 + }, + { + "epoch": 0.11560102301790282, + "grad_norm": 27.239303588867188, + "learning_rate": 5.754475703324808e-07, + "loss": 0.314, + "mean_token_accuracy": 0.8874155282974243, + "num_tokens": 718457.0, + "step": 226 + }, + { + "epoch": 0.11611253196930946, + "grad_norm": 28.186582565307617, + "learning_rate": 5.78005115089514e-07, + "loss": 0.4128, + "mean_token_accuracy": 0.8460063934326172, + "num_tokens": 721104.0, + "step": 227 + }, + { + "epoch": 0.11662404092071611, + "grad_norm": 9.561694145202637, + "learning_rate": 5.805626598465472e-07, + "loss": 0.211, + "mean_token_accuracy": 0.9126982688903809, + "num_tokens": 725047.0, + "step": 228 + }, + { + "epoch": 0.11713554987212277, + "grad_norm": 13.392457962036133, + "learning_rate": 5.831202046035805e-07, + "loss": 0.3255, + "mean_token_accuracy": 0.8881731033325195, + "num_tokens": 728457.0, + "step": 229 + }, + { + "epoch": 0.11764705882352941, + "grad_norm": 18.89466667175293, + "learning_rate": 5.856777493606137e-07, + "loss": 0.2532, + "mean_token_accuracy": 0.887466311454773, + "num_tokens": 731050.0, + "step": 230 + }, + { + "epoch": 0.11815856777493607, + "grad_norm": 11.421552658081055, + "learning_rate": 5.88235294117647e-07, + "loss": 0.278, + "mean_token_accuracy": 0.8759908676147461, + "num_tokens": 734766.0, + "step": 231 + }, + { + "epoch": 0.11867007672634271, + "grad_norm": 17.716773986816406, + "learning_rate": 5.907928388746802e-07, + "loss": 0.4019, + "mean_token_accuracy": 0.8388150930404663, + "num_tokens": 737929.0, + "step": 232 + }, + { + "epoch": 0.11918158567774936, + "grad_norm": 9.89438533782959, + "learning_rate": 5.933503836317135e-07, + "loss": 0.3021, + "mean_token_accuracy": 0.8975797891616821, + "num_tokens": 741701.0, + "step": 233 + }, + { + "epoch": 0.119693094629156, + "grad_norm": 30.164215087890625, + "learning_rate": 5.959079283887468e-07, + "loss": 0.287, + "mean_token_accuracy": 0.8958166837692261, + "num_tokens": 744369.0, + "step": 234 + }, + { + "epoch": 0.12020460358056266, + "grad_norm": 25.05753517150879, + "learning_rate": 5.9846547314578e-07, + "loss": 0.4557, + "mean_token_accuracy": 0.8301535844802856, + "num_tokens": 746839.0, + "step": 235 + }, + { + "epoch": 0.12071611253196932, + "grad_norm": 22.358558654785156, + "learning_rate": 6.010230179028132e-07, + "loss": 0.3016, + "mean_token_accuracy": 0.8988915681838989, + "num_tokens": 749942.0, + "step": 236 + }, + { + "epoch": 0.12122762148337596, + "grad_norm": 21.809249877929688, + "learning_rate": 6.035805626598464e-07, + "loss": 0.3142, + "mean_token_accuracy": 0.8785640001296997, + "num_tokens": 753318.0, + "step": 237 + }, + { + "epoch": 0.12173913043478261, + "grad_norm": 7.454619407653809, + "learning_rate": 6.061381074168797e-07, + "loss": 0.158, + "mean_token_accuracy": 0.9247511625289917, + "num_tokens": 756859.0, + "step": 238 + }, + { + "epoch": 0.12225063938618926, + "grad_norm": 16.895065307617188, + "learning_rate": 6.08695652173913e-07, + "loss": 0.3208, + "mean_token_accuracy": 0.8662222027778625, + "num_tokens": 760417.0, + "step": 239 + }, + { + "epoch": 0.12276214833759591, + "grad_norm": 13.004997253417969, + "learning_rate": 6.112531969309462e-07, + "loss": 0.3354, + "mean_token_accuracy": 0.8621222972869873, + "num_tokens": 764250.0, + "step": 240 + }, + { + "epoch": 0.12327365728900255, + "grad_norm": 14.804743766784668, + "learning_rate": 6.138107416879795e-07, + "loss": 0.25, + "mean_token_accuracy": 0.8880070447921753, + "num_tokens": 766953.0, + "step": 241 + }, + { + "epoch": 0.12378516624040921, + "grad_norm": 21.370433807373047, + "learning_rate": 6.163682864450127e-07, + "loss": 0.275, + "mean_token_accuracy": 0.8684628009796143, + "num_tokens": 770420.0, + "step": 242 + }, + { + "epoch": 0.12429667519181585, + "grad_norm": 20.187814712524414, + "learning_rate": 6.18925831202046e-07, + "loss": 0.2998, + "mean_token_accuracy": 0.8940629363059998, + "num_tokens": 773381.0, + "step": 243 + }, + { + "epoch": 0.1248081841432225, + "grad_norm": 15.050924301147461, + "learning_rate": 6.214833759590793e-07, + "loss": 0.3306, + "mean_token_accuracy": 0.8741259574890137, + "num_tokens": 776588.0, + "step": 244 + }, + { + "epoch": 0.12531969309462915, + "grad_norm": 22.089214324951172, + "learning_rate": 6.240409207161124e-07, + "loss": 0.3238, + "mean_token_accuracy": 0.8638997673988342, + "num_tokens": 780150.0, + "step": 245 + }, + { + "epoch": 0.12583120204603582, + "grad_norm": 18.05229949951172, + "learning_rate": 6.265984654731457e-07, + "loss": 0.3774, + "mean_token_accuracy": 0.863777756690979, + "num_tokens": 783124.0, + "step": 246 + }, + { + "epoch": 0.12634271099744246, + "grad_norm": 18.770727157592773, + "learning_rate": 6.291560102301789e-07, + "loss": 0.2695, + "mean_token_accuracy": 0.8788148164749146, + "num_tokens": 785791.0, + "step": 247 + }, + { + "epoch": 0.1268542199488491, + "grad_norm": 9.526939392089844, + "learning_rate": 6.317135549872122e-07, + "loss": 0.2686, + "mean_token_accuracy": 0.89682936668396, + "num_tokens": 789589.0, + "step": 248 + }, + { + "epoch": 0.12736572890025574, + "grad_norm": 22.92203140258789, + "learning_rate": 6.342710997442455e-07, + "loss": 0.3265, + "mean_token_accuracy": 0.8561132550239563, + "num_tokens": 791919.0, + "step": 249 + }, + { + "epoch": 0.1278772378516624, + "grad_norm": 27.121856689453125, + "learning_rate": 6.368286445012787e-07, + "loss": 0.5255, + "mean_token_accuracy": 0.8190746307373047, + "num_tokens": 794290.0, + "step": 250 + }, + { + "epoch": 0.12838874680306905, + "grad_norm": 16.43126106262207, + "learning_rate": 6.39386189258312e-07, + "loss": 0.2122, + "mean_token_accuracy": 0.913859486579895, + "num_tokens": 798427.0, + "step": 251 + }, + { + "epoch": 0.1289002557544757, + "grad_norm": 20.392574310302734, + "learning_rate": 6.419437340153451e-07, + "loss": 0.2857, + "mean_token_accuracy": 0.8876726627349854, + "num_tokens": 801278.0, + "step": 252 + }, + { + "epoch": 0.12941176470588237, + "grad_norm": 11.736381530761719, + "learning_rate": 6.445012787723784e-07, + "loss": 0.2843, + "mean_token_accuracy": 0.8747060298919678, + "num_tokens": 805239.0, + "step": 253 + }, + { + "epoch": 0.129923273657289, + "grad_norm": 11.574703216552734, + "learning_rate": 6.470588235294117e-07, + "loss": 0.25, + "mean_token_accuracy": 0.8860891461372375, + "num_tokens": 808768.0, + "step": 254 + }, + { + "epoch": 0.13043478260869565, + "grad_norm": 15.884101867675781, + "learning_rate": 6.496163682864449e-07, + "loss": 0.3003, + "mean_token_accuracy": 0.8860281705856323, + "num_tokens": 811990.0, + "step": 255 + }, + { + "epoch": 0.1309462915601023, + "grad_norm": 7.028905868530273, + "learning_rate": 6.521739130434782e-07, + "loss": 0.142, + "mean_token_accuracy": 0.9200819134712219, + "num_tokens": 817499.0, + "step": 256 + }, + { + "epoch": 0.13145780051150896, + "grad_norm": 8.530285835266113, + "learning_rate": 6.547314578005115e-07, + "loss": 0.1997, + "mean_token_accuracy": 0.9069508910179138, + "num_tokens": 821402.0, + "step": 257 + }, + { + "epoch": 0.1319693094629156, + "grad_norm": 16.016950607299805, + "learning_rate": 6.572890025575447e-07, + "loss": 0.2846, + "mean_token_accuracy": 0.8946903944015503, + "num_tokens": 824282.0, + "step": 258 + }, + { + "epoch": 0.13248081841432224, + "grad_norm": 21.79111099243164, + "learning_rate": 6.59846547314578e-07, + "loss": 0.2895, + "mean_token_accuracy": 0.8990781307220459, + "num_tokens": 826837.0, + "step": 259 + }, + { + "epoch": 0.1329923273657289, + "grad_norm": 16.920082092285156, + "learning_rate": 6.624040920716112e-07, + "loss": 0.2532, + "mean_token_accuracy": 0.8987568616867065, + "num_tokens": 829841.0, + "step": 260 + }, + { + "epoch": 0.13350383631713555, + "grad_norm": 18.730304718017578, + "learning_rate": 6.649616368286444e-07, + "loss": 0.3826, + "mean_token_accuracy": 0.8412432670593262, + "num_tokens": 832812.0, + "step": 261 + }, + { + "epoch": 0.1340153452685422, + "grad_norm": 11.533071517944336, + "learning_rate": 6.675191815856777e-07, + "loss": 0.3859, + "mean_token_accuracy": 0.8577761650085449, + "num_tokens": 836825.0, + "step": 262 + }, + { + "epoch": 0.13452685421994884, + "grad_norm": 15.528999328613281, + "learning_rate": 6.700767263427109e-07, + "loss": 0.3338, + "mean_token_accuracy": 0.8641093969345093, + "num_tokens": 839913.0, + "step": 263 + }, + { + "epoch": 0.1350383631713555, + "grad_norm": 12.037776947021484, + "learning_rate": 6.726342710997442e-07, + "loss": 0.1737, + "mean_token_accuracy": 0.94732666015625, + "num_tokens": 843317.0, + "step": 264 + }, + { + "epoch": 0.13554987212276215, + "grad_norm": 17.167919158935547, + "learning_rate": 6.751918158567774e-07, + "loss": 0.4041, + "mean_token_accuracy": 0.8267875909805298, + "num_tokens": 846064.0, + "step": 265 + }, + { + "epoch": 0.1360613810741688, + "grad_norm": 13.371946334838867, + "learning_rate": 6.777493606138107e-07, + "loss": 0.2727, + "mean_token_accuracy": 0.8808144927024841, + "num_tokens": 850407.0, + "step": 266 + }, + { + "epoch": 0.13657289002557546, + "grad_norm": 12.046252250671387, + "learning_rate": 6.80306905370844e-07, + "loss": 0.2509, + "mean_token_accuracy": 0.9103795289993286, + "num_tokens": 853260.0, + "step": 267 + }, + { + "epoch": 0.1370843989769821, + "grad_norm": 12.268072128295898, + "learning_rate": 6.828644501278772e-07, + "loss": 0.2486, + "mean_token_accuracy": 0.9107768535614014, + "num_tokens": 856064.0, + "step": 268 + }, + { + "epoch": 0.13759590792838874, + "grad_norm": 19.33623695373535, + "learning_rate": 6.854219948849104e-07, + "loss": 0.3506, + "mean_token_accuracy": 0.8628673553466797, + "num_tokens": 858647.0, + "step": 269 + }, + { + "epoch": 0.13810741687979539, + "grad_norm": 9.359271049499512, + "learning_rate": 6.879795396419436e-07, + "loss": 0.3045, + "mean_token_accuracy": 0.8997613191604614, + "num_tokens": 863071.0, + "step": 270 + }, + { + "epoch": 0.13861892583120206, + "grad_norm": 11.255187034606934, + "learning_rate": 6.905370843989769e-07, + "loss": 0.2363, + "mean_token_accuracy": 0.8955159187316895, + "num_tokens": 866681.0, + "step": 271 + }, + { + "epoch": 0.1391304347826087, + "grad_norm": 11.082009315490723, + "learning_rate": 6.930946291560102e-07, + "loss": 0.2511, + "mean_token_accuracy": 0.8889261484146118, + "num_tokens": 870147.0, + "step": 272 + }, + { + "epoch": 0.13964194373401534, + "grad_norm": 16.1816349029541, + "learning_rate": 6.956521739130434e-07, + "loss": 0.352, + "mean_token_accuracy": 0.8387446403503418, + "num_tokens": 872785.0, + "step": 273 + }, + { + "epoch": 0.140153452685422, + "grad_norm": 21.130413055419922, + "learning_rate": 6.982097186700767e-07, + "loss": 0.2511, + "mean_token_accuracy": 0.9063255786895752, + "num_tokens": 875955.0, + "step": 274 + }, + { + "epoch": 0.14066496163682865, + "grad_norm": 18.60016632080078, + "learning_rate": 7.007672634271099e-07, + "loss": 0.2344, + "mean_token_accuracy": 0.8928755521774292, + "num_tokens": 878792.0, + "step": 275 + }, + { + "epoch": 0.1411764705882353, + "grad_norm": 13.90630054473877, + "learning_rate": 7.033248081841432e-07, + "loss": 0.2884, + "mean_token_accuracy": 0.8711792230606079, + "num_tokens": 881800.0, + "step": 276 + }, + { + "epoch": 0.14168797953964193, + "grad_norm": 12.276054382324219, + "learning_rate": 7.058823529411765e-07, + "loss": 0.2663, + "mean_token_accuracy": 0.8735630512237549, + "num_tokens": 885391.0, + "step": 277 + }, + { + "epoch": 0.1421994884910486, + "grad_norm": 24.737289428710938, + "learning_rate": 7.084398976982096e-07, + "loss": 0.2834, + "mean_token_accuracy": 0.880832314491272, + "num_tokens": 888876.0, + "step": 278 + }, + { + "epoch": 0.14271099744245525, + "grad_norm": 16.31096839904785, + "learning_rate": 7.109974424552429e-07, + "loss": 0.2638, + "mean_token_accuracy": 0.901459813117981, + "num_tokens": 892467.0, + "step": 279 + }, + { + "epoch": 0.1432225063938619, + "grad_norm": 8.897708892822266, + "learning_rate": 7.135549872122761e-07, + "loss": 0.3099, + "mean_token_accuracy": 0.8744910955429077, + "num_tokens": 897409.0, + "step": 280 + }, + { + "epoch": 0.14373401534526853, + "grad_norm": 25.039897918701172, + "learning_rate": 7.161125319693094e-07, + "loss": 0.3209, + "mean_token_accuracy": 0.8569847345352173, + "num_tokens": 900136.0, + "step": 281 + }, + { + "epoch": 0.1442455242966752, + "grad_norm": 24.281524658203125, + "learning_rate": 7.186700767263427e-07, + "loss": 0.3673, + "mean_token_accuracy": 0.8752604722976685, + "num_tokens": 902687.0, + "step": 282 + }, + { + "epoch": 0.14475703324808184, + "grad_norm": 8.588457107543945, + "learning_rate": 7.212276214833759e-07, + "loss": 0.2636, + "mean_token_accuracy": 0.8818404078483582, + "num_tokens": 906266.0, + "step": 283 + }, + { + "epoch": 0.14526854219948848, + "grad_norm": 11.956930160522461, + "learning_rate": 7.237851662404092e-07, + "loss": 0.4243, + "mean_token_accuracy": 0.828006386756897, + "num_tokens": 911037.0, + "step": 284 + }, + { + "epoch": 0.14578005115089515, + "grad_norm": 14.331524848937988, + "learning_rate": 7.263427109974423e-07, + "loss": 0.2814, + "mean_token_accuracy": 0.8993332982063293, + "num_tokens": 914465.0, + "step": 285 + }, + { + "epoch": 0.1462915601023018, + "grad_norm": 7.917057037353516, + "learning_rate": 7.289002557544756e-07, + "loss": 0.2184, + "mean_token_accuracy": 0.8688901662826538, + "num_tokens": 919090.0, + "step": 286 + }, + { + "epoch": 0.14680306905370843, + "grad_norm": 15.900273323059082, + "learning_rate": 7.314578005115089e-07, + "loss": 0.323, + "mean_token_accuracy": 0.8828432559967041, + "num_tokens": 922148.0, + "step": 287 + }, + { + "epoch": 0.14731457800511508, + "grad_norm": 12.376312255859375, + "learning_rate": 7.340153452685421e-07, + "loss": 0.2925, + "mean_token_accuracy": 0.8932837247848511, + "num_tokens": 925217.0, + "step": 288 + }, + { + "epoch": 0.14782608695652175, + "grad_norm": 14.779565811157227, + "learning_rate": 7.365728900255754e-07, + "loss": 0.2924, + "mean_token_accuracy": 0.8773035407066345, + "num_tokens": 928302.0, + "step": 289 + }, + { + "epoch": 0.1483375959079284, + "grad_norm": 16.22716522216797, + "learning_rate": 7.391304347826086e-07, + "loss": 0.2789, + "mean_token_accuracy": 0.8602189421653748, + "num_tokens": 931948.0, + "step": 290 + }, + { + "epoch": 0.14884910485933503, + "grad_norm": 19.546478271484375, + "learning_rate": 7.416879795396419e-07, + "loss": 0.2554, + "mean_token_accuracy": 0.8894336223602295, + "num_tokens": 934857.0, + "step": 291 + }, + { + "epoch": 0.1493606138107417, + "grad_norm": 14.140451431274414, + "learning_rate": 7.442455242966752e-07, + "loss": 0.2804, + "mean_token_accuracy": 0.8704689741134644, + "num_tokens": 937541.0, + "step": 292 + }, + { + "epoch": 0.14987212276214834, + "grad_norm": 19.065580368041992, + "learning_rate": 7.468030690537084e-07, + "loss": 0.3823, + "mean_token_accuracy": 0.8702987432479858, + "num_tokens": 940568.0, + "step": 293 + }, + { + "epoch": 0.15038363171355498, + "grad_norm": 15.041227340698242, + "learning_rate": 7.493606138107416e-07, + "loss": 0.2911, + "mean_token_accuracy": 0.8802947998046875, + "num_tokens": 943313.0, + "step": 294 + }, + { + "epoch": 0.15089514066496162, + "grad_norm": 18.25023078918457, + "learning_rate": 7.519181585677748e-07, + "loss": 0.2675, + "mean_token_accuracy": 0.8945796489715576, + "num_tokens": 946073.0, + "step": 295 + }, + { + "epoch": 0.1514066496163683, + "grad_norm": 21.196495056152344, + "learning_rate": 7.544757033248081e-07, + "loss": 0.3262, + "mean_token_accuracy": 0.8689770698547363, + "num_tokens": 948751.0, + "step": 296 + }, + { + "epoch": 0.15191815856777494, + "grad_norm": 13.502689361572266, + "learning_rate": 7.570332480818414e-07, + "loss": 0.247, + "mean_token_accuracy": 0.8598753213882446, + "num_tokens": 951703.0, + "step": 297 + }, + { + "epoch": 0.15242966751918158, + "grad_norm": 13.584320068359375, + "learning_rate": 7.595907928388746e-07, + "loss": 0.2016, + "mean_token_accuracy": 0.8836255669593811, + "num_tokens": 954304.0, + "step": 298 + }, + { + "epoch": 0.15294117647058825, + "grad_norm": 15.543497085571289, + "learning_rate": 7.621483375959079e-07, + "loss": 0.3398, + "mean_token_accuracy": 0.8844573497772217, + "num_tokens": 957297.0, + "step": 299 + }, + { + "epoch": 0.1534526854219949, + "grad_norm": 14.280072212219238, + "learning_rate": 7.647058823529411e-07, + "loss": 0.2543, + "mean_token_accuracy": 0.902655303478241, + "num_tokens": 959938.0, + "step": 300 + }, + { + "epoch": 0.15396419437340153, + "grad_norm": 12.879986763000488, + "learning_rate": 7.672634271099744e-07, + "loss": 0.328, + "mean_token_accuracy": 0.8914279937744141, + "num_tokens": 963332.0, + "step": 301 + }, + { + "epoch": 0.15447570332480817, + "grad_norm": 16.95353126525879, + "learning_rate": 7.698209718670076e-07, + "loss": 0.4025, + "mean_token_accuracy": 0.8525075912475586, + "num_tokens": 966147.0, + "step": 302 + }, + { + "epoch": 0.15498721227621484, + "grad_norm": 10.019353866577148, + "learning_rate": 7.723785166240408e-07, + "loss": 0.2079, + "mean_token_accuracy": 0.8794631958007812, + "num_tokens": 970414.0, + "step": 303 + }, + { + "epoch": 0.15549872122762148, + "grad_norm": 19.842836380004883, + "learning_rate": 7.749360613810741e-07, + "loss": 0.2934, + "mean_token_accuracy": 0.8831413388252258, + "num_tokens": 973639.0, + "step": 304 + }, + { + "epoch": 0.15601023017902813, + "grad_norm": 15.053909301757812, + "learning_rate": 7.774936061381073e-07, + "loss": 0.2522, + "mean_token_accuracy": 0.8872917890548706, + "num_tokens": 976430.0, + "step": 305 + }, + { + "epoch": 0.1565217391304348, + "grad_norm": 17.543821334838867, + "learning_rate": 7.800511508951406e-07, + "loss": 0.3115, + "mean_token_accuracy": 0.8831071853637695, + "num_tokens": 979147.0, + "step": 306 + }, + { + "epoch": 0.15703324808184144, + "grad_norm": 15.78184700012207, + "learning_rate": 7.826086956521739e-07, + "loss": 0.215, + "mean_token_accuracy": 0.9262404441833496, + "num_tokens": 982170.0, + "step": 307 + }, + { + "epoch": 0.15754475703324808, + "grad_norm": 6.9840826988220215, + "learning_rate": 7.851662404092071e-07, + "loss": 0.2972, + "mean_token_accuracy": 0.890216588973999, + "num_tokens": 987752.0, + "step": 308 + }, + { + "epoch": 0.15805626598465472, + "grad_norm": 14.912068367004395, + "learning_rate": 7.877237851662404e-07, + "loss": 0.2932, + "mean_token_accuracy": 0.8482984304428101, + "num_tokens": 990353.0, + "step": 309 + }, + { + "epoch": 0.1585677749360614, + "grad_norm": 19.925132751464844, + "learning_rate": 7.902813299232737e-07, + "loss": 0.2733, + "mean_token_accuracy": 0.9034287929534912, + "num_tokens": 993767.0, + "step": 310 + }, + { + "epoch": 0.15907928388746803, + "grad_norm": 35.86583709716797, + "learning_rate": 7.928388746803068e-07, + "loss": 0.337, + "mean_token_accuracy": 0.8682563900947571, + "num_tokens": 996991.0, + "step": 311 + }, + { + "epoch": 0.15959079283887467, + "grad_norm": 17.077655792236328, + "learning_rate": 7.953964194373401e-07, + "loss": 0.2899, + "mean_token_accuracy": 0.896345317363739, + "num_tokens": 999968.0, + "step": 312 + }, + { + "epoch": 0.16010230179028134, + "grad_norm": 19.258651733398438, + "learning_rate": 7.979539641943733e-07, + "loss": 0.3175, + "mean_token_accuracy": 0.876115083694458, + "num_tokens": 1002706.0, + "step": 313 + }, + { + "epoch": 0.16061381074168798, + "grad_norm": 16.342382431030273, + "learning_rate": 8.005115089514066e-07, + "loss": 0.2705, + "mean_token_accuracy": 0.8952701091766357, + "num_tokens": 1006329.0, + "step": 314 + }, + { + "epoch": 0.16112531969309463, + "grad_norm": 20.44878387451172, + "learning_rate": 8.030690537084399e-07, + "loss": 0.3497, + "mean_token_accuracy": 0.8542380332946777, + "num_tokens": 1009178.0, + "step": 315 + }, + { + "epoch": 0.16163682864450127, + "grad_norm": 22.846771240234375, + "learning_rate": 8.056265984654731e-07, + "loss": 0.3954, + "mean_token_accuracy": 0.8467920422554016, + "num_tokens": 1012243.0, + "step": 316 + }, + { + "epoch": 0.16214833759590794, + "grad_norm": 24.46397590637207, + "learning_rate": 8.081841432225064e-07, + "loss": 0.3309, + "mean_token_accuracy": 0.8674427270889282, + "num_tokens": 1015257.0, + "step": 317 + }, + { + "epoch": 0.16265984654731458, + "grad_norm": 16.019071578979492, + "learning_rate": 8.107416879795395e-07, + "loss": 0.317, + "mean_token_accuracy": 0.8670022487640381, + "num_tokens": 1019293.0, + "step": 318 + }, + { + "epoch": 0.16317135549872122, + "grad_norm": 20.253305435180664, + "learning_rate": 8.132992327365728e-07, + "loss": 0.3504, + "mean_token_accuracy": 0.8863744735717773, + "num_tokens": 1022366.0, + "step": 319 + }, + { + "epoch": 0.1636828644501279, + "grad_norm": 15.677396774291992, + "learning_rate": 8.158567774936061e-07, + "loss": 0.3559, + "mean_token_accuracy": 0.8386479616165161, + "num_tokens": 1024964.0, + "step": 320 + }, + { + "epoch": 0.16419437340153453, + "grad_norm": 13.238309860229492, + "learning_rate": 8.184143222506393e-07, + "loss": 0.3184, + "mean_token_accuracy": 0.8731343746185303, + "num_tokens": 1027769.0, + "step": 321 + }, + { + "epoch": 0.16470588235294117, + "grad_norm": 10.625545501708984, + "learning_rate": 8.209718670076726e-07, + "loss": 0.3447, + "mean_token_accuracy": 0.8884180784225464, + "num_tokens": 1031513.0, + "step": 322 + }, + { + "epoch": 0.16521739130434782, + "grad_norm": 20.707937240600586, + "learning_rate": 8.235294117647058e-07, + "loss": 0.3039, + "mean_token_accuracy": 0.8801056146621704, + "num_tokens": 1034626.0, + "step": 323 + }, + { + "epoch": 0.16572890025575449, + "grad_norm": 15.446698188781738, + "learning_rate": 8.260869565217391e-07, + "loss": 0.3321, + "mean_token_accuracy": 0.8725600242614746, + "num_tokens": 1038200.0, + "step": 324 + }, + { + "epoch": 0.16624040920716113, + "grad_norm": 15.977676391601562, + "learning_rate": 8.286445012787724e-07, + "loss": 0.3681, + "mean_token_accuracy": 0.861161470413208, + "num_tokens": 1041115.0, + "step": 325 + }, + { + "epoch": 0.16675191815856777, + "grad_norm": 11.971078872680664, + "learning_rate": 8.312020460358056e-07, + "loss": 0.2918, + "mean_token_accuracy": 0.8741958141326904, + "num_tokens": 1044920.0, + "step": 326 + }, + { + "epoch": 0.1672634271099744, + "grad_norm": 8.769084930419922, + "learning_rate": 8.337595907928388e-07, + "loss": 0.2441, + "mean_token_accuracy": 0.9002683758735657, + "num_tokens": 1048532.0, + "step": 327 + }, + { + "epoch": 0.16777493606138108, + "grad_norm": 21.2575626373291, + "learning_rate": 8.36317135549872e-07, + "loss": 0.3144, + "mean_token_accuracy": 0.8780736923217773, + "num_tokens": 1051542.0, + "step": 328 + }, + { + "epoch": 0.16828644501278772, + "grad_norm": 19.77461051940918, + "learning_rate": 8.388746803069053e-07, + "loss": 0.2876, + "mean_token_accuracy": 0.8840268850326538, + "num_tokens": 1054652.0, + "step": 329 + }, + { + "epoch": 0.16879795396419436, + "grad_norm": 18.050304412841797, + "learning_rate": 8.414322250639386e-07, + "loss": 0.348, + "mean_token_accuracy": 0.8520166873931885, + "num_tokens": 1056870.0, + "step": 330 + }, + { + "epoch": 0.16930946291560103, + "grad_norm": 17.560741424560547, + "learning_rate": 8.439897698209718e-07, + "loss": 0.3415, + "mean_token_accuracy": 0.867695689201355, + "num_tokens": 1059365.0, + "step": 331 + }, + { + "epoch": 0.16982097186700768, + "grad_norm": 10.636878967285156, + "learning_rate": 8.465473145780051e-07, + "loss": 0.2866, + "mean_token_accuracy": 0.8710038661956787, + "num_tokens": 1062819.0, + "step": 332 + }, + { + "epoch": 0.17033248081841432, + "grad_norm": 11.122025489807129, + "learning_rate": 8.491048593350383e-07, + "loss": 0.2862, + "mean_token_accuracy": 0.8580281734466553, + "num_tokens": 1065832.0, + "step": 333 + }, + { + "epoch": 0.17084398976982096, + "grad_norm": 14.64178466796875, + "learning_rate": 8.516624040920716e-07, + "loss": 0.3237, + "mean_token_accuracy": 0.8683099746704102, + "num_tokens": 1069618.0, + "step": 334 + }, + { + "epoch": 0.17135549872122763, + "grad_norm": 19.371198654174805, + "learning_rate": 8.542199488491048e-07, + "loss": 0.3364, + "mean_token_accuracy": 0.8537746667861938, + "num_tokens": 1072068.0, + "step": 335 + }, + { + "epoch": 0.17186700767263427, + "grad_norm": 17.37316131591797, + "learning_rate": 8.56777493606138e-07, + "loss": 0.2873, + "mean_token_accuracy": 0.8807929754257202, + "num_tokens": 1075030.0, + "step": 336 + }, + { + "epoch": 0.1723785166240409, + "grad_norm": 14.843271255493164, + "learning_rate": 8.593350383631713e-07, + "loss": 0.2654, + "mean_token_accuracy": 0.9045373201370239, + "num_tokens": 1078132.0, + "step": 337 + }, + { + "epoch": 0.17289002557544758, + "grad_norm": 17.413463592529297, + "learning_rate": 8.618925831202045e-07, + "loss": 0.293, + "mean_token_accuracy": 0.8793296217918396, + "num_tokens": 1080658.0, + "step": 338 + }, + { + "epoch": 0.17340153452685422, + "grad_norm": 13.317368507385254, + "learning_rate": 8.644501278772378e-07, + "loss": 0.256, + "mean_token_accuracy": 0.8853883743286133, + "num_tokens": 1083542.0, + "step": 339 + }, + { + "epoch": 0.17391304347826086, + "grad_norm": 19.120595932006836, + "learning_rate": 8.670076726342711e-07, + "loss": 0.3163, + "mean_token_accuracy": 0.8489685654640198, + "num_tokens": 1086965.0, + "step": 340 + }, + { + "epoch": 0.1744245524296675, + "grad_norm": 14.273585319519043, + "learning_rate": 8.695652173913043e-07, + "loss": 0.3368, + "mean_token_accuracy": 0.8463470935821533, + "num_tokens": 1089889.0, + "step": 341 + }, + { + "epoch": 0.17493606138107418, + "grad_norm": 16.788606643676758, + "learning_rate": 8.721227621483376e-07, + "loss": 0.2725, + "mean_token_accuracy": 0.9075159430503845, + "num_tokens": 1092756.0, + "step": 342 + }, + { + "epoch": 0.17544757033248082, + "grad_norm": 17.631547927856445, + "learning_rate": 8.746803069053707e-07, + "loss": 0.3261, + "mean_token_accuracy": 0.8571901917457581, + "num_tokens": 1095367.0, + "step": 343 + }, + { + "epoch": 0.17595907928388746, + "grad_norm": 11.809229850769043, + "learning_rate": 8.77237851662404e-07, + "loss": 0.2837, + "mean_token_accuracy": 0.8778702020645142, + "num_tokens": 1098473.0, + "step": 344 + }, + { + "epoch": 0.17647058823529413, + "grad_norm": 18.219318389892578, + "learning_rate": 8.797953964194373e-07, + "loss": 0.3366, + "mean_token_accuracy": 0.8726704716682434, + "num_tokens": 1101276.0, + "step": 345 + }, + { + "epoch": 0.17698209718670077, + "grad_norm": 13.334338188171387, + "learning_rate": 8.823529411764705e-07, + "loss": 0.2373, + "mean_token_accuracy": 0.9029194712638855, + "num_tokens": 1104329.0, + "step": 346 + }, + { + "epoch": 0.1774936061381074, + "grad_norm": 12.936488151550293, + "learning_rate": 8.849104859335038e-07, + "loss": 0.2831, + "mean_token_accuracy": 0.8918833136558533, + "num_tokens": 1107612.0, + "step": 347 + }, + { + "epoch": 0.17800511508951405, + "grad_norm": 18.71854591369629, + "learning_rate": 8.87468030690537e-07, + "loss": 0.4048, + "mean_token_accuracy": 0.8455857634544373, + "num_tokens": 1110827.0, + "step": 348 + }, + { + "epoch": 0.17851662404092072, + "grad_norm": 10.147542953491211, + "learning_rate": 8.900255754475703e-07, + "loss": 0.2958, + "mean_token_accuracy": 0.889761209487915, + "num_tokens": 1115776.0, + "step": 349 + }, + { + "epoch": 0.17902813299232737, + "grad_norm": 15.601264953613281, + "learning_rate": 8.925831202046036e-07, + "loss": 0.3487, + "mean_token_accuracy": 0.8747429847717285, + "num_tokens": 1118060.0, + "step": 350 + }, + { + "epoch": 0.179539641943734, + "grad_norm": 16.326635360717773, + "learning_rate": 8.951406649616368e-07, + "loss": 0.4108, + "mean_token_accuracy": 0.8508123159408569, + "num_tokens": 1120891.0, + "step": 351 + }, + { + "epoch": 0.18005115089514068, + "grad_norm": 10.532147407531738, + "learning_rate": 8.9769820971867e-07, + "loss": 0.4006, + "mean_token_accuracy": 0.8885059952735901, + "num_tokens": 1124259.0, + "step": 352 + }, + { + "epoch": 0.18056265984654732, + "grad_norm": 17.00088882446289, + "learning_rate": 9.002557544757032e-07, + "loss": 0.3691, + "mean_token_accuracy": 0.860910177230835, + "num_tokens": 1126945.0, + "step": 353 + }, + { + "epoch": 0.18107416879795396, + "grad_norm": 13.946377754211426, + "learning_rate": 9.028132992327365e-07, + "loss": 0.2588, + "mean_token_accuracy": 0.8830157518386841, + "num_tokens": 1129851.0, + "step": 354 + }, + { + "epoch": 0.1815856777493606, + "grad_norm": 18.38866424560547, + "learning_rate": 9.053708439897698e-07, + "loss": 0.4197, + "mean_token_accuracy": 0.8341412544250488, + "num_tokens": 1132582.0, + "step": 355 + }, + { + "epoch": 0.18209718670076727, + "grad_norm": 16.470291137695312, + "learning_rate": 9.07928388746803e-07, + "loss": 0.2498, + "mean_token_accuracy": 0.8898026943206787, + "num_tokens": 1135504.0, + "step": 356 + }, + { + "epoch": 0.1826086956521739, + "grad_norm": 9.159273147583008, + "learning_rate": 9.104859335038363e-07, + "loss": 0.2523, + "mean_token_accuracy": 0.9159303307533264, + "num_tokens": 1139834.0, + "step": 357 + }, + { + "epoch": 0.18312020460358056, + "grad_norm": 15.577966690063477, + "learning_rate": 9.130434782608695e-07, + "loss": 0.3024, + "mean_token_accuracy": 0.8775466680526733, + "num_tokens": 1142984.0, + "step": 358 + }, + { + "epoch": 0.18363171355498722, + "grad_norm": 12.573100090026855, + "learning_rate": 9.156010230179028e-07, + "loss": 0.3715, + "mean_token_accuracy": 0.8662092685699463, + "num_tokens": 1146658.0, + "step": 359 + }, + { + "epoch": 0.18414322250639387, + "grad_norm": 13.494993209838867, + "learning_rate": 9.18158567774936e-07, + "loss": 0.2377, + "mean_token_accuracy": 0.9096786975860596, + "num_tokens": 1149862.0, + "step": 360 + }, + { + "epoch": 0.1846547314578005, + "grad_norm": 15.522153854370117, + "learning_rate": 9.207161125319692e-07, + "loss": 0.3109, + "mean_token_accuracy": 0.9004313945770264, + "num_tokens": 1153479.0, + "step": 361 + }, + { + "epoch": 0.18516624040920715, + "grad_norm": 16.20472526550293, + "learning_rate": 9.232736572890025e-07, + "loss": 0.3133, + "mean_token_accuracy": 0.8764342665672302, + "num_tokens": 1156853.0, + "step": 362 + }, + { + "epoch": 0.18567774936061382, + "grad_norm": 12.974690437316895, + "learning_rate": 9.258312020460357e-07, + "loss": 0.3424, + "mean_token_accuracy": 0.8715150952339172, + "num_tokens": 1160930.0, + "step": 363 + }, + { + "epoch": 0.18618925831202046, + "grad_norm": 19.1595458984375, + "learning_rate": 9.28388746803069e-07, + "loss": 0.3164, + "mean_token_accuracy": 0.8797683119773865, + "num_tokens": 1163294.0, + "step": 364 + }, + { + "epoch": 0.1867007672634271, + "grad_norm": 12.637186050415039, + "learning_rate": 9.309462915601023e-07, + "loss": 0.3485, + "mean_token_accuracy": 0.8791938424110413, + "num_tokens": 1166593.0, + "step": 365 + }, + { + "epoch": 0.18721227621483377, + "grad_norm": 18.439165115356445, + "learning_rate": 9.335038363171355e-07, + "loss": 0.321, + "mean_token_accuracy": 0.8907080292701721, + "num_tokens": 1169701.0, + "step": 366 + }, + { + "epoch": 0.18772378516624041, + "grad_norm": 8.373260498046875, + "learning_rate": 9.360613810741688e-07, + "loss": 0.2663, + "mean_token_accuracy": 0.9051944613456726, + "num_tokens": 1173631.0, + "step": 367 + }, + { + "epoch": 0.18823529411764706, + "grad_norm": 18.899538040161133, + "learning_rate": 9.38618925831202e-07, + "loss": 0.3665, + "mean_token_accuracy": 0.8603608012199402, + "num_tokens": 1176004.0, + "step": 368 + }, + { + "epoch": 0.1887468030690537, + "grad_norm": 17.151033401489258, + "learning_rate": 9.411764705882352e-07, + "loss": 0.3561, + "mean_token_accuracy": 0.865958034992218, + "num_tokens": 1179120.0, + "step": 369 + }, + { + "epoch": 0.18925831202046037, + "grad_norm": 11.476635932922363, + "learning_rate": 9.437340153452685e-07, + "loss": 0.1839, + "mean_token_accuracy": 0.9289584159851074, + "num_tokens": 1181828.0, + "step": 370 + }, + { + "epoch": 0.189769820971867, + "grad_norm": 18.503381729125977, + "learning_rate": 9.462915601023017e-07, + "loss": 0.3154, + "mean_token_accuracy": 0.8788929581642151, + "num_tokens": 1184535.0, + "step": 371 + }, + { + "epoch": 0.19028132992327365, + "grad_norm": 15.359516143798828, + "learning_rate": 9.48849104859335e-07, + "loss": 0.2736, + "mean_token_accuracy": 0.8596996665000916, + "num_tokens": 1188075.0, + "step": 372 + }, + { + "epoch": 0.19079283887468032, + "grad_norm": 12.031140327453613, + "learning_rate": 9.514066496163683e-07, + "loss": 0.2958, + "mean_token_accuracy": 0.8937944173812866, + "num_tokens": 1191456.0, + "step": 373 + }, + { + "epoch": 0.19130434782608696, + "grad_norm": 10.152610778808594, + "learning_rate": 9.539641943734016e-07, + "loss": 0.1772, + "mean_token_accuracy": 0.9068412780761719, + "num_tokens": 1194400.0, + "step": 374 + }, + { + "epoch": 0.1918158567774936, + "grad_norm": 16.6165828704834, + "learning_rate": 9.565217391304349e-07, + "loss": 0.3102, + "mean_token_accuracy": 0.9026028513908386, + "num_tokens": 1197978.0, + "step": 375 + }, + { + "epoch": 0.19232736572890025, + "grad_norm": 12.34303092956543, + "learning_rate": 9.59079283887468e-07, + "loss": 0.3215, + "mean_token_accuracy": 0.860179603099823, + "num_tokens": 1202380.0, + "step": 376 + }, + { + "epoch": 0.19283887468030692, + "grad_norm": 12.664880752563477, + "learning_rate": 9.616368286445012e-07, + "loss": 0.3962, + "mean_token_accuracy": 0.8743681311607361, + "num_tokens": 1205644.0, + "step": 377 + }, + { + "epoch": 0.19335038363171356, + "grad_norm": 8.762524604797363, + "learning_rate": 9.641943734015345e-07, + "loss": 0.2413, + "mean_token_accuracy": 0.9044547080993652, + "num_tokens": 1208739.0, + "step": 378 + }, + { + "epoch": 0.1938618925831202, + "grad_norm": 14.506214141845703, + "learning_rate": 9.667519181585678e-07, + "loss": 0.2291, + "mean_token_accuracy": 0.8973813056945801, + "num_tokens": 1211427.0, + "step": 379 + }, + { + "epoch": 0.19437340153452684, + "grad_norm": 19.761001586914062, + "learning_rate": 9.693094629156011e-07, + "loss": 0.2682, + "mean_token_accuracy": 0.898030161857605, + "num_tokens": 1214230.0, + "step": 380 + }, + { + "epoch": 0.1948849104859335, + "grad_norm": 14.514360427856445, + "learning_rate": 9.718670076726342e-07, + "loss": 0.2955, + "mean_token_accuracy": 0.8701151013374329, + "num_tokens": 1217595.0, + "step": 381 + }, + { + "epoch": 0.19539641943734015, + "grad_norm": 7.906139850616455, + "learning_rate": 9.744245524296675e-07, + "loss": 0.2715, + "mean_token_accuracy": 0.8788046836853027, + "num_tokens": 1222290.0, + "step": 382 + }, + { + "epoch": 0.1959079283887468, + "grad_norm": 23.877222061157227, + "learning_rate": 9.769820971867008e-07, + "loss": 0.3306, + "mean_token_accuracy": 0.8920998573303223, + "num_tokens": 1225574.0, + "step": 383 + }, + { + "epoch": 0.19641943734015346, + "grad_norm": 13.109369277954102, + "learning_rate": 9.79539641943734e-07, + "loss": 0.2909, + "mean_token_accuracy": 0.8907207250595093, + "num_tokens": 1228593.0, + "step": 384 + }, + { + "epoch": 0.1969309462915601, + "grad_norm": 11.29456615447998, + "learning_rate": 9.820971867007673e-07, + "loss": 0.3406, + "mean_token_accuracy": 0.8613995313644409, + "num_tokens": 1232090.0, + "step": 385 + }, + { + "epoch": 0.19744245524296675, + "grad_norm": 15.552072525024414, + "learning_rate": 9.846547314578004e-07, + "loss": 0.3243, + "mean_token_accuracy": 0.9153829216957092, + "num_tokens": 1234985.0, + "step": 386 + }, + { + "epoch": 0.1979539641943734, + "grad_norm": 18.095626831054688, + "learning_rate": 9.872122762148337e-07, + "loss": 0.3336, + "mean_token_accuracy": 0.8585290908813477, + "num_tokens": 1237732.0, + "step": 387 + }, + { + "epoch": 0.19846547314578006, + "grad_norm": 15.351041793823242, + "learning_rate": 9.89769820971867e-07, + "loss": 0.2944, + "mean_token_accuracy": 0.8934398889541626, + "num_tokens": 1240399.0, + "step": 388 + }, + { + "epoch": 0.1989769820971867, + "grad_norm": 13.938977241516113, + "learning_rate": 9.923273657289e-07, + "loss": 0.3201, + "mean_token_accuracy": 0.8795569539070129, + "num_tokens": 1243827.0, + "step": 389 + }, + { + "epoch": 0.19948849104859334, + "grad_norm": 15.601459503173828, + "learning_rate": 9.948849104859336e-07, + "loss": 0.3883, + "mean_token_accuracy": 0.8455557823181152, + "num_tokens": 1246414.0, + "step": 390 + }, + { + "epoch": 0.2, + "grad_norm": 19.497802734375, + "learning_rate": 9.974424552429667e-07, + "loss": 0.3122, + "mean_token_accuracy": 0.87461918592453, + "num_tokens": 1249515.0, + "step": 391 + }, + { + "epoch": 0.20051150895140665, + "grad_norm": 11.073195457458496, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8627398610115051, + "num_tokens": 1253456.0, + "step": 392 + }, + { + "epoch": 0.2010230179028133, + "grad_norm": 9.696197509765625, + "learning_rate": 1e-06, + "loss": 0.2447, + "mean_token_accuracy": 0.887937068939209, + "num_tokens": 1256865.0, + "step": 393 + }, + { + "epoch": 0.20153452685421994, + "grad_norm": 10.812300682067871, + "learning_rate": 1e-06, + "loss": 0.2322, + "mean_token_accuracy": 0.9109159111976624, + "num_tokens": 1260262.0, + "step": 394 + }, + { + "epoch": 0.2020460358056266, + "grad_norm": 10.564620971679688, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.867943286895752, + "num_tokens": 1263769.0, + "step": 395 + }, + { + "epoch": 0.20255754475703325, + "grad_norm": 17.54439926147461, + "learning_rate": 1e-06, + "loss": 0.3922, + "mean_token_accuracy": 0.8333864212036133, + "num_tokens": 1266815.0, + "step": 396 + }, + { + "epoch": 0.2030690537084399, + "grad_norm": 14.573796272277832, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8757822513580322, + "num_tokens": 1269592.0, + "step": 397 + }, + { + "epoch": 0.20358056265984656, + "grad_norm": 12.618227005004883, + "learning_rate": 1e-06, + "loss": 0.2689, + "mean_token_accuracy": 0.9015680551528931, + "num_tokens": 1272284.0, + "step": 398 + }, + { + "epoch": 0.2040920716112532, + "grad_norm": 14.820114135742188, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8756434917449951, + "num_tokens": 1275068.0, + "step": 399 + }, + { + "epoch": 0.20460358056265984, + "grad_norm": 14.3322172164917, + "learning_rate": 1e-06, + "loss": 0.3816, + "mean_token_accuracy": 0.8649807572364807, + "num_tokens": 1279310.0, + "step": 400 + }, + { + "epoch": 0.20511508951406648, + "grad_norm": 12.664986610412598, + "learning_rate": 1e-06, + "loss": 0.3808, + "mean_token_accuracy": 0.8541327118873596, + "num_tokens": 1282438.0, + "step": 401 + }, + { + "epoch": 0.20562659846547315, + "grad_norm": 12.737696647644043, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.8729633092880249, + "num_tokens": 1286252.0, + "step": 402 + }, + { + "epoch": 0.2061381074168798, + "grad_norm": 12.327568054199219, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.8885443210601807, + "num_tokens": 1289011.0, + "step": 403 + }, + { + "epoch": 0.20664961636828644, + "grad_norm": 12.812989234924316, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8786247372627258, + "num_tokens": 1292228.0, + "step": 404 + }, + { + "epoch": 0.2071611253196931, + "grad_norm": 13.962796211242676, + "learning_rate": 1e-06, + "loss": 0.2504, + "mean_token_accuracy": 0.908928394317627, + "num_tokens": 1295197.0, + "step": 405 + }, + { + "epoch": 0.20767263427109975, + "grad_norm": 11.7476167678833, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8798178434371948, + "num_tokens": 1298355.0, + "step": 406 + }, + { + "epoch": 0.2081841432225064, + "grad_norm": 9.797943115234375, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.87217116355896, + "num_tokens": 1301647.0, + "step": 407 + }, + { + "epoch": 0.20869565217391303, + "grad_norm": 12.290892601013184, + "learning_rate": 1e-06, + "loss": 0.2576, + "mean_token_accuracy": 0.8944055438041687, + "num_tokens": 1305588.0, + "step": 408 + }, + { + "epoch": 0.2092071611253197, + "grad_norm": 7.242301940917969, + "learning_rate": 1e-06, + "loss": 0.2031, + "mean_token_accuracy": 0.8929107189178467, + "num_tokens": 1309997.0, + "step": 409 + }, + { + "epoch": 0.20971867007672634, + "grad_norm": 22.260393142700195, + "learning_rate": 1e-06, + "loss": 0.3165, + "mean_token_accuracy": 0.9058746099472046, + "num_tokens": 1313205.0, + "step": 410 + }, + { + "epoch": 0.21023017902813299, + "grad_norm": 14.174097061157227, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8861746191978455, + "num_tokens": 1317118.0, + "step": 411 + }, + { + "epoch": 0.21074168797953965, + "grad_norm": 15.004929542541504, + "learning_rate": 1e-06, + "loss": 0.2823, + "mean_token_accuracy": 0.8863715529441833, + "num_tokens": 1320230.0, + "step": 412 + }, + { + "epoch": 0.2112531969309463, + "grad_norm": 11.56860637664795, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8518421649932861, + "num_tokens": 1323752.0, + "step": 413 + }, + { + "epoch": 0.21176470588235294, + "grad_norm": 11.712552070617676, + "learning_rate": 1e-06, + "loss": 0.288, + "mean_token_accuracy": 0.8753690719604492, + "num_tokens": 1326756.0, + "step": 414 + }, + { + "epoch": 0.21227621483375958, + "grad_norm": 15.390015602111816, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.9027931690216064, + "num_tokens": 1330151.0, + "step": 415 + }, + { + "epoch": 0.21278772378516625, + "grad_norm": 14.863924026489258, + "learning_rate": 1e-06, + "loss": 0.4497, + "mean_token_accuracy": 0.8460139036178589, + "num_tokens": 1333512.0, + "step": 416 + }, + { + "epoch": 0.2132992327365729, + "grad_norm": 20.827604293823242, + "learning_rate": 1e-06, + "loss": 0.3852, + "mean_token_accuracy": 0.8590009212493896, + "num_tokens": 1336768.0, + "step": 417 + }, + { + "epoch": 0.21381074168797953, + "grad_norm": 11.287870407104492, + "learning_rate": 1e-06, + "loss": 0.2685, + "mean_token_accuracy": 0.8867309093475342, + "num_tokens": 1340692.0, + "step": 418 + }, + { + "epoch": 0.2143222506393862, + "grad_norm": 10.520566940307617, + "learning_rate": 1e-06, + "loss": 0.2712, + "mean_token_accuracy": 0.8852665424346924, + "num_tokens": 1344314.0, + "step": 419 + }, + { + "epoch": 0.21483375959079284, + "grad_norm": 14.710827827453613, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.870856761932373, + "num_tokens": 1347972.0, + "step": 420 + }, + { + "epoch": 0.21534526854219949, + "grad_norm": 13.986008644104004, + "learning_rate": 1e-06, + "loss": 0.2615, + "mean_token_accuracy": 0.9064996242523193, + "num_tokens": 1351573.0, + "step": 421 + }, + { + "epoch": 0.21585677749360613, + "grad_norm": 12.31541919708252, + "learning_rate": 1e-06, + "loss": 0.2365, + "mean_token_accuracy": 0.9095855355262756, + "num_tokens": 1353987.0, + "step": 422 + }, + { + "epoch": 0.2163682864450128, + "grad_norm": 16.50943374633789, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8541903495788574, + "num_tokens": 1357020.0, + "step": 423 + }, + { + "epoch": 0.21687979539641944, + "grad_norm": 11.994645118713379, + "learning_rate": 1e-06, + "loss": 0.2776, + "mean_token_accuracy": 0.8968881368637085, + "num_tokens": 1360308.0, + "step": 424 + }, + { + "epoch": 0.21739130434782608, + "grad_norm": 12.25501537322998, + "learning_rate": 1e-06, + "loss": 0.3579, + "mean_token_accuracy": 0.8549478054046631, + "num_tokens": 1364556.0, + "step": 425 + }, + { + "epoch": 0.21790281329923275, + "grad_norm": 14.481485366821289, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8685290813446045, + "num_tokens": 1367672.0, + "step": 426 + }, + { + "epoch": 0.2184143222506394, + "grad_norm": 14.678394317626953, + "learning_rate": 1e-06, + "loss": 0.3124, + "mean_token_accuracy": 0.8549331426620483, + "num_tokens": 1370834.0, + "step": 427 + }, + { + "epoch": 0.21892583120204603, + "grad_norm": 11.07945442199707, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.8742744326591492, + "num_tokens": 1373922.0, + "step": 428 + }, + { + "epoch": 0.21943734015345268, + "grad_norm": 15.761552810668945, + "learning_rate": 1e-06, + "loss": 0.2406, + "mean_token_accuracy": 0.890803337097168, + "num_tokens": 1377981.0, + "step": 429 + }, + { + "epoch": 0.21994884910485935, + "grad_norm": 13.952115058898926, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8586075305938721, + "num_tokens": 1380740.0, + "step": 430 + }, + { + "epoch": 0.220460358056266, + "grad_norm": 10.019248962402344, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8363426923751831, + "num_tokens": 1384326.0, + "step": 431 + }, + { + "epoch": 0.22097186700767263, + "grad_norm": 9.538802146911621, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.865118682384491, + "num_tokens": 1387326.0, + "step": 432 + }, + { + "epoch": 0.22148337595907927, + "grad_norm": 9.447925567626953, + "learning_rate": 1e-06, + "loss": 0.2493, + "mean_token_accuracy": 0.8870115280151367, + "num_tokens": 1390953.0, + "step": 433 + }, + { + "epoch": 0.22199488491048594, + "grad_norm": 13.742980003356934, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8671133518218994, + "num_tokens": 1394323.0, + "step": 434 + }, + { + "epoch": 0.22250639386189258, + "grad_norm": 11.75737476348877, + "learning_rate": 1e-06, + "loss": 0.3086, + "mean_token_accuracy": 0.8761314749717712, + "num_tokens": 1397612.0, + "step": 435 + }, + { + "epoch": 0.22301790281329922, + "grad_norm": 15.46864128112793, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8761159181594849, + "num_tokens": 1400326.0, + "step": 436 + }, + { + "epoch": 0.2235294117647059, + "grad_norm": 13.739323616027832, + "learning_rate": 1e-06, + "loss": 0.278, + "mean_token_accuracy": 0.8701079487800598, + "num_tokens": 1403477.0, + "step": 437 + }, + { + "epoch": 0.22404092071611253, + "grad_norm": 23.63274383544922, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8686922192573547, + "num_tokens": 1406316.0, + "step": 438 + }, + { + "epoch": 0.22455242966751918, + "grad_norm": 16.734752655029297, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8615848422050476, + "num_tokens": 1409468.0, + "step": 439 + }, + { + "epoch": 0.22506393861892582, + "grad_norm": 20.776731491088867, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8881409764289856, + "num_tokens": 1412303.0, + "step": 440 + }, + { + "epoch": 0.2255754475703325, + "grad_norm": 14.366899490356445, + "learning_rate": 1e-06, + "loss": 0.3531, + "mean_token_accuracy": 0.8812082409858704, + "num_tokens": 1415232.0, + "step": 441 + }, + { + "epoch": 0.22608695652173913, + "grad_norm": 19.69586753845215, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8848989009857178, + "num_tokens": 1418426.0, + "step": 442 + }, + { + "epoch": 0.22659846547314577, + "grad_norm": 16.53937530517578, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8974993824958801, + "num_tokens": 1421474.0, + "step": 443 + }, + { + "epoch": 0.22710997442455244, + "grad_norm": 14.044533729553223, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.877251386642456, + "num_tokens": 1424671.0, + "step": 444 + }, + { + "epoch": 0.22762148337595908, + "grad_norm": 11.964879035949707, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8909577131271362, + "num_tokens": 1428030.0, + "step": 445 + }, + { + "epoch": 0.22813299232736572, + "grad_norm": 8.606834411621094, + "learning_rate": 1e-06, + "loss": 0.1911, + "mean_token_accuracy": 0.9270161986351013, + "num_tokens": 1431898.0, + "step": 446 + }, + { + "epoch": 0.22864450127877237, + "grad_norm": 12.616013526916504, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8622230291366577, + "num_tokens": 1435602.0, + "step": 447 + }, + { + "epoch": 0.22915601023017904, + "grad_norm": 16.92999839782715, + "learning_rate": 1e-06, + "loss": 0.2665, + "mean_token_accuracy": 0.8962302207946777, + "num_tokens": 1438483.0, + "step": 448 + }, + { + "epoch": 0.22966751918158568, + "grad_norm": 18.951295852661133, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8553330898284912, + "num_tokens": 1441614.0, + "step": 449 + }, + { + "epoch": 0.23017902813299232, + "grad_norm": 11.028697967529297, + "learning_rate": 1e-06, + "loss": 0.2962, + "mean_token_accuracy": 0.9056555032730103, + "num_tokens": 1444679.0, + "step": 450 + }, + { + "epoch": 0.230690537084399, + "grad_norm": 13.839302062988281, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8684096336364746, + "num_tokens": 1447152.0, + "step": 451 + }, + { + "epoch": 0.23120204603580563, + "grad_norm": 9.234912872314453, + "learning_rate": 1e-06, + "loss": 0.2644, + "mean_token_accuracy": 0.8976245522499084, + "num_tokens": 1450524.0, + "step": 452 + }, + { + "epoch": 0.23171355498721227, + "grad_norm": 12.04763412475586, + "learning_rate": 1e-06, + "loss": 0.2921, + "mean_token_accuracy": 0.8709707260131836, + "num_tokens": 1454176.0, + "step": 453 + }, + { + "epoch": 0.23222506393861891, + "grad_norm": 16.77342987060547, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8795716762542725, + "num_tokens": 1457086.0, + "step": 454 + }, + { + "epoch": 0.23273657289002558, + "grad_norm": 14.852314949035645, + "learning_rate": 1e-06, + "loss": 0.2745, + "mean_token_accuracy": 0.8978085517883301, + "num_tokens": 1459893.0, + "step": 455 + }, + { + "epoch": 0.23324808184143223, + "grad_norm": 10.250783920288086, + "learning_rate": 1e-06, + "loss": 0.2952, + "mean_token_accuracy": 0.8699127435684204, + "num_tokens": 1463399.0, + "step": 456 + }, + { + "epoch": 0.23375959079283887, + "grad_norm": 15.141324996948242, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8635789752006531, + "num_tokens": 1466335.0, + "step": 457 + }, + { + "epoch": 0.23427109974424554, + "grad_norm": 17.413454055786133, + "learning_rate": 1e-06, + "loss": 0.4346, + "mean_token_accuracy": 0.8402832746505737, + "num_tokens": 1470123.0, + "step": 458 + }, + { + "epoch": 0.23478260869565218, + "grad_norm": 10.212882041931152, + "learning_rate": 1e-06, + "loss": 0.3496, + "mean_token_accuracy": 0.8769175410270691, + "num_tokens": 1473765.0, + "step": 459 + }, + { + "epoch": 0.23529411764705882, + "grad_norm": 9.934209823608398, + "learning_rate": 1e-06, + "loss": 0.2783, + "mean_token_accuracy": 0.9020060896873474, + "num_tokens": 1476807.0, + "step": 460 + }, + { + "epoch": 0.23580562659846546, + "grad_norm": 12.605216026306152, + "learning_rate": 1e-06, + "loss": 0.4015, + "mean_token_accuracy": 0.8394349813461304, + "num_tokens": 1479692.0, + "step": 461 + }, + { + "epoch": 0.23631713554987213, + "grad_norm": 16.679914474487305, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8705731630325317, + "num_tokens": 1482733.0, + "step": 462 + }, + { + "epoch": 0.23682864450127877, + "grad_norm": 12.509939193725586, + "learning_rate": 1e-06, + "loss": 0.3194, + "mean_token_accuracy": 0.8739874362945557, + "num_tokens": 1486316.0, + "step": 463 + }, + { + "epoch": 0.23734015345268542, + "grad_norm": 9.711560249328613, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.8824620842933655, + "num_tokens": 1489680.0, + "step": 464 + }, + { + "epoch": 0.23785166240409208, + "grad_norm": 17.28227424621582, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8618347644805908, + "num_tokens": 1492636.0, + "step": 465 + }, + { + "epoch": 0.23836317135549873, + "grad_norm": 8.32713508605957, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.8834016919136047, + "num_tokens": 1496961.0, + "step": 466 + }, + { + "epoch": 0.23887468030690537, + "grad_norm": 16.88022804260254, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8682305812835693, + "num_tokens": 1500528.0, + "step": 467 + }, + { + "epoch": 0.239386189258312, + "grad_norm": 12.504081726074219, + "learning_rate": 1e-06, + "loss": 0.2801, + "mean_token_accuracy": 0.8899301290512085, + "num_tokens": 1503327.0, + "step": 468 + }, + { + "epoch": 0.23989769820971868, + "grad_norm": 11.413683891296387, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8684209585189819, + "num_tokens": 1506304.0, + "step": 469 + }, + { + "epoch": 0.24040920716112532, + "grad_norm": 12.146827697753906, + "learning_rate": 1e-06, + "loss": 0.3823, + "mean_token_accuracy": 0.8475160002708435, + "num_tokens": 1509764.0, + "step": 470 + }, + { + "epoch": 0.24092071611253196, + "grad_norm": 16.067367553710938, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8572134375572205, + "num_tokens": 1512895.0, + "step": 471 + }, + { + "epoch": 0.24143222506393863, + "grad_norm": 13.39111614227295, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8730241060256958, + "num_tokens": 1515606.0, + "step": 472 + }, + { + "epoch": 0.24194373401534527, + "grad_norm": 7.8253607749938965, + "learning_rate": 1e-06, + "loss": 0.1998, + "mean_token_accuracy": 0.9047308564186096, + "num_tokens": 1519254.0, + "step": 473 + }, + { + "epoch": 0.24245524296675192, + "grad_norm": 11.00867748260498, + "learning_rate": 1e-06, + "loss": 0.2532, + "mean_token_accuracy": 0.8768486976623535, + "num_tokens": 1522402.0, + "step": 474 + }, + { + "epoch": 0.24296675191815856, + "grad_norm": 16.444007873535156, + "learning_rate": 1e-06, + "loss": 0.4143, + "mean_token_accuracy": 0.8497911691665649, + "num_tokens": 1525076.0, + "step": 475 + }, + { + "epoch": 0.24347826086956523, + "grad_norm": 13.398669242858887, + "learning_rate": 1e-06, + "loss": 0.3188, + "mean_token_accuracy": 0.8710530400276184, + "num_tokens": 1527397.0, + "step": 476 + }, + { + "epoch": 0.24398976982097187, + "grad_norm": 11.21585750579834, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.8760433197021484, + "num_tokens": 1530667.0, + "step": 477 + }, + { + "epoch": 0.2445012787723785, + "grad_norm": 16.914581298828125, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8970941305160522, + "num_tokens": 1534135.0, + "step": 478 + }, + { + "epoch": 0.24501278772378515, + "grad_norm": 11.17978286743164, + "learning_rate": 1e-06, + "loss": 0.2851, + "mean_token_accuracy": 0.8685699701309204, + "num_tokens": 1537234.0, + "step": 479 + }, + { + "epoch": 0.24552429667519182, + "grad_norm": 15.762043952941895, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8469974994659424, + "num_tokens": 1539804.0, + "step": 480 + }, + { + "epoch": 0.24603580562659846, + "grad_norm": 11.493070602416992, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.875587522983551, + "num_tokens": 1543437.0, + "step": 481 + }, + { + "epoch": 0.2465473145780051, + "grad_norm": 12.70046329498291, + "learning_rate": 1e-06, + "loss": 0.5002, + "mean_token_accuracy": 0.8026427030563354, + "num_tokens": 1546679.0, + "step": 482 + }, + { + "epoch": 0.24705882352941178, + "grad_norm": 10.457542419433594, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8936280012130737, + "num_tokens": 1550256.0, + "step": 483 + }, + { + "epoch": 0.24757033248081842, + "grad_norm": 11.865777969360352, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8598408102989197, + "num_tokens": 1554470.0, + "step": 484 + }, + { + "epoch": 0.24808184143222506, + "grad_norm": 14.767586708068848, + "learning_rate": 1e-06, + "loss": 0.3479, + "mean_token_accuracy": 0.8797599077224731, + "num_tokens": 1557342.0, + "step": 485 + }, + { + "epoch": 0.2485933503836317, + "grad_norm": 8.808748245239258, + "learning_rate": 1e-06, + "loss": 0.2634, + "mean_token_accuracy": 0.8947266340255737, + "num_tokens": 1561303.0, + "step": 486 + }, + { + "epoch": 0.24910485933503837, + "grad_norm": 14.84607982635498, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8851885795593262, + "num_tokens": 1564093.0, + "step": 487 + }, + { + "epoch": 0.249616368286445, + "grad_norm": 16.077674865722656, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.8916463851928711, + "num_tokens": 1566697.0, + "step": 488 + }, + { + "epoch": 0.2501278772378517, + "grad_norm": 10.433249473571777, + "learning_rate": 1e-06, + "loss": 0.2388, + "mean_token_accuracy": 0.8978537321090698, + "num_tokens": 1569304.0, + "step": 489 + }, + { + "epoch": 0.2506393861892583, + "grad_norm": 17.980148315429688, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8445977568626404, + "num_tokens": 1572198.0, + "step": 490 + }, + { + "epoch": 0.25115089514066496, + "grad_norm": 14.962204933166504, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8606356382369995, + "num_tokens": 1575452.0, + "step": 491 + }, + { + "epoch": 0.25166240409207163, + "grad_norm": 15.775679588317871, + "learning_rate": 1e-06, + "loss": 0.394, + "mean_token_accuracy": 0.8578532934188843, + "num_tokens": 1578403.0, + "step": 492 + }, + { + "epoch": 0.25217391304347825, + "grad_norm": 10.059922218322754, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8876512050628662, + "num_tokens": 1582193.0, + "step": 493 + }, + { + "epoch": 0.2526854219948849, + "grad_norm": 6.27842378616333, + "learning_rate": 1e-06, + "loss": 0.229, + "mean_token_accuracy": 0.9039496183395386, + "num_tokens": 1587105.0, + "step": 494 + }, + { + "epoch": 0.2531969309462916, + "grad_norm": 15.793968200683594, + "learning_rate": 1e-06, + "loss": 0.3622, + "mean_token_accuracy": 0.8589081764221191, + "num_tokens": 1590412.0, + "step": 495 + }, + { + "epoch": 0.2537084398976982, + "grad_norm": 15.723666191101074, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8805711269378662, + "num_tokens": 1592977.0, + "step": 496 + }, + { + "epoch": 0.25421994884910487, + "grad_norm": 17.8271541595459, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8794233798980713, + "num_tokens": 1595626.0, + "step": 497 + }, + { + "epoch": 0.2547314578005115, + "grad_norm": 14.178645133972168, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.8473532199859619, + "num_tokens": 1599169.0, + "step": 498 + }, + { + "epoch": 0.25524296675191815, + "grad_norm": 12.251031875610352, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8479810953140259, + "num_tokens": 1602997.0, + "step": 499 + }, + { + "epoch": 0.2557544757033248, + "grad_norm": 18.068927764892578, + "learning_rate": 1e-06, + "loss": 0.4308, + "mean_token_accuracy": 0.8217931985855103, + "num_tokens": 1606030.0, + "step": 500 + }, + { + "epoch": 0.25626598465473144, + "grad_norm": 9.698293685913086, + "learning_rate": 1e-06, + "loss": 0.2681, + "mean_token_accuracy": 0.8915220499038696, + "num_tokens": 1609573.0, + "step": 501 + }, + { + "epoch": 0.2567774936061381, + "grad_norm": 9.931620597839355, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8348516821861267, + "num_tokens": 1612991.0, + "step": 502 + }, + { + "epoch": 0.2572890025575448, + "grad_norm": 7.564299583435059, + "learning_rate": 1e-06, + "loss": 0.2398, + "mean_token_accuracy": 0.8976997137069702, + "num_tokens": 1616788.0, + "step": 503 + }, + { + "epoch": 0.2578005115089514, + "grad_norm": 11.853876113891602, + "learning_rate": 1e-06, + "loss": 0.284, + "mean_token_accuracy": 0.9093705415725708, + "num_tokens": 1620970.0, + "step": 504 + }, + { + "epoch": 0.25831202046035806, + "grad_norm": 14.889935493469238, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.8827328085899353, + "num_tokens": 1623602.0, + "step": 505 + }, + { + "epoch": 0.25882352941176473, + "grad_norm": 11.13202953338623, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8732191920280457, + "num_tokens": 1626893.0, + "step": 506 + }, + { + "epoch": 0.25933503836317134, + "grad_norm": 15.156326293945312, + "learning_rate": 1e-06, + "loss": 0.2412, + "mean_token_accuracy": 0.8919957280158997, + "num_tokens": 1629908.0, + "step": 507 + }, + { + "epoch": 0.259846547314578, + "grad_norm": 14.01004695892334, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8871732950210571, + "num_tokens": 1632709.0, + "step": 508 + }, + { + "epoch": 0.2603580562659846, + "grad_norm": 15.685450553894043, + "learning_rate": 1e-06, + "loss": 0.3546, + "mean_token_accuracy": 0.8708140254020691, + "num_tokens": 1635813.0, + "step": 509 + }, + { + "epoch": 0.2608695652173913, + "grad_norm": 14.460762023925781, + "learning_rate": 1e-06, + "loss": 0.2845, + "mean_token_accuracy": 0.8868285417556763, + "num_tokens": 1638956.0, + "step": 510 + }, + { + "epoch": 0.26138107416879797, + "grad_norm": 15.371318817138672, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.9056193828582764, + "num_tokens": 1642094.0, + "step": 511 + }, + { + "epoch": 0.2618925831202046, + "grad_norm": 8.433154106140137, + "learning_rate": 1e-06, + "loss": 0.2639, + "mean_token_accuracy": 0.8598539233207703, + "num_tokens": 1645781.0, + "step": 512 + }, + { + "epoch": 0.26240409207161125, + "grad_norm": 12.271040916442871, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.8599205613136292, + "num_tokens": 1649093.0, + "step": 513 + }, + { + "epoch": 0.2629156010230179, + "grad_norm": 16.11471939086914, + "learning_rate": 1e-06, + "loss": 0.4022, + "mean_token_accuracy": 0.8500322699546814, + "num_tokens": 1652269.0, + "step": 514 + }, + { + "epoch": 0.26342710997442453, + "grad_norm": 17.338958740234375, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8597656488418579, + "num_tokens": 1655010.0, + "step": 515 + }, + { + "epoch": 0.2639386189258312, + "grad_norm": 10.713608741760254, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8525800704956055, + "num_tokens": 1658173.0, + "step": 516 + }, + { + "epoch": 0.2644501278772379, + "grad_norm": 13.806838989257812, + "learning_rate": 1e-06, + "loss": 0.3708, + "mean_token_accuracy": 0.8618186712265015, + "num_tokens": 1661311.0, + "step": 517 + }, + { + "epoch": 0.2649616368286445, + "grad_norm": 6.406898021697998, + "learning_rate": 1e-06, + "loss": 0.238, + "mean_token_accuracy": 0.9007923603057861, + "num_tokens": 1666345.0, + "step": 518 + }, + { + "epoch": 0.26547314578005116, + "grad_norm": 11.126029014587402, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.9074433445930481, + "num_tokens": 1669423.0, + "step": 519 + }, + { + "epoch": 0.2659846547314578, + "grad_norm": 13.881749153137207, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8850342035293579, + "num_tokens": 1673152.0, + "step": 520 + }, + { + "epoch": 0.26649616368286444, + "grad_norm": 13.974784851074219, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8647503852844238, + "num_tokens": 1676361.0, + "step": 521 + }, + { + "epoch": 0.2670076726342711, + "grad_norm": 11.033149719238281, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.9045165777206421, + "num_tokens": 1679481.0, + "step": 522 + }, + { + "epoch": 0.2675191815856777, + "grad_norm": 7.558069705963135, + "learning_rate": 1e-06, + "loss": 0.2323, + "mean_token_accuracy": 0.919087290763855, + "num_tokens": 1682870.0, + "step": 523 + }, + { + "epoch": 0.2680306905370844, + "grad_norm": 10.978789329528809, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8464416265487671, + "num_tokens": 1685218.0, + "step": 524 + }, + { + "epoch": 0.26854219948849106, + "grad_norm": 21.582948684692383, + "learning_rate": 1e-06, + "loss": 0.2413, + "mean_token_accuracy": 0.9144904017448425, + "num_tokens": 1688097.0, + "step": 525 + }, + { + "epoch": 0.2690537084398977, + "grad_norm": 12.07419490814209, + "learning_rate": 1e-06, + "loss": 0.2721, + "mean_token_accuracy": 0.8880397081375122, + "num_tokens": 1691422.0, + "step": 526 + }, + { + "epoch": 0.26956521739130435, + "grad_norm": 9.644283294677734, + "learning_rate": 1e-06, + "loss": 0.2895, + "mean_token_accuracy": 0.8597363233566284, + "num_tokens": 1694893.0, + "step": 527 + }, + { + "epoch": 0.270076726342711, + "grad_norm": 3.058785915374756, + "learning_rate": 1e-06, + "loss": 0.0645, + "mean_token_accuracy": 0.9267978072166443, + "num_tokens": 1700024.0, + "step": 528 + }, + { + "epoch": 0.27058823529411763, + "grad_norm": 9.278042793273926, + "learning_rate": 1e-06, + "loss": 0.1941, + "mean_token_accuracy": 0.910812258720398, + "num_tokens": 1703455.0, + "step": 529 + }, + { + "epoch": 0.2710997442455243, + "grad_norm": 14.794896125793457, + "learning_rate": 1e-06, + "loss": 0.3987, + "mean_token_accuracy": 0.853733241558075, + "num_tokens": 1706837.0, + "step": 530 + }, + { + "epoch": 0.27161125319693097, + "grad_norm": 17.562475204467773, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8675461411476135, + "num_tokens": 1709608.0, + "step": 531 + }, + { + "epoch": 0.2721227621483376, + "grad_norm": 20.2573299407959, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.8528096079826355, + "num_tokens": 1712563.0, + "step": 532 + }, + { + "epoch": 0.27263427109974425, + "grad_norm": 15.03721809387207, + "learning_rate": 1e-06, + "loss": 0.4052, + "mean_token_accuracy": 0.8682149648666382, + "num_tokens": 1715441.0, + "step": 533 + }, + { + "epoch": 0.2731457800511509, + "grad_norm": 8.007268905639648, + "learning_rate": 1e-06, + "loss": 0.2549, + "mean_token_accuracy": 0.8959417343139648, + "num_tokens": 1719856.0, + "step": 534 + }, + { + "epoch": 0.27365728900255754, + "grad_norm": 13.329854011535645, + "learning_rate": 1e-06, + "loss": 0.3926, + "mean_token_accuracy": 0.8404567241668701, + "num_tokens": 1722455.0, + "step": 535 + }, + { + "epoch": 0.2741687979539642, + "grad_norm": 12.101184844970703, + "learning_rate": 1e-06, + "loss": 0.2738, + "mean_token_accuracy": 0.8774422407150269, + "num_tokens": 1726163.0, + "step": 536 + }, + { + "epoch": 0.2746803069053708, + "grad_norm": 11.289080619812012, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8845313787460327, + "num_tokens": 1729884.0, + "step": 537 + }, + { + "epoch": 0.2751918158567775, + "grad_norm": 14.348565101623535, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8758071660995483, + "num_tokens": 1732868.0, + "step": 538 + }, + { + "epoch": 0.27570332480818416, + "grad_norm": 13.407430648803711, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.8824278116226196, + "num_tokens": 1735477.0, + "step": 539 + }, + { + "epoch": 0.27621483375959077, + "grad_norm": 14.135065078735352, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.883101761341095, + "num_tokens": 1738169.0, + "step": 540 + }, + { + "epoch": 0.27672634271099744, + "grad_norm": 16.163713455200195, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8753073215484619, + "num_tokens": 1741347.0, + "step": 541 + }, + { + "epoch": 0.2772378516624041, + "grad_norm": 13.915772438049316, + "learning_rate": 1e-06, + "loss": 0.3184, + "mean_token_accuracy": 0.8818732500076294, + "num_tokens": 1744050.0, + "step": 542 + }, + { + "epoch": 0.2777493606138107, + "grad_norm": 13.709809303283691, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8955281972885132, + "num_tokens": 1747250.0, + "step": 543 + }, + { + "epoch": 0.2782608695652174, + "grad_norm": 8.1741304397583, + "learning_rate": 1e-06, + "loss": 0.1912, + "mean_token_accuracy": 0.9043216705322266, + "num_tokens": 1751219.0, + "step": 544 + }, + { + "epoch": 0.27877237851662406, + "grad_norm": 16.17558479309082, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.855955958366394, + "num_tokens": 1753775.0, + "step": 545 + }, + { + "epoch": 0.2792838874680307, + "grad_norm": 7.268914222717285, + "learning_rate": 1e-06, + "loss": 0.2157, + "mean_token_accuracy": 0.8964036703109741, + "num_tokens": 1757696.0, + "step": 546 + }, + { + "epoch": 0.27979539641943735, + "grad_norm": 13.87177562713623, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8645429611206055, + "num_tokens": 1760480.0, + "step": 547 + }, + { + "epoch": 0.280306905370844, + "grad_norm": 10.937453269958496, + "learning_rate": 1e-06, + "loss": 0.3372, + "mean_token_accuracy": 0.8661988377571106, + "num_tokens": 1764125.0, + "step": 548 + }, + { + "epoch": 0.28081841432225063, + "grad_norm": 14.290433883666992, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8793525099754333, + "num_tokens": 1767419.0, + "step": 549 + }, + { + "epoch": 0.2813299232736573, + "grad_norm": 15.838776588439941, + "learning_rate": 1e-06, + "loss": 0.2587, + "mean_token_accuracy": 0.9121124744415283, + "num_tokens": 1770421.0, + "step": 550 + }, + { + "epoch": 0.2818414322250639, + "grad_norm": 17.494462966918945, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8689502477645874, + "num_tokens": 1773757.0, + "step": 551 + }, + { + "epoch": 0.2823529411764706, + "grad_norm": 12.45329475402832, + "learning_rate": 1e-06, + "loss": 0.2465, + "mean_token_accuracy": 0.909424364566803, + "num_tokens": 1776608.0, + "step": 552 + }, + { + "epoch": 0.28286445012787725, + "grad_norm": 9.564762115478516, + "learning_rate": 1e-06, + "loss": 0.2644, + "mean_token_accuracy": 0.8947610855102539, + "num_tokens": 1780824.0, + "step": 553 + }, + { + "epoch": 0.28337595907928387, + "grad_norm": 12.662968635559082, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8523430824279785, + "num_tokens": 1784809.0, + "step": 554 + }, + { + "epoch": 0.28388746803069054, + "grad_norm": 15.03979778289795, + "learning_rate": 1e-06, + "loss": 0.3977, + "mean_token_accuracy": 0.8528634905815125, + "num_tokens": 1787963.0, + "step": 555 + }, + { + "epoch": 0.2843989769820972, + "grad_norm": 11.830971717834473, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8822925686836243, + "num_tokens": 1791463.0, + "step": 556 + }, + { + "epoch": 0.2849104859335038, + "grad_norm": 12.698100090026855, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8938568830490112, + "num_tokens": 1794400.0, + "step": 557 + }, + { + "epoch": 0.2854219948849105, + "grad_norm": 16.633516311645508, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8727911710739136, + "num_tokens": 1797693.0, + "step": 558 + }, + { + "epoch": 0.28593350383631716, + "grad_norm": 7.550976276397705, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8777979612350464, + "num_tokens": 1801967.0, + "step": 559 + }, + { + "epoch": 0.2864450127877238, + "grad_norm": 12.202840805053711, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.9015238881111145, + "num_tokens": 1804840.0, + "step": 560 + }, + { + "epoch": 0.28695652173913044, + "grad_norm": 9.690847396850586, + "learning_rate": 1e-06, + "loss": 0.2205, + "mean_token_accuracy": 0.895723819732666, + "num_tokens": 1807824.0, + "step": 561 + }, + { + "epoch": 0.28746803069053706, + "grad_norm": 10.332216262817383, + "learning_rate": 1e-06, + "loss": 0.2873, + "mean_token_accuracy": 0.8916831612586975, + "num_tokens": 1811160.0, + "step": 562 + }, + { + "epoch": 0.2879795396419437, + "grad_norm": 9.033621788024902, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.881218433380127, + "num_tokens": 1814386.0, + "step": 563 + }, + { + "epoch": 0.2884910485933504, + "grad_norm": 10.510766983032227, + "learning_rate": 1e-06, + "loss": 0.3205, + "mean_token_accuracy": 0.8582417964935303, + "num_tokens": 1817855.0, + "step": 564 + }, + { + "epoch": 0.289002557544757, + "grad_norm": 10.843559265136719, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8803493976593018, + "num_tokens": 1821594.0, + "step": 565 + }, + { + "epoch": 0.2895140664961637, + "grad_norm": 10.990818977355957, + "learning_rate": 1e-06, + "loss": 0.2406, + "mean_token_accuracy": 0.8951648473739624, + "num_tokens": 1824391.0, + "step": 566 + }, + { + "epoch": 0.29002557544757035, + "grad_norm": 12.183963775634766, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8714807033538818, + "num_tokens": 1827518.0, + "step": 567 + }, + { + "epoch": 0.29053708439897696, + "grad_norm": 10.91976547241211, + "learning_rate": 1e-06, + "loss": 0.2884, + "mean_token_accuracy": 0.9021462202072144, + "num_tokens": 1830463.0, + "step": 568 + }, + { + "epoch": 0.29104859335038363, + "grad_norm": 9.653498649597168, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8714511394500732, + "num_tokens": 1834181.0, + "step": 569 + }, + { + "epoch": 0.2915601023017903, + "grad_norm": 12.505758285522461, + "learning_rate": 1e-06, + "loss": 0.3787, + "mean_token_accuracy": 0.8472919464111328, + "num_tokens": 1837171.0, + "step": 570 + }, + { + "epoch": 0.2920716112531969, + "grad_norm": 12.319246292114258, + "learning_rate": 1e-06, + "loss": 0.3231, + "mean_token_accuracy": 0.8813568353652954, + "num_tokens": 1839881.0, + "step": 571 + }, + { + "epoch": 0.2925831202046036, + "grad_norm": 20.728246688842773, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8847751617431641, + "num_tokens": 1843014.0, + "step": 572 + }, + { + "epoch": 0.29309462915601026, + "grad_norm": 10.614629745483398, + "learning_rate": 1e-06, + "loss": 0.2807, + "mean_token_accuracy": 0.8924933671951294, + "num_tokens": 1845595.0, + "step": 573 + }, + { + "epoch": 0.29360613810741687, + "grad_norm": 9.864729881286621, + "learning_rate": 1e-06, + "loss": 0.2643, + "mean_token_accuracy": 0.8961693048477173, + "num_tokens": 1848494.0, + "step": 574 + }, + { + "epoch": 0.29411764705882354, + "grad_norm": 17.995866775512695, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8251160383224487, + "num_tokens": 1851864.0, + "step": 575 + }, + { + "epoch": 0.29462915601023015, + "grad_norm": 13.086151123046875, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8653826713562012, + "num_tokens": 1855570.0, + "step": 576 + }, + { + "epoch": 0.2951406649616368, + "grad_norm": 12.077271461486816, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8523810505867004, + "num_tokens": 1858511.0, + "step": 577 + }, + { + "epoch": 0.2956521739130435, + "grad_norm": 16.645896911621094, + "learning_rate": 1e-06, + "loss": 0.4317, + "mean_token_accuracy": 0.8344398736953735, + "num_tokens": 1861414.0, + "step": 578 + }, + { + "epoch": 0.2961636828644501, + "grad_norm": 13.101356506347656, + "learning_rate": 1e-06, + "loss": 0.3286, + "mean_token_accuracy": 0.878421425819397, + "num_tokens": 1864035.0, + "step": 579 + }, + { + "epoch": 0.2966751918158568, + "grad_norm": 11.83151626586914, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8573206663131714, + "num_tokens": 1866962.0, + "step": 580 + }, + { + "epoch": 0.29718670076726345, + "grad_norm": 9.712897300720215, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8603866100311279, + "num_tokens": 1870848.0, + "step": 581 + }, + { + "epoch": 0.29769820971867006, + "grad_norm": 11.959774017333984, + "learning_rate": 1e-06, + "loss": 0.3882, + "mean_token_accuracy": 0.8346256017684937, + "num_tokens": 1874232.0, + "step": 582 + }, + { + "epoch": 0.29820971867007673, + "grad_norm": 12.704811096191406, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8587859869003296, + "num_tokens": 1877223.0, + "step": 583 + }, + { + "epoch": 0.2987212276214834, + "grad_norm": 11.330497741699219, + "learning_rate": 1e-06, + "loss": 0.3454, + "mean_token_accuracy": 0.8738391995429993, + "num_tokens": 1880447.0, + "step": 584 + }, + { + "epoch": 0.29923273657289, + "grad_norm": 10.615103721618652, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8891997933387756, + "num_tokens": 1883727.0, + "step": 585 + }, + { + "epoch": 0.2997442455242967, + "grad_norm": 10.719282150268555, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8751413822174072, + "num_tokens": 1886457.0, + "step": 586 + }, + { + "epoch": 0.30025575447570335, + "grad_norm": 11.265795707702637, + "learning_rate": 1e-06, + "loss": 0.3605, + "mean_token_accuracy": 0.8644086122512817, + "num_tokens": 1889480.0, + "step": 587 + }, + { + "epoch": 0.30076726342710997, + "grad_norm": 9.566096305847168, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8839631080627441, + "num_tokens": 1893578.0, + "step": 588 + }, + { + "epoch": 0.30127877237851663, + "grad_norm": 13.362357139587402, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8666911721229553, + "num_tokens": 1896424.0, + "step": 589 + }, + { + "epoch": 0.30179028132992325, + "grad_norm": 11.724580764770508, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.8677473068237305, + "num_tokens": 1900278.0, + "step": 590 + }, + { + "epoch": 0.3023017902813299, + "grad_norm": 10.563050270080566, + "learning_rate": 1e-06, + "loss": 0.2801, + "mean_token_accuracy": 0.8858944773674011, + "num_tokens": 1902761.0, + "step": 591 + }, + { + "epoch": 0.3028132992327366, + "grad_norm": 11.774596214294434, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8713995814323425, + "num_tokens": 1905965.0, + "step": 592 + }, + { + "epoch": 0.3033248081841432, + "grad_norm": 13.59554672241211, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8756445646286011, + "num_tokens": 1909024.0, + "step": 593 + }, + { + "epoch": 0.30383631713554987, + "grad_norm": 14.093494415283203, + "learning_rate": 1e-06, + "loss": 0.337, + "mean_token_accuracy": 0.8685897588729858, + "num_tokens": 1911867.0, + "step": 594 + }, + { + "epoch": 0.30434782608695654, + "grad_norm": 8.267583847045898, + "learning_rate": 1e-06, + "loss": 0.274, + "mean_token_accuracy": 0.8796796798706055, + "num_tokens": 1915525.0, + "step": 595 + }, + { + "epoch": 0.30485933503836316, + "grad_norm": 11.02059268951416, + "learning_rate": 1e-06, + "loss": 0.2852, + "mean_token_accuracy": 0.8656569719314575, + "num_tokens": 1919119.0, + "step": 596 + }, + { + "epoch": 0.3053708439897698, + "grad_norm": 17.64804458618164, + "learning_rate": 1e-06, + "loss": 0.3976, + "mean_token_accuracy": 0.8606032729148865, + "num_tokens": 1921851.0, + "step": 597 + }, + { + "epoch": 0.3058823529411765, + "grad_norm": 10.231327056884766, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8483633995056152, + "num_tokens": 1925289.0, + "step": 598 + }, + { + "epoch": 0.3063938618925831, + "grad_norm": 8.06957721710205, + "learning_rate": 1e-06, + "loss": 0.259, + "mean_token_accuracy": 0.8802536129951477, + "num_tokens": 1929066.0, + "step": 599 + }, + { + "epoch": 0.3069053708439898, + "grad_norm": 13.976158142089844, + "learning_rate": 1e-06, + "loss": 0.2007, + "mean_token_accuracy": 0.9195317625999451, + "num_tokens": 1932452.0, + "step": 600 + }, + { + "epoch": 0.30741687979539645, + "grad_norm": 15.250338554382324, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.860145092010498, + "num_tokens": 1935073.0, + "step": 601 + }, + { + "epoch": 0.30792838874680306, + "grad_norm": 10.535677909851074, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.892890214920044, + "num_tokens": 1938591.0, + "step": 602 + }, + { + "epoch": 0.30843989769820973, + "grad_norm": 12.838874816894531, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8848855495452881, + "num_tokens": 1941926.0, + "step": 603 + }, + { + "epoch": 0.30895140664961634, + "grad_norm": 12.910635948181152, + "learning_rate": 1e-06, + "loss": 0.3728, + "mean_token_accuracy": 0.8678164482116699, + "num_tokens": 1944771.0, + "step": 604 + }, + { + "epoch": 0.309462915601023, + "grad_norm": 12.991207122802734, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.8645254373550415, + "num_tokens": 1948124.0, + "step": 605 + }, + { + "epoch": 0.3099744245524297, + "grad_norm": 11.325337409973145, + "learning_rate": 1e-06, + "loss": 0.3324, + "mean_token_accuracy": 0.8721424341201782, + "num_tokens": 1950782.0, + "step": 606 + }, + { + "epoch": 0.3104859335038363, + "grad_norm": 13.492304801940918, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8791749477386475, + "num_tokens": 1954431.0, + "step": 607 + }, + { + "epoch": 0.31099744245524297, + "grad_norm": 7.587202072143555, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8395544290542603, + "num_tokens": 1957741.0, + "step": 608 + }, + { + "epoch": 0.31150895140664964, + "grad_norm": 13.962121963500977, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8442134857177734, + "num_tokens": 1960683.0, + "step": 609 + }, + { + "epoch": 0.31202046035805625, + "grad_norm": 10.227664947509766, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8867729902267456, + "num_tokens": 1964607.0, + "step": 610 + }, + { + "epoch": 0.3125319693094629, + "grad_norm": 12.546529769897461, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8962774872779846, + "num_tokens": 1967278.0, + "step": 611 + }, + { + "epoch": 0.3130434782608696, + "grad_norm": 15.813627243041992, + "learning_rate": 1e-06, + "loss": 0.3812, + "mean_token_accuracy": 0.8400591611862183, + "num_tokens": 1970426.0, + "step": 612 + }, + { + "epoch": 0.3135549872122762, + "grad_norm": 15.427966117858887, + "learning_rate": 1e-06, + "loss": 0.3863, + "mean_token_accuracy": 0.8335683345794678, + "num_tokens": 1973135.0, + "step": 613 + }, + { + "epoch": 0.3140664961636829, + "grad_norm": 8.984301567077637, + "learning_rate": 1e-06, + "loss": 0.2559, + "mean_token_accuracy": 0.9082856178283691, + "num_tokens": 1976805.0, + "step": 614 + }, + { + "epoch": 0.3145780051150895, + "grad_norm": 12.15277099609375, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.9297782182693481, + "num_tokens": 1980097.0, + "step": 615 + }, + { + "epoch": 0.31508951406649616, + "grad_norm": 16.425283432006836, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8583484888076782, + "num_tokens": 1982961.0, + "step": 616 + }, + { + "epoch": 0.3156010230179028, + "grad_norm": 9.599771499633789, + "learning_rate": 1e-06, + "loss": 0.2812, + "mean_token_accuracy": 0.890623152256012, + "num_tokens": 1986495.0, + "step": 617 + }, + { + "epoch": 0.31611253196930944, + "grad_norm": 13.027787208557129, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8950393199920654, + "num_tokens": 1989330.0, + "step": 618 + }, + { + "epoch": 0.3166240409207161, + "grad_norm": 13.388157844543457, + "learning_rate": 1e-06, + "loss": 0.4255, + "mean_token_accuracy": 0.8410759568214417, + "num_tokens": 1992979.0, + "step": 619 + }, + { + "epoch": 0.3171355498721228, + "grad_norm": 14.199203491210938, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.831788182258606, + "num_tokens": 1996215.0, + "step": 620 + }, + { + "epoch": 0.3176470588235294, + "grad_norm": 15.732568740844727, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.875708818435669, + "num_tokens": 1999475.0, + "step": 621 + }, + { + "epoch": 0.31815856777493606, + "grad_norm": 15.186005592346191, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8703300952911377, + "num_tokens": 2002445.0, + "step": 622 + }, + { + "epoch": 0.31867007672634273, + "grad_norm": 12.316145896911621, + "learning_rate": 1e-06, + "loss": 0.3614, + "mean_token_accuracy": 0.8396055698394775, + "num_tokens": 2005331.0, + "step": 623 + }, + { + "epoch": 0.31918158567774935, + "grad_norm": 11.40450668334961, + "learning_rate": 1e-06, + "loss": 0.3685, + "mean_token_accuracy": 0.8511098623275757, + "num_tokens": 2008077.0, + "step": 624 + }, + { + "epoch": 0.319693094629156, + "grad_norm": 12.02751350402832, + "learning_rate": 1e-06, + "loss": 0.2182, + "mean_token_accuracy": 0.9060527086257935, + "num_tokens": 2010805.0, + "step": 625 + }, + { + "epoch": 0.3202046035805627, + "grad_norm": 12.127095222473145, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8573061227798462, + "num_tokens": 2014238.0, + "step": 626 + }, + { + "epoch": 0.3207161125319693, + "grad_norm": 15.456155776977539, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.8500684499740601, + "num_tokens": 2016761.0, + "step": 627 + }, + { + "epoch": 0.32122762148337597, + "grad_norm": 9.009317398071289, + "learning_rate": 1e-06, + "loss": 0.3457, + "mean_token_accuracy": 0.8750191926956177, + "num_tokens": 2020836.0, + "step": 628 + }, + { + "epoch": 0.3217391304347826, + "grad_norm": 8.267858505249023, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8762250542640686, + "num_tokens": 2024879.0, + "step": 629 + }, + { + "epoch": 0.32225063938618925, + "grad_norm": 13.861729621887207, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.864265501499176, + "num_tokens": 2027673.0, + "step": 630 + }, + { + "epoch": 0.3227621483375959, + "grad_norm": 13.215876579284668, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8821467757225037, + "num_tokens": 2030762.0, + "step": 631 + }, + { + "epoch": 0.32327365728900254, + "grad_norm": 10.578253746032715, + "learning_rate": 1e-06, + "loss": 0.3113, + "mean_token_accuracy": 0.8613710999488831, + "num_tokens": 2034823.0, + "step": 632 + }, + { + "epoch": 0.3237851662404092, + "grad_norm": 12.520264625549316, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8807361721992493, + "num_tokens": 2037783.0, + "step": 633 + }, + { + "epoch": 0.3242966751918159, + "grad_norm": 9.286970138549805, + "learning_rate": 1e-06, + "loss": 0.2243, + "mean_token_accuracy": 0.8985353708267212, + "num_tokens": 2041019.0, + "step": 634 + }, + { + "epoch": 0.3248081841432225, + "grad_norm": 15.534914016723633, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8743188381195068, + "num_tokens": 2044460.0, + "step": 635 + }, + { + "epoch": 0.32531969309462916, + "grad_norm": 12.45059585571289, + "learning_rate": 1e-06, + "loss": 0.3711, + "mean_token_accuracy": 0.850994348526001, + "num_tokens": 2047082.0, + "step": 636 + }, + { + "epoch": 0.32583120204603583, + "grad_norm": 11.383479118347168, + "learning_rate": 1e-06, + "loss": 0.2294, + "mean_token_accuracy": 0.8816633224487305, + "num_tokens": 2050136.0, + "step": 637 + }, + { + "epoch": 0.32634271099744244, + "grad_norm": 7.580070972442627, + "learning_rate": 1e-06, + "loss": 0.2101, + "mean_token_accuracy": 0.8913281559944153, + "num_tokens": 2054053.0, + "step": 638 + }, + { + "epoch": 0.3268542199488491, + "grad_norm": 13.371691703796387, + "learning_rate": 1e-06, + "loss": 0.3251, + "mean_token_accuracy": 0.8846929669380188, + "num_tokens": 2057659.0, + "step": 639 + }, + { + "epoch": 0.3273657289002558, + "grad_norm": 14.87271785736084, + "learning_rate": 1e-06, + "loss": 0.3621, + "mean_token_accuracy": 0.8615937829017639, + "num_tokens": 2060419.0, + "step": 640 + }, + { + "epoch": 0.3278772378516624, + "grad_norm": 14.473946571350098, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.8563730120658875, + "num_tokens": 2063365.0, + "step": 641 + }, + { + "epoch": 0.32838874680306906, + "grad_norm": 13.186681747436523, + "learning_rate": 1e-06, + "loss": 0.3913, + "mean_token_accuracy": 0.8645204901695251, + "num_tokens": 2066969.0, + "step": 642 + }, + { + "epoch": 0.3289002557544757, + "grad_norm": 11.842896461486816, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8839818239212036, + "num_tokens": 2069685.0, + "step": 643 + }, + { + "epoch": 0.32941176470588235, + "grad_norm": 10.501236915588379, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.8731555342674255, + "num_tokens": 2073019.0, + "step": 644 + }, + { + "epoch": 0.329923273657289, + "grad_norm": 10.454621315002441, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8744878172874451, + "num_tokens": 2076364.0, + "step": 645 + }, + { + "epoch": 0.33043478260869563, + "grad_norm": 11.008440017700195, + "learning_rate": 1e-06, + "loss": 0.3075, + "mean_token_accuracy": 0.8836504817008972, + "num_tokens": 2079413.0, + "step": 646 + }, + { + "epoch": 0.3309462915601023, + "grad_norm": 12.137876510620117, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8667197823524475, + "num_tokens": 2082384.0, + "step": 647 + }, + { + "epoch": 0.33145780051150897, + "grad_norm": 10.723073959350586, + "learning_rate": 1e-06, + "loss": 0.2534, + "mean_token_accuracy": 0.8815286159515381, + "num_tokens": 2085719.0, + "step": 648 + }, + { + "epoch": 0.3319693094629156, + "grad_norm": 17.275957107543945, + "learning_rate": 1e-06, + "loss": 0.4619, + "mean_token_accuracy": 0.8559491634368896, + "num_tokens": 2088489.0, + "step": 649 + }, + { + "epoch": 0.33248081841432225, + "grad_norm": 8.870601654052734, + "learning_rate": 1e-06, + "loss": 0.3406, + "mean_token_accuracy": 0.8784782290458679, + "num_tokens": 2092146.0, + "step": 650 + }, + { + "epoch": 0.3329923273657289, + "grad_norm": 13.39759635925293, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8546271324157715, + "num_tokens": 2094953.0, + "step": 651 + }, + { + "epoch": 0.33350383631713554, + "grad_norm": 8.737892150878906, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8684357404708862, + "num_tokens": 2098898.0, + "step": 652 + }, + { + "epoch": 0.3340153452685422, + "grad_norm": 13.229005813598633, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8872843384742737, + "num_tokens": 2101877.0, + "step": 653 + }, + { + "epoch": 0.3345268542199488, + "grad_norm": 8.991033554077148, + "learning_rate": 1e-06, + "loss": 0.2113, + "mean_token_accuracy": 0.9028555750846863, + "num_tokens": 2104935.0, + "step": 654 + }, + { + "epoch": 0.3350383631713555, + "grad_norm": 13.443801879882812, + "learning_rate": 1e-06, + "loss": 0.3449, + "mean_token_accuracy": 0.8633553981781006, + "num_tokens": 2108034.0, + "step": 655 + }, + { + "epoch": 0.33554987212276216, + "grad_norm": 11.760781288146973, + "learning_rate": 1e-06, + "loss": 0.3253, + "mean_token_accuracy": 0.8811332583427429, + "num_tokens": 2110973.0, + "step": 656 + }, + { + "epoch": 0.3360613810741688, + "grad_norm": 10.31169605255127, + "learning_rate": 1e-06, + "loss": 0.2691, + "mean_token_accuracy": 0.8923220634460449, + "num_tokens": 2114129.0, + "step": 657 + }, + { + "epoch": 0.33657289002557544, + "grad_norm": 11.14243221282959, + "learning_rate": 1e-06, + "loss": 0.3753, + "mean_token_accuracy": 0.8514940738677979, + "num_tokens": 2117545.0, + "step": 658 + }, + { + "epoch": 0.3370843989769821, + "grad_norm": 14.96920394897461, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8405372500419617, + "num_tokens": 2120139.0, + "step": 659 + }, + { + "epoch": 0.3375959079283887, + "grad_norm": 13.12793254852295, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8711297512054443, + "num_tokens": 2123272.0, + "step": 660 + }, + { + "epoch": 0.3381074168797954, + "grad_norm": 13.517767906188965, + "learning_rate": 1e-06, + "loss": 0.2572, + "mean_token_accuracy": 0.8930150866508484, + "num_tokens": 2126198.0, + "step": 661 + }, + { + "epoch": 0.33861892583120207, + "grad_norm": 14.096695899963379, + "learning_rate": 1e-06, + "loss": 0.3369, + "mean_token_accuracy": 0.8714191913604736, + "num_tokens": 2128731.0, + "step": 662 + }, + { + "epoch": 0.3391304347826087, + "grad_norm": 14.324934005737305, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8678382039070129, + "num_tokens": 2131762.0, + "step": 663 + }, + { + "epoch": 0.33964194373401535, + "grad_norm": 11.344512939453125, + "learning_rate": 1e-06, + "loss": 0.3289, + "mean_token_accuracy": 0.8689365386962891, + "num_tokens": 2135014.0, + "step": 664 + }, + { + "epoch": 0.340153452685422, + "grad_norm": 15.046608924865723, + "learning_rate": 1e-06, + "loss": 0.3439, + "mean_token_accuracy": 0.8636367321014404, + "num_tokens": 2137793.0, + "step": 665 + }, + { + "epoch": 0.34066496163682863, + "grad_norm": 8.16331672668457, + "learning_rate": 1e-06, + "loss": 0.2493, + "mean_token_accuracy": 0.8923305869102478, + "num_tokens": 2142088.0, + "step": 666 + }, + { + "epoch": 0.3411764705882353, + "grad_norm": 12.093399047851562, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8876876831054688, + "num_tokens": 2145074.0, + "step": 667 + }, + { + "epoch": 0.3416879795396419, + "grad_norm": 8.93895149230957, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8744669556617737, + "num_tokens": 2149130.0, + "step": 668 + }, + { + "epoch": 0.3421994884910486, + "grad_norm": 12.39262580871582, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8557522296905518, + "num_tokens": 2152885.0, + "step": 669 + }, + { + "epoch": 0.34271099744245526, + "grad_norm": 12.953524589538574, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8703027963638306, + "num_tokens": 2155581.0, + "step": 670 + }, + { + "epoch": 0.34322250639386187, + "grad_norm": 9.828596115112305, + "learning_rate": 1e-06, + "loss": 0.2224, + "mean_token_accuracy": 0.9175544381141663, + "num_tokens": 2158534.0, + "step": 671 + }, + { + "epoch": 0.34373401534526854, + "grad_norm": 7.6670918464660645, + "learning_rate": 1e-06, + "loss": 0.2614, + "mean_token_accuracy": 0.8810169696807861, + "num_tokens": 2162258.0, + "step": 672 + }, + { + "epoch": 0.3442455242966752, + "grad_norm": 13.489228248596191, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.8605577349662781, + "num_tokens": 2165138.0, + "step": 673 + }, + { + "epoch": 0.3447570332480818, + "grad_norm": 13.604670524597168, + "learning_rate": 1e-06, + "loss": 0.4336, + "mean_token_accuracy": 0.8403220176696777, + "num_tokens": 2168539.0, + "step": 674 + }, + { + "epoch": 0.3452685421994885, + "grad_norm": 9.783918380737305, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8576656579971313, + "num_tokens": 2172180.0, + "step": 675 + }, + { + "epoch": 0.34578005115089516, + "grad_norm": 14.848692893981934, + "learning_rate": 1e-06, + "loss": 0.4436, + "mean_token_accuracy": 0.8414238691329956, + "num_tokens": 2174681.0, + "step": 676 + }, + { + "epoch": 0.3462915601023018, + "grad_norm": 8.042329788208008, + "learning_rate": 1e-06, + "loss": 0.2634, + "mean_token_accuracy": 0.9071872234344482, + "num_tokens": 2178519.0, + "step": 677 + }, + { + "epoch": 0.34680306905370845, + "grad_norm": 9.040654182434082, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.8795674443244934, + "num_tokens": 2182127.0, + "step": 678 + }, + { + "epoch": 0.3473145780051151, + "grad_norm": 14.326780319213867, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.8794478178024292, + "num_tokens": 2185454.0, + "step": 679 + }, + { + "epoch": 0.34782608695652173, + "grad_norm": 10.86933708190918, + "learning_rate": 1e-06, + "loss": 0.2928, + "mean_token_accuracy": 0.8998289108276367, + "num_tokens": 2188323.0, + "step": 680 + }, + { + "epoch": 0.3483375959079284, + "grad_norm": 12.352703094482422, + "learning_rate": 1e-06, + "loss": 0.3589, + "mean_token_accuracy": 0.8518187403678894, + "num_tokens": 2191730.0, + "step": 681 + }, + { + "epoch": 0.348849104859335, + "grad_norm": 9.603046417236328, + "learning_rate": 1e-06, + "loss": 0.3472, + "mean_token_accuracy": 0.8719912767410278, + "num_tokens": 2195131.0, + "step": 682 + }, + { + "epoch": 0.3493606138107417, + "grad_norm": 5.799684047698975, + "learning_rate": 1e-06, + "loss": 0.1128, + "mean_token_accuracy": 0.9291893243789673, + "num_tokens": 2199075.0, + "step": 683 + }, + { + "epoch": 0.34987212276214835, + "grad_norm": 12.349932670593262, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.9104875922203064, + "num_tokens": 2202070.0, + "step": 684 + }, + { + "epoch": 0.35038363171355497, + "grad_norm": 11.671791076660156, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8507446050643921, + "num_tokens": 2205099.0, + "step": 685 + }, + { + "epoch": 0.35089514066496164, + "grad_norm": 10.10104751586914, + "learning_rate": 1e-06, + "loss": 0.3, + "mean_token_accuracy": 0.8735082745552063, + "num_tokens": 2208172.0, + "step": 686 + }, + { + "epoch": 0.3514066496163683, + "grad_norm": 14.947441101074219, + "learning_rate": 1e-06, + "loss": 0.3196, + "mean_token_accuracy": 0.8651621341705322, + "num_tokens": 2211120.0, + "step": 687 + }, + { + "epoch": 0.3519181585677749, + "grad_norm": 8.292336463928223, + "learning_rate": 1e-06, + "loss": 0.2367, + "mean_token_accuracy": 0.8958795070648193, + "num_tokens": 2214964.0, + "step": 688 + }, + { + "epoch": 0.3524296675191816, + "grad_norm": 13.448694229125977, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8933272957801819, + "num_tokens": 2217732.0, + "step": 689 + }, + { + "epoch": 0.35294117647058826, + "grad_norm": 10.295429229736328, + "learning_rate": 1e-06, + "loss": 0.2252, + "mean_token_accuracy": 0.9141883850097656, + "num_tokens": 2220159.0, + "step": 690 + }, + { + "epoch": 0.35345268542199487, + "grad_norm": 10.293424606323242, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8603801727294922, + "num_tokens": 2224165.0, + "step": 691 + }, + { + "epoch": 0.35396419437340154, + "grad_norm": 14.172266960144043, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8708329200744629, + "num_tokens": 2226954.0, + "step": 692 + }, + { + "epoch": 0.3544757033248082, + "grad_norm": 12.12340259552002, + "learning_rate": 1e-06, + "loss": 0.334, + "mean_token_accuracy": 0.888018012046814, + "num_tokens": 2229560.0, + "step": 693 + }, + { + "epoch": 0.3549872122762148, + "grad_norm": 12.143366813659668, + "learning_rate": 1e-06, + "loss": 0.2593, + "mean_token_accuracy": 0.8978109359741211, + "num_tokens": 2232458.0, + "step": 694 + }, + { + "epoch": 0.3554987212276215, + "grad_norm": 12.367873191833496, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8932687044143677, + "num_tokens": 2235259.0, + "step": 695 + }, + { + "epoch": 0.3560102301790281, + "grad_norm": 12.239860534667969, + "learning_rate": 1e-06, + "loss": 0.3801, + "mean_token_accuracy": 0.8820523619651794, + "num_tokens": 2238654.0, + "step": 696 + }, + { + "epoch": 0.3565217391304348, + "grad_norm": 10.32213020324707, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8776949644088745, + "num_tokens": 2241459.0, + "step": 697 + }, + { + "epoch": 0.35703324808184145, + "grad_norm": 6.500899791717529, + "learning_rate": 1e-06, + "loss": 0.2715, + "mean_token_accuracy": 0.9132015705108643, + "num_tokens": 2246347.0, + "step": 698 + }, + { + "epoch": 0.35754475703324806, + "grad_norm": 14.089311599731445, + "learning_rate": 1e-06, + "loss": 0.4122, + "mean_token_accuracy": 0.8603634834289551, + "num_tokens": 2249320.0, + "step": 699 + }, + { + "epoch": 0.35805626598465473, + "grad_norm": 10.852264404296875, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.887044370174408, + "num_tokens": 2252922.0, + "step": 700 + }, + { + "epoch": 0.3585677749360614, + "grad_norm": 13.23531723022461, + "learning_rate": 1e-06, + "loss": 0.364, + "mean_token_accuracy": 0.8597767949104309, + "num_tokens": 2255675.0, + "step": 701 + }, + { + "epoch": 0.359079283887468, + "grad_norm": 9.899959564208984, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8405709266662598, + "num_tokens": 2259679.0, + "step": 702 + }, + { + "epoch": 0.3595907928388747, + "grad_norm": 13.546935081481934, + "learning_rate": 1e-06, + "loss": 0.3633, + "mean_token_accuracy": 0.8652259111404419, + "num_tokens": 2262828.0, + "step": 703 + }, + { + "epoch": 0.36010230179028135, + "grad_norm": 9.507216453552246, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.9004150629043579, + "num_tokens": 2266615.0, + "step": 704 + }, + { + "epoch": 0.36061381074168797, + "grad_norm": 11.591544151306152, + "learning_rate": 1e-06, + "loss": 0.408, + "mean_token_accuracy": 0.8525815010070801, + "num_tokens": 2269334.0, + "step": 705 + }, + { + "epoch": 0.36112531969309464, + "grad_norm": 10.368247985839844, + "learning_rate": 1e-06, + "loss": 0.2994, + "mean_token_accuracy": 0.8924715518951416, + "num_tokens": 2272588.0, + "step": 706 + }, + { + "epoch": 0.36163682864450125, + "grad_norm": 5.5975189208984375, + "learning_rate": 1e-06, + "loss": 0.1903, + "mean_token_accuracy": 0.8660679459571838, + "num_tokens": 2277493.0, + "step": 707 + }, + { + "epoch": 0.3621483375959079, + "grad_norm": 10.724620819091797, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.8811455965042114, + "num_tokens": 2280987.0, + "step": 708 + }, + { + "epoch": 0.3626598465473146, + "grad_norm": 10.761877059936523, + "learning_rate": 1e-06, + "loss": 0.3568, + "mean_token_accuracy": 0.8535566329956055, + "num_tokens": 2283385.0, + "step": 709 + }, + { + "epoch": 0.3631713554987212, + "grad_norm": 13.677326202392578, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8913857340812683, + "num_tokens": 2285759.0, + "step": 710 + }, + { + "epoch": 0.3636828644501279, + "grad_norm": 13.854697227478027, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.892447829246521, + "num_tokens": 2288706.0, + "step": 711 + }, + { + "epoch": 0.36419437340153454, + "grad_norm": 12.701376914978027, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8273634910583496, + "num_tokens": 2291658.0, + "step": 712 + }, + { + "epoch": 0.36470588235294116, + "grad_norm": 11.129890441894531, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.8710660934448242, + "num_tokens": 2295388.0, + "step": 713 + }, + { + "epoch": 0.3652173913043478, + "grad_norm": 15.367339134216309, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8598296046257019, + "num_tokens": 2298475.0, + "step": 714 + }, + { + "epoch": 0.3657289002557545, + "grad_norm": 13.291051864624023, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8742356300354004, + "num_tokens": 2301417.0, + "step": 715 + }, + { + "epoch": 0.3662404092071611, + "grad_norm": 10.165493965148926, + "learning_rate": 1e-06, + "loss": 0.2627, + "mean_token_accuracy": 0.8645349144935608, + "num_tokens": 2304252.0, + "step": 716 + }, + { + "epoch": 0.3667519181585678, + "grad_norm": 12.942008018493652, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8701609969139099, + "num_tokens": 2307562.0, + "step": 717 + }, + { + "epoch": 0.36726342710997445, + "grad_norm": 8.90731143951416, + "learning_rate": 1e-06, + "loss": 0.2553, + "mean_token_accuracy": 0.9007599949836731, + "num_tokens": 2310652.0, + "step": 718 + }, + { + "epoch": 0.36777493606138106, + "grad_norm": 11.554854393005371, + "learning_rate": 1e-06, + "loss": 0.3228, + "mean_token_accuracy": 0.8825085163116455, + "num_tokens": 2313941.0, + "step": 719 + }, + { + "epoch": 0.36828644501278773, + "grad_norm": 10.647043228149414, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8568165302276611, + "num_tokens": 2317621.0, + "step": 720 + }, + { + "epoch": 0.36879795396419435, + "grad_norm": 8.492293357849121, + "learning_rate": 1e-06, + "loss": 0.2213, + "mean_token_accuracy": 0.87404865026474, + "num_tokens": 2320361.0, + "step": 721 + }, + { + "epoch": 0.369309462915601, + "grad_norm": 10.665771484375, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8641471862792969, + "num_tokens": 2323988.0, + "step": 722 + }, + { + "epoch": 0.3698209718670077, + "grad_norm": 12.763575553894043, + "learning_rate": 1e-06, + "loss": 0.2656, + "mean_token_accuracy": 0.8925448060035706, + "num_tokens": 2326884.0, + "step": 723 + }, + { + "epoch": 0.3703324808184143, + "grad_norm": 13.695013999938965, + "learning_rate": 1e-06, + "loss": 0.3104, + "mean_token_accuracy": 0.8762590289115906, + "num_tokens": 2329909.0, + "step": 724 + }, + { + "epoch": 0.37084398976982097, + "grad_norm": 12.61507797241211, + "learning_rate": 1e-06, + "loss": 0.4552, + "mean_token_accuracy": 0.8545339107513428, + "num_tokens": 2332766.0, + "step": 725 + }, + { + "epoch": 0.37135549872122764, + "grad_norm": 15.761796951293945, + "learning_rate": 1e-06, + "loss": 0.371, + "mean_token_accuracy": 0.8597894906997681, + "num_tokens": 2336028.0, + "step": 726 + }, + { + "epoch": 0.37186700767263425, + "grad_norm": 12.9581880569458, + "learning_rate": 1e-06, + "loss": 0.4037, + "mean_token_accuracy": 0.8484875559806824, + "num_tokens": 2339444.0, + "step": 727 + }, + { + "epoch": 0.3723785166240409, + "grad_norm": 8.901715278625488, + "learning_rate": 1e-06, + "loss": 0.2186, + "mean_token_accuracy": 0.8878698945045471, + "num_tokens": 2342351.0, + "step": 728 + }, + { + "epoch": 0.3728900255754476, + "grad_norm": 12.614337921142578, + "learning_rate": 1e-06, + "loss": 0.2701, + "mean_token_accuracy": 0.8722273707389832, + "num_tokens": 2345551.0, + "step": 729 + }, + { + "epoch": 0.3734015345268542, + "grad_norm": 7.388822555541992, + "learning_rate": 1e-06, + "loss": 0.253, + "mean_token_accuracy": 0.884419322013855, + "num_tokens": 2350328.0, + "step": 730 + }, + { + "epoch": 0.3739130434782609, + "grad_norm": 12.443246841430664, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8692762851715088, + "num_tokens": 2353007.0, + "step": 731 + }, + { + "epoch": 0.37442455242966755, + "grad_norm": 11.858120918273926, + "learning_rate": 1e-06, + "loss": 0.3975, + "mean_token_accuracy": 0.8372101783752441, + "num_tokens": 2356421.0, + "step": 732 + }, + { + "epoch": 0.37493606138107416, + "grad_norm": 11.878129959106445, + "learning_rate": 1e-06, + "loss": 0.2412, + "mean_token_accuracy": 0.8997386693954468, + "num_tokens": 2358694.0, + "step": 733 + }, + { + "epoch": 0.37544757033248083, + "grad_norm": 7.357983112335205, + "learning_rate": 1e-06, + "loss": 0.2183, + "mean_token_accuracy": 0.879306972026825, + "num_tokens": 2362280.0, + "step": 734 + }, + { + "epoch": 0.37595907928388744, + "grad_norm": 10.160401344299316, + "learning_rate": 1e-06, + "loss": 0.3441, + "mean_token_accuracy": 0.8368372321128845, + "num_tokens": 2365270.0, + "step": 735 + }, + { + "epoch": 0.3764705882352941, + "grad_norm": 8.196702003479004, + "learning_rate": 1e-06, + "loss": 0.2235, + "mean_token_accuracy": 0.9150710105895996, + "num_tokens": 2368289.0, + "step": 736 + }, + { + "epoch": 0.3769820971867008, + "grad_norm": 9.725262641906738, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8474714756011963, + "num_tokens": 2370946.0, + "step": 737 + }, + { + "epoch": 0.3774936061381074, + "grad_norm": 15.466564178466797, + "learning_rate": 1e-06, + "loss": 0.4301, + "mean_token_accuracy": 0.8345612287521362, + "num_tokens": 2374080.0, + "step": 738 + }, + { + "epoch": 0.37800511508951407, + "grad_norm": 14.184393882751465, + "learning_rate": 1e-06, + "loss": 0.4584, + "mean_token_accuracy": 0.8245385885238647, + "num_tokens": 2377610.0, + "step": 739 + }, + { + "epoch": 0.37851662404092073, + "grad_norm": 7.777725696563721, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.9107701778411865, + "num_tokens": 2381344.0, + "step": 740 + }, + { + "epoch": 0.37902813299232735, + "grad_norm": 14.431168556213379, + "learning_rate": 1e-06, + "loss": 0.3776, + "mean_token_accuracy": 0.8606500625610352, + "num_tokens": 2384385.0, + "step": 741 + }, + { + "epoch": 0.379539641943734, + "grad_norm": 13.32978630065918, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8744112253189087, + "num_tokens": 2387576.0, + "step": 742 + }, + { + "epoch": 0.3800511508951407, + "grad_norm": 12.70350170135498, + "learning_rate": 1e-06, + "loss": 0.3407, + "mean_token_accuracy": 0.8739145994186401, + "num_tokens": 2390675.0, + "step": 743 + }, + { + "epoch": 0.3805626598465473, + "grad_norm": 7.477239608764648, + "learning_rate": 1e-06, + "loss": 0.208, + "mean_token_accuracy": 0.9044224619865417, + "num_tokens": 2394007.0, + "step": 744 + }, + { + "epoch": 0.38107416879795397, + "grad_norm": 13.10552978515625, + "learning_rate": 1e-06, + "loss": 0.3392, + "mean_token_accuracy": 0.8659080266952515, + "num_tokens": 2396605.0, + "step": 745 + }, + { + "epoch": 0.38158567774936064, + "grad_norm": 11.786264419555664, + "learning_rate": 1e-06, + "loss": 0.2895, + "mean_token_accuracy": 0.882695734500885, + "num_tokens": 2399458.0, + "step": 746 + }, + { + "epoch": 0.38209718670076726, + "grad_norm": 12.335841178894043, + "learning_rate": 1e-06, + "loss": 0.3395, + "mean_token_accuracy": 0.8755210638046265, + "num_tokens": 2402938.0, + "step": 747 + }, + { + "epoch": 0.3826086956521739, + "grad_norm": 13.163167953491211, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8517718315124512, + "num_tokens": 2405598.0, + "step": 748 + }, + { + "epoch": 0.38312020460358054, + "grad_norm": 12.19955062866211, + "learning_rate": 1e-06, + "loss": 0.3545, + "mean_token_accuracy": 0.8719124794006348, + "num_tokens": 2408019.0, + "step": 749 + }, + { + "epoch": 0.3836317135549872, + "grad_norm": 16.780946731567383, + "learning_rate": 1e-06, + "loss": 0.3765, + "mean_token_accuracy": 0.8597394227981567, + "num_tokens": 2411254.0, + "step": 750 + }, + { + "epoch": 0.3841432225063939, + "grad_norm": 9.877854347229004, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8752623796463013, + "num_tokens": 2414920.0, + "step": 751 + }, + { + "epoch": 0.3846547314578005, + "grad_norm": 10.321187019348145, + "learning_rate": 1e-06, + "loss": 0.3352, + "mean_token_accuracy": 0.8496031761169434, + "num_tokens": 2418510.0, + "step": 752 + }, + { + "epoch": 0.38516624040920716, + "grad_norm": 14.407057762145996, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8758354187011719, + "num_tokens": 2421271.0, + "step": 753 + }, + { + "epoch": 0.38567774936061383, + "grad_norm": 6.463123798370361, + "learning_rate": 1e-06, + "loss": 0.2476, + "mean_token_accuracy": 0.8968883752822876, + "num_tokens": 2426072.0, + "step": 754 + }, + { + "epoch": 0.38618925831202044, + "grad_norm": 14.849924087524414, + "learning_rate": 1e-06, + "loss": 0.4076, + "mean_token_accuracy": 0.834703266620636, + "num_tokens": 2428259.0, + "step": 755 + }, + { + "epoch": 0.3867007672634271, + "grad_norm": 13.747321128845215, + "learning_rate": 1e-06, + "loss": 0.3917, + "mean_token_accuracy": 0.8328269124031067, + "num_tokens": 2431596.0, + "step": 756 + }, + { + "epoch": 0.3872122762148338, + "grad_norm": 11.587017059326172, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8909512162208557, + "num_tokens": 2434099.0, + "step": 757 + }, + { + "epoch": 0.3877237851662404, + "grad_norm": 13.084529876708984, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.861039400100708, + "num_tokens": 2436646.0, + "step": 758 + }, + { + "epoch": 0.38823529411764707, + "grad_norm": 11.537817001342773, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8442738652229309, + "num_tokens": 2440282.0, + "step": 759 + }, + { + "epoch": 0.3887468030690537, + "grad_norm": 16.296171188354492, + "learning_rate": 1e-06, + "loss": 0.4087, + "mean_token_accuracy": 0.8553575277328491, + "num_tokens": 2443125.0, + "step": 760 + }, + { + "epoch": 0.38925831202046035, + "grad_norm": 16.900365829467773, + "learning_rate": 1e-06, + "loss": 0.4398, + "mean_token_accuracy": 0.8216776251792908, + "num_tokens": 2445869.0, + "step": 761 + }, + { + "epoch": 0.389769820971867, + "grad_norm": 12.579811096191406, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8582453727722168, + "num_tokens": 2448914.0, + "step": 762 + }, + { + "epoch": 0.39028132992327363, + "grad_norm": 13.604323387145996, + "learning_rate": 1e-06, + "loss": 0.4224, + "mean_token_accuracy": 0.8586398363113403, + "num_tokens": 2452026.0, + "step": 763 + }, + { + "epoch": 0.3907928388746803, + "grad_norm": 11.887103080749512, + "learning_rate": 1e-06, + "loss": 0.367, + "mean_token_accuracy": 0.8715530037879944, + "num_tokens": 2455385.0, + "step": 764 + }, + { + "epoch": 0.391304347826087, + "grad_norm": 9.897648811340332, + "learning_rate": 1e-06, + "loss": 0.3273, + "mean_token_accuracy": 0.8498885035514832, + "num_tokens": 2458743.0, + "step": 765 + }, + { + "epoch": 0.3918158567774936, + "grad_norm": 12.73377513885498, + "learning_rate": 1e-06, + "loss": 0.42, + "mean_token_accuracy": 0.8483631610870361, + "num_tokens": 2461506.0, + "step": 766 + }, + { + "epoch": 0.39232736572890026, + "grad_norm": 8.710630416870117, + "learning_rate": 1e-06, + "loss": 0.2651, + "mean_token_accuracy": 0.8904064893722534, + "num_tokens": 2465317.0, + "step": 767 + }, + { + "epoch": 0.3928388746803069, + "grad_norm": 8.866681098937988, + "learning_rate": 1e-06, + "loss": 0.2173, + "mean_token_accuracy": 0.9057878851890564, + "num_tokens": 2468685.0, + "step": 768 + }, + { + "epoch": 0.39335038363171354, + "grad_norm": 11.989212989807129, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8676427602767944, + "num_tokens": 2471295.0, + "step": 769 + }, + { + "epoch": 0.3938618925831202, + "grad_norm": 12.270661354064941, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8581636548042297, + "num_tokens": 2474447.0, + "step": 770 + }, + { + "epoch": 0.3943734015345269, + "grad_norm": 11.872435569763184, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.8535171747207642, + "num_tokens": 2477763.0, + "step": 771 + }, + { + "epoch": 0.3948849104859335, + "grad_norm": 10.93403434753418, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.878096878528595, + "num_tokens": 2480697.0, + "step": 772 + }, + { + "epoch": 0.39539641943734016, + "grad_norm": 12.369491577148438, + "learning_rate": 1e-06, + "loss": 0.4554, + "mean_token_accuracy": 0.8336338996887207, + "num_tokens": 2483947.0, + "step": 773 + }, + { + "epoch": 0.3959079283887468, + "grad_norm": 10.077775955200195, + "learning_rate": 1e-06, + "loss": 0.302, + "mean_token_accuracy": 0.8826087713241577, + "num_tokens": 2487178.0, + "step": 774 + }, + { + "epoch": 0.39641943734015345, + "grad_norm": 10.537418365478516, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8913246393203735, + "num_tokens": 2490486.0, + "step": 775 + }, + { + "epoch": 0.3969309462915601, + "grad_norm": 12.395389556884766, + "learning_rate": 1e-06, + "loss": 0.395, + "mean_token_accuracy": 0.8413717150688171, + "num_tokens": 2494337.0, + "step": 776 + }, + { + "epoch": 0.39744245524296673, + "grad_norm": 7.862729549407959, + "learning_rate": 1e-06, + "loss": 0.2671, + "mean_token_accuracy": 0.8804572820663452, + "num_tokens": 2497599.0, + "step": 777 + }, + { + "epoch": 0.3979539641943734, + "grad_norm": 19.256956100463867, + "learning_rate": 1e-06, + "loss": 0.4257, + "mean_token_accuracy": 0.8355220556259155, + "num_tokens": 2500048.0, + "step": 778 + }, + { + "epoch": 0.39846547314578007, + "grad_norm": 12.24603271484375, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.855269193649292, + "num_tokens": 2503210.0, + "step": 779 + }, + { + "epoch": 0.3989769820971867, + "grad_norm": 10.7739839553833, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8494989275932312, + "num_tokens": 2506051.0, + "step": 780 + }, + { + "epoch": 0.39948849104859335, + "grad_norm": 7.798818111419678, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8783869743347168, + "num_tokens": 2510528.0, + "step": 781 + }, + { + "epoch": 0.4, + "grad_norm": 9.308228492736816, + "learning_rate": 1e-06, + "loss": 0.331, + "mean_token_accuracy": 0.886665940284729, + "num_tokens": 2513763.0, + "step": 782 + }, + { + "epoch": 0.40051150895140664, + "grad_norm": 8.57072925567627, + "learning_rate": 1e-06, + "loss": 0.2656, + "mean_token_accuracy": 0.8702093362808228, + "num_tokens": 2517356.0, + "step": 783 + }, + { + "epoch": 0.4010230179028133, + "grad_norm": 9.436847686767578, + "learning_rate": 1e-06, + "loss": 0.3839, + "mean_token_accuracy": 0.8598650693893433, + "num_tokens": 2520920.0, + "step": 784 + }, + { + "epoch": 0.40153452685422, + "grad_norm": 9.909207344055176, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.863916277885437, + "num_tokens": 2524762.0, + "step": 785 + }, + { + "epoch": 0.4020460358056266, + "grad_norm": 10.40085220336914, + "learning_rate": 1e-06, + "loss": 0.2541, + "mean_token_accuracy": 0.8913169503211975, + "num_tokens": 2527736.0, + "step": 786 + }, + { + "epoch": 0.40255754475703326, + "grad_norm": 10.712691307067871, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.900741457939148, + "num_tokens": 2530953.0, + "step": 787 + }, + { + "epoch": 0.4030690537084399, + "grad_norm": 8.397249221801758, + "learning_rate": 1e-06, + "loss": 0.2596, + "mean_token_accuracy": 0.8937783241271973, + "num_tokens": 2534677.0, + "step": 788 + }, + { + "epoch": 0.40358056265984654, + "grad_norm": 11.16357421875, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.8720211386680603, + "num_tokens": 2538084.0, + "step": 789 + }, + { + "epoch": 0.4040920716112532, + "grad_norm": 11.56660270690918, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.87165367603302, + "num_tokens": 2541360.0, + "step": 790 + }, + { + "epoch": 0.4046035805626598, + "grad_norm": 8.451704025268555, + "learning_rate": 1e-06, + "loss": 0.2596, + "mean_token_accuracy": 0.8969577550888062, + "num_tokens": 2544581.0, + "step": 791 + }, + { + "epoch": 0.4051150895140665, + "grad_norm": 14.03293514251709, + "learning_rate": 1e-06, + "loss": 0.4288, + "mean_token_accuracy": 0.849962592124939, + "num_tokens": 2547348.0, + "step": 792 + }, + { + "epoch": 0.40562659846547316, + "grad_norm": 12.143998146057129, + "learning_rate": 1e-06, + "loss": 0.3519, + "mean_token_accuracy": 0.8704688549041748, + "num_tokens": 2550667.0, + "step": 793 + }, + { + "epoch": 0.4061381074168798, + "grad_norm": 7.551026821136475, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8937927484512329, + "num_tokens": 2554038.0, + "step": 794 + }, + { + "epoch": 0.40664961636828645, + "grad_norm": 10.736122131347656, + "learning_rate": 1e-06, + "loss": 0.3578, + "mean_token_accuracy": 0.8536503314971924, + "num_tokens": 2556448.0, + "step": 795 + }, + { + "epoch": 0.4071611253196931, + "grad_norm": 13.732342720031738, + "learning_rate": 1e-06, + "loss": 0.4237, + "mean_token_accuracy": 0.8294909596443176, + "num_tokens": 2559429.0, + "step": 796 + }, + { + "epoch": 0.40767263427109973, + "grad_norm": 13.858304023742676, + "learning_rate": 1e-06, + "loss": 0.4183, + "mean_token_accuracy": 0.8327470421791077, + "num_tokens": 2562707.0, + "step": 797 + }, + { + "epoch": 0.4081841432225064, + "grad_norm": 12.909836769104004, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8634666204452515, + "num_tokens": 2565507.0, + "step": 798 + }, + { + "epoch": 0.40869565217391307, + "grad_norm": 8.71622085571289, + "learning_rate": 1e-06, + "loss": 0.2585, + "mean_token_accuracy": 0.8661707043647766, + "num_tokens": 2569551.0, + "step": 799 + }, + { + "epoch": 0.4092071611253197, + "grad_norm": 6.661313056945801, + "learning_rate": 1e-06, + "loss": 0.2307, + "mean_token_accuracy": 0.9000266194343567, + "num_tokens": 2573569.0, + "step": 800 + }, + { + "epoch": 0.40971867007672635, + "grad_norm": 14.589411735534668, + "learning_rate": 1e-06, + "loss": 0.301, + "mean_token_accuracy": 0.8786396980285645, + "num_tokens": 2576803.0, + "step": 801 + }, + { + "epoch": 0.41023017902813297, + "grad_norm": 12.488948822021484, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8850224018096924, + "num_tokens": 2580403.0, + "step": 802 + }, + { + "epoch": 0.41074168797953964, + "grad_norm": 9.43191146850586, + "learning_rate": 1e-06, + "loss": 0.3114, + "mean_token_accuracy": 0.8616297245025635, + "num_tokens": 2583925.0, + "step": 803 + }, + { + "epoch": 0.4112531969309463, + "grad_norm": 10.305671691894531, + "learning_rate": 1e-06, + "loss": 0.3245, + "mean_token_accuracy": 0.8489900231361389, + "num_tokens": 2586692.0, + "step": 804 + }, + { + "epoch": 0.4117647058823529, + "grad_norm": 11.340248107910156, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8925039768218994, + "num_tokens": 2590558.0, + "step": 805 + }, + { + "epoch": 0.4122762148337596, + "grad_norm": 12.533100128173828, + "learning_rate": 1e-06, + "loss": 0.3202, + "mean_token_accuracy": 0.8801335096359253, + "num_tokens": 2592984.0, + "step": 806 + }, + { + "epoch": 0.41278772378516626, + "grad_norm": 12.824984550476074, + "learning_rate": 1e-06, + "loss": 0.4296, + "mean_token_accuracy": 0.8383547067642212, + "num_tokens": 2595920.0, + "step": 807 + }, + { + "epoch": 0.4132992327365729, + "grad_norm": 14.632420539855957, + "learning_rate": 1e-06, + "loss": 0.3736, + "mean_token_accuracy": 0.8770462274551392, + "num_tokens": 2598655.0, + "step": 808 + }, + { + "epoch": 0.41381074168797954, + "grad_norm": 12.263317108154297, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8667185306549072, + "num_tokens": 2602106.0, + "step": 809 + }, + { + "epoch": 0.4143222506393862, + "grad_norm": 13.78665828704834, + "learning_rate": 1e-06, + "loss": 0.423, + "mean_token_accuracy": 0.8516675233840942, + "num_tokens": 2605727.0, + "step": 810 + }, + { + "epoch": 0.4148337595907928, + "grad_norm": 9.7965669631958, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.9039721488952637, + "num_tokens": 2609093.0, + "step": 811 + }, + { + "epoch": 0.4153452685421995, + "grad_norm": 7.218213081359863, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8671448230743408, + "num_tokens": 2613749.0, + "step": 812 + }, + { + "epoch": 0.4158567774936061, + "grad_norm": 10.11300277709961, + "learning_rate": 1e-06, + "loss": 0.3529, + "mean_token_accuracy": 0.8489594459533691, + "num_tokens": 2617507.0, + "step": 813 + }, + { + "epoch": 0.4163682864450128, + "grad_norm": 11.124874114990234, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8488260507583618, + "num_tokens": 2620973.0, + "step": 814 + }, + { + "epoch": 0.41687979539641945, + "grad_norm": 11.305093765258789, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8637939691543579, + "num_tokens": 2624195.0, + "step": 815 + }, + { + "epoch": 0.41739130434782606, + "grad_norm": 21.82972526550293, + "learning_rate": 1e-06, + "loss": 0.4204, + "mean_token_accuracy": 0.8335391879081726, + "num_tokens": 2626987.0, + "step": 816 + }, + { + "epoch": 0.41790281329923273, + "grad_norm": 11.284955978393555, + "learning_rate": 1e-06, + "loss": 0.2885, + "mean_token_accuracy": 0.8959527611732483, + "num_tokens": 2629669.0, + "step": 817 + }, + { + "epoch": 0.4184143222506394, + "grad_norm": 13.673121452331543, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8449498414993286, + "num_tokens": 2633228.0, + "step": 818 + }, + { + "epoch": 0.418925831202046, + "grad_norm": 12.195951461791992, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8840965628623962, + "num_tokens": 2636557.0, + "step": 819 + }, + { + "epoch": 0.4194373401534527, + "grad_norm": 11.277207374572754, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.894667387008667, + "num_tokens": 2639460.0, + "step": 820 + }, + { + "epoch": 0.41994884910485936, + "grad_norm": 12.601073265075684, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8822051286697388, + "num_tokens": 2642643.0, + "step": 821 + }, + { + "epoch": 0.42046035805626597, + "grad_norm": 8.846500396728516, + "learning_rate": 1e-06, + "loss": 0.1913, + "mean_token_accuracy": 0.9269968271255493, + "num_tokens": 2645803.0, + "step": 822 + }, + { + "epoch": 0.42097186700767264, + "grad_norm": 13.969371795654297, + "learning_rate": 1e-06, + "loss": 0.4926, + "mean_token_accuracy": 0.8228101134300232, + "num_tokens": 2648899.0, + "step": 823 + }, + { + "epoch": 0.4214833759590793, + "grad_norm": 11.716463088989258, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8545897006988525, + "num_tokens": 2651306.0, + "step": 824 + }, + { + "epoch": 0.4219948849104859, + "grad_norm": 16.091163635253906, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.866828978061676, + "num_tokens": 2653775.0, + "step": 825 + }, + { + "epoch": 0.4225063938618926, + "grad_norm": 7.834726810455322, + "learning_rate": 1e-06, + "loss": 0.2872, + "mean_token_accuracy": 0.8874789476394653, + "num_tokens": 2657280.0, + "step": 826 + }, + { + "epoch": 0.4230179028132992, + "grad_norm": 11.311782836914062, + "learning_rate": 1e-06, + "loss": 0.3841, + "mean_token_accuracy": 0.8288345336914062, + "num_tokens": 2661030.0, + "step": 827 + }, + { + "epoch": 0.4235294117647059, + "grad_norm": 11.241264343261719, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8248502612113953, + "num_tokens": 2664111.0, + "step": 828 + }, + { + "epoch": 0.42404092071611255, + "grad_norm": 15.046639442443848, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8851001262664795, + "num_tokens": 2667009.0, + "step": 829 + }, + { + "epoch": 0.42455242966751916, + "grad_norm": 9.744989395141602, + "learning_rate": 1e-06, + "loss": 0.2802, + "mean_token_accuracy": 0.8956931829452515, + "num_tokens": 2670111.0, + "step": 830 + }, + { + "epoch": 0.42506393861892583, + "grad_norm": 11.481727600097656, + "learning_rate": 1e-06, + "loss": 0.2971, + "mean_token_accuracy": 0.8998780250549316, + "num_tokens": 2673235.0, + "step": 831 + }, + { + "epoch": 0.4255754475703325, + "grad_norm": 14.07820987701416, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8555784821510315, + "num_tokens": 2676265.0, + "step": 832 + }, + { + "epoch": 0.4260869565217391, + "grad_norm": 13.454007148742676, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8797883987426758, + "num_tokens": 2678949.0, + "step": 833 + }, + { + "epoch": 0.4265984654731458, + "grad_norm": 8.937646865844727, + "learning_rate": 1e-06, + "loss": 0.2623, + "mean_token_accuracy": 0.875735878944397, + "num_tokens": 2682342.0, + "step": 834 + }, + { + "epoch": 0.42710997442455245, + "grad_norm": 8.093568801879883, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8839364051818848, + "num_tokens": 2686858.0, + "step": 835 + }, + { + "epoch": 0.42762148337595907, + "grad_norm": 7.984218120574951, + "learning_rate": 1e-06, + "loss": 0.2918, + "mean_token_accuracy": 0.8717662692070007, + "num_tokens": 2690694.0, + "step": 836 + }, + { + "epoch": 0.42813299232736574, + "grad_norm": 10.612974166870117, + "learning_rate": 1e-06, + "loss": 0.2536, + "mean_token_accuracy": 0.8957259654998779, + "num_tokens": 2693651.0, + "step": 837 + }, + { + "epoch": 0.4286445012787724, + "grad_norm": 15.238377571105957, + "learning_rate": 1e-06, + "loss": 0.4117, + "mean_token_accuracy": 0.8402559757232666, + "num_tokens": 2696353.0, + "step": 838 + }, + { + "epoch": 0.429156010230179, + "grad_norm": 11.484664916992188, + "learning_rate": 1e-06, + "loss": 0.4094, + "mean_token_accuracy": 0.8715421557426453, + "num_tokens": 2699543.0, + "step": 839 + }, + { + "epoch": 0.4296675191815857, + "grad_norm": 15.943119049072266, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8909001350402832, + "num_tokens": 2702351.0, + "step": 840 + }, + { + "epoch": 0.4301790281329923, + "grad_norm": 7.825136661529541, + "learning_rate": 1e-06, + "loss": 0.3001, + "mean_token_accuracy": 0.8850977420806885, + "num_tokens": 2706305.0, + "step": 841 + }, + { + "epoch": 0.43069053708439897, + "grad_norm": 15.670710563659668, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.8953860998153687, + "num_tokens": 2709586.0, + "step": 842 + }, + { + "epoch": 0.43120204603580564, + "grad_norm": 7.820193767547607, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9060537219047546, + "num_tokens": 2713435.0, + "step": 843 + }, + { + "epoch": 0.43171355498721226, + "grad_norm": 12.084263801574707, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8803712129592896, + "num_tokens": 2716719.0, + "step": 844 + }, + { + "epoch": 0.4322250639386189, + "grad_norm": 11.038049697875977, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8637933731079102, + "num_tokens": 2719720.0, + "step": 845 + }, + { + "epoch": 0.4327365728900256, + "grad_norm": 9.011344909667969, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8391794562339783, + "num_tokens": 2722559.0, + "step": 846 + }, + { + "epoch": 0.4332480818414322, + "grad_norm": 12.326299667358398, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8746428489685059, + "num_tokens": 2725790.0, + "step": 847 + }, + { + "epoch": 0.4337595907928389, + "grad_norm": 13.108488082885742, + "learning_rate": 1e-06, + "loss": 0.3903, + "mean_token_accuracy": 0.8374954462051392, + "num_tokens": 2728843.0, + "step": 848 + }, + { + "epoch": 0.43427109974424555, + "grad_norm": 14.192893028259277, + "learning_rate": 1e-06, + "loss": 0.4119, + "mean_token_accuracy": 0.8377382755279541, + "num_tokens": 2732177.0, + "step": 849 + }, + { + "epoch": 0.43478260869565216, + "grad_norm": 18.206052780151367, + "learning_rate": 1e-06, + "loss": 0.3632, + "mean_token_accuracy": 0.8452451229095459, + "num_tokens": 2734525.0, + "step": 850 + }, + { + "epoch": 0.43529411764705883, + "grad_norm": 11.012189865112305, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8887546062469482, + "num_tokens": 2737474.0, + "step": 851 + }, + { + "epoch": 0.4358056265984655, + "grad_norm": 9.541834831237793, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.8590027093887329, + "num_tokens": 2741347.0, + "step": 852 + }, + { + "epoch": 0.4363171355498721, + "grad_norm": 17.671255111694336, + "learning_rate": 1e-06, + "loss": 0.415, + "mean_token_accuracy": 0.8426784873008728, + "num_tokens": 2743909.0, + "step": 853 + }, + { + "epoch": 0.4368286445012788, + "grad_norm": 13.713323593139648, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8879892826080322, + "num_tokens": 2747068.0, + "step": 854 + }, + { + "epoch": 0.4373401534526854, + "grad_norm": 10.759109497070312, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8440834879875183, + "num_tokens": 2750537.0, + "step": 855 + }, + { + "epoch": 0.43785166240409207, + "grad_norm": 13.749089241027832, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8654704093933105, + "num_tokens": 2753268.0, + "step": 856 + }, + { + "epoch": 0.43836317135549874, + "grad_norm": 12.013143539428711, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8404425382614136, + "num_tokens": 2756526.0, + "step": 857 + }, + { + "epoch": 0.43887468030690535, + "grad_norm": 11.835307121276855, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8686637878417969, + "num_tokens": 2759466.0, + "step": 858 + }, + { + "epoch": 0.439386189258312, + "grad_norm": 10.139033317565918, + "learning_rate": 1e-06, + "loss": 0.318, + "mean_token_accuracy": 0.8921761512756348, + "num_tokens": 2762191.0, + "step": 859 + }, + { + "epoch": 0.4398976982097187, + "grad_norm": 13.071488380432129, + "learning_rate": 1e-06, + "loss": 0.4438, + "mean_token_accuracy": 0.8512457609176636, + "num_tokens": 2765635.0, + "step": 860 + }, + { + "epoch": 0.4404092071611253, + "grad_norm": 13.109183311462402, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8849855065345764, + "num_tokens": 2768931.0, + "step": 861 + }, + { + "epoch": 0.440920716112532, + "grad_norm": 11.74361801147461, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8641557693481445, + "num_tokens": 2771837.0, + "step": 862 + }, + { + "epoch": 0.44143222506393864, + "grad_norm": 11.900647163391113, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8684237003326416, + "num_tokens": 2775206.0, + "step": 863 + }, + { + "epoch": 0.44194373401534526, + "grad_norm": 14.210750579833984, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8304340839385986, + "num_tokens": 2778062.0, + "step": 864 + }, + { + "epoch": 0.4424552429667519, + "grad_norm": 7.069910049438477, + "learning_rate": 1e-06, + "loss": 0.2392, + "mean_token_accuracy": 0.8895266056060791, + "num_tokens": 2782419.0, + "step": 865 + }, + { + "epoch": 0.44296675191815854, + "grad_norm": 10.402308464050293, + "learning_rate": 1e-06, + "loss": 0.2525, + "mean_token_accuracy": 0.9171864986419678, + "num_tokens": 2785080.0, + "step": 866 + }, + { + "epoch": 0.4434782608695652, + "grad_norm": 11.09431266784668, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.8422414660453796, + "num_tokens": 2788428.0, + "step": 867 + }, + { + "epoch": 0.4439897698209719, + "grad_norm": 11.28905963897705, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.9042673110961914, + "num_tokens": 2790861.0, + "step": 868 + }, + { + "epoch": 0.4445012787723785, + "grad_norm": 14.298320770263672, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8500727415084839, + "num_tokens": 2793812.0, + "step": 869 + }, + { + "epoch": 0.44501278772378516, + "grad_norm": 7.714026927947998, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8665850162506104, + "num_tokens": 2797887.0, + "step": 870 + }, + { + "epoch": 0.44552429667519183, + "grad_norm": 10.991291046142578, + "learning_rate": 1e-06, + "loss": 0.3456, + "mean_token_accuracy": 0.8777404427528381, + "num_tokens": 2800433.0, + "step": 871 + }, + { + "epoch": 0.44603580562659845, + "grad_norm": 9.928397178649902, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8915415406227112, + "num_tokens": 2803865.0, + "step": 872 + }, + { + "epoch": 0.4465473145780051, + "grad_norm": 12.34248161315918, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8743884563446045, + "num_tokens": 2806248.0, + "step": 873 + }, + { + "epoch": 0.4470588235294118, + "grad_norm": 12.106684684753418, + "learning_rate": 1e-06, + "loss": 0.3953, + "mean_token_accuracy": 0.8404736518859863, + "num_tokens": 2809079.0, + "step": 874 + }, + { + "epoch": 0.4475703324808184, + "grad_norm": 11.637381553649902, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8680799603462219, + "num_tokens": 2811797.0, + "step": 875 + }, + { + "epoch": 0.44808184143222507, + "grad_norm": 13.132768630981445, + "learning_rate": 1e-06, + "loss": 0.4536, + "mean_token_accuracy": 0.8082146644592285, + "num_tokens": 2815256.0, + "step": 876 + }, + { + "epoch": 0.44859335038363174, + "grad_norm": 7.637244701385498, + "learning_rate": 1e-06, + "loss": 0.225, + "mean_token_accuracy": 0.8842298984527588, + "num_tokens": 2819318.0, + "step": 877 + }, + { + "epoch": 0.44910485933503835, + "grad_norm": 13.01534366607666, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8842536211013794, + "num_tokens": 2822901.0, + "step": 878 + }, + { + "epoch": 0.449616368286445, + "grad_norm": 12.238357543945312, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.874203622341156, + "num_tokens": 2825865.0, + "step": 879 + }, + { + "epoch": 0.45012787723785164, + "grad_norm": 14.206753730773926, + "learning_rate": 1e-06, + "loss": 0.3797, + "mean_token_accuracy": 0.8732781410217285, + "num_tokens": 2828931.0, + "step": 880 + }, + { + "epoch": 0.4506393861892583, + "grad_norm": 8.198290824890137, + "learning_rate": 1e-06, + "loss": 0.263, + "mean_token_accuracy": 0.866453230381012, + "num_tokens": 2832997.0, + "step": 881 + }, + { + "epoch": 0.451150895140665, + "grad_norm": 8.781259536743164, + "learning_rate": 1e-06, + "loss": 0.3147, + "mean_token_accuracy": 0.8873196840286255, + "num_tokens": 2836386.0, + "step": 882 + }, + { + "epoch": 0.4516624040920716, + "grad_norm": 12.613022804260254, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8944642543792725, + "num_tokens": 2839956.0, + "step": 883 + }, + { + "epoch": 0.45217391304347826, + "grad_norm": 10.433046340942383, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.877253532409668, + "num_tokens": 2843811.0, + "step": 884 + }, + { + "epoch": 0.45268542199488493, + "grad_norm": 16.963237762451172, + "learning_rate": 1e-06, + "loss": 0.474, + "mean_token_accuracy": 0.8286255598068237, + "num_tokens": 2846688.0, + "step": 885 + }, + { + "epoch": 0.45319693094629154, + "grad_norm": 11.943642616271973, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8727601766586304, + "num_tokens": 2850225.0, + "step": 886 + }, + { + "epoch": 0.4537084398976982, + "grad_norm": 10.047855377197266, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8684277534484863, + "num_tokens": 2853592.0, + "step": 887 + }, + { + "epoch": 0.4542199488491049, + "grad_norm": 9.708282470703125, + "learning_rate": 1e-06, + "loss": 0.2986, + "mean_token_accuracy": 0.8830285668373108, + "num_tokens": 2856119.0, + "step": 888 + }, + { + "epoch": 0.4547314578005115, + "grad_norm": 9.44277286529541, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8654972314834595, + "num_tokens": 2860340.0, + "step": 889 + }, + { + "epoch": 0.45524296675191817, + "grad_norm": 11.715938568115234, + "learning_rate": 1e-06, + "loss": 0.3181, + "mean_token_accuracy": 0.8715125918388367, + "num_tokens": 2862945.0, + "step": 890 + }, + { + "epoch": 0.45575447570332484, + "grad_norm": 9.362715721130371, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8937624096870422, + "num_tokens": 2866137.0, + "step": 891 + }, + { + "epoch": 0.45626598465473145, + "grad_norm": 10.40349292755127, + "learning_rate": 1e-06, + "loss": 0.3757, + "mean_token_accuracy": 0.8522293567657471, + "num_tokens": 2869883.0, + "step": 892 + }, + { + "epoch": 0.4567774936061381, + "grad_norm": 13.936174392700195, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8643739223480225, + "num_tokens": 2872720.0, + "step": 893 + }, + { + "epoch": 0.45728900255754473, + "grad_norm": 10.683292388916016, + "learning_rate": 1e-06, + "loss": 0.2803, + "mean_token_accuracy": 0.9022043347358704, + "num_tokens": 2876077.0, + "step": 894 + }, + { + "epoch": 0.4578005115089514, + "grad_norm": 12.130156517028809, + "learning_rate": 1e-06, + "loss": 0.3221, + "mean_token_accuracy": 0.8852667808532715, + "num_tokens": 2878848.0, + "step": 895 + }, + { + "epoch": 0.45831202046035807, + "grad_norm": 15.141876220703125, + "learning_rate": 1e-06, + "loss": 0.2583, + "mean_token_accuracy": 0.9105632305145264, + "num_tokens": 2882157.0, + "step": 896 + }, + { + "epoch": 0.4588235294117647, + "grad_norm": 11.40127944946289, + "learning_rate": 1e-06, + "loss": 0.2702, + "mean_token_accuracy": 0.9042760133743286, + "num_tokens": 2885584.0, + "step": 897 + }, + { + "epoch": 0.45933503836317136, + "grad_norm": 18.868257522583008, + "learning_rate": 1e-06, + "loss": 0.4224, + "mean_token_accuracy": 0.8246454000473022, + "num_tokens": 2887943.0, + "step": 898 + }, + { + "epoch": 0.459846547314578, + "grad_norm": 16.991918563842773, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.8675976991653442, + "num_tokens": 2891127.0, + "step": 899 + }, + { + "epoch": 0.46035805626598464, + "grad_norm": 11.799626350402832, + "learning_rate": 1e-06, + "loss": 0.3869, + "mean_token_accuracy": 0.8458914756774902, + "num_tokens": 2894321.0, + "step": 900 + }, + { + "epoch": 0.4608695652173913, + "grad_norm": 14.149956703186035, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8839595317840576, + "num_tokens": 2896641.0, + "step": 901 + }, + { + "epoch": 0.461381074168798, + "grad_norm": 10.460779190063477, + "learning_rate": 1e-06, + "loss": 0.2658, + "mean_token_accuracy": 0.9013607501983643, + "num_tokens": 2899972.0, + "step": 902 + }, + { + "epoch": 0.4618925831202046, + "grad_norm": 13.542315483093262, + "learning_rate": 1e-06, + "loss": 0.3237, + "mean_token_accuracy": 0.8582552671432495, + "num_tokens": 2903069.0, + "step": 903 + }, + { + "epoch": 0.46240409207161126, + "grad_norm": 13.85715389251709, + "learning_rate": 1e-06, + "loss": 0.3292, + "mean_token_accuracy": 0.8624086380004883, + "num_tokens": 2905473.0, + "step": 904 + }, + { + "epoch": 0.4629156010230179, + "grad_norm": 12.896639823913574, + "learning_rate": 1e-06, + "loss": 0.356, + "mean_token_accuracy": 0.8564426898956299, + "num_tokens": 2908342.0, + "step": 905 + }, + { + "epoch": 0.46342710997442454, + "grad_norm": 10.481925010681152, + "learning_rate": 1e-06, + "loss": 0.2974, + "mean_token_accuracy": 0.8580273389816284, + "num_tokens": 2911353.0, + "step": 906 + }, + { + "epoch": 0.4639386189258312, + "grad_norm": 8.968584060668945, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8727476596832275, + "num_tokens": 2915255.0, + "step": 907 + }, + { + "epoch": 0.46445012787723783, + "grad_norm": 11.928292274475098, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8718550801277161, + "num_tokens": 2917587.0, + "step": 908 + }, + { + "epoch": 0.4649616368286445, + "grad_norm": 11.881031036376953, + "learning_rate": 1e-06, + "loss": 0.3617, + "mean_token_accuracy": 0.8616429567337036, + "num_tokens": 2920338.0, + "step": 909 + }, + { + "epoch": 0.46547314578005117, + "grad_norm": 22.10897445678711, + "learning_rate": 1e-06, + "loss": 0.4228, + "mean_token_accuracy": 0.8459236025810242, + "num_tokens": 2923228.0, + "step": 910 + }, + { + "epoch": 0.4659846547314578, + "grad_norm": 14.14470100402832, + "learning_rate": 1e-06, + "loss": 0.4208, + "mean_token_accuracy": 0.8284231424331665, + "num_tokens": 2926607.0, + "step": 911 + }, + { + "epoch": 0.46649616368286445, + "grad_norm": 14.863109588623047, + "learning_rate": 1e-06, + "loss": 0.373, + "mean_token_accuracy": 0.8859142065048218, + "num_tokens": 2929267.0, + "step": 912 + }, + { + "epoch": 0.4670076726342711, + "grad_norm": 13.182780265808105, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8584602475166321, + "num_tokens": 2932128.0, + "step": 913 + }, + { + "epoch": 0.46751918158567773, + "grad_norm": 9.489856719970703, + "learning_rate": 1e-06, + "loss": 0.2818, + "mean_token_accuracy": 0.8905086517333984, + "num_tokens": 2935464.0, + "step": 914 + }, + { + "epoch": 0.4680306905370844, + "grad_norm": 16.96414566040039, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8677945733070374, + "num_tokens": 2938140.0, + "step": 915 + }, + { + "epoch": 0.4685421994884911, + "grad_norm": 10.540396690368652, + "learning_rate": 1e-06, + "loss": 0.2433, + "mean_token_accuracy": 0.8942321538925171, + "num_tokens": 2941612.0, + "step": 916 + }, + { + "epoch": 0.4690537084398977, + "grad_norm": 9.78998851776123, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8847475051879883, + "num_tokens": 2944822.0, + "step": 917 + }, + { + "epoch": 0.46956521739130436, + "grad_norm": 21.66231346130371, + "learning_rate": 1e-06, + "loss": 0.3648, + "mean_token_accuracy": 0.8479377627372742, + "num_tokens": 2947594.0, + "step": 918 + }, + { + "epoch": 0.47007672634271097, + "grad_norm": 13.543991088867188, + "learning_rate": 1e-06, + "loss": 0.3693, + "mean_token_accuracy": 0.8574816584587097, + "num_tokens": 2950497.0, + "step": 919 + }, + { + "epoch": 0.47058823529411764, + "grad_norm": 16.033876419067383, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8634313941001892, + "num_tokens": 2952868.0, + "step": 920 + }, + { + "epoch": 0.4710997442455243, + "grad_norm": 14.050460815429688, + "learning_rate": 1e-06, + "loss": 0.3788, + "mean_token_accuracy": 0.8225568532943726, + "num_tokens": 2955322.0, + "step": 921 + }, + { + "epoch": 0.4716112531969309, + "grad_norm": 9.801735877990723, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.8733581900596619, + "num_tokens": 2958213.0, + "step": 922 + }, + { + "epoch": 0.4721227621483376, + "grad_norm": 12.593173027038574, + "learning_rate": 1e-06, + "loss": 0.3123, + "mean_token_accuracy": 0.8985493183135986, + "num_tokens": 2960936.0, + "step": 923 + }, + { + "epoch": 0.47263427109974426, + "grad_norm": 18.592195510864258, + "learning_rate": 1e-06, + "loss": 0.3846, + "mean_token_accuracy": 0.8546795845031738, + "num_tokens": 2964234.0, + "step": 924 + }, + { + "epoch": 0.4731457800511509, + "grad_norm": 13.83154296875, + "learning_rate": 1e-06, + "loss": 0.4792, + "mean_token_accuracy": 0.8257318735122681, + "num_tokens": 2967210.0, + "step": 925 + }, + { + "epoch": 0.47365728900255755, + "grad_norm": 11.770318031311035, + "learning_rate": 1e-06, + "loss": 0.3336, + "mean_token_accuracy": 0.9003199338912964, + "num_tokens": 2970753.0, + "step": 926 + }, + { + "epoch": 0.4741687979539642, + "grad_norm": 10.347370147705078, + "learning_rate": 1e-06, + "loss": 0.2275, + "mean_token_accuracy": 0.9095898866653442, + "num_tokens": 2973772.0, + "step": 927 + }, + { + "epoch": 0.47468030690537083, + "grad_norm": 12.387580871582031, + "learning_rate": 1e-06, + "loss": 0.4354, + "mean_token_accuracy": 0.8332762122154236, + "num_tokens": 2977167.0, + "step": 928 + }, + { + "epoch": 0.4751918158567775, + "grad_norm": 9.28970718383789, + "learning_rate": 1e-06, + "loss": 0.2458, + "mean_token_accuracy": 0.9013727903366089, + "num_tokens": 2980349.0, + "step": 929 + }, + { + "epoch": 0.47570332480818417, + "grad_norm": 10.462084770202637, + "learning_rate": 1e-06, + "loss": 0.2385, + "mean_token_accuracy": 0.9115036129951477, + "num_tokens": 2983158.0, + "step": 930 + }, + { + "epoch": 0.4762148337595908, + "grad_norm": 13.245957374572754, + "learning_rate": 1e-06, + "loss": 0.3429, + "mean_token_accuracy": 0.8724484443664551, + "num_tokens": 2985873.0, + "step": 931 + }, + { + "epoch": 0.47672634271099745, + "grad_norm": 12.530412673950195, + "learning_rate": 1e-06, + "loss": 0.4217, + "mean_token_accuracy": 0.8626120090484619, + "num_tokens": 2989499.0, + "step": 932 + }, + { + "epoch": 0.47723785166240407, + "grad_norm": 8.145061492919922, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8935018181800842, + "num_tokens": 2994032.0, + "step": 933 + }, + { + "epoch": 0.47774936061381074, + "grad_norm": 15.919651985168457, + "learning_rate": 1e-06, + "loss": 0.329, + "mean_token_accuracy": 0.8827689290046692, + "num_tokens": 2996828.0, + "step": 934 + }, + { + "epoch": 0.4782608695652174, + "grad_norm": 8.654165267944336, + "learning_rate": 1e-06, + "loss": 0.2311, + "mean_token_accuracy": 0.9072872996330261, + "num_tokens": 3000001.0, + "step": 935 + }, + { + "epoch": 0.478772378516624, + "grad_norm": 10.116979598999023, + "learning_rate": 1e-06, + "loss": 0.2134, + "mean_token_accuracy": 0.9311456680297852, + "num_tokens": 3002932.0, + "step": 936 + }, + { + "epoch": 0.4792838874680307, + "grad_norm": 12.19992446899414, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.9039953947067261, + "num_tokens": 3005951.0, + "step": 937 + }, + { + "epoch": 0.47979539641943736, + "grad_norm": 10.344828605651855, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8648958206176758, + "num_tokens": 3009600.0, + "step": 938 + }, + { + "epoch": 0.480306905370844, + "grad_norm": 14.56040096282959, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8666714429855347, + "num_tokens": 3011977.0, + "step": 939 + }, + { + "epoch": 0.48081841432225064, + "grad_norm": 12.25909423828125, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.8756367564201355, + "num_tokens": 3015332.0, + "step": 940 + }, + { + "epoch": 0.4813299232736573, + "grad_norm": 12.322773933410645, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8926877975463867, + "num_tokens": 3018524.0, + "step": 941 + }, + { + "epoch": 0.4818414322250639, + "grad_norm": 9.87955093383789, + "learning_rate": 1e-06, + "loss": 0.3022, + "mean_token_accuracy": 0.8517705798149109, + "num_tokens": 3021685.0, + "step": 942 + }, + { + "epoch": 0.4823529411764706, + "grad_norm": 12.652046203613281, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8711777925491333, + "num_tokens": 3024515.0, + "step": 943 + }, + { + "epoch": 0.48286445012787726, + "grad_norm": 11.97408676147461, + "learning_rate": 1e-06, + "loss": 0.3682, + "mean_token_accuracy": 0.8557884693145752, + "num_tokens": 3027774.0, + "step": 944 + }, + { + "epoch": 0.4833759590792839, + "grad_norm": 13.431666374206543, + "learning_rate": 1e-06, + "loss": 0.4305, + "mean_token_accuracy": 0.8469985127449036, + "num_tokens": 3030668.0, + "step": 945 + }, + { + "epoch": 0.48388746803069055, + "grad_norm": 12.30853271484375, + "learning_rate": 1e-06, + "loss": 0.3743, + "mean_token_accuracy": 0.8636653423309326, + "num_tokens": 3033883.0, + "step": 946 + }, + { + "epoch": 0.48439897698209716, + "grad_norm": 8.239209175109863, + "learning_rate": 1e-06, + "loss": 0.3871, + "mean_token_accuracy": 0.8537560105323792, + "num_tokens": 3037926.0, + "step": 947 + }, + { + "epoch": 0.48491048593350383, + "grad_norm": 12.795426368713379, + "learning_rate": 1e-06, + "loss": 0.4406, + "mean_token_accuracy": 0.8387296199798584, + "num_tokens": 3040572.0, + "step": 948 + }, + { + "epoch": 0.4854219948849105, + "grad_norm": 11.507695198059082, + "learning_rate": 1e-06, + "loss": 0.3747, + "mean_token_accuracy": 0.8647042512893677, + "num_tokens": 3044430.0, + "step": 949 + }, + { + "epoch": 0.4859335038363171, + "grad_norm": 12.35075855255127, + "learning_rate": 1e-06, + "loss": 0.4253, + "mean_token_accuracy": 0.8439821004867554, + "num_tokens": 3047677.0, + "step": 950 + }, + { + "epoch": 0.4864450127877238, + "grad_norm": 11.988993644714355, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8761758804321289, + "num_tokens": 3050377.0, + "step": 951 + }, + { + "epoch": 0.48695652173913045, + "grad_norm": 13.338907241821289, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.9016526937484741, + "num_tokens": 3053069.0, + "step": 952 + }, + { + "epoch": 0.48746803069053707, + "grad_norm": 13.102737426757812, + "learning_rate": 1e-06, + "loss": 0.37, + "mean_token_accuracy": 0.8453854322433472, + "num_tokens": 3055949.0, + "step": 953 + }, + { + "epoch": 0.48797953964194374, + "grad_norm": 15.024629592895508, + "learning_rate": 1e-06, + "loss": 0.4097, + "mean_token_accuracy": 0.8442329168319702, + "num_tokens": 3058426.0, + "step": 954 + }, + { + "epoch": 0.4884910485933504, + "grad_norm": 6.911631107330322, + "learning_rate": 1e-06, + "loss": 0.2413, + "mean_token_accuracy": 0.9039894938468933, + "num_tokens": 3061879.0, + "step": 955 + }, + { + "epoch": 0.489002557544757, + "grad_norm": 9.889756202697754, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8559821248054504, + "num_tokens": 3064704.0, + "step": 956 + }, + { + "epoch": 0.4895140664961637, + "grad_norm": 11.00394058227539, + "learning_rate": 1e-06, + "loss": 0.2991, + "mean_token_accuracy": 0.8912968039512634, + "num_tokens": 3068594.0, + "step": 957 + }, + { + "epoch": 0.4900255754475703, + "grad_norm": 9.529709815979004, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8581268787384033, + "num_tokens": 3071534.0, + "step": 958 + }, + { + "epoch": 0.490537084398977, + "grad_norm": 10.823709487915039, + "learning_rate": 1e-06, + "loss": 0.3031, + "mean_token_accuracy": 0.8964917063713074, + "num_tokens": 3074618.0, + "step": 959 + }, + { + "epoch": 0.49104859335038364, + "grad_norm": 14.157295227050781, + "learning_rate": 1e-06, + "loss": 0.4791, + "mean_token_accuracy": 0.832872748374939, + "num_tokens": 3077987.0, + "step": 960 + }, + { + "epoch": 0.49156010230179026, + "grad_norm": 11.327789306640625, + "learning_rate": 1e-06, + "loss": 0.3474, + "mean_token_accuracy": 0.8886510133743286, + "num_tokens": 3080904.0, + "step": 961 + }, + { + "epoch": 0.4920716112531969, + "grad_norm": 11.609646797180176, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8643538951873779, + "num_tokens": 3084375.0, + "step": 962 + }, + { + "epoch": 0.4925831202046036, + "grad_norm": 10.377692222595215, + "learning_rate": 1e-06, + "loss": 0.2487, + "mean_token_accuracy": 0.8862117528915405, + "num_tokens": 3087834.0, + "step": 963 + }, + { + "epoch": 0.4930946291560102, + "grad_norm": 12.201619148254395, + "learning_rate": 1e-06, + "loss": 0.44, + "mean_token_accuracy": 0.8684810400009155, + "num_tokens": 3091107.0, + "step": 964 + }, + { + "epoch": 0.4936061381074169, + "grad_norm": 14.957257270812988, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8667834997177124, + "num_tokens": 3094531.0, + "step": 965 + }, + { + "epoch": 0.49411764705882355, + "grad_norm": 9.3212251663208, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8703012466430664, + "num_tokens": 3097807.0, + "step": 966 + }, + { + "epoch": 0.49462915601023016, + "grad_norm": 15.117452621459961, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.8670969009399414, + "num_tokens": 3101049.0, + "step": 967 + }, + { + "epoch": 0.49514066496163683, + "grad_norm": 12.999300003051758, + "learning_rate": 1e-06, + "loss": 0.4294, + "mean_token_accuracy": 0.8274656534194946, + "num_tokens": 3104096.0, + "step": 968 + }, + { + "epoch": 0.4956521739130435, + "grad_norm": 8.499798774719238, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8804267048835754, + "num_tokens": 3108194.0, + "step": 969 + }, + { + "epoch": 0.4961636828644501, + "grad_norm": 11.010818481445312, + "learning_rate": 1e-06, + "loss": 0.3523, + "mean_token_accuracy": 0.8644294142723083, + "num_tokens": 3111286.0, + "step": 970 + }, + { + "epoch": 0.4966751918158568, + "grad_norm": 11.4700927734375, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.9034624695777893, + "num_tokens": 3114378.0, + "step": 971 + }, + { + "epoch": 0.4971867007672634, + "grad_norm": 11.468038558959961, + "learning_rate": 1e-06, + "loss": 0.4393, + "mean_token_accuracy": 0.8464052677154541, + "num_tokens": 3117521.0, + "step": 972 + }, + { + "epoch": 0.49769820971867007, + "grad_norm": 11.164506912231445, + "learning_rate": 1e-06, + "loss": 0.3277, + "mean_token_accuracy": 0.8735452890396118, + "num_tokens": 3120733.0, + "step": 973 + }, + { + "epoch": 0.49820971867007674, + "grad_norm": 10.644013404846191, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.8858639597892761, + "num_tokens": 3123745.0, + "step": 974 + }, + { + "epoch": 0.49872122762148335, + "grad_norm": 11.762138366699219, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8642209768295288, + "num_tokens": 3126046.0, + "step": 975 + }, + { + "epoch": 0.49923273657289, + "grad_norm": 12.293657302856445, + "learning_rate": 1e-06, + "loss": 0.2654, + "mean_token_accuracy": 0.9014085531234741, + "num_tokens": 3128955.0, + "step": 976 + }, + { + "epoch": 0.4997442455242967, + "grad_norm": 9.774767875671387, + "learning_rate": 1e-06, + "loss": 0.3503, + "mean_token_accuracy": 0.893223762512207, + "num_tokens": 3132103.0, + "step": 977 + }, + { + "epoch": 0.5002557544757034, + "grad_norm": 13.300708770751953, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.9044313430786133, + "num_tokens": 3135048.0, + "step": 978 + }, + { + "epoch": 0.50076726342711, + "grad_norm": 10.66950798034668, + "learning_rate": 1e-06, + "loss": 0.351, + "mean_token_accuracy": 0.883882999420166, + "num_tokens": 3138755.0, + "step": 979 + }, + { + "epoch": 0.5012787723785166, + "grad_norm": 22.885448455810547, + "learning_rate": 1e-06, + "loss": 0.2512, + "mean_token_accuracy": 0.8945340514183044, + "num_tokens": 3142431.0, + "step": 980 + }, + { + "epoch": 0.5017902813299233, + "grad_norm": 11.037796974182129, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8612267971038818, + "num_tokens": 3145852.0, + "step": 981 + }, + { + "epoch": 0.5023017902813299, + "grad_norm": 11.26646614074707, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8753182888031006, + "num_tokens": 3149301.0, + "step": 982 + }, + { + "epoch": 0.5028132992327365, + "grad_norm": 13.180606842041016, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8894568681716919, + "num_tokens": 3151948.0, + "step": 983 + }, + { + "epoch": 0.5033248081841433, + "grad_norm": 10.67347240447998, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8718405365943909, + "num_tokens": 3155056.0, + "step": 984 + }, + { + "epoch": 0.5038363171355499, + "grad_norm": 11.814398765563965, + "learning_rate": 1e-06, + "loss": 0.3315, + "mean_token_accuracy": 0.8893407583236694, + "num_tokens": 3157500.0, + "step": 985 + }, + { + "epoch": 0.5043478260869565, + "grad_norm": 8.756986618041992, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.9030624628067017, + "num_tokens": 3160225.0, + "step": 986 + }, + { + "epoch": 0.5048593350383632, + "grad_norm": 8.203181266784668, + "learning_rate": 1e-06, + "loss": 0.2617, + "mean_token_accuracy": 0.8939878940582275, + "num_tokens": 3163414.0, + "step": 987 + }, + { + "epoch": 0.5053708439897698, + "grad_norm": 13.249443054199219, + "learning_rate": 1e-06, + "loss": 0.3521, + "mean_token_accuracy": 0.8571264743804932, + "num_tokens": 3166783.0, + "step": 988 + }, + { + "epoch": 0.5058823529411764, + "grad_norm": 12.564422607421875, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.8671267032623291, + "num_tokens": 3169905.0, + "step": 989 + }, + { + "epoch": 0.5063938618925832, + "grad_norm": 14.071937561035156, + "learning_rate": 1e-06, + "loss": 0.3766, + "mean_token_accuracy": 0.8543511629104614, + "num_tokens": 3172795.0, + "step": 990 + }, + { + "epoch": 0.5069053708439898, + "grad_norm": 10.90072250366211, + "learning_rate": 1e-06, + "loss": 0.3485, + "mean_token_accuracy": 0.8721176385879517, + "num_tokens": 3175883.0, + "step": 991 + }, + { + "epoch": 0.5074168797953964, + "grad_norm": 10.20435905456543, + "learning_rate": 1e-06, + "loss": 0.3487, + "mean_token_accuracy": 0.8743531703948975, + "num_tokens": 3179960.0, + "step": 992 + }, + { + "epoch": 0.507928388746803, + "grad_norm": 10.038641929626465, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.85045325756073, + "num_tokens": 3183789.0, + "step": 993 + }, + { + "epoch": 0.5084398976982097, + "grad_norm": 7.992009162902832, + "learning_rate": 1e-06, + "loss": 0.2212, + "mean_token_accuracy": 0.8767416477203369, + "num_tokens": 3187646.0, + "step": 994 + }, + { + "epoch": 0.5089514066496164, + "grad_norm": 9.796868324279785, + "learning_rate": 1e-06, + "loss": 0.2709, + "mean_token_accuracy": 0.8969448804855347, + "num_tokens": 3191449.0, + "step": 995 + }, + { + "epoch": 0.509462915601023, + "grad_norm": 8.67127513885498, + "learning_rate": 1e-06, + "loss": 0.2817, + "mean_token_accuracy": 0.8886064887046814, + "num_tokens": 3194709.0, + "step": 996 + }, + { + "epoch": 0.5099744245524297, + "grad_norm": 13.604639053344727, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8724319934844971, + "num_tokens": 3197488.0, + "step": 997 + }, + { + "epoch": 0.5104859335038363, + "grad_norm": 11.619154930114746, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.875438392162323, + "num_tokens": 3200281.0, + "step": 998 + }, + { + "epoch": 0.5109974424552429, + "grad_norm": 11.309974670410156, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8476709127426147, + "num_tokens": 3203091.0, + "step": 999 + }, + { + "epoch": 0.5115089514066496, + "grad_norm": 12.17459774017334, + "learning_rate": 1e-06, + "loss": 0.3539, + "mean_token_accuracy": 0.8827847242355347, + "num_tokens": 3206106.0, + "step": 1000 + }, + { + "epoch": 0.5120204603580563, + "grad_norm": 9.45663070678711, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8861998915672302, + "num_tokens": 3209981.0, + "step": 1001 + }, + { + "epoch": 0.5125319693094629, + "grad_norm": 9.674986839294434, + "learning_rate": 1e-06, + "loss": 0.3723, + "mean_token_accuracy": 0.8633290529251099, + "num_tokens": 3213281.0, + "step": 1002 + }, + { + "epoch": 0.5130434782608696, + "grad_norm": 9.232489585876465, + "learning_rate": 1e-06, + "loss": 0.2079, + "mean_token_accuracy": 0.9054157733917236, + "num_tokens": 3216649.0, + "step": 1003 + }, + { + "epoch": 0.5135549872122762, + "grad_norm": 10.66522216796875, + "learning_rate": 1e-06, + "loss": 0.385, + "mean_token_accuracy": 0.8569345474243164, + "num_tokens": 3219002.0, + "step": 1004 + }, + { + "epoch": 0.5140664961636828, + "grad_norm": 10.10362434387207, + "learning_rate": 1e-06, + "loss": 0.3588, + "mean_token_accuracy": 0.8693905472755432, + "num_tokens": 3222050.0, + "step": 1005 + }, + { + "epoch": 0.5145780051150896, + "grad_norm": 8.32437515258789, + "learning_rate": 1e-06, + "loss": 0.3142, + "mean_token_accuracy": 0.8685088753700256, + "num_tokens": 3226265.0, + "step": 1006 + }, + { + "epoch": 0.5150895140664962, + "grad_norm": 7.415220260620117, + "learning_rate": 1e-06, + "loss": 0.2164, + "mean_token_accuracy": 0.9150919318199158, + "num_tokens": 3230011.0, + "step": 1007 + }, + { + "epoch": 0.5156010230179028, + "grad_norm": 10.754469871520996, + "learning_rate": 1e-06, + "loss": 0.2648, + "mean_token_accuracy": 0.8990311622619629, + "num_tokens": 3232712.0, + "step": 1008 + }, + { + "epoch": 0.5161125319693095, + "grad_norm": 11.488266944885254, + "learning_rate": 1e-06, + "loss": 0.3748, + "mean_token_accuracy": 0.8839737176895142, + "num_tokens": 3235657.0, + "step": 1009 + }, + { + "epoch": 0.5166240409207161, + "grad_norm": 14.987966537475586, + "learning_rate": 1e-06, + "loss": 0.332, + "mean_token_accuracy": 0.8497023582458496, + "num_tokens": 3237955.0, + "step": 1010 + }, + { + "epoch": 0.5171355498721227, + "grad_norm": 10.587514877319336, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8447462320327759, + "num_tokens": 3242681.0, + "step": 1011 + }, + { + "epoch": 0.5176470588235295, + "grad_norm": 13.486703872680664, + "learning_rate": 1e-06, + "loss": 0.4238, + "mean_token_accuracy": 0.8197755217552185, + "num_tokens": 3245279.0, + "step": 1012 + }, + { + "epoch": 0.5181585677749361, + "grad_norm": 8.315016746520996, + "learning_rate": 1e-06, + "loss": 0.2678, + "mean_token_accuracy": 0.876562237739563, + "num_tokens": 3248744.0, + "step": 1013 + }, + { + "epoch": 0.5186700767263427, + "grad_norm": 12.506719589233398, + "learning_rate": 1e-06, + "loss": 0.4079, + "mean_token_accuracy": 0.8509550094604492, + "num_tokens": 3251543.0, + "step": 1014 + }, + { + "epoch": 0.5191815856777494, + "grad_norm": 10.929181098937988, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.9070258140563965, + "num_tokens": 3255164.0, + "step": 1015 + }, + { + "epoch": 0.519693094629156, + "grad_norm": 8.266942024230957, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8593504428863525, + "num_tokens": 3258891.0, + "step": 1016 + }, + { + "epoch": 0.5202046035805626, + "grad_norm": 8.850625991821289, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8626301884651184, + "num_tokens": 3263023.0, + "step": 1017 + }, + { + "epoch": 0.5207161125319693, + "grad_norm": 9.588022232055664, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.8967305421829224, + "num_tokens": 3265971.0, + "step": 1018 + }, + { + "epoch": 0.521227621483376, + "grad_norm": 13.282846450805664, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.879308819770813, + "num_tokens": 3268889.0, + "step": 1019 + }, + { + "epoch": 0.5217391304347826, + "grad_norm": 14.202546119689941, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.8690648078918457, + "num_tokens": 3272006.0, + "step": 1020 + }, + { + "epoch": 0.5222506393861892, + "grad_norm": 7.914816379547119, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8777766823768616, + "num_tokens": 3276099.0, + "step": 1021 + }, + { + "epoch": 0.5227621483375959, + "grad_norm": 8.46007251739502, + "learning_rate": 1e-06, + "loss": 0.1957, + "mean_token_accuracy": 0.8855264186859131, + "num_tokens": 3279498.0, + "step": 1022 + }, + { + "epoch": 0.5232736572890025, + "grad_norm": 10.301488876342773, + "learning_rate": 1e-06, + "loss": 0.2297, + "mean_token_accuracy": 0.8977056741714478, + "num_tokens": 3282711.0, + "step": 1023 + }, + { + "epoch": 0.5237851662404092, + "grad_norm": 12.64526653289795, + "learning_rate": 1e-06, + "loss": 0.2997, + "mean_token_accuracy": 0.8823347091674805, + "num_tokens": 3285377.0, + "step": 1024 + }, + { + "epoch": 0.5242966751918159, + "grad_norm": 13.11957836151123, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.85963374376297, + "num_tokens": 3288367.0, + "step": 1025 + }, + { + "epoch": 0.5248081841432225, + "grad_norm": 11.656230926513672, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.822425365447998, + "num_tokens": 3291951.0, + "step": 1026 + }, + { + "epoch": 0.5253196930946291, + "grad_norm": 12.90439510345459, + "learning_rate": 1e-06, + "loss": 0.3763, + "mean_token_accuracy": 0.8505256175994873, + "num_tokens": 3294410.0, + "step": 1027 + }, + { + "epoch": 0.5258312020460358, + "grad_norm": 11.90691089630127, + "learning_rate": 1e-06, + "loss": 0.3346, + "mean_token_accuracy": 0.8699103593826294, + "num_tokens": 3297443.0, + "step": 1028 + }, + { + "epoch": 0.5263427109974425, + "grad_norm": 11.010714530944824, + "learning_rate": 1e-06, + "loss": 0.3189, + "mean_token_accuracy": 0.8532537817955017, + "num_tokens": 3301034.0, + "step": 1029 + }, + { + "epoch": 0.5268542199488491, + "grad_norm": 7.851024150848389, + "learning_rate": 1e-06, + "loss": 0.2363, + "mean_token_accuracy": 0.9057563543319702, + "num_tokens": 3304200.0, + "step": 1030 + }, + { + "epoch": 0.5273657289002558, + "grad_norm": 11.354597091674805, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8692073822021484, + "num_tokens": 3308108.0, + "step": 1031 + }, + { + "epoch": 0.5278772378516624, + "grad_norm": 14.25631332397461, + "learning_rate": 1e-06, + "loss": 0.387, + "mean_token_accuracy": 0.8429135084152222, + "num_tokens": 3311187.0, + "step": 1032 + }, + { + "epoch": 0.528388746803069, + "grad_norm": 10.28518295288086, + "learning_rate": 1e-06, + "loss": 0.289, + "mean_token_accuracy": 0.8869210481643677, + "num_tokens": 3315715.0, + "step": 1033 + }, + { + "epoch": 0.5289002557544757, + "grad_norm": 13.052826881408691, + "learning_rate": 1e-06, + "loss": 0.4068, + "mean_token_accuracy": 0.8279911279678345, + "num_tokens": 3318576.0, + "step": 1034 + }, + { + "epoch": 0.5294117647058824, + "grad_norm": 11.853252410888672, + "learning_rate": 1e-06, + "loss": 0.3157, + "mean_token_accuracy": 0.8849778175354004, + "num_tokens": 3321693.0, + "step": 1035 + }, + { + "epoch": 0.529923273657289, + "grad_norm": 16.210073471069336, + "learning_rate": 1e-06, + "loss": 0.3894, + "mean_token_accuracy": 0.8533130288124084, + "num_tokens": 3325186.0, + "step": 1036 + }, + { + "epoch": 0.5304347826086957, + "grad_norm": 12.654520034790039, + "learning_rate": 1e-06, + "loss": 0.3571, + "mean_token_accuracy": 0.869036078453064, + "num_tokens": 3328039.0, + "step": 1037 + }, + { + "epoch": 0.5309462915601023, + "grad_norm": 9.063246726989746, + "learning_rate": 1e-06, + "loss": 0.2838, + "mean_token_accuracy": 0.8859761953353882, + "num_tokens": 3331506.0, + "step": 1038 + }, + { + "epoch": 0.5314578005115089, + "grad_norm": 11.171963691711426, + "learning_rate": 1e-06, + "loss": 0.2849, + "mean_token_accuracy": 0.9051767587661743, + "num_tokens": 3334273.0, + "step": 1039 + }, + { + "epoch": 0.5319693094629157, + "grad_norm": 13.534765243530273, + "learning_rate": 1e-06, + "loss": 0.3947, + "mean_token_accuracy": 0.85411536693573, + "num_tokens": 3336640.0, + "step": 1040 + }, + { + "epoch": 0.5324808184143223, + "grad_norm": 3.922973871231079, + "learning_rate": 1e-06, + "loss": 0.1493, + "mean_token_accuracy": 0.90967857837677, + "num_tokens": 3342229.0, + "step": 1041 + }, + { + "epoch": 0.5329923273657289, + "grad_norm": 12.106030464172363, + "learning_rate": 1e-06, + "loss": 0.3935, + "mean_token_accuracy": 0.8502762317657471, + "num_tokens": 3345327.0, + "step": 1042 + }, + { + "epoch": 0.5335038363171356, + "grad_norm": 14.652471542358398, + "learning_rate": 1e-06, + "loss": 0.4193, + "mean_token_accuracy": 0.8401981592178345, + "num_tokens": 3348991.0, + "step": 1043 + }, + { + "epoch": 0.5340153452685422, + "grad_norm": 9.136303901672363, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.908119797706604, + "num_tokens": 3351905.0, + "step": 1044 + }, + { + "epoch": 0.5345268542199488, + "grad_norm": 9.237889289855957, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8914244771003723, + "num_tokens": 3355544.0, + "step": 1045 + }, + { + "epoch": 0.5350383631713554, + "grad_norm": 10.064491271972656, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.8869485855102539, + "num_tokens": 3358673.0, + "step": 1046 + }, + { + "epoch": 0.5355498721227622, + "grad_norm": 10.346622467041016, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8528505563735962, + "num_tokens": 3361572.0, + "step": 1047 + }, + { + "epoch": 0.5360613810741688, + "grad_norm": 10.275368690490723, + "learning_rate": 1e-06, + "loss": 0.4023, + "mean_token_accuracy": 0.8484382629394531, + "num_tokens": 3364915.0, + "step": 1048 + }, + { + "epoch": 0.5365728900255754, + "grad_norm": 16.675498962402344, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8973503112792969, + "num_tokens": 3367970.0, + "step": 1049 + }, + { + "epoch": 0.5370843989769821, + "grad_norm": 13.174642562866211, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8856555223464966, + "num_tokens": 3371563.0, + "step": 1050 + }, + { + "epoch": 0.5375959079283887, + "grad_norm": 9.140716552734375, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.8892649412155151, + "num_tokens": 3374754.0, + "step": 1051 + }, + { + "epoch": 0.5381074168797954, + "grad_norm": 9.776262283325195, + "learning_rate": 1e-06, + "loss": 0.2897, + "mean_token_accuracy": 0.8790663480758667, + "num_tokens": 3377925.0, + "step": 1052 + }, + { + "epoch": 0.5386189258312021, + "grad_norm": 8.95678424835205, + "learning_rate": 1e-06, + "loss": 0.2761, + "mean_token_accuracy": 0.9133662581443787, + "num_tokens": 3380977.0, + "step": 1053 + }, + { + "epoch": 0.5391304347826087, + "grad_norm": 9.168599128723145, + "learning_rate": 1e-06, + "loss": 0.2547, + "mean_token_accuracy": 0.9105772376060486, + "num_tokens": 3384821.0, + "step": 1054 + }, + { + "epoch": 0.5396419437340153, + "grad_norm": 9.680779457092285, + "learning_rate": 1e-06, + "loss": 0.2991, + "mean_token_accuracy": 0.8541803359985352, + "num_tokens": 3388166.0, + "step": 1055 + }, + { + "epoch": 0.540153452685422, + "grad_norm": 11.272835731506348, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.8788042068481445, + "num_tokens": 3391328.0, + "step": 1056 + }, + { + "epoch": 0.5406649616368286, + "grad_norm": 9.950818061828613, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.9014953374862671, + "num_tokens": 3396122.0, + "step": 1057 + }, + { + "epoch": 0.5411764705882353, + "grad_norm": 13.951508522033691, + "learning_rate": 1e-06, + "loss": 0.311, + "mean_token_accuracy": 0.8808168172836304, + "num_tokens": 3398750.0, + "step": 1058 + }, + { + "epoch": 0.541687979539642, + "grad_norm": 11.464447975158691, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8863642811775208, + "num_tokens": 3401710.0, + "step": 1059 + }, + { + "epoch": 0.5421994884910486, + "grad_norm": 10.194596290588379, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8775277733802795, + "num_tokens": 3404705.0, + "step": 1060 + }, + { + "epoch": 0.5427109974424552, + "grad_norm": 9.935873031616211, + "learning_rate": 1e-06, + "loss": 0.3626, + "mean_token_accuracy": 0.8524566888809204, + "num_tokens": 3408477.0, + "step": 1061 + }, + { + "epoch": 0.5432225063938619, + "grad_norm": 10.852006912231445, + "learning_rate": 1e-06, + "loss": 0.391, + "mean_token_accuracy": 0.8481231927871704, + "num_tokens": 3411767.0, + "step": 1062 + }, + { + "epoch": 0.5437340153452686, + "grad_norm": 12.006669998168945, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8801931142807007, + "num_tokens": 3415179.0, + "step": 1063 + }, + { + "epoch": 0.5442455242966752, + "grad_norm": 9.707732200622559, + "learning_rate": 1e-06, + "loss": 0.2655, + "mean_token_accuracy": 0.8953284621238708, + "num_tokens": 3418083.0, + "step": 1064 + }, + { + "epoch": 0.5447570332480819, + "grad_norm": 14.453017234802246, + "learning_rate": 1e-06, + "loss": 0.4706, + "mean_token_accuracy": 0.8227800130844116, + "num_tokens": 3420777.0, + "step": 1065 + }, + { + "epoch": 0.5452685421994885, + "grad_norm": 8.945930480957031, + "learning_rate": 1e-06, + "loss": 0.2645, + "mean_token_accuracy": 0.8819423317909241, + "num_tokens": 3423814.0, + "step": 1066 + }, + { + "epoch": 0.5457800511508951, + "grad_norm": 10.175992965698242, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8594931960105896, + "num_tokens": 3427565.0, + "step": 1067 + }, + { + "epoch": 0.5462915601023018, + "grad_norm": 13.12817096710205, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8652607202529907, + "num_tokens": 3430666.0, + "step": 1068 + }, + { + "epoch": 0.5468030690537085, + "grad_norm": 11.01089859008789, + "learning_rate": 1e-06, + "loss": 0.3858, + "mean_token_accuracy": 0.8605099320411682, + "num_tokens": 3433383.0, + "step": 1069 + }, + { + "epoch": 0.5473145780051151, + "grad_norm": 14.776579856872559, + "learning_rate": 1e-06, + "loss": 0.3794, + "mean_token_accuracy": 0.8509079217910767, + "num_tokens": 3436138.0, + "step": 1070 + }, + { + "epoch": 0.5478260869565217, + "grad_norm": 11.285945892333984, + "learning_rate": 1e-06, + "loss": 0.3488, + "mean_token_accuracy": 0.8680742979049683, + "num_tokens": 3438385.0, + "step": 1071 + }, + { + "epoch": 0.5483375959079284, + "grad_norm": 11.250843048095703, + "learning_rate": 1e-06, + "loss": 0.3333, + "mean_token_accuracy": 0.8633264899253845, + "num_tokens": 3442030.0, + "step": 1072 + }, + { + "epoch": 0.548849104859335, + "grad_norm": 12.92923641204834, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8842928409576416, + "num_tokens": 3445046.0, + "step": 1073 + }, + { + "epoch": 0.5493606138107416, + "grad_norm": 14.293557167053223, + "learning_rate": 1e-06, + "loss": 0.3274, + "mean_token_accuracy": 0.8780151605606079, + "num_tokens": 3449059.0, + "step": 1074 + }, + { + "epoch": 0.5498721227621484, + "grad_norm": 10.903390884399414, + "learning_rate": 1e-06, + "loss": 0.4074, + "mean_token_accuracy": 0.8779761791229248, + "num_tokens": 3452299.0, + "step": 1075 + }, + { + "epoch": 0.550383631713555, + "grad_norm": 6.573505401611328, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.876438319683075, + "num_tokens": 3456862.0, + "step": 1076 + }, + { + "epoch": 0.5508951406649616, + "grad_norm": 10.304394721984863, + "learning_rate": 1e-06, + "loss": 0.3042, + "mean_token_accuracy": 0.852203369140625, + "num_tokens": 3459755.0, + "step": 1077 + }, + { + "epoch": 0.5514066496163683, + "grad_norm": 9.977828025817871, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8783536553382874, + "num_tokens": 3462941.0, + "step": 1078 + }, + { + "epoch": 0.5519181585677749, + "grad_norm": 13.247413635253906, + "learning_rate": 1e-06, + "loss": 0.4343, + "mean_token_accuracy": 0.8284920454025269, + "num_tokens": 3465674.0, + "step": 1079 + }, + { + "epoch": 0.5524296675191815, + "grad_norm": 10.562975883483887, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8609258532524109, + "num_tokens": 3469062.0, + "step": 1080 + }, + { + "epoch": 0.5529411764705883, + "grad_norm": 11.042271614074707, + "learning_rate": 1e-06, + "loss": 0.3636, + "mean_token_accuracy": 0.8775452971458435, + "num_tokens": 3472032.0, + "step": 1081 + }, + { + "epoch": 0.5534526854219949, + "grad_norm": 9.162189483642578, + "learning_rate": 1e-06, + "loss": 0.3365, + "mean_token_accuracy": 0.8596882820129395, + "num_tokens": 3476087.0, + "step": 1082 + }, + { + "epoch": 0.5539641943734015, + "grad_norm": 11.559773445129395, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8551957607269287, + "num_tokens": 3478903.0, + "step": 1083 + }, + { + "epoch": 0.5544757033248082, + "grad_norm": 11.091499328613281, + "learning_rate": 1e-06, + "loss": 0.3798, + "mean_token_accuracy": 0.8584734201431274, + "num_tokens": 3482269.0, + "step": 1084 + }, + { + "epoch": 0.5549872122762148, + "grad_norm": 7.578518390655518, + "learning_rate": 1e-06, + "loss": 0.2561, + "mean_token_accuracy": 0.8766546249389648, + "num_tokens": 3486005.0, + "step": 1085 + }, + { + "epoch": 0.5554987212276215, + "grad_norm": 11.275805473327637, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8690553307533264, + "num_tokens": 3489031.0, + "step": 1086 + }, + { + "epoch": 0.5560102301790282, + "grad_norm": 13.149421691894531, + "learning_rate": 1e-06, + "loss": 0.3749, + "mean_token_accuracy": 0.8605058789253235, + "num_tokens": 3491991.0, + "step": 1087 + }, + { + "epoch": 0.5565217391304348, + "grad_norm": 11.564826965332031, + "learning_rate": 1e-06, + "loss": 0.4307, + "mean_token_accuracy": 0.8304476141929626, + "num_tokens": 3494649.0, + "step": 1088 + }, + { + "epoch": 0.5570332480818414, + "grad_norm": 12.499238014221191, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8669577240943909, + "num_tokens": 3498289.0, + "step": 1089 + }, + { + "epoch": 0.5575447570332481, + "grad_norm": 8.0123929977417, + "learning_rate": 1e-06, + "loss": 0.2307, + "mean_token_accuracy": 0.8956987261772156, + "num_tokens": 3501092.0, + "step": 1090 + }, + { + "epoch": 0.5580562659846547, + "grad_norm": 10.8207426071167, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8635135889053345, + "num_tokens": 3503896.0, + "step": 1091 + }, + { + "epoch": 0.5585677749360614, + "grad_norm": 14.799336433410645, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8490543961524963, + "num_tokens": 3507678.0, + "step": 1092 + }, + { + "epoch": 0.5590792838874681, + "grad_norm": 14.3640718460083, + "learning_rate": 1e-06, + "loss": 0.4289, + "mean_token_accuracy": 0.8269736766815186, + "num_tokens": 3511010.0, + "step": 1093 + }, + { + "epoch": 0.5595907928388747, + "grad_norm": 11.398890495300293, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8738685846328735, + "num_tokens": 3513978.0, + "step": 1094 + }, + { + "epoch": 0.5601023017902813, + "grad_norm": 10.813648223876953, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8745825886726379, + "num_tokens": 3517866.0, + "step": 1095 + }, + { + "epoch": 0.560613810741688, + "grad_norm": 9.244976997375488, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8648173809051514, + "num_tokens": 3520549.0, + "step": 1096 + }, + { + "epoch": 0.5611253196930946, + "grad_norm": 9.490327835083008, + "learning_rate": 1e-06, + "loss": 0.2736, + "mean_token_accuracy": 0.9006826877593994, + "num_tokens": 3523568.0, + "step": 1097 + }, + { + "epoch": 0.5616368286445013, + "grad_norm": 8.142496109008789, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8823137283325195, + "num_tokens": 3527899.0, + "step": 1098 + }, + { + "epoch": 0.5621483375959079, + "grad_norm": 9.929887771606445, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8633518218994141, + "num_tokens": 3531292.0, + "step": 1099 + }, + { + "epoch": 0.5626598465473146, + "grad_norm": 6.9588518142700195, + "learning_rate": 1e-06, + "loss": 0.2574, + "mean_token_accuracy": 0.8898847103118896, + "num_tokens": 3535656.0, + "step": 1100 + }, + { + "epoch": 0.5631713554987212, + "grad_norm": 14.554916381835938, + "learning_rate": 1e-06, + "loss": 0.3942, + "mean_token_accuracy": 0.8496435880661011, + "num_tokens": 3538777.0, + "step": 1101 + }, + { + "epoch": 0.5636828644501278, + "grad_norm": 10.81091022491455, + "learning_rate": 1e-06, + "loss": 0.393, + "mean_token_accuracy": 0.8483219146728516, + "num_tokens": 3542167.0, + "step": 1102 + }, + { + "epoch": 0.5641943734015346, + "grad_norm": 7.62358283996582, + "learning_rate": 1e-06, + "loss": 0.2378, + "mean_token_accuracy": 0.8881635665893555, + "num_tokens": 3546045.0, + "step": 1103 + }, + { + "epoch": 0.5647058823529412, + "grad_norm": 7.7374396324157715, + "learning_rate": 1e-06, + "loss": 0.2536, + "mean_token_accuracy": 0.8916771411895752, + "num_tokens": 3550733.0, + "step": 1104 + }, + { + "epoch": 0.5652173913043478, + "grad_norm": 11.928784370422363, + "learning_rate": 1e-06, + "loss": 0.3171, + "mean_token_accuracy": 0.8841986060142517, + "num_tokens": 3553509.0, + "step": 1105 + }, + { + "epoch": 0.5657289002557545, + "grad_norm": 12.250648498535156, + "learning_rate": 1e-06, + "loss": 0.3845, + "mean_token_accuracy": 0.8156589269638062, + "num_tokens": 3556457.0, + "step": 1106 + }, + { + "epoch": 0.5662404092071611, + "grad_norm": 8.455331802368164, + "learning_rate": 1e-06, + "loss": 0.2316, + "mean_token_accuracy": 0.905347466468811, + "num_tokens": 3560325.0, + "step": 1107 + }, + { + "epoch": 0.5667519181585677, + "grad_norm": 10.34080696105957, + "learning_rate": 1e-06, + "loss": 0.361, + "mean_token_accuracy": 0.868186354637146, + "num_tokens": 3563322.0, + "step": 1108 + }, + { + "epoch": 0.5672634271099745, + "grad_norm": 10.127676010131836, + "learning_rate": 1e-06, + "loss": 0.3769, + "mean_token_accuracy": 0.8852823376655579, + "num_tokens": 3566728.0, + "step": 1109 + }, + { + "epoch": 0.5677749360613811, + "grad_norm": 10.495680809020996, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8763812780380249, + "num_tokens": 3570077.0, + "step": 1110 + }, + { + "epoch": 0.5682864450127877, + "grad_norm": 11.085817337036133, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8649650812149048, + "num_tokens": 3572674.0, + "step": 1111 + }, + { + "epoch": 0.5687979539641944, + "grad_norm": 15.389176368713379, + "learning_rate": 1e-06, + "loss": 0.4105, + "mean_token_accuracy": 0.8485181331634521, + "num_tokens": 3575812.0, + "step": 1112 + }, + { + "epoch": 0.569309462915601, + "grad_norm": 11.935484886169434, + "learning_rate": 1e-06, + "loss": 0.3959, + "mean_token_accuracy": 0.8514500260353088, + "num_tokens": 3578559.0, + "step": 1113 + }, + { + "epoch": 0.5698209718670076, + "grad_norm": 12.3976411819458, + "learning_rate": 1e-06, + "loss": 0.3484, + "mean_token_accuracy": 0.8664188385009766, + "num_tokens": 3581684.0, + "step": 1114 + }, + { + "epoch": 0.5703324808184144, + "grad_norm": 12.540820121765137, + "learning_rate": 1e-06, + "loss": 0.4503, + "mean_token_accuracy": 0.8418182134628296, + "num_tokens": 3584413.0, + "step": 1115 + }, + { + "epoch": 0.570843989769821, + "grad_norm": 8.790332794189453, + "learning_rate": 1e-06, + "loss": 0.2515, + "mean_token_accuracy": 0.8795485496520996, + "num_tokens": 3587825.0, + "step": 1116 + }, + { + "epoch": 0.5713554987212276, + "grad_norm": 11.561161041259766, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8687339425086975, + "num_tokens": 3590708.0, + "step": 1117 + }, + { + "epoch": 0.5718670076726343, + "grad_norm": 12.125725746154785, + "learning_rate": 1e-06, + "loss": 0.407, + "mean_token_accuracy": 0.822291374206543, + "num_tokens": 3593693.0, + "step": 1118 + }, + { + "epoch": 0.5723785166240409, + "grad_norm": 9.2493896484375, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8923349380493164, + "num_tokens": 3597203.0, + "step": 1119 + }, + { + "epoch": 0.5728900255754475, + "grad_norm": 11.483939170837402, + "learning_rate": 1e-06, + "loss": 0.3241, + "mean_token_accuracy": 0.8867213726043701, + "num_tokens": 3599811.0, + "step": 1120 + }, + { + "epoch": 0.5734015345268543, + "grad_norm": 13.985563278198242, + "learning_rate": 1e-06, + "loss": 0.3313, + "mean_token_accuracy": 0.8763322830200195, + "num_tokens": 3602713.0, + "step": 1121 + }, + { + "epoch": 0.5739130434782609, + "grad_norm": 11.220341682434082, + "learning_rate": 1e-06, + "loss": 0.4091, + "mean_token_accuracy": 0.8381960391998291, + "num_tokens": 3605611.0, + "step": 1122 + }, + { + "epoch": 0.5744245524296675, + "grad_norm": 9.999303817749023, + "learning_rate": 1e-06, + "loss": 0.35, + "mean_token_accuracy": 0.8330889940261841, + "num_tokens": 3609459.0, + "step": 1123 + }, + { + "epoch": 0.5749360613810741, + "grad_norm": 11.652371406555176, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.9011613726615906, + "num_tokens": 3612985.0, + "step": 1124 + }, + { + "epoch": 0.5754475703324808, + "grad_norm": 14.38115119934082, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8798344731330872, + "num_tokens": 3616182.0, + "step": 1125 + }, + { + "epoch": 0.5759590792838875, + "grad_norm": 8.475350379943848, + "learning_rate": 1e-06, + "loss": 0.2248, + "mean_token_accuracy": 0.8789963126182556, + "num_tokens": 3618687.0, + "step": 1126 + }, + { + "epoch": 0.5764705882352941, + "grad_norm": 15.309759140014648, + "learning_rate": 1e-06, + "loss": 0.3634, + "mean_token_accuracy": 0.8799245357513428, + "num_tokens": 3621851.0, + "step": 1127 + }, + { + "epoch": 0.5769820971867008, + "grad_norm": 9.63990592956543, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8608914017677307, + "num_tokens": 3625723.0, + "step": 1128 + }, + { + "epoch": 0.5774936061381074, + "grad_norm": 9.517413139343262, + "learning_rate": 1e-06, + "loss": 0.2872, + "mean_token_accuracy": 0.9035037159919739, + "num_tokens": 3629860.0, + "step": 1129 + }, + { + "epoch": 0.578005115089514, + "grad_norm": 12.329913139343262, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8853532671928406, + "num_tokens": 3632503.0, + "step": 1130 + }, + { + "epoch": 0.5785166240409207, + "grad_norm": 14.573844909667969, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.8629859685897827, + "num_tokens": 3636089.0, + "step": 1131 + }, + { + "epoch": 0.5790281329923274, + "grad_norm": 12.203182220458984, + "learning_rate": 1e-06, + "loss": 0.4256, + "mean_token_accuracy": 0.818617582321167, + "num_tokens": 3639892.0, + "step": 1132 + }, + { + "epoch": 0.579539641943734, + "grad_norm": 8.594698905944824, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8930453658103943, + "num_tokens": 3644310.0, + "step": 1133 + }, + { + "epoch": 0.5800511508951407, + "grad_norm": 9.657590866088867, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.8866003751754761, + "num_tokens": 3647189.0, + "step": 1134 + }, + { + "epoch": 0.5805626598465473, + "grad_norm": 6.69843053817749, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.87799072265625, + "num_tokens": 3651473.0, + "step": 1135 + }, + { + "epoch": 0.5810741687979539, + "grad_norm": 11.214253425598145, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.839472234249115, + "num_tokens": 3654345.0, + "step": 1136 + }, + { + "epoch": 0.5815856777493607, + "grad_norm": 10.306328773498535, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8828489780426025, + "num_tokens": 3657670.0, + "step": 1137 + }, + { + "epoch": 0.5820971867007673, + "grad_norm": 11.125374794006348, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8770114183425903, + "num_tokens": 3660929.0, + "step": 1138 + }, + { + "epoch": 0.5826086956521739, + "grad_norm": 11.938135147094727, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8868998289108276, + "num_tokens": 3663753.0, + "step": 1139 + }, + { + "epoch": 0.5831202046035806, + "grad_norm": 16.34882926940918, + "learning_rate": 1e-06, + "loss": 0.4695, + "mean_token_accuracy": 0.8464576005935669, + "num_tokens": 3667077.0, + "step": 1140 + }, + { + "epoch": 0.5836317135549872, + "grad_norm": 7.546816825866699, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.864054262638092, + "num_tokens": 3671692.0, + "step": 1141 + }, + { + "epoch": 0.5841432225063938, + "grad_norm": 9.13259506225586, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.868971586227417, + "num_tokens": 3674973.0, + "step": 1142 + }, + { + "epoch": 0.5846547314578006, + "grad_norm": 10.935068130493164, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.8608812093734741, + "num_tokens": 3678348.0, + "step": 1143 + }, + { + "epoch": 0.5851662404092072, + "grad_norm": 16.92277717590332, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8710730671882629, + "num_tokens": 3681569.0, + "step": 1144 + }, + { + "epoch": 0.5856777493606138, + "grad_norm": 10.196342468261719, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8738024234771729, + "num_tokens": 3684563.0, + "step": 1145 + }, + { + "epoch": 0.5861892583120205, + "grad_norm": 14.292884826660156, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8829096555709839, + "num_tokens": 3688075.0, + "step": 1146 + }, + { + "epoch": 0.5867007672634271, + "grad_norm": 9.353959083557129, + "learning_rate": 1e-06, + "loss": 0.2627, + "mean_token_accuracy": 0.8763470649719238, + "num_tokens": 3691155.0, + "step": 1147 + }, + { + "epoch": 0.5872122762148337, + "grad_norm": 10.315488815307617, + "learning_rate": 1e-06, + "loss": 0.426, + "mean_token_accuracy": 0.8472956418991089, + "num_tokens": 3694499.0, + "step": 1148 + }, + { + "epoch": 0.5877237851662405, + "grad_norm": 8.3024263381958, + "learning_rate": 1e-06, + "loss": 0.3713, + "mean_token_accuracy": 0.8490402102470398, + "num_tokens": 3698524.0, + "step": 1149 + }, + { + "epoch": 0.5882352941176471, + "grad_norm": 14.3652925491333, + "learning_rate": 1e-06, + "loss": 0.3857, + "mean_token_accuracy": 0.8496336936950684, + "num_tokens": 3702255.0, + "step": 1150 + }, + { + "epoch": 0.5887468030690537, + "grad_norm": 10.334202766418457, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8661426305770874, + "num_tokens": 3705466.0, + "step": 1151 + }, + { + "epoch": 0.5892583120204603, + "grad_norm": 7.38871431350708, + "learning_rate": 1e-06, + "loss": 0.2423, + "mean_token_accuracy": 0.8984506726264954, + "num_tokens": 3708280.0, + "step": 1152 + }, + { + "epoch": 0.589769820971867, + "grad_norm": 9.873029708862305, + "learning_rate": 1e-06, + "loss": 0.3214, + "mean_token_accuracy": 0.8755739331245422, + "num_tokens": 3710981.0, + "step": 1153 + }, + { + "epoch": 0.5902813299232736, + "grad_norm": 10.957919120788574, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8980433940887451, + "num_tokens": 3714277.0, + "step": 1154 + }, + { + "epoch": 0.5907928388746803, + "grad_norm": 13.691415786743164, + "learning_rate": 1e-06, + "loss": 0.3795, + "mean_token_accuracy": 0.8690457344055176, + "num_tokens": 3717123.0, + "step": 1155 + }, + { + "epoch": 0.591304347826087, + "grad_norm": 9.298113822937012, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8678231239318848, + "num_tokens": 3721005.0, + "step": 1156 + }, + { + "epoch": 0.5918158567774936, + "grad_norm": 10.516525268554688, + "learning_rate": 1e-06, + "loss": 0.3326, + "mean_token_accuracy": 0.8807300925254822, + "num_tokens": 3724023.0, + "step": 1157 + }, + { + "epoch": 0.5923273657289002, + "grad_norm": 9.455720901489258, + "learning_rate": 1e-06, + "loss": 0.2413, + "mean_token_accuracy": 0.917529821395874, + "num_tokens": 3727000.0, + "step": 1158 + }, + { + "epoch": 0.5928388746803069, + "grad_norm": 10.225021362304688, + "learning_rate": 1e-06, + "loss": 0.3956, + "mean_token_accuracy": 0.8680474758148193, + "num_tokens": 3730378.0, + "step": 1159 + }, + { + "epoch": 0.5933503836317136, + "grad_norm": 10.478256225585938, + "learning_rate": 1e-06, + "loss": 0.3143, + "mean_token_accuracy": 0.8715623617172241, + "num_tokens": 3733148.0, + "step": 1160 + }, + { + "epoch": 0.5938618925831202, + "grad_norm": 10.772371292114258, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8827259540557861, + "num_tokens": 3736844.0, + "step": 1161 + }, + { + "epoch": 0.5943734015345269, + "grad_norm": 14.454479217529297, + "learning_rate": 1e-06, + "loss": 0.4854, + "mean_token_accuracy": 0.8355168104171753, + "num_tokens": 3739800.0, + "step": 1162 + }, + { + "epoch": 0.5948849104859335, + "grad_norm": 12.53548812866211, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.887832522392273, + "num_tokens": 3742521.0, + "step": 1163 + }, + { + "epoch": 0.5953964194373401, + "grad_norm": 11.852282524108887, + "learning_rate": 1e-06, + "loss": 0.3939, + "mean_token_accuracy": 0.8576551675796509, + "num_tokens": 3745397.0, + "step": 1164 + }, + { + "epoch": 0.5959079283887468, + "grad_norm": 20.696102142333984, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8685722351074219, + "num_tokens": 3748389.0, + "step": 1165 + }, + { + "epoch": 0.5964194373401535, + "grad_norm": 12.305212020874023, + "learning_rate": 1e-06, + "loss": 0.39, + "mean_token_accuracy": 0.857035219669342, + "num_tokens": 3750837.0, + "step": 1166 + }, + { + "epoch": 0.5969309462915601, + "grad_norm": 12.45964527130127, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.8685457706451416, + "num_tokens": 3753739.0, + "step": 1167 + }, + { + "epoch": 0.5974424552429668, + "grad_norm": 10.322422981262207, + "learning_rate": 1e-06, + "loss": 0.278, + "mean_token_accuracy": 0.900351345539093, + "num_tokens": 3756235.0, + "step": 1168 + }, + { + "epoch": 0.5979539641943734, + "grad_norm": 11.418726921081543, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.8804512023925781, + "num_tokens": 3759152.0, + "step": 1169 + }, + { + "epoch": 0.59846547314578, + "grad_norm": 9.399277687072754, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.8773841857910156, + "num_tokens": 3762023.0, + "step": 1170 + }, + { + "epoch": 0.5989769820971867, + "grad_norm": 11.49421215057373, + "learning_rate": 1e-06, + "loss": 0.3707, + "mean_token_accuracy": 0.8557829260826111, + "num_tokens": 3764853.0, + "step": 1171 + }, + { + "epoch": 0.5994884910485934, + "grad_norm": 12.238399505615234, + "learning_rate": 1e-06, + "loss": 0.3453, + "mean_token_accuracy": 0.8654708862304688, + "num_tokens": 3768096.0, + "step": 1172 + }, + { + "epoch": 0.6, + "grad_norm": 10.28905200958252, + "learning_rate": 1e-06, + "loss": 0.3687, + "mean_token_accuracy": 0.8497236967086792, + "num_tokens": 3770973.0, + "step": 1173 + }, + { + "epoch": 0.6005115089514067, + "grad_norm": 8.941548347473145, + "learning_rate": 1e-06, + "loss": 0.357, + "mean_token_accuracy": 0.8575128316879272, + "num_tokens": 3774670.0, + "step": 1174 + }, + { + "epoch": 0.6010230179028133, + "grad_norm": 7.868617534637451, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.8649881482124329, + "num_tokens": 3778503.0, + "step": 1175 + }, + { + "epoch": 0.6015345268542199, + "grad_norm": 14.370011329650879, + "learning_rate": 1e-06, + "loss": 0.4047, + "mean_token_accuracy": 0.8529391884803772, + "num_tokens": 3781469.0, + "step": 1176 + }, + { + "epoch": 0.6020460358056265, + "grad_norm": 8.830658912658691, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8909140825271606, + "num_tokens": 3783991.0, + "step": 1177 + }, + { + "epoch": 0.6025575447570333, + "grad_norm": 11.139373779296875, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.8632049560546875, + "num_tokens": 3786726.0, + "step": 1178 + }, + { + "epoch": 0.6030690537084399, + "grad_norm": 12.909387588500977, + "learning_rate": 1e-06, + "loss": 0.3981, + "mean_token_accuracy": 0.847350537776947, + "num_tokens": 3789352.0, + "step": 1179 + }, + { + "epoch": 0.6035805626598465, + "grad_norm": 14.290514945983887, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8810735940933228, + "num_tokens": 3792123.0, + "step": 1180 + }, + { + "epoch": 0.6040920716112532, + "grad_norm": 12.409744262695312, + "learning_rate": 1e-06, + "loss": 0.3105, + "mean_token_accuracy": 0.8872981667518616, + "num_tokens": 3795294.0, + "step": 1181 + }, + { + "epoch": 0.6046035805626598, + "grad_norm": 7.84775447845459, + "learning_rate": 1e-06, + "loss": 0.2035, + "mean_token_accuracy": 0.9046599864959717, + "num_tokens": 3798287.0, + "step": 1182 + }, + { + "epoch": 0.6051150895140665, + "grad_norm": 11.412379264831543, + "learning_rate": 1e-06, + "loss": 0.3702, + "mean_token_accuracy": 0.8569371700286865, + "num_tokens": 3801765.0, + "step": 1183 + }, + { + "epoch": 0.6056265984654732, + "grad_norm": 8.768613815307617, + "learning_rate": 1e-06, + "loss": 0.2697, + "mean_token_accuracy": 0.9034188389778137, + "num_tokens": 3805303.0, + "step": 1184 + }, + { + "epoch": 0.6061381074168798, + "grad_norm": 8.28930377960205, + "learning_rate": 1e-06, + "loss": 0.3024, + "mean_token_accuracy": 0.8471777439117432, + "num_tokens": 3808712.0, + "step": 1185 + }, + { + "epoch": 0.6066496163682864, + "grad_norm": 7.960793972015381, + "learning_rate": 1e-06, + "loss": 0.2752, + "mean_token_accuracy": 0.9014302492141724, + "num_tokens": 3813133.0, + "step": 1186 + }, + { + "epoch": 0.6071611253196931, + "grad_norm": 8.68681526184082, + "learning_rate": 1e-06, + "loss": 0.2603, + "mean_token_accuracy": 0.8891686201095581, + "num_tokens": 3815813.0, + "step": 1187 + }, + { + "epoch": 0.6076726342710997, + "grad_norm": 13.872517585754395, + "learning_rate": 1e-06, + "loss": 0.4528, + "mean_token_accuracy": 0.8382340669631958, + "num_tokens": 3818425.0, + "step": 1188 + }, + { + "epoch": 0.6081841432225064, + "grad_norm": 16.44076919555664, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8599863648414612, + "num_tokens": 3820878.0, + "step": 1189 + }, + { + "epoch": 0.6086956521739131, + "grad_norm": 12.28260326385498, + "learning_rate": 1e-06, + "loss": 0.4086, + "mean_token_accuracy": 0.8358621001243591, + "num_tokens": 3823523.0, + "step": 1190 + }, + { + "epoch": 0.6092071611253197, + "grad_norm": 11.426657676696777, + "learning_rate": 1e-06, + "loss": 0.3646, + "mean_token_accuracy": 0.8552541732788086, + "num_tokens": 3826488.0, + "step": 1191 + }, + { + "epoch": 0.6097186700767263, + "grad_norm": 10.902342796325684, + "learning_rate": 1e-06, + "loss": 0.2986, + "mean_token_accuracy": 0.887954592704773, + "num_tokens": 3829494.0, + "step": 1192 + }, + { + "epoch": 0.610230179028133, + "grad_norm": 13.560700416564941, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8677123785018921, + "num_tokens": 3832767.0, + "step": 1193 + }, + { + "epoch": 0.6107416879795396, + "grad_norm": 7.780962944030762, + "learning_rate": 1e-06, + "loss": 0.335, + "mean_token_accuracy": 0.8386998176574707, + "num_tokens": 3837022.0, + "step": 1194 + }, + { + "epoch": 0.6112531969309463, + "grad_norm": 12.032415390014648, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.890953004360199, + "num_tokens": 3839690.0, + "step": 1195 + }, + { + "epoch": 0.611764705882353, + "grad_norm": 11.61088752746582, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.8737046718597412, + "num_tokens": 3843079.0, + "step": 1196 + }, + { + "epoch": 0.6122762148337596, + "grad_norm": 10.207722663879395, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.874363124370575, + "num_tokens": 3846260.0, + "step": 1197 + }, + { + "epoch": 0.6127877237851662, + "grad_norm": 7.859294891357422, + "learning_rate": 1e-06, + "loss": 0.2547, + "mean_token_accuracy": 0.870315432548523, + "num_tokens": 3849281.0, + "step": 1198 + }, + { + "epoch": 0.6132992327365729, + "grad_norm": 10.657724380493164, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.8963853716850281, + "num_tokens": 3852348.0, + "step": 1199 + }, + { + "epoch": 0.6138107416879796, + "grad_norm": 10.789295196533203, + "learning_rate": 1e-06, + "loss": 0.3507, + "mean_token_accuracy": 0.8784264922142029, + "num_tokens": 3855635.0, + "step": 1200 + }, + { + "epoch": 0.6143222506393862, + "grad_norm": 12.96703815460205, + "learning_rate": 1e-06, + "loss": 0.4732, + "mean_token_accuracy": 0.8191946744918823, + "num_tokens": 3859010.0, + "step": 1201 + }, + { + "epoch": 0.6148337595907929, + "grad_norm": 9.84344482421875, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8740864992141724, + "num_tokens": 3861498.0, + "step": 1202 + }, + { + "epoch": 0.6153452685421995, + "grad_norm": 9.747871398925781, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8557623624801636, + "num_tokens": 3864583.0, + "step": 1203 + }, + { + "epoch": 0.6158567774936061, + "grad_norm": 14.88211727142334, + "learning_rate": 1e-06, + "loss": 0.4072, + "mean_token_accuracy": 0.8688552379608154, + "num_tokens": 3867374.0, + "step": 1204 + }, + { + "epoch": 0.6163682864450127, + "grad_norm": 11.055643081665039, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8386242389678955, + "num_tokens": 3870816.0, + "step": 1205 + }, + { + "epoch": 0.6168797953964195, + "grad_norm": 13.635407447814941, + "learning_rate": 1e-06, + "loss": 0.3796, + "mean_token_accuracy": 0.8528604507446289, + "num_tokens": 3873797.0, + "step": 1206 + }, + { + "epoch": 0.6173913043478261, + "grad_norm": 11.258234977722168, + "learning_rate": 1e-06, + "loss": 0.3478, + "mean_token_accuracy": 0.8796788454055786, + "num_tokens": 3876388.0, + "step": 1207 + }, + { + "epoch": 0.6179028132992327, + "grad_norm": 9.611861228942871, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8522601127624512, + "num_tokens": 3879543.0, + "step": 1208 + }, + { + "epoch": 0.6184143222506394, + "grad_norm": 14.070998191833496, + "learning_rate": 1e-06, + "loss": 0.3885, + "mean_token_accuracy": 0.8376564979553223, + "num_tokens": 3882820.0, + "step": 1209 + }, + { + "epoch": 0.618925831202046, + "grad_norm": 9.848541259765625, + "learning_rate": 1e-06, + "loss": 0.3997, + "mean_token_accuracy": 0.8646836280822754, + "num_tokens": 3886190.0, + "step": 1210 + }, + { + "epoch": 0.6194373401534526, + "grad_norm": 12.269155502319336, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8720510005950928, + "num_tokens": 3888979.0, + "step": 1211 + }, + { + "epoch": 0.6199488491048594, + "grad_norm": 9.745532035827637, + "learning_rate": 1e-06, + "loss": 0.3119, + "mean_token_accuracy": 0.8833258152008057, + "num_tokens": 3891854.0, + "step": 1212 + }, + { + "epoch": 0.620460358056266, + "grad_norm": 10.458219528198242, + "learning_rate": 1e-06, + "loss": 0.3524, + "mean_token_accuracy": 0.8633208274841309, + "num_tokens": 3895164.0, + "step": 1213 + }, + { + "epoch": 0.6209718670076726, + "grad_norm": 12.977638244628906, + "learning_rate": 1e-06, + "loss": 0.4279, + "mean_token_accuracy": 0.8681859374046326, + "num_tokens": 3897896.0, + "step": 1214 + }, + { + "epoch": 0.6214833759590793, + "grad_norm": 11.424116134643555, + "learning_rate": 1e-06, + "loss": 0.2979, + "mean_token_accuracy": 0.8754730224609375, + "num_tokens": 3901892.0, + "step": 1215 + }, + { + "epoch": 0.6219948849104859, + "grad_norm": 10.976190567016602, + "learning_rate": 1e-06, + "loss": 0.3185, + "mean_token_accuracy": 0.8602217435836792, + "num_tokens": 3905039.0, + "step": 1216 + }, + { + "epoch": 0.6225063938618925, + "grad_norm": 9.375754356384277, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8557100892066956, + "num_tokens": 3908324.0, + "step": 1217 + }, + { + "epoch": 0.6230179028132993, + "grad_norm": 12.297677040100098, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.8664782047271729, + "num_tokens": 3911096.0, + "step": 1218 + }, + { + "epoch": 0.6235294117647059, + "grad_norm": 9.092273712158203, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8871626853942871, + "num_tokens": 3914806.0, + "step": 1219 + }, + { + "epoch": 0.6240409207161125, + "grad_norm": 12.5184965133667, + "learning_rate": 1e-06, + "loss": 0.4785, + "mean_token_accuracy": 0.8246667385101318, + "num_tokens": 3917825.0, + "step": 1220 + }, + { + "epoch": 0.6245524296675192, + "grad_norm": 8.849267959594727, + "learning_rate": 1e-06, + "loss": 0.3246, + "mean_token_accuracy": 0.8449758291244507, + "num_tokens": 3922289.0, + "step": 1221 + }, + { + "epoch": 0.6250639386189258, + "grad_norm": 9.242704391479492, + "learning_rate": 1e-06, + "loss": 0.3409, + "mean_token_accuracy": 0.9028418660163879, + "num_tokens": 3926308.0, + "step": 1222 + }, + { + "epoch": 0.6255754475703325, + "grad_norm": 12.469637870788574, + "learning_rate": 1e-06, + "loss": 0.4101, + "mean_token_accuracy": 0.8273317217826843, + "num_tokens": 3928868.0, + "step": 1223 + }, + { + "epoch": 0.6260869565217392, + "grad_norm": 8.576268196105957, + "learning_rate": 1e-06, + "loss": 0.3741, + "mean_token_accuracy": 0.8385591506958008, + "num_tokens": 3933572.0, + "step": 1224 + }, + { + "epoch": 0.6265984654731458, + "grad_norm": 9.452706336975098, + "learning_rate": 1e-06, + "loss": 0.2654, + "mean_token_accuracy": 0.8748040199279785, + "num_tokens": 3936796.0, + "step": 1225 + }, + { + "epoch": 0.6271099744245524, + "grad_norm": 11.330033302307129, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.8914318084716797, + "num_tokens": 3939910.0, + "step": 1226 + }, + { + "epoch": 0.6276214833759591, + "grad_norm": 14.152462005615234, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8582308292388916, + "num_tokens": 3942673.0, + "step": 1227 + }, + { + "epoch": 0.6281329923273657, + "grad_norm": 16.508731842041016, + "learning_rate": 1e-06, + "loss": 0.4262, + "mean_token_accuracy": 0.8229575753211975, + "num_tokens": 3945250.0, + "step": 1228 + }, + { + "epoch": 0.6286445012787724, + "grad_norm": 13.079713821411133, + "learning_rate": 1e-06, + "loss": 0.2709, + "mean_token_accuracy": 0.891451358795166, + "num_tokens": 3948868.0, + "step": 1229 + }, + { + "epoch": 0.629156010230179, + "grad_norm": 10.077868461608887, + "learning_rate": 1e-06, + "loss": 0.3278, + "mean_token_accuracy": 0.9079709649085999, + "num_tokens": 3952523.0, + "step": 1230 + }, + { + "epoch": 0.6296675191815857, + "grad_norm": 14.594223976135254, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8594171404838562, + "num_tokens": 3956191.0, + "step": 1231 + }, + { + "epoch": 0.6301790281329923, + "grad_norm": 13.813314437866211, + "learning_rate": 1e-06, + "loss": 0.4483, + "mean_token_accuracy": 0.8387018442153931, + "num_tokens": 3959089.0, + "step": 1232 + }, + { + "epoch": 0.6306905370843989, + "grad_norm": 9.605485916137695, + "learning_rate": 1e-06, + "loss": 0.3574, + "mean_token_accuracy": 0.8451111912727356, + "num_tokens": 3962838.0, + "step": 1233 + }, + { + "epoch": 0.6312020460358057, + "grad_norm": 15.302794456481934, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8679344654083252, + "num_tokens": 3965561.0, + "step": 1234 + }, + { + "epoch": 0.6317135549872123, + "grad_norm": 9.858880996704102, + "learning_rate": 1e-06, + "loss": 0.3256, + "mean_token_accuracy": 0.8578156232833862, + "num_tokens": 3968478.0, + "step": 1235 + }, + { + "epoch": 0.6322250639386189, + "grad_norm": 12.604254722595215, + "learning_rate": 1e-06, + "loss": 0.3378, + "mean_token_accuracy": 0.8776736259460449, + "num_tokens": 3971135.0, + "step": 1236 + }, + { + "epoch": 0.6327365728900256, + "grad_norm": 13.093145370483398, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8638942837715149, + "num_tokens": 3974208.0, + "step": 1237 + }, + { + "epoch": 0.6332480818414322, + "grad_norm": 12.232504844665527, + "learning_rate": 1e-06, + "loss": 0.4153, + "mean_token_accuracy": 0.8408013582229614, + "num_tokens": 3976988.0, + "step": 1238 + }, + { + "epoch": 0.6337595907928388, + "grad_norm": 9.537192344665527, + "learning_rate": 1e-06, + "loss": 0.3083, + "mean_token_accuracy": 0.8624410629272461, + "num_tokens": 3980719.0, + "step": 1239 + }, + { + "epoch": 0.6342710997442456, + "grad_norm": 9.537771224975586, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.898473858833313, + "num_tokens": 3983774.0, + "step": 1240 + }, + { + "epoch": 0.6347826086956522, + "grad_norm": 12.006194114685059, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8743434548377991, + "num_tokens": 3986474.0, + "step": 1241 + }, + { + "epoch": 0.6352941176470588, + "grad_norm": 6.3404107093811035, + "learning_rate": 1e-06, + "loss": 0.2018, + "mean_token_accuracy": 0.9127280712127686, + "num_tokens": 3989383.0, + "step": 1242 + }, + { + "epoch": 0.6358056265984655, + "grad_norm": 11.318603515625, + "learning_rate": 1e-06, + "loss": 0.4093, + "mean_token_accuracy": 0.8887819647789001, + "num_tokens": 3992891.0, + "step": 1243 + }, + { + "epoch": 0.6363171355498721, + "grad_norm": 10.518275260925293, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8809239268302917, + "num_tokens": 3995608.0, + "step": 1244 + }, + { + "epoch": 0.6368286445012787, + "grad_norm": 8.468523979187012, + "learning_rate": 1e-06, + "loss": 0.2667, + "mean_token_accuracy": 0.90129154920578, + "num_tokens": 3999839.0, + "step": 1245 + }, + { + "epoch": 0.6373401534526855, + "grad_norm": 10.295276641845703, + "learning_rate": 1e-06, + "loss": 0.3961, + "mean_token_accuracy": 0.8554266691207886, + "num_tokens": 4003227.0, + "step": 1246 + }, + { + "epoch": 0.6378516624040921, + "grad_norm": 10.620259284973145, + "learning_rate": 1e-06, + "loss": 0.3226, + "mean_token_accuracy": 0.8560187220573425, + "num_tokens": 4006370.0, + "step": 1247 + }, + { + "epoch": 0.6383631713554987, + "grad_norm": 11.68477725982666, + "learning_rate": 1e-06, + "loss": 0.2861, + "mean_token_accuracy": 0.8738769292831421, + "num_tokens": 4009400.0, + "step": 1248 + }, + { + "epoch": 0.6388746803069054, + "grad_norm": 13.499238014221191, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8805080652236938, + "num_tokens": 4012384.0, + "step": 1249 + }, + { + "epoch": 0.639386189258312, + "grad_norm": 9.68447208404541, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8891477584838867, + "num_tokens": 4015675.0, + "step": 1250 + }, + { + "epoch": 0.6398976982097186, + "grad_norm": 11.649696350097656, + "learning_rate": 1e-06, + "loss": 0.4306, + "mean_token_accuracy": 0.8350684642791748, + "num_tokens": 4018649.0, + "step": 1251 + }, + { + "epoch": 0.6404092071611254, + "grad_norm": 10.330309867858887, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8917169570922852, + "num_tokens": 4021345.0, + "step": 1252 + }, + { + "epoch": 0.640920716112532, + "grad_norm": 11.677372932434082, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.847533106803894, + "num_tokens": 4024089.0, + "step": 1253 + }, + { + "epoch": 0.6414322250639386, + "grad_norm": 10.031149864196777, + "learning_rate": 1e-06, + "loss": 0.3046, + "mean_token_accuracy": 0.8585271835327148, + "num_tokens": 4027318.0, + "step": 1254 + }, + { + "epoch": 0.6419437340153452, + "grad_norm": 14.309308052062988, + "learning_rate": 1e-06, + "loss": 0.3993, + "mean_token_accuracy": 0.8624383807182312, + "num_tokens": 4029827.0, + "step": 1255 + }, + { + "epoch": 0.6424552429667519, + "grad_norm": 10.200884819030762, + "learning_rate": 1e-06, + "loss": 0.4008, + "mean_token_accuracy": 0.8494430780410767, + "num_tokens": 4033151.0, + "step": 1256 + }, + { + "epoch": 0.6429667519181586, + "grad_norm": 9.600420951843262, + "learning_rate": 1e-06, + "loss": 0.4188, + "mean_token_accuracy": 0.8366366028785706, + "num_tokens": 4037782.0, + "step": 1257 + }, + { + "epoch": 0.6434782608695652, + "grad_norm": 14.630768775939941, + "learning_rate": 1e-06, + "loss": 0.4571, + "mean_token_accuracy": 0.8301141858100891, + "num_tokens": 4040674.0, + "step": 1258 + }, + { + "epoch": 0.6439897698209719, + "grad_norm": 12.349617958068848, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8633977174758911, + "num_tokens": 4044049.0, + "step": 1259 + }, + { + "epoch": 0.6445012787723785, + "grad_norm": 11.036015510559082, + "learning_rate": 1e-06, + "loss": 0.294, + "mean_token_accuracy": 0.8956751823425293, + "num_tokens": 4047239.0, + "step": 1260 + }, + { + "epoch": 0.6450127877237851, + "grad_norm": 10.036813735961914, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8823615312576294, + "num_tokens": 4051698.0, + "step": 1261 + }, + { + "epoch": 0.6455242966751918, + "grad_norm": 8.82288932800293, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8744019865989685, + "num_tokens": 4054683.0, + "step": 1262 + }, + { + "epoch": 0.6460358056265985, + "grad_norm": 9.008418083190918, + "learning_rate": 1e-06, + "loss": 0.2478, + "mean_token_accuracy": 0.895815908908844, + "num_tokens": 4058023.0, + "step": 1263 + }, + { + "epoch": 0.6465473145780051, + "grad_norm": 9.292862892150879, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.8780657052993774, + "num_tokens": 4061527.0, + "step": 1264 + }, + { + "epoch": 0.6470588235294118, + "grad_norm": 10.279597282409668, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8984898924827576, + "num_tokens": 4064540.0, + "step": 1265 + }, + { + "epoch": 0.6475703324808184, + "grad_norm": 13.264421463012695, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8738516569137573, + "num_tokens": 4067118.0, + "step": 1266 + }, + { + "epoch": 0.648081841432225, + "grad_norm": 10.30312728881836, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8665091395378113, + "num_tokens": 4070741.0, + "step": 1267 + }, + { + "epoch": 0.6485933503836318, + "grad_norm": 17.13248062133789, + "learning_rate": 1e-06, + "loss": 0.3213, + "mean_token_accuracy": 0.8679739832878113, + "num_tokens": 4073916.0, + "step": 1268 + }, + { + "epoch": 0.6491048593350384, + "grad_norm": 14.096405982971191, + "learning_rate": 1e-06, + "loss": 0.4195, + "mean_token_accuracy": 0.8494715690612793, + "num_tokens": 4076922.0, + "step": 1269 + }, + { + "epoch": 0.649616368286445, + "grad_norm": 9.619511604309082, + "learning_rate": 1e-06, + "loss": 0.2915, + "mean_token_accuracy": 0.868065357208252, + "num_tokens": 4080536.0, + "step": 1270 + }, + { + "epoch": 0.6501278772378517, + "grad_norm": 8.202055931091309, + "learning_rate": 1e-06, + "loss": 0.2355, + "mean_token_accuracy": 0.8811783790588379, + "num_tokens": 4083232.0, + "step": 1271 + }, + { + "epoch": 0.6506393861892583, + "grad_norm": 11.703287124633789, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8486140370368958, + "num_tokens": 4085579.0, + "step": 1272 + }, + { + "epoch": 0.6511508951406649, + "grad_norm": 9.362549781799316, + "learning_rate": 1e-06, + "loss": 0.2967, + "mean_token_accuracy": 0.874351978302002, + "num_tokens": 4088768.0, + "step": 1273 + }, + { + "epoch": 0.6516624040920717, + "grad_norm": 12.730229377746582, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8631933927536011, + "num_tokens": 4092250.0, + "step": 1274 + }, + { + "epoch": 0.6521739130434783, + "grad_norm": 14.818039894104004, + "learning_rate": 1e-06, + "loss": 0.3901, + "mean_token_accuracy": 0.8632419109344482, + "num_tokens": 4094829.0, + "step": 1275 + }, + { + "epoch": 0.6526854219948849, + "grad_norm": 11.4035062789917, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8746525049209595, + "num_tokens": 4097648.0, + "step": 1276 + }, + { + "epoch": 0.6531969309462916, + "grad_norm": 10.689445495605469, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8775131702423096, + "num_tokens": 4101130.0, + "step": 1277 + }, + { + "epoch": 0.6537084398976982, + "grad_norm": 8.835206031799316, + "learning_rate": 1e-06, + "loss": 0.3265, + "mean_token_accuracy": 0.8807447552680969, + "num_tokens": 4103470.0, + "step": 1278 + }, + { + "epoch": 0.6542199488491048, + "grad_norm": 11.89098072052002, + "learning_rate": 1e-06, + "loss": 0.3696, + "mean_token_accuracy": 0.8503659963607788, + "num_tokens": 4106377.0, + "step": 1279 + }, + { + "epoch": 0.6547314578005116, + "grad_norm": 9.668766975402832, + "learning_rate": 1e-06, + "loss": 0.3366, + "mean_token_accuracy": 0.8665227293968201, + "num_tokens": 4109562.0, + "step": 1280 + }, + { + "epoch": 0.6552429667519182, + "grad_norm": 13.122664451599121, + "learning_rate": 1e-06, + "loss": 0.4074, + "mean_token_accuracy": 0.852183997631073, + "num_tokens": 4113329.0, + "step": 1281 + }, + { + "epoch": 0.6557544757033248, + "grad_norm": 10.48562240600586, + "learning_rate": 1e-06, + "loss": 0.2718, + "mean_token_accuracy": 0.8892161846160889, + "num_tokens": 4116460.0, + "step": 1282 + }, + { + "epoch": 0.6562659846547314, + "grad_norm": 11.3262300491333, + "learning_rate": 1e-06, + "loss": 0.4239, + "mean_token_accuracy": 0.8229724764823914, + "num_tokens": 4120458.0, + "step": 1283 + }, + { + "epoch": 0.6567774936061381, + "grad_norm": 11.20020580291748, + "learning_rate": 1e-06, + "loss": 0.3818, + "mean_token_accuracy": 0.8451340198516846, + "num_tokens": 4123142.0, + "step": 1284 + }, + { + "epoch": 0.6572890025575447, + "grad_norm": 10.303257942199707, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8643233776092529, + "num_tokens": 4126762.0, + "step": 1285 + }, + { + "epoch": 0.6578005115089514, + "grad_norm": 11.552943229675293, + "learning_rate": 1e-06, + "loss": 0.3293, + "mean_token_accuracy": 0.8771601915359497, + "num_tokens": 4130145.0, + "step": 1286 + }, + { + "epoch": 0.6583120204603581, + "grad_norm": 9.450655937194824, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8551141023635864, + "num_tokens": 4133451.0, + "step": 1287 + }, + { + "epoch": 0.6588235294117647, + "grad_norm": 20.056413650512695, + "learning_rate": 1e-06, + "loss": 0.458, + "mean_token_accuracy": 0.8374632596969604, + "num_tokens": 4136304.0, + "step": 1288 + }, + { + "epoch": 0.6593350383631713, + "grad_norm": 8.947237968444824, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8609842658042908, + "num_tokens": 4139599.0, + "step": 1289 + }, + { + "epoch": 0.659846547314578, + "grad_norm": 10.354096412658691, + "learning_rate": 1e-06, + "loss": 0.3718, + "mean_token_accuracy": 0.8452842831611633, + "num_tokens": 4143097.0, + "step": 1290 + }, + { + "epoch": 0.6603580562659846, + "grad_norm": 6.954370975494385, + "learning_rate": 1e-06, + "loss": 0.2226, + "mean_token_accuracy": 0.8851746320724487, + "num_tokens": 4146551.0, + "step": 1291 + }, + { + "epoch": 0.6608695652173913, + "grad_norm": 12.349481582641602, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.880494236946106, + "num_tokens": 4149611.0, + "step": 1292 + }, + { + "epoch": 0.661381074168798, + "grad_norm": 9.357484817504883, + "learning_rate": 1e-06, + "loss": 0.3853, + "mean_token_accuracy": 0.8207080960273743, + "num_tokens": 4153744.0, + "step": 1293 + }, + { + "epoch": 0.6618925831202046, + "grad_norm": 9.367287635803223, + "learning_rate": 1e-06, + "loss": 0.2436, + "mean_token_accuracy": 0.8847532272338867, + "num_tokens": 4156412.0, + "step": 1294 + }, + { + "epoch": 0.6624040920716112, + "grad_norm": 8.364608764648438, + "learning_rate": 1e-06, + "loss": 0.3161, + "mean_token_accuracy": 0.8706939816474915, + "num_tokens": 4160006.0, + "step": 1295 + }, + { + "epoch": 0.6629156010230179, + "grad_norm": 12.608636856079102, + "learning_rate": 1e-06, + "loss": 0.3652, + "mean_token_accuracy": 0.8831836581230164, + "num_tokens": 4163375.0, + "step": 1296 + }, + { + "epoch": 0.6634271099744246, + "grad_norm": 9.706138610839844, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8585764765739441, + "num_tokens": 4166096.0, + "step": 1297 + }, + { + "epoch": 0.6639386189258312, + "grad_norm": 9.59891414642334, + "learning_rate": 1e-06, + "loss": 0.2534, + "mean_token_accuracy": 0.909267783164978, + "num_tokens": 4168889.0, + "step": 1298 + }, + { + "epoch": 0.6644501278772379, + "grad_norm": 9.319005966186523, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8793442249298096, + "num_tokens": 4173565.0, + "step": 1299 + }, + { + "epoch": 0.6649616368286445, + "grad_norm": 9.715263366699219, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8822182416915894, + "num_tokens": 4176578.0, + "step": 1300 + }, + { + "epoch": 0.6654731457800511, + "grad_norm": 5.408201217651367, + "learning_rate": 1e-06, + "loss": 0.1397, + "mean_token_accuracy": 0.8995860815048218, + "num_tokens": 4181339.0, + "step": 1301 + }, + { + "epoch": 0.6659846547314578, + "grad_norm": 11.893036842346191, + "learning_rate": 1e-06, + "loss": 0.3925, + "mean_token_accuracy": 0.8632230162620544, + "num_tokens": 4183585.0, + "step": 1302 + }, + { + "epoch": 0.6664961636828645, + "grad_norm": 10.660306930541992, + "learning_rate": 1e-06, + "loss": 0.3767, + "mean_token_accuracy": 0.8603694438934326, + "num_tokens": 4186660.0, + "step": 1303 + }, + { + "epoch": 0.6670076726342711, + "grad_norm": 8.369747161865234, + "learning_rate": 1e-06, + "loss": 0.3375, + "mean_token_accuracy": 0.8787676095962524, + "num_tokens": 4190248.0, + "step": 1304 + }, + { + "epoch": 0.6675191815856778, + "grad_norm": 11.002497673034668, + "learning_rate": 1e-06, + "loss": 0.4014, + "mean_token_accuracy": 0.8623042106628418, + "num_tokens": 4193435.0, + "step": 1305 + }, + { + "epoch": 0.6680306905370844, + "grad_norm": 10.256441116333008, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8746579885482788, + "num_tokens": 4197318.0, + "step": 1306 + }, + { + "epoch": 0.668542199488491, + "grad_norm": 12.982004165649414, + "learning_rate": 1e-06, + "loss": 0.3761, + "mean_token_accuracy": 0.84002685546875, + "num_tokens": 4199591.0, + "step": 1307 + }, + { + "epoch": 0.6690537084398976, + "grad_norm": 10.098322868347168, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.8893937468528748, + "num_tokens": 4202447.0, + "step": 1308 + }, + { + "epoch": 0.6695652173913044, + "grad_norm": 10.483053207397461, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8779987692832947, + "num_tokens": 4205000.0, + "step": 1309 + }, + { + "epoch": 0.670076726342711, + "grad_norm": 10.909928321838379, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8707313537597656, + "num_tokens": 4208488.0, + "step": 1310 + }, + { + "epoch": 0.6705882352941176, + "grad_norm": 8.94771957397461, + "learning_rate": 1e-06, + "loss": 0.2246, + "mean_token_accuracy": 0.918169379234314, + "num_tokens": 4211206.0, + "step": 1311 + }, + { + "epoch": 0.6710997442455243, + "grad_norm": 11.74841022491455, + "learning_rate": 1e-06, + "loss": 0.3931, + "mean_token_accuracy": 0.8454799652099609, + "num_tokens": 4214677.0, + "step": 1312 + }, + { + "epoch": 0.6716112531969309, + "grad_norm": 12.078686714172363, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.8708317279815674, + "num_tokens": 4218360.0, + "step": 1313 + }, + { + "epoch": 0.6721227621483375, + "grad_norm": 10.48958969116211, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8829506635665894, + "num_tokens": 4221204.0, + "step": 1314 + }, + { + "epoch": 0.6726342710997443, + "grad_norm": 7.754645347595215, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.9032129645347595, + "num_tokens": 4223815.0, + "step": 1315 + }, + { + "epoch": 0.6731457800511509, + "grad_norm": 12.098584175109863, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8796732425689697, + "num_tokens": 4226137.0, + "step": 1316 + }, + { + "epoch": 0.6736572890025575, + "grad_norm": 11.401557922363281, + "learning_rate": 1e-06, + "loss": 0.3864, + "mean_token_accuracy": 0.8514106273651123, + "num_tokens": 4229375.0, + "step": 1317 + }, + { + "epoch": 0.6741687979539642, + "grad_norm": 8.565624237060547, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8882774114608765, + "num_tokens": 4232104.0, + "step": 1318 + }, + { + "epoch": 0.6746803069053708, + "grad_norm": 10.007513046264648, + "learning_rate": 1e-06, + "loss": 0.2728, + "mean_token_accuracy": 0.9048153162002563, + "num_tokens": 4235321.0, + "step": 1319 + }, + { + "epoch": 0.6751918158567775, + "grad_norm": 10.650408744812012, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8449912667274475, + "num_tokens": 4238777.0, + "step": 1320 + }, + { + "epoch": 0.6757033248081842, + "grad_norm": 11.609333992004395, + "learning_rate": 1e-06, + "loss": 0.3912, + "mean_token_accuracy": 0.8610537052154541, + "num_tokens": 4241526.0, + "step": 1321 + }, + { + "epoch": 0.6762148337595908, + "grad_norm": 12.117566108703613, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8729913234710693, + "num_tokens": 4245042.0, + "step": 1322 + }, + { + "epoch": 0.6767263427109974, + "grad_norm": 10.557023048400879, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8729380369186401, + "num_tokens": 4247932.0, + "step": 1323 + }, + { + "epoch": 0.6772378516624041, + "grad_norm": 15.653379440307617, + "learning_rate": 1e-06, + "loss": 0.4309, + "mean_token_accuracy": 0.8364406228065491, + "num_tokens": 4250900.0, + "step": 1324 + }, + { + "epoch": 0.6777493606138107, + "grad_norm": 12.063697814941406, + "learning_rate": 1e-06, + "loss": 0.4258, + "mean_token_accuracy": 0.8684910535812378, + "num_tokens": 4253875.0, + "step": 1325 + }, + { + "epoch": 0.6782608695652174, + "grad_norm": 11.573298454284668, + "learning_rate": 1e-06, + "loss": 0.3865, + "mean_token_accuracy": 0.8534279465675354, + "num_tokens": 4257079.0, + "step": 1326 + }, + { + "epoch": 0.6787723785166241, + "grad_norm": 15.67264175415039, + "learning_rate": 1e-06, + "loss": 0.4529, + "mean_token_accuracy": 0.8563243746757507, + "num_tokens": 4259865.0, + "step": 1327 + }, + { + "epoch": 0.6792838874680307, + "grad_norm": 15.799976348876953, + "learning_rate": 1e-06, + "loss": 0.3888, + "mean_token_accuracy": 0.8592345714569092, + "num_tokens": 4263083.0, + "step": 1328 + }, + { + "epoch": 0.6797953964194373, + "grad_norm": 8.358271598815918, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.8943663835525513, + "num_tokens": 4266551.0, + "step": 1329 + }, + { + "epoch": 0.680306905370844, + "grad_norm": 7.9154181480407715, + "learning_rate": 1e-06, + "loss": 0.2799, + "mean_token_accuracy": 0.8673769235610962, + "num_tokens": 4270270.0, + "step": 1330 + }, + { + "epoch": 0.6808184143222507, + "grad_norm": 10.263226509094238, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.867318868637085, + "num_tokens": 4274599.0, + "step": 1331 + }, + { + "epoch": 0.6813299232736573, + "grad_norm": 9.310510635375977, + "learning_rate": 1e-06, + "loss": 0.363, + "mean_token_accuracy": 0.861941933631897, + "num_tokens": 4278123.0, + "step": 1332 + }, + { + "epoch": 0.681841432225064, + "grad_norm": 9.557873725891113, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.9075155258178711, + "num_tokens": 4282062.0, + "step": 1333 + }, + { + "epoch": 0.6823529411764706, + "grad_norm": 12.337969779968262, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8290508985519409, + "num_tokens": 4285252.0, + "step": 1334 + }, + { + "epoch": 0.6828644501278772, + "grad_norm": 10.688824653625488, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8768067955970764, + "num_tokens": 4288389.0, + "step": 1335 + }, + { + "epoch": 0.6833759590792838, + "grad_norm": 15.786665916442871, + "learning_rate": 1e-06, + "loss": 0.4482, + "mean_token_accuracy": 0.8703786134719849, + "num_tokens": 4291658.0, + "step": 1336 + }, + { + "epoch": 0.6838874680306906, + "grad_norm": 10.574238777160645, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8769915699958801, + "num_tokens": 4294836.0, + "step": 1337 + }, + { + "epoch": 0.6843989769820972, + "grad_norm": 11.116721153259277, + "learning_rate": 1e-06, + "loss": 0.4296, + "mean_token_accuracy": 0.8584549427032471, + "num_tokens": 4297389.0, + "step": 1338 + }, + { + "epoch": 0.6849104859335038, + "grad_norm": 8.761027336120605, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8610930442810059, + "num_tokens": 4300403.0, + "step": 1339 + }, + { + "epoch": 0.6854219948849105, + "grad_norm": 13.315133094787598, + "learning_rate": 1e-06, + "loss": 0.4454, + "mean_token_accuracy": 0.8304353952407837, + "num_tokens": 4303453.0, + "step": 1340 + }, + { + "epoch": 0.6859335038363171, + "grad_norm": 11.171506881713867, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8789963722229004, + "num_tokens": 4306212.0, + "step": 1341 + }, + { + "epoch": 0.6864450127877237, + "grad_norm": 13.354803085327148, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.884636640548706, + "num_tokens": 4309704.0, + "step": 1342 + }, + { + "epoch": 0.6869565217391305, + "grad_norm": 8.938164710998535, + "learning_rate": 1e-06, + "loss": 0.3576, + "mean_token_accuracy": 0.8817752599716187, + "num_tokens": 4314263.0, + "step": 1343 + }, + { + "epoch": 0.6874680306905371, + "grad_norm": 9.161478042602539, + "learning_rate": 1e-06, + "loss": 0.2337, + "mean_token_accuracy": 0.8953856229782104, + "num_tokens": 4317892.0, + "step": 1344 + }, + { + "epoch": 0.6879795396419437, + "grad_norm": 8.256343841552734, + "learning_rate": 1e-06, + "loss": 0.1915, + "mean_token_accuracy": 0.9277629852294922, + "num_tokens": 4321169.0, + "step": 1345 + }, + { + "epoch": 0.6884910485933504, + "grad_norm": 9.05270004272461, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8643615245819092, + "num_tokens": 4324482.0, + "step": 1346 + }, + { + "epoch": 0.689002557544757, + "grad_norm": 10.679677963256836, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8543557524681091, + "num_tokens": 4327666.0, + "step": 1347 + }, + { + "epoch": 0.6895140664961636, + "grad_norm": 10.120573997497559, + "learning_rate": 1e-06, + "loss": 0.3432, + "mean_token_accuracy": 0.8481442332267761, + "num_tokens": 4331148.0, + "step": 1348 + }, + { + "epoch": 0.6900255754475704, + "grad_norm": 9.484328269958496, + "learning_rate": 1e-06, + "loss": 0.3211, + "mean_token_accuracy": 0.889230489730835, + "num_tokens": 4334233.0, + "step": 1349 + }, + { + "epoch": 0.690537084398977, + "grad_norm": 8.166836738586426, + "learning_rate": 1e-06, + "loss": 0.2858, + "mean_token_accuracy": 0.8809307813644409, + "num_tokens": 4338034.0, + "step": 1350 + }, + { + "epoch": 0.6910485933503836, + "grad_norm": 8.117595672607422, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8659146428108215, + "num_tokens": 4341983.0, + "step": 1351 + }, + { + "epoch": 0.6915601023017903, + "grad_norm": 8.914999008178711, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.8933382630348206, + "num_tokens": 4345387.0, + "step": 1352 + }, + { + "epoch": 0.6920716112531969, + "grad_norm": 13.678491592407227, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.8368951082229614, + "num_tokens": 4349229.0, + "step": 1353 + }, + { + "epoch": 0.6925831202046036, + "grad_norm": 10.422210693359375, + "learning_rate": 1e-06, + "loss": 0.4054, + "mean_token_accuracy": 0.8244720101356506, + "num_tokens": 4352430.0, + "step": 1354 + }, + { + "epoch": 0.6930946291560103, + "grad_norm": 13.437305450439453, + "learning_rate": 1e-06, + "loss": 0.3837, + "mean_token_accuracy": 0.8708065748214722, + "num_tokens": 4355694.0, + "step": 1355 + }, + { + "epoch": 0.6936061381074169, + "grad_norm": 11.992202758789062, + "learning_rate": 1e-06, + "loss": 0.3715, + "mean_token_accuracy": 0.8555386066436768, + "num_tokens": 4359120.0, + "step": 1356 + }, + { + "epoch": 0.6941176470588235, + "grad_norm": 10.584592819213867, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.855395495891571, + "num_tokens": 4362024.0, + "step": 1357 + }, + { + "epoch": 0.6946291560102302, + "grad_norm": 11.86050033569336, + "learning_rate": 1e-06, + "loss": 0.4029, + "mean_token_accuracy": 0.8428738117218018, + "num_tokens": 4365138.0, + "step": 1358 + }, + { + "epoch": 0.6951406649616368, + "grad_norm": 9.519515037536621, + "learning_rate": 1e-06, + "loss": 0.3987, + "mean_token_accuracy": 0.8448227643966675, + "num_tokens": 4368821.0, + "step": 1359 + }, + { + "epoch": 0.6956521739130435, + "grad_norm": 11.048541069030762, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8546985387802124, + "num_tokens": 4371686.0, + "step": 1360 + }, + { + "epoch": 0.6961636828644501, + "grad_norm": 10.776700973510742, + "learning_rate": 1e-06, + "loss": 0.4074, + "mean_token_accuracy": 0.8388689756393433, + "num_tokens": 4375120.0, + "step": 1361 + }, + { + "epoch": 0.6966751918158568, + "grad_norm": 9.699110984802246, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.86008620262146, + "num_tokens": 4377873.0, + "step": 1362 + }, + { + "epoch": 0.6971867007672634, + "grad_norm": 8.841136932373047, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8468316793441772, + "num_tokens": 4381099.0, + "step": 1363 + }, + { + "epoch": 0.69769820971867, + "grad_norm": 12.68734073638916, + "learning_rate": 1e-06, + "loss": 0.4478, + "mean_token_accuracy": 0.8120115995407104, + "num_tokens": 4384146.0, + "step": 1364 + }, + { + "epoch": 0.6982097186700768, + "grad_norm": 13.440479278564453, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.835088849067688, + "num_tokens": 4387275.0, + "step": 1365 + }, + { + "epoch": 0.6987212276214834, + "grad_norm": 9.486404418945312, + "learning_rate": 1e-06, + "loss": 0.2905, + "mean_token_accuracy": 0.8911110162734985, + "num_tokens": 4390202.0, + "step": 1366 + }, + { + "epoch": 0.69923273657289, + "grad_norm": 13.832581520080566, + "learning_rate": 1e-06, + "loss": 0.4434, + "mean_token_accuracy": 0.8349443674087524, + "num_tokens": 4392878.0, + "step": 1367 + }, + { + "epoch": 0.6997442455242967, + "grad_norm": 8.419797897338867, + "learning_rate": 1e-06, + "loss": 0.2366, + "mean_token_accuracy": 0.9065784215927124, + "num_tokens": 4396269.0, + "step": 1368 + }, + { + "epoch": 0.7002557544757033, + "grad_norm": 10.474448204040527, + "learning_rate": 1e-06, + "loss": 0.3177, + "mean_token_accuracy": 0.8766422271728516, + "num_tokens": 4399055.0, + "step": 1369 + }, + { + "epoch": 0.7007672634271099, + "grad_norm": 13.160551071166992, + "learning_rate": 1e-06, + "loss": 0.4212, + "mean_token_accuracy": 0.8462790250778198, + "num_tokens": 4401798.0, + "step": 1370 + }, + { + "epoch": 0.7012787723785167, + "grad_norm": 10.890052795410156, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.9153440594673157, + "num_tokens": 4404335.0, + "step": 1371 + }, + { + "epoch": 0.7017902813299233, + "grad_norm": 10.987016677856445, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8875036239624023, + "num_tokens": 4407381.0, + "step": 1372 + }, + { + "epoch": 0.7023017902813299, + "grad_norm": 10.204764366149902, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8742294311523438, + "num_tokens": 4411270.0, + "step": 1373 + }, + { + "epoch": 0.7028132992327366, + "grad_norm": 12.431354522705078, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8851885795593262, + "num_tokens": 4414833.0, + "step": 1374 + }, + { + "epoch": 0.7033248081841432, + "grad_norm": 10.559626579284668, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.867095947265625, + "num_tokens": 4418334.0, + "step": 1375 + }, + { + "epoch": 0.7038363171355498, + "grad_norm": 15.375250816345215, + "learning_rate": 1e-06, + "loss": 0.3819, + "mean_token_accuracy": 0.8630595207214355, + "num_tokens": 4421387.0, + "step": 1376 + }, + { + "epoch": 0.7043478260869566, + "grad_norm": 8.360330581665039, + "learning_rate": 1e-06, + "loss": 0.2779, + "mean_token_accuracy": 0.879486620426178, + "num_tokens": 4424378.0, + "step": 1377 + }, + { + "epoch": 0.7048593350383632, + "grad_norm": 13.136882781982422, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8816231489181519, + "num_tokens": 4427799.0, + "step": 1378 + }, + { + "epoch": 0.7053708439897698, + "grad_norm": 11.38169002532959, + "learning_rate": 1e-06, + "loss": 0.3774, + "mean_token_accuracy": 0.8618716597557068, + "num_tokens": 4430373.0, + "step": 1379 + }, + { + "epoch": 0.7058823529411765, + "grad_norm": 6.6255364418029785, + "learning_rate": 1e-06, + "loss": 0.2845, + "mean_token_accuracy": 0.8534618020057678, + "num_tokens": 4435723.0, + "step": 1380 + }, + { + "epoch": 0.7063938618925831, + "grad_norm": 9.981968879699707, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8950836062431335, + "num_tokens": 4439048.0, + "step": 1381 + }, + { + "epoch": 0.7069053708439897, + "grad_norm": 12.69753360748291, + "learning_rate": 1e-06, + "loss": 0.4241, + "mean_token_accuracy": 0.8343503475189209, + "num_tokens": 4442266.0, + "step": 1382 + }, + { + "epoch": 0.7074168797953965, + "grad_norm": 10.45250129699707, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8666239976882935, + "num_tokens": 4445665.0, + "step": 1383 + }, + { + "epoch": 0.7079283887468031, + "grad_norm": 11.878055572509766, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.862655520439148, + "num_tokens": 4448235.0, + "step": 1384 + }, + { + "epoch": 0.7084398976982097, + "grad_norm": 6.626915454864502, + "learning_rate": 1e-06, + "loss": 0.1826, + "mean_token_accuracy": 0.8993443250656128, + "num_tokens": 4451608.0, + "step": 1385 + }, + { + "epoch": 0.7089514066496164, + "grad_norm": 9.821795463562012, + "learning_rate": 1e-06, + "loss": 0.2531, + "mean_token_accuracy": 0.8805984854698181, + "num_tokens": 4454774.0, + "step": 1386 + }, + { + "epoch": 0.709462915601023, + "grad_norm": 10.615224838256836, + "learning_rate": 1e-06, + "loss": 0.2772, + "mean_token_accuracy": 0.898436427116394, + "num_tokens": 4458030.0, + "step": 1387 + }, + { + "epoch": 0.7099744245524297, + "grad_norm": 8.064520835876465, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.8955210447311401, + "num_tokens": 4461027.0, + "step": 1388 + }, + { + "epoch": 0.7104859335038363, + "grad_norm": 11.643417358398438, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8830069303512573, + "num_tokens": 4464831.0, + "step": 1389 + }, + { + "epoch": 0.710997442455243, + "grad_norm": 11.18470287322998, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8434773683547974, + "num_tokens": 4467757.0, + "step": 1390 + }, + { + "epoch": 0.7115089514066496, + "grad_norm": 9.738129615783691, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.8758857846260071, + "num_tokens": 4470765.0, + "step": 1391 + }, + { + "epoch": 0.7120204603580562, + "grad_norm": 13.318821907043457, + "learning_rate": 1e-06, + "loss": 0.3719, + "mean_token_accuracy": 0.8729260563850403, + "num_tokens": 4473686.0, + "step": 1392 + }, + { + "epoch": 0.7125319693094629, + "grad_norm": 10.676469802856445, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8506865501403809, + "num_tokens": 4478184.0, + "step": 1393 + }, + { + "epoch": 0.7130434782608696, + "grad_norm": 12.61536979675293, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.862602710723877, + "num_tokens": 4481360.0, + "step": 1394 + }, + { + "epoch": 0.7135549872122762, + "grad_norm": 12.028992652893066, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8832195997238159, + "num_tokens": 4484646.0, + "step": 1395 + }, + { + "epoch": 0.7140664961636829, + "grad_norm": 10.656644821166992, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8377348780632019, + "num_tokens": 4487675.0, + "step": 1396 + }, + { + "epoch": 0.7145780051150895, + "grad_norm": 10.825737953186035, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8607847690582275, + "num_tokens": 4491008.0, + "step": 1397 + }, + { + "epoch": 0.7150895140664961, + "grad_norm": 11.406759262084961, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8829237222671509, + "num_tokens": 4495317.0, + "step": 1398 + }, + { + "epoch": 0.7156010230179028, + "grad_norm": 9.926431655883789, + "learning_rate": 1e-06, + "loss": 0.38, + "mean_token_accuracy": 0.8633925318717957, + "num_tokens": 4499275.0, + "step": 1399 + }, + { + "epoch": 0.7161125319693095, + "grad_norm": 10.306974411010742, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8486754894256592, + "num_tokens": 4502619.0, + "step": 1400 + }, + { + "epoch": 0.7166240409207161, + "grad_norm": 9.32587718963623, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.9046347737312317, + "num_tokens": 4505288.0, + "step": 1401 + }, + { + "epoch": 0.7171355498721228, + "grad_norm": 17.472627639770508, + "learning_rate": 1e-06, + "loss": 0.3833, + "mean_token_accuracy": 0.8617192506790161, + "num_tokens": 4509561.0, + "step": 1402 + }, + { + "epoch": 0.7176470588235294, + "grad_norm": 10.667862892150879, + "learning_rate": 1e-06, + "loss": 0.3498, + "mean_token_accuracy": 0.8745883703231812, + "num_tokens": 4512794.0, + "step": 1403 + }, + { + "epoch": 0.718158567774936, + "grad_norm": 7.353420257568359, + "learning_rate": 1e-06, + "loss": 0.2301, + "mean_token_accuracy": 0.9049769639968872, + "num_tokens": 4516447.0, + "step": 1404 + }, + { + "epoch": 0.7186700767263428, + "grad_norm": 11.928954124450684, + "learning_rate": 1e-06, + "loss": 0.3349, + "mean_token_accuracy": 0.8699750900268555, + "num_tokens": 4519488.0, + "step": 1405 + }, + { + "epoch": 0.7191815856777494, + "grad_norm": 9.389585494995117, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.8785735964775085, + "num_tokens": 4522805.0, + "step": 1406 + }, + { + "epoch": 0.719693094629156, + "grad_norm": 9.781326293945312, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.9014933109283447, + "num_tokens": 4525661.0, + "step": 1407 + }, + { + "epoch": 0.7202046035805627, + "grad_norm": 9.390595436096191, + "learning_rate": 1e-06, + "loss": 0.2911, + "mean_token_accuracy": 0.8784722089767456, + "num_tokens": 4528447.0, + "step": 1408 + }, + { + "epoch": 0.7207161125319693, + "grad_norm": 12.10556411743164, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.8632714152336121, + "num_tokens": 4531313.0, + "step": 1409 + }, + { + "epoch": 0.7212276214833759, + "grad_norm": 12.024921417236328, + "learning_rate": 1e-06, + "loss": 0.3525, + "mean_token_accuracy": 0.8814564943313599, + "num_tokens": 4534738.0, + "step": 1410 + }, + { + "epoch": 0.7217391304347827, + "grad_norm": 8.061948776245117, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8797262907028198, + "num_tokens": 4538346.0, + "step": 1411 + }, + { + "epoch": 0.7222506393861893, + "grad_norm": 8.232439041137695, + "learning_rate": 1e-06, + "loss": 0.2345, + "mean_token_accuracy": 0.8723012208938599, + "num_tokens": 4541371.0, + "step": 1412 + }, + { + "epoch": 0.7227621483375959, + "grad_norm": 10.130982398986816, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8640300035476685, + "num_tokens": 4544175.0, + "step": 1413 + }, + { + "epoch": 0.7232736572890025, + "grad_norm": 13.067787170410156, + "learning_rate": 1e-06, + "loss": 0.3645, + "mean_token_accuracy": 0.8642820119857788, + "num_tokens": 4546728.0, + "step": 1414 + }, + { + "epoch": 0.7237851662404092, + "grad_norm": 9.370174407958984, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8751945495605469, + "num_tokens": 4549733.0, + "step": 1415 + }, + { + "epoch": 0.7242966751918158, + "grad_norm": 9.282221794128418, + "learning_rate": 1e-06, + "loss": 0.343, + "mean_token_accuracy": 0.8509749174118042, + "num_tokens": 4553136.0, + "step": 1416 + }, + { + "epoch": 0.7248081841432225, + "grad_norm": 12.610220909118652, + "learning_rate": 1e-06, + "loss": 0.3704, + "mean_token_accuracy": 0.8640908002853394, + "num_tokens": 4555866.0, + "step": 1417 + }, + { + "epoch": 0.7253196930946292, + "grad_norm": 9.756088256835938, + "learning_rate": 1e-06, + "loss": 0.2665, + "mean_token_accuracy": 0.8849856853485107, + "num_tokens": 4558654.0, + "step": 1418 + }, + { + "epoch": 0.7258312020460358, + "grad_norm": 11.07321834564209, + "learning_rate": 1e-06, + "loss": 0.2681, + "mean_token_accuracy": 0.8923999667167664, + "num_tokens": 4562536.0, + "step": 1419 + }, + { + "epoch": 0.7263427109974424, + "grad_norm": 8.882501602172852, + "learning_rate": 1e-06, + "loss": 0.3258, + "mean_token_accuracy": 0.8714868426322937, + "num_tokens": 4566234.0, + "step": 1420 + }, + { + "epoch": 0.7268542199488491, + "grad_norm": 12.70562744140625, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.9060897827148438, + "num_tokens": 4569186.0, + "step": 1421 + }, + { + "epoch": 0.7273657289002557, + "grad_norm": 11.675084114074707, + "learning_rate": 1e-06, + "loss": 0.3363, + "mean_token_accuracy": 0.87484210729599, + "num_tokens": 4571805.0, + "step": 1422 + }, + { + "epoch": 0.7278772378516624, + "grad_norm": 10.043526649475098, + "learning_rate": 1e-06, + "loss": 0.3616, + "mean_token_accuracy": 0.8891775608062744, + "num_tokens": 4574712.0, + "step": 1423 + }, + { + "epoch": 0.7283887468030691, + "grad_norm": 9.059711456298828, + "learning_rate": 1e-06, + "loss": 0.369, + "mean_token_accuracy": 0.8711800575256348, + "num_tokens": 4578350.0, + "step": 1424 + }, + { + "epoch": 0.7289002557544757, + "grad_norm": 8.967986106872559, + "learning_rate": 1e-06, + "loss": 0.346, + "mean_token_accuracy": 0.8730541467666626, + "num_tokens": 4581951.0, + "step": 1425 + }, + { + "epoch": 0.7294117647058823, + "grad_norm": 10.029874801635742, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8929277658462524, + "num_tokens": 4584650.0, + "step": 1426 + }, + { + "epoch": 0.729923273657289, + "grad_norm": 10.689305305480957, + "learning_rate": 1e-06, + "loss": 0.3069, + "mean_token_accuracy": 0.880101203918457, + "num_tokens": 4587716.0, + "step": 1427 + }, + { + "epoch": 0.7304347826086957, + "grad_norm": 10.748004913330078, + "learning_rate": 1e-06, + "loss": 0.3357, + "mean_token_accuracy": 0.8792475461959839, + "num_tokens": 4590812.0, + "step": 1428 + }, + { + "epoch": 0.7309462915601023, + "grad_norm": 10.458259582519531, + "learning_rate": 1e-06, + "loss": 0.3367, + "mean_token_accuracy": 0.8747764229774475, + "num_tokens": 4593864.0, + "step": 1429 + }, + { + "epoch": 0.731457800511509, + "grad_norm": 9.39875316619873, + "learning_rate": 1e-06, + "loss": 0.316, + "mean_token_accuracy": 0.8715863823890686, + "num_tokens": 4597053.0, + "step": 1430 + }, + { + "epoch": 0.7319693094629156, + "grad_norm": 11.181739807128906, + "learning_rate": 1e-06, + "loss": 0.3167, + "mean_token_accuracy": 0.8815178871154785, + "num_tokens": 4600597.0, + "step": 1431 + }, + { + "epoch": 0.7324808184143222, + "grad_norm": 10.57309627532959, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8858749866485596, + "num_tokens": 4604010.0, + "step": 1432 + }, + { + "epoch": 0.732992327365729, + "grad_norm": 9.813411712646484, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.8505665063858032, + "num_tokens": 4607260.0, + "step": 1433 + }, + { + "epoch": 0.7335038363171356, + "grad_norm": 7.2586259841918945, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.8592438697814941, + "num_tokens": 4610840.0, + "step": 1434 + }, + { + "epoch": 0.7340153452685422, + "grad_norm": 9.383245468139648, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8587493300437927, + "num_tokens": 4614198.0, + "step": 1435 + }, + { + "epoch": 0.7345268542199489, + "grad_norm": 10.581398963928223, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.886861264705658, + "num_tokens": 4617284.0, + "step": 1436 + }, + { + "epoch": 0.7350383631713555, + "grad_norm": 10.561820030212402, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.8617796897888184, + "num_tokens": 4620932.0, + "step": 1437 + }, + { + "epoch": 0.7355498721227621, + "grad_norm": 20.596088409423828, + "learning_rate": 1e-06, + "loss": 0.4834, + "mean_token_accuracy": 0.8750030398368835, + "num_tokens": 4623551.0, + "step": 1438 + }, + { + "epoch": 0.7360613810741689, + "grad_norm": 9.734308242797852, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8594328761100769, + "num_tokens": 4627046.0, + "step": 1439 + }, + { + "epoch": 0.7365728900255755, + "grad_norm": 11.263108253479004, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8520959615707397, + "num_tokens": 4630169.0, + "step": 1440 + }, + { + "epoch": 0.7370843989769821, + "grad_norm": 6.641489028930664, + "learning_rate": 1e-06, + "loss": 0.2272, + "mean_token_accuracy": 0.8986624479293823, + "num_tokens": 4634268.0, + "step": 1441 + }, + { + "epoch": 0.7375959079283887, + "grad_norm": 11.401461601257324, + "learning_rate": 1e-06, + "loss": 0.319, + "mean_token_accuracy": 0.8828974962234497, + "num_tokens": 4637655.0, + "step": 1442 + }, + { + "epoch": 0.7381074168797954, + "grad_norm": 14.886895179748535, + "learning_rate": 1e-06, + "loss": 0.242, + "mean_token_accuracy": 0.9227713942527771, + "num_tokens": 4640711.0, + "step": 1443 + }, + { + "epoch": 0.738618925831202, + "grad_norm": 12.413936614990234, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8583149909973145, + "num_tokens": 4643713.0, + "step": 1444 + }, + { + "epoch": 0.7391304347826086, + "grad_norm": 8.658496856689453, + "learning_rate": 1e-06, + "loss": 0.2905, + "mean_token_accuracy": 0.8871780037879944, + "num_tokens": 4646501.0, + "step": 1445 + }, + { + "epoch": 0.7396419437340154, + "grad_norm": 13.606736183166504, + "learning_rate": 1e-06, + "loss": 0.4752, + "mean_token_accuracy": 0.7913315296173096, + "num_tokens": 4648993.0, + "step": 1446 + }, + { + "epoch": 0.740153452685422, + "grad_norm": 9.948553085327148, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.8792394399642944, + "num_tokens": 4652032.0, + "step": 1447 + }, + { + "epoch": 0.7406649616368286, + "grad_norm": 11.61540412902832, + "learning_rate": 1e-06, + "loss": 0.4028, + "mean_token_accuracy": 0.8381882905960083, + "num_tokens": 4655149.0, + "step": 1448 + }, + { + "epoch": 0.7411764705882353, + "grad_norm": 12.750521659851074, + "learning_rate": 1e-06, + "loss": 0.4182, + "mean_token_accuracy": 0.8583055138587952, + "num_tokens": 4658546.0, + "step": 1449 + }, + { + "epoch": 0.7416879795396419, + "grad_norm": 11.530418395996094, + "learning_rate": 1e-06, + "loss": 0.3391, + "mean_token_accuracy": 0.8975876569747925, + "num_tokens": 4661079.0, + "step": 1450 + }, + { + "epoch": 0.7421994884910486, + "grad_norm": 10.44747257232666, + "learning_rate": 1e-06, + "loss": 0.3058, + "mean_token_accuracy": 0.8846142292022705, + "num_tokens": 4664360.0, + "step": 1451 + }, + { + "epoch": 0.7427109974424553, + "grad_norm": 12.154128074645996, + "learning_rate": 1e-06, + "loss": 0.4438, + "mean_token_accuracy": 0.8183348178863525, + "num_tokens": 4667285.0, + "step": 1452 + }, + { + "epoch": 0.7432225063938619, + "grad_norm": 15.698531150817871, + "learning_rate": 1e-06, + "loss": 0.4415, + "mean_token_accuracy": 0.8428903818130493, + "num_tokens": 4670023.0, + "step": 1453 + }, + { + "epoch": 0.7437340153452685, + "grad_norm": 10.671659469604492, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8716350793838501, + "num_tokens": 4672873.0, + "step": 1454 + }, + { + "epoch": 0.7442455242966752, + "grad_norm": 9.480812072753906, + "learning_rate": 1e-06, + "loss": 0.3348, + "mean_token_accuracy": 0.8635941743850708, + "num_tokens": 4675675.0, + "step": 1455 + }, + { + "epoch": 0.7447570332480818, + "grad_norm": 12.512433052062988, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8361864686012268, + "num_tokens": 4678270.0, + "step": 1456 + }, + { + "epoch": 0.7452685421994885, + "grad_norm": 10.627717971801758, + "learning_rate": 1e-06, + "loss": 0.4021, + "mean_token_accuracy": 0.867680013179779, + "num_tokens": 4680670.0, + "step": 1457 + }, + { + "epoch": 0.7457800511508952, + "grad_norm": 12.101768493652344, + "learning_rate": 1e-06, + "loss": 0.3427, + "mean_token_accuracy": 0.8682612180709839, + "num_tokens": 4683802.0, + "step": 1458 + }, + { + "epoch": 0.7462915601023018, + "grad_norm": 10.560627937316895, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8816061019897461, + "num_tokens": 4686359.0, + "step": 1459 + }, + { + "epoch": 0.7468030690537084, + "grad_norm": 11.230610847473145, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8621763586997986, + "num_tokens": 4689766.0, + "step": 1460 + }, + { + "epoch": 0.7473145780051151, + "grad_norm": 12.041868209838867, + "learning_rate": 1e-06, + "loss": 0.4177, + "mean_token_accuracy": 0.8638549447059631, + "num_tokens": 4692786.0, + "step": 1461 + }, + { + "epoch": 0.7478260869565218, + "grad_norm": 10.453149795532227, + "learning_rate": 1e-06, + "loss": 0.3824, + "mean_token_accuracy": 0.8599926233291626, + "num_tokens": 4695898.0, + "step": 1462 + }, + { + "epoch": 0.7483375959079284, + "grad_norm": 10.407557487487793, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8684885501861572, + "num_tokens": 4699406.0, + "step": 1463 + }, + { + "epoch": 0.7488491048593351, + "grad_norm": 12.486167907714844, + "learning_rate": 1e-06, + "loss": 0.3897, + "mean_token_accuracy": 0.8653281927108765, + "num_tokens": 4702503.0, + "step": 1464 + }, + { + "epoch": 0.7493606138107417, + "grad_norm": 11.915102005004883, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.9055607318878174, + "num_tokens": 4705159.0, + "step": 1465 + }, + { + "epoch": 0.7498721227621483, + "grad_norm": 11.042435646057129, + "learning_rate": 1e-06, + "loss": 0.269, + "mean_token_accuracy": 0.9001621007919312, + "num_tokens": 4708069.0, + "step": 1466 + }, + { + "epoch": 0.7503836317135549, + "grad_norm": 9.469669342041016, + "learning_rate": 1e-06, + "loss": 0.2342, + "mean_token_accuracy": 0.9043421745300293, + "num_tokens": 4711170.0, + "step": 1467 + }, + { + "epoch": 0.7508951406649617, + "grad_norm": 10.499608993530273, + "learning_rate": 1e-06, + "loss": 0.3477, + "mean_token_accuracy": 0.8340808749198914, + "num_tokens": 4714614.0, + "step": 1468 + }, + { + "epoch": 0.7514066496163683, + "grad_norm": 12.765420913696289, + "learning_rate": 1e-06, + "loss": 0.4433, + "mean_token_accuracy": 0.8041545152664185, + "num_tokens": 4718215.0, + "step": 1469 + }, + { + "epoch": 0.7519181585677749, + "grad_norm": 6.091392517089844, + "learning_rate": 1e-06, + "loss": 0.1657, + "mean_token_accuracy": 0.9132242798805237, + "num_tokens": 4721115.0, + "step": 1470 + }, + { + "epoch": 0.7524296675191816, + "grad_norm": 13.518455505371094, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8387110233306885, + "num_tokens": 4723926.0, + "step": 1471 + }, + { + "epoch": 0.7529411764705882, + "grad_norm": 9.809378623962402, + "learning_rate": 1e-06, + "loss": 0.2788, + "mean_token_accuracy": 0.8920150399208069, + "num_tokens": 4726621.0, + "step": 1472 + }, + { + "epoch": 0.7534526854219948, + "grad_norm": 10.749932289123535, + "learning_rate": 1e-06, + "loss": 0.3396, + "mean_token_accuracy": 0.8726205825805664, + "num_tokens": 4730497.0, + "step": 1473 + }, + { + "epoch": 0.7539641943734016, + "grad_norm": 5.131331920623779, + "learning_rate": 1e-06, + "loss": 0.2462, + "mean_token_accuracy": 0.873564600944519, + "num_tokens": 4735527.0, + "step": 1474 + }, + { + "epoch": 0.7544757033248082, + "grad_norm": 13.124236106872559, + "learning_rate": 1e-06, + "loss": 0.3573, + "mean_token_accuracy": 0.8638793230056763, + "num_tokens": 4738791.0, + "step": 1475 + }, + { + "epoch": 0.7549872122762148, + "grad_norm": 7.875625133514404, + "learning_rate": 1e-06, + "loss": 0.2454, + "mean_token_accuracy": 0.8985511660575867, + "num_tokens": 4742558.0, + "step": 1476 + }, + { + "epoch": 0.7554987212276215, + "grad_norm": 8.63218879699707, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8985509276390076, + "num_tokens": 4746971.0, + "step": 1477 + }, + { + "epoch": 0.7560102301790281, + "grad_norm": 8.116559982299805, + "learning_rate": 1e-06, + "loss": 0.2248, + "mean_token_accuracy": 0.8882913589477539, + "num_tokens": 4750532.0, + "step": 1478 + }, + { + "epoch": 0.7565217391304347, + "grad_norm": 13.081381797790527, + "learning_rate": 1e-06, + "loss": 0.3436, + "mean_token_accuracy": 0.8763084411621094, + "num_tokens": 4753807.0, + "step": 1479 + }, + { + "epoch": 0.7570332480818415, + "grad_norm": 6.8785881996154785, + "learning_rate": 1e-06, + "loss": 0.3483, + "mean_token_accuracy": 0.85792076587677, + "num_tokens": 4758731.0, + "step": 1480 + }, + { + "epoch": 0.7575447570332481, + "grad_norm": 8.689020156860352, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8762243986129761, + "num_tokens": 4761945.0, + "step": 1481 + }, + { + "epoch": 0.7580562659846547, + "grad_norm": 10.628013610839844, + "learning_rate": 1e-06, + "loss": 0.2697, + "mean_token_accuracy": 0.9066451787948608, + "num_tokens": 4765068.0, + "step": 1482 + }, + { + "epoch": 0.7585677749360614, + "grad_norm": 9.223523139953613, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.8835288286209106, + "num_tokens": 4768938.0, + "step": 1483 + }, + { + "epoch": 0.759079283887468, + "grad_norm": 13.833648681640625, + "learning_rate": 1e-06, + "loss": 0.4393, + "mean_token_accuracy": 0.8547313213348389, + "num_tokens": 4771275.0, + "step": 1484 + }, + { + "epoch": 0.7595907928388747, + "grad_norm": 9.547795295715332, + "learning_rate": 1e-06, + "loss": 0.3946, + "mean_token_accuracy": 0.8551748991012573, + "num_tokens": 4773997.0, + "step": 1485 + }, + { + "epoch": 0.7601023017902814, + "grad_norm": 10.261980056762695, + "learning_rate": 1e-06, + "loss": 0.3577, + "mean_token_accuracy": 0.8660130500793457, + "num_tokens": 4777317.0, + "step": 1486 + }, + { + "epoch": 0.760613810741688, + "grad_norm": 10.480224609375, + "learning_rate": 1e-06, + "loss": 0.4251, + "mean_token_accuracy": 0.841187596321106, + "num_tokens": 4780459.0, + "step": 1487 + }, + { + "epoch": 0.7611253196930946, + "grad_norm": 11.94742488861084, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8786859512329102, + "num_tokens": 4783834.0, + "step": 1488 + }, + { + "epoch": 0.7616368286445013, + "grad_norm": 13.729394912719727, + "learning_rate": 1e-06, + "loss": 0.4118, + "mean_token_accuracy": 0.8549573421478271, + "num_tokens": 4787364.0, + "step": 1489 + }, + { + "epoch": 0.7621483375959079, + "grad_norm": 11.247786521911621, + "learning_rate": 1e-06, + "loss": 0.4432, + "mean_token_accuracy": 0.852325975894928, + "num_tokens": 4790538.0, + "step": 1490 + }, + { + "epoch": 0.7626598465473146, + "grad_norm": 10.716842651367188, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8697808980941772, + "num_tokens": 4793497.0, + "step": 1491 + }, + { + "epoch": 0.7631713554987213, + "grad_norm": 10.919100761413574, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8824184536933899, + "num_tokens": 4796423.0, + "step": 1492 + }, + { + "epoch": 0.7636828644501279, + "grad_norm": 9.6129789352417, + "learning_rate": 1e-06, + "loss": 0.3009, + "mean_token_accuracy": 0.8732922077178955, + "num_tokens": 4799559.0, + "step": 1493 + }, + { + "epoch": 0.7641943734015345, + "grad_norm": 13.84692668914795, + "learning_rate": 1e-06, + "loss": 0.3287, + "mean_token_accuracy": 0.871100664138794, + "num_tokens": 4802359.0, + "step": 1494 + }, + { + "epoch": 0.7647058823529411, + "grad_norm": 8.39726448059082, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8691015243530273, + "num_tokens": 4806363.0, + "step": 1495 + }, + { + "epoch": 0.7652173913043478, + "grad_norm": 10.751205444335938, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.9026222229003906, + "num_tokens": 4808968.0, + "step": 1496 + }, + { + "epoch": 0.7657289002557545, + "grad_norm": 10.855826377868652, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.887755274772644, + "num_tokens": 4812639.0, + "step": 1497 + }, + { + "epoch": 0.7662404092071611, + "grad_norm": 9.975476264953613, + "learning_rate": 1e-06, + "loss": 0.2893, + "mean_token_accuracy": 0.8900687098503113, + "num_tokens": 4816736.0, + "step": 1498 + }, + { + "epoch": 0.7667519181585678, + "grad_norm": 10.780116081237793, + "learning_rate": 1e-06, + "loss": 0.4107, + "mean_token_accuracy": 0.8396096229553223, + "num_tokens": 4819744.0, + "step": 1499 + }, + { + "epoch": 0.7672634271099744, + "grad_norm": 9.355888366699219, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.86527419090271, + "num_tokens": 4822530.0, + "step": 1500 + }, + { + "epoch": 0.767774936061381, + "grad_norm": 12.112872123718262, + "learning_rate": 1e-06, + "loss": 0.3462, + "mean_token_accuracy": 0.8864724040031433, + "num_tokens": 4825512.0, + "step": 1501 + }, + { + "epoch": 0.7682864450127878, + "grad_norm": 11.265092849731445, + "learning_rate": 1e-06, + "loss": 0.3361, + "mean_token_accuracy": 0.8681430816650391, + "num_tokens": 4828258.0, + "step": 1502 + }, + { + "epoch": 0.7687979539641944, + "grad_norm": 8.983792304992676, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8716779351234436, + "num_tokens": 4831241.0, + "step": 1503 + }, + { + "epoch": 0.769309462915601, + "grad_norm": 8.194734573364258, + "learning_rate": 1e-06, + "loss": 0.2478, + "mean_token_accuracy": 0.9022904634475708, + "num_tokens": 4834100.0, + "step": 1504 + }, + { + "epoch": 0.7698209718670077, + "grad_norm": 9.642915725708008, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8746521472930908, + "num_tokens": 4838212.0, + "step": 1505 + }, + { + "epoch": 0.7703324808184143, + "grad_norm": 10.491279602050781, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8763197660446167, + "num_tokens": 4842669.0, + "step": 1506 + }, + { + "epoch": 0.7708439897698209, + "grad_norm": 11.25154972076416, + "learning_rate": 1e-06, + "loss": 0.376, + "mean_token_accuracy": 0.849622368812561, + "num_tokens": 4845482.0, + "step": 1507 + }, + { + "epoch": 0.7713554987212277, + "grad_norm": 14.23435115814209, + "learning_rate": 1e-06, + "loss": 0.3701, + "mean_token_accuracy": 0.8767958283424377, + "num_tokens": 4848299.0, + "step": 1508 + }, + { + "epoch": 0.7718670076726343, + "grad_norm": 14.481354713439941, + "learning_rate": 1e-06, + "loss": 0.484, + "mean_token_accuracy": 0.8181565999984741, + "num_tokens": 4852792.0, + "step": 1509 + }, + { + "epoch": 0.7723785166240409, + "grad_norm": 11.37488842010498, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.868480384349823, + "num_tokens": 4855556.0, + "step": 1510 + }, + { + "epoch": 0.7728900255754476, + "grad_norm": 10.934660911560059, + "learning_rate": 1e-06, + "loss": 0.3683, + "mean_token_accuracy": 0.8703486919403076, + "num_tokens": 4858744.0, + "step": 1511 + }, + { + "epoch": 0.7734015345268542, + "grad_norm": 11.919289588928223, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8410472869873047, + "num_tokens": 4862158.0, + "step": 1512 + }, + { + "epoch": 0.7739130434782608, + "grad_norm": 7.745144844055176, + "learning_rate": 1e-06, + "loss": 0.2497, + "mean_token_accuracy": 0.8813697099685669, + "num_tokens": 4865961.0, + "step": 1513 + }, + { + "epoch": 0.7744245524296676, + "grad_norm": 14.070585250854492, + "learning_rate": 1e-06, + "loss": 0.4631, + "mean_token_accuracy": 0.8452889323234558, + "num_tokens": 4868757.0, + "step": 1514 + }, + { + "epoch": 0.7749360613810742, + "grad_norm": 11.469379425048828, + "learning_rate": 1e-06, + "loss": 0.4144, + "mean_token_accuracy": 0.8265860080718994, + "num_tokens": 4871474.0, + "step": 1515 + }, + { + "epoch": 0.7754475703324808, + "grad_norm": 8.201169967651367, + "learning_rate": 1e-06, + "loss": 0.3612, + "mean_token_accuracy": 0.8882495164871216, + "num_tokens": 4876071.0, + "step": 1516 + }, + { + "epoch": 0.7759590792838875, + "grad_norm": 9.611371994018555, + "learning_rate": 1e-06, + "loss": 0.291, + "mean_token_accuracy": 0.863662600517273, + "num_tokens": 4879132.0, + "step": 1517 + }, + { + "epoch": 0.7764705882352941, + "grad_norm": 9.28987979888916, + "learning_rate": 1e-06, + "loss": 0.354, + "mean_token_accuracy": 0.8825045824050903, + "num_tokens": 4882698.0, + "step": 1518 + }, + { + "epoch": 0.7769820971867007, + "grad_norm": 8.30471420288086, + "learning_rate": 1e-06, + "loss": 0.242, + "mean_token_accuracy": 0.8858762979507446, + "num_tokens": 4886571.0, + "step": 1519 + }, + { + "epoch": 0.7774936061381074, + "grad_norm": 9.876708984375, + "learning_rate": 1e-06, + "loss": 0.3339, + "mean_token_accuracy": 0.870831310749054, + "num_tokens": 4889261.0, + "step": 1520 + }, + { + "epoch": 0.7780051150895141, + "grad_norm": 9.366384506225586, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8925001621246338, + "num_tokens": 4892643.0, + "step": 1521 + }, + { + "epoch": 0.7785166240409207, + "grad_norm": 8.87664794921875, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.8725605010986328, + "num_tokens": 4895819.0, + "step": 1522 + }, + { + "epoch": 0.7790281329923273, + "grad_norm": 12.061918258666992, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8736807107925415, + "num_tokens": 4899583.0, + "step": 1523 + }, + { + "epoch": 0.779539641943734, + "grad_norm": 11.413463592529297, + "learning_rate": 1e-06, + "loss": 0.355, + "mean_token_accuracy": 0.8641119003295898, + "num_tokens": 4902709.0, + "step": 1524 + }, + { + "epoch": 0.7800511508951407, + "grad_norm": 9.63621711730957, + "learning_rate": 1e-06, + "loss": 0.3452, + "mean_token_accuracy": 0.87453293800354, + "num_tokens": 4905755.0, + "step": 1525 + }, + { + "epoch": 0.7805626598465473, + "grad_norm": 13.884368896484375, + "learning_rate": 1e-06, + "loss": 0.4498, + "mean_token_accuracy": 0.8455442190170288, + "num_tokens": 4908685.0, + "step": 1526 + }, + { + "epoch": 0.781074168797954, + "grad_norm": 14.202569961547852, + "learning_rate": 1e-06, + "loss": 0.4489, + "mean_token_accuracy": 0.8129010200500488, + "num_tokens": 4911174.0, + "step": 1527 + }, + { + "epoch": 0.7815856777493606, + "grad_norm": 12.53344440460205, + "learning_rate": 1e-06, + "loss": 0.4176, + "mean_token_accuracy": 0.8577576875686646, + "num_tokens": 4913810.0, + "step": 1528 + }, + { + "epoch": 0.7820971867007672, + "grad_norm": 7.376351833343506, + "learning_rate": 1e-06, + "loss": 0.3303, + "mean_token_accuracy": 0.8595905303955078, + "num_tokens": 4918439.0, + "step": 1529 + }, + { + "epoch": 0.782608695652174, + "grad_norm": 11.21518611907959, + "learning_rate": 1e-06, + "loss": 0.2746, + "mean_token_accuracy": 0.9032741785049438, + "num_tokens": 4920981.0, + "step": 1530 + }, + { + "epoch": 0.7831202046035806, + "grad_norm": 11.490998268127441, + "learning_rate": 1e-06, + "loss": 0.3982, + "mean_token_accuracy": 0.8640164732933044, + "num_tokens": 4924964.0, + "step": 1531 + }, + { + "epoch": 0.7836317135549872, + "grad_norm": 11.342489242553711, + "learning_rate": 1e-06, + "loss": 0.3417, + "mean_token_accuracy": 0.8715810775756836, + "num_tokens": 4927854.0, + "step": 1532 + }, + { + "epoch": 0.7841432225063939, + "grad_norm": 10.552193641662598, + "learning_rate": 1e-06, + "loss": 0.4058, + "mean_token_accuracy": 0.8556089997291565, + "num_tokens": 4930600.0, + "step": 1533 + }, + { + "epoch": 0.7846547314578005, + "grad_norm": 10.654037475585938, + "learning_rate": 1e-06, + "loss": 0.4294, + "mean_token_accuracy": 0.8271262645721436, + "num_tokens": 4933546.0, + "step": 1534 + }, + { + "epoch": 0.7851662404092071, + "grad_norm": 6.562899589538574, + "learning_rate": 1e-06, + "loss": 0.2073, + "mean_token_accuracy": 0.8922282457351685, + "num_tokens": 4936092.0, + "step": 1535 + }, + { + "epoch": 0.7856777493606139, + "grad_norm": 13.574783325195312, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8900843858718872, + "num_tokens": 4939226.0, + "step": 1536 + }, + { + "epoch": 0.7861892583120205, + "grad_norm": 9.866961479187012, + "learning_rate": 1e-06, + "loss": 0.3296, + "mean_token_accuracy": 0.8687929511070251, + "num_tokens": 4942235.0, + "step": 1537 + }, + { + "epoch": 0.7867007672634271, + "grad_norm": 12.231143951416016, + "learning_rate": 1e-06, + "loss": 0.377, + "mean_token_accuracy": 0.8409225940704346, + "num_tokens": 4945165.0, + "step": 1538 + }, + { + "epoch": 0.7872122762148338, + "grad_norm": 11.042398452758789, + "learning_rate": 1e-06, + "loss": 0.3676, + "mean_token_accuracy": 0.8803489208221436, + "num_tokens": 4948372.0, + "step": 1539 + }, + { + "epoch": 0.7877237851662404, + "grad_norm": 10.362146377563477, + "learning_rate": 1e-06, + "loss": 0.2746, + "mean_token_accuracy": 0.893146276473999, + "num_tokens": 4950986.0, + "step": 1540 + }, + { + "epoch": 0.788235294117647, + "grad_norm": 12.04506778717041, + "learning_rate": 1e-06, + "loss": 0.3515, + "mean_token_accuracy": 0.8679083585739136, + "num_tokens": 4954044.0, + "step": 1541 + }, + { + "epoch": 0.7887468030690538, + "grad_norm": 11.622941970825195, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.9007585644721985, + "num_tokens": 4957281.0, + "step": 1542 + }, + { + "epoch": 0.7892583120204604, + "grad_norm": 10.992376327514648, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.9093605875968933, + "num_tokens": 4960187.0, + "step": 1543 + }, + { + "epoch": 0.789769820971867, + "grad_norm": 7.847912788391113, + "learning_rate": 1e-06, + "loss": 0.239, + "mean_token_accuracy": 0.9019119143486023, + "num_tokens": 4963955.0, + "step": 1544 + }, + { + "epoch": 0.7902813299232737, + "grad_norm": 12.453612327575684, + "learning_rate": 1e-06, + "loss": 0.3786, + "mean_token_accuracy": 0.876849889755249, + "num_tokens": 4966978.0, + "step": 1545 + }, + { + "epoch": 0.7907928388746803, + "grad_norm": 13.370856285095215, + "learning_rate": 1e-06, + "loss": 0.437, + "mean_token_accuracy": 0.8681290149688721, + "num_tokens": 4970215.0, + "step": 1546 + }, + { + "epoch": 0.7913043478260869, + "grad_norm": 9.898555755615234, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8816763162612915, + "num_tokens": 4974451.0, + "step": 1547 + }, + { + "epoch": 0.7918158567774936, + "grad_norm": 10.7514066696167, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8803966045379639, + "num_tokens": 4977400.0, + "step": 1548 + }, + { + "epoch": 0.7923273657289003, + "grad_norm": 13.649922370910645, + "learning_rate": 1e-06, + "loss": 0.4155, + "mean_token_accuracy": 0.8321906328201294, + "num_tokens": 4979972.0, + "step": 1549 + }, + { + "epoch": 0.7928388746803069, + "grad_norm": 8.706512451171875, + "learning_rate": 1e-06, + "loss": 0.3225, + "mean_token_accuracy": 0.8758577108383179, + "num_tokens": 4984007.0, + "step": 1550 + }, + { + "epoch": 0.7933503836317135, + "grad_norm": 7.194654941558838, + "learning_rate": 1e-06, + "loss": 0.3343, + "mean_token_accuracy": 0.8482149839401245, + "num_tokens": 4988034.0, + "step": 1551 + }, + { + "epoch": 0.7938618925831202, + "grad_norm": 9.156620979309082, + "learning_rate": 1e-06, + "loss": 0.2518, + "mean_token_accuracy": 0.8900492191314697, + "num_tokens": 4991295.0, + "step": 1552 + }, + { + "epoch": 0.7943734015345268, + "grad_norm": 6.984818458557129, + "learning_rate": 1e-06, + "loss": 0.2456, + "mean_token_accuracy": 0.889854907989502, + "num_tokens": 4994890.0, + "step": 1553 + }, + { + "epoch": 0.7948849104859335, + "grad_norm": 11.807022094726562, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8513464331626892, + "num_tokens": 4998400.0, + "step": 1554 + }, + { + "epoch": 0.7953964194373402, + "grad_norm": 11.428057670593262, + "learning_rate": 1e-06, + "loss": 0.3866, + "mean_token_accuracy": 0.8505865335464478, + "num_tokens": 5002094.0, + "step": 1555 + }, + { + "epoch": 0.7959079283887468, + "grad_norm": 10.803607940673828, + "learning_rate": 1e-06, + "loss": 0.3547, + "mean_token_accuracy": 0.8620779514312744, + "num_tokens": 5004600.0, + "step": 1556 + }, + { + "epoch": 0.7964194373401534, + "grad_norm": 12.999496459960938, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8672720193862915, + "num_tokens": 5007895.0, + "step": 1557 + }, + { + "epoch": 0.7969309462915601, + "grad_norm": 13.1753568649292, + "learning_rate": 1e-06, + "loss": 0.4034, + "mean_token_accuracy": 0.8430761694908142, + "num_tokens": 5010881.0, + "step": 1558 + }, + { + "epoch": 0.7974424552429668, + "grad_norm": 8.801138877868652, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8752965331077576, + "num_tokens": 5013884.0, + "step": 1559 + }, + { + "epoch": 0.7979539641943734, + "grad_norm": 9.377269744873047, + "learning_rate": 1e-06, + "loss": 0.2819, + "mean_token_accuracy": 0.8848604559898376, + "num_tokens": 5017668.0, + "step": 1560 + }, + { + "epoch": 0.7984654731457801, + "grad_norm": 13.184809684753418, + "learning_rate": 1e-06, + "loss": 0.4272, + "mean_token_accuracy": 0.8440558910369873, + "num_tokens": 5020571.0, + "step": 1561 + }, + { + "epoch": 0.7989769820971867, + "grad_norm": 13.886914253234863, + "learning_rate": 1e-06, + "loss": 0.4382, + "mean_token_accuracy": 0.8538205027580261, + "num_tokens": 5023951.0, + "step": 1562 + }, + { + "epoch": 0.7994884910485933, + "grad_norm": 9.507210731506348, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8430963754653931, + "num_tokens": 5026413.0, + "step": 1563 + }, + { + "epoch": 0.8, + "grad_norm": 9.83483600616455, + "learning_rate": 1e-06, + "loss": 0.3465, + "mean_token_accuracy": 0.8784113526344299, + "num_tokens": 5030223.0, + "step": 1564 + }, + { + "epoch": 0.8005115089514067, + "grad_norm": 9.427374839782715, + "learning_rate": 1e-06, + "loss": 0.3337, + "mean_token_accuracy": 0.8598601818084717, + "num_tokens": 5032882.0, + "step": 1565 + }, + { + "epoch": 0.8010230179028133, + "grad_norm": 12.475510597229004, + "learning_rate": 1e-06, + "loss": 0.4838, + "mean_token_accuracy": 0.8351278305053711, + "num_tokens": 5035614.0, + "step": 1566 + }, + { + "epoch": 0.80153452685422, + "grad_norm": 14.02868366241455, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8857123851776123, + "num_tokens": 5038954.0, + "step": 1567 + }, + { + "epoch": 0.8020460358056266, + "grad_norm": 10.473823547363281, + "learning_rate": 1e-06, + "loss": 0.3506, + "mean_token_accuracy": 0.8649780750274658, + "num_tokens": 5041724.0, + "step": 1568 + }, + { + "epoch": 0.8025575447570332, + "grad_norm": 12.752495765686035, + "learning_rate": 1e-06, + "loss": 0.3783, + "mean_token_accuracy": 0.8685314655303955, + "num_tokens": 5044863.0, + "step": 1569 + }, + { + "epoch": 0.80306905370844, + "grad_norm": 10.668329238891602, + "learning_rate": 1e-06, + "loss": 0.3565, + "mean_token_accuracy": 0.8623363971710205, + "num_tokens": 5048024.0, + "step": 1570 + }, + { + "epoch": 0.8035805626598466, + "grad_norm": 12.396618843078613, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8587952852249146, + "num_tokens": 5050863.0, + "step": 1571 + }, + { + "epoch": 0.8040920716112532, + "grad_norm": 12.21134090423584, + "learning_rate": 1e-06, + "loss": 0.3827, + "mean_token_accuracy": 0.8784373998641968, + "num_tokens": 5053650.0, + "step": 1572 + }, + { + "epoch": 0.8046035805626598, + "grad_norm": 11.07675552368164, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8723011612892151, + "num_tokens": 5056600.0, + "step": 1573 + }, + { + "epoch": 0.8051150895140665, + "grad_norm": 10.014707565307617, + "learning_rate": 1e-06, + "loss": 0.3502, + "mean_token_accuracy": 0.8819128274917603, + "num_tokens": 5059838.0, + "step": 1574 + }, + { + "epoch": 0.8056265984654731, + "grad_norm": 7.236202716827393, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.9030452966690063, + "num_tokens": 5064243.0, + "step": 1575 + }, + { + "epoch": 0.8061381074168797, + "grad_norm": 10.223894119262695, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.9089970588684082, + "num_tokens": 5067667.0, + "step": 1576 + }, + { + "epoch": 0.8066496163682865, + "grad_norm": 13.838016510009766, + "learning_rate": 1e-06, + "loss": 0.4221, + "mean_token_accuracy": 0.8711608648300171, + "num_tokens": 5070624.0, + "step": 1577 + }, + { + "epoch": 0.8071611253196931, + "grad_norm": 12.344620704650879, + "learning_rate": 1e-06, + "loss": 0.4026, + "mean_token_accuracy": 0.8734393119812012, + "num_tokens": 5074155.0, + "step": 1578 + }, + { + "epoch": 0.8076726342710997, + "grad_norm": 7.432308197021484, + "learning_rate": 1e-06, + "loss": 0.2337, + "mean_token_accuracy": 0.8947790265083313, + "num_tokens": 5077165.0, + "step": 1579 + }, + { + "epoch": 0.8081841432225064, + "grad_norm": 10.540308952331543, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.9008231163024902, + "num_tokens": 5080020.0, + "step": 1580 + }, + { + "epoch": 0.808695652173913, + "grad_norm": 11.880084991455078, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8586865663528442, + "num_tokens": 5083389.0, + "step": 1581 + }, + { + "epoch": 0.8092071611253197, + "grad_norm": 10.983241081237793, + "learning_rate": 1e-06, + "loss": 0.3564, + "mean_token_accuracy": 0.8635936379432678, + "num_tokens": 5086766.0, + "step": 1582 + }, + { + "epoch": 0.8097186700767264, + "grad_norm": 11.511302947998047, + "learning_rate": 1e-06, + "loss": 0.3594, + "mean_token_accuracy": 0.8952059745788574, + "num_tokens": 5089599.0, + "step": 1583 + }, + { + "epoch": 0.810230179028133, + "grad_norm": 12.551594734191895, + "learning_rate": 1e-06, + "loss": 0.3517, + "mean_token_accuracy": 0.8665984869003296, + "num_tokens": 5092507.0, + "step": 1584 + }, + { + "epoch": 0.8107416879795396, + "grad_norm": 10.452446937561035, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8683521747589111, + "num_tokens": 5096134.0, + "step": 1585 + }, + { + "epoch": 0.8112531969309463, + "grad_norm": 11.203338623046875, + "learning_rate": 1e-06, + "loss": 0.3968, + "mean_token_accuracy": 0.8645825386047363, + "num_tokens": 5099409.0, + "step": 1586 + }, + { + "epoch": 0.8117647058823529, + "grad_norm": 9.619794845581055, + "learning_rate": 1e-06, + "loss": 0.2943, + "mean_token_accuracy": 0.8934091925621033, + "num_tokens": 5102002.0, + "step": 1587 + }, + { + "epoch": 0.8122762148337596, + "grad_norm": 10.3101806640625, + "learning_rate": 1e-06, + "loss": 0.2977, + "mean_token_accuracy": 0.9145536422729492, + "num_tokens": 5105200.0, + "step": 1588 + }, + { + "epoch": 0.8127877237851663, + "grad_norm": 8.169625282287598, + "learning_rate": 1e-06, + "loss": 0.339, + "mean_token_accuracy": 0.8458442091941833, + "num_tokens": 5109260.0, + "step": 1589 + }, + { + "epoch": 0.8132992327365729, + "grad_norm": 11.573602676391602, + "learning_rate": 1e-06, + "loss": 0.3149, + "mean_token_accuracy": 0.862554132938385, + "num_tokens": 5111921.0, + "step": 1590 + }, + { + "epoch": 0.8138107416879795, + "grad_norm": 8.445290565490723, + "learning_rate": 1e-06, + "loss": 0.2978, + "mean_token_accuracy": 0.8489028215408325, + "num_tokens": 5115558.0, + "step": 1591 + }, + { + "epoch": 0.8143222506393862, + "grad_norm": 10.276301383972168, + "learning_rate": 1e-06, + "loss": 0.3125, + "mean_token_accuracy": 0.8928134441375732, + "num_tokens": 5119332.0, + "step": 1592 + }, + { + "epoch": 0.8148337595907928, + "grad_norm": 10.612527847290039, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.8761010766029358, + "num_tokens": 5122686.0, + "step": 1593 + }, + { + "epoch": 0.8153452685421995, + "grad_norm": 8.877145767211914, + "learning_rate": 1e-06, + "loss": 0.3199, + "mean_token_accuracy": 0.8769176602363586, + "num_tokens": 5125780.0, + "step": 1594 + }, + { + "epoch": 0.8158567774936062, + "grad_norm": 10.100449562072754, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8668944835662842, + "num_tokens": 5129005.0, + "step": 1595 + }, + { + "epoch": 0.8163682864450128, + "grad_norm": 8.762825012207031, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8790775537490845, + "num_tokens": 5132426.0, + "step": 1596 + }, + { + "epoch": 0.8168797953964194, + "grad_norm": 12.180489540100098, + "learning_rate": 1e-06, + "loss": 0.5588, + "mean_token_accuracy": 0.7812751531600952, + "num_tokens": 5135336.0, + "step": 1597 + }, + { + "epoch": 0.8173913043478261, + "grad_norm": 11.745646476745605, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8620535731315613, + "num_tokens": 5138105.0, + "step": 1598 + }, + { + "epoch": 0.8179028132992328, + "grad_norm": 7.830503940582275, + "learning_rate": 1e-06, + "loss": 0.2982, + "mean_token_accuracy": 0.8583951592445374, + "num_tokens": 5142133.0, + "step": 1599 + }, + { + "epoch": 0.8184143222506394, + "grad_norm": 9.237008094787598, + "learning_rate": 1e-06, + "loss": 0.2734, + "mean_token_accuracy": 0.8661343455314636, + "num_tokens": 5145270.0, + "step": 1600 + }, + { + "epoch": 0.818925831202046, + "grad_norm": 8.932146072387695, + "learning_rate": 1e-06, + "loss": 0.3811, + "mean_token_accuracy": 0.8531109690666199, + "num_tokens": 5147823.0, + "step": 1601 + }, + { + "epoch": 0.8194373401534527, + "grad_norm": 11.95020580291748, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8716084361076355, + "num_tokens": 5150919.0, + "step": 1602 + }, + { + "epoch": 0.8199488491048593, + "grad_norm": 12.099647521972656, + "learning_rate": 1e-06, + "loss": 0.4169, + "mean_token_accuracy": 0.8432409763336182, + "num_tokens": 5154167.0, + "step": 1603 + }, + { + "epoch": 0.8204603580562659, + "grad_norm": 8.75119686126709, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.8811772465705872, + "num_tokens": 5158561.0, + "step": 1604 + }, + { + "epoch": 0.8209718670076727, + "grad_norm": 9.906003952026367, + "learning_rate": 1e-06, + "loss": 0.3792, + "mean_token_accuracy": 0.8449647426605225, + "num_tokens": 5162262.0, + "step": 1605 + }, + { + "epoch": 0.8214833759590793, + "grad_norm": 11.10822582244873, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8732730150222778, + "num_tokens": 5165799.0, + "step": 1606 + }, + { + "epoch": 0.8219948849104859, + "grad_norm": 12.395425796508789, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8360179662704468, + "num_tokens": 5169050.0, + "step": 1607 + }, + { + "epoch": 0.8225063938618926, + "grad_norm": 9.351714134216309, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.9059219360351562, + "num_tokens": 5171926.0, + "step": 1608 + }, + { + "epoch": 0.8230179028132992, + "grad_norm": 11.500663757324219, + "learning_rate": 1e-06, + "loss": 0.4533, + "mean_token_accuracy": 0.8118718266487122, + "num_tokens": 5174797.0, + "step": 1609 + }, + { + "epoch": 0.8235294117647058, + "grad_norm": 9.351899147033691, + "learning_rate": 1e-06, + "loss": 0.3875, + "mean_token_accuracy": 0.8530128002166748, + "num_tokens": 5178076.0, + "step": 1610 + }, + { + "epoch": 0.8240409207161126, + "grad_norm": 12.958797454833984, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8733140230178833, + "num_tokens": 5181266.0, + "step": 1611 + }, + { + "epoch": 0.8245524296675192, + "grad_norm": 7.282625198364258, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8673220872879028, + "num_tokens": 5185410.0, + "step": 1612 + }, + { + "epoch": 0.8250639386189258, + "grad_norm": 12.74129867553711, + "learning_rate": 1e-06, + "loss": 0.2715, + "mean_token_accuracy": 0.8820769786834717, + "num_tokens": 5188137.0, + "step": 1613 + }, + { + "epoch": 0.8255754475703325, + "grad_norm": 12.145611763000488, + "learning_rate": 1e-06, + "loss": 0.2533, + "mean_token_accuracy": 0.8939349055290222, + "num_tokens": 5190667.0, + "step": 1614 + }, + { + "epoch": 0.8260869565217391, + "grad_norm": 10.066375732421875, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8727210760116577, + "num_tokens": 5193344.0, + "step": 1615 + }, + { + "epoch": 0.8265984654731457, + "grad_norm": 11.722862243652344, + "learning_rate": 1e-06, + "loss": 0.3272, + "mean_token_accuracy": 0.8685218691825867, + "num_tokens": 5196052.0, + "step": 1616 + }, + { + "epoch": 0.8271099744245525, + "grad_norm": 10.675392150878906, + "learning_rate": 1e-06, + "loss": 0.3264, + "mean_token_accuracy": 0.8683376312255859, + "num_tokens": 5199819.0, + "step": 1617 + }, + { + "epoch": 0.8276214833759591, + "grad_norm": 10.938132286071777, + "learning_rate": 1e-06, + "loss": 0.375, + "mean_token_accuracy": 0.8676736354827881, + "num_tokens": 5203072.0, + "step": 1618 + }, + { + "epoch": 0.8281329923273657, + "grad_norm": 7.251138210296631, + "learning_rate": 1e-06, + "loss": 0.2413, + "mean_token_accuracy": 0.881714940071106, + "num_tokens": 5206109.0, + "step": 1619 + }, + { + "epoch": 0.8286445012787724, + "grad_norm": 12.532036781311035, + "learning_rate": 1e-06, + "loss": 0.4406, + "mean_token_accuracy": 0.8348543643951416, + "num_tokens": 5209512.0, + "step": 1620 + }, + { + "epoch": 0.829156010230179, + "grad_norm": 8.977534294128418, + "learning_rate": 1e-06, + "loss": 0.235, + "mean_token_accuracy": 0.9060393571853638, + "num_tokens": 5212525.0, + "step": 1621 + }, + { + "epoch": 0.8296675191815857, + "grad_norm": 13.97496223449707, + "learning_rate": 1e-06, + "loss": 0.4208, + "mean_token_accuracy": 0.8511277437210083, + "num_tokens": 5215826.0, + "step": 1622 + }, + { + "epoch": 0.8301790281329924, + "grad_norm": 7.519045829772949, + "learning_rate": 1e-06, + "loss": 0.2542, + "mean_token_accuracy": 0.8963285684585571, + "num_tokens": 5219289.0, + "step": 1623 + }, + { + "epoch": 0.830690537084399, + "grad_norm": 10.76604175567627, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8813347220420837, + "num_tokens": 5221895.0, + "step": 1624 + }, + { + "epoch": 0.8312020460358056, + "grad_norm": 7.48518705368042, + "learning_rate": 1e-06, + "loss": 0.2649, + "mean_token_accuracy": 0.9077960252761841, + "num_tokens": 5225701.0, + "step": 1625 + }, + { + "epoch": 0.8317135549872122, + "grad_norm": 6.851404190063477, + "learning_rate": 1e-06, + "loss": 0.26, + "mean_token_accuracy": 0.8776540160179138, + "num_tokens": 5229413.0, + "step": 1626 + }, + { + "epoch": 0.832225063938619, + "grad_norm": 9.430646896362305, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8823477029800415, + "num_tokens": 5232570.0, + "step": 1627 + }, + { + "epoch": 0.8327365728900256, + "grad_norm": 12.358142852783203, + "learning_rate": 1e-06, + "loss": 0.3746, + "mean_token_accuracy": 0.8378756642341614, + "num_tokens": 5235158.0, + "step": 1628 + }, + { + "epoch": 0.8332480818414322, + "grad_norm": 9.541687965393066, + "learning_rate": 1e-06, + "loss": 0.3132, + "mean_token_accuracy": 0.8934683799743652, + "num_tokens": 5238718.0, + "step": 1629 + }, + { + "epoch": 0.8337595907928389, + "grad_norm": 12.812853813171387, + "learning_rate": 1e-06, + "loss": 0.4585, + "mean_token_accuracy": 0.8654358386993408, + "num_tokens": 5242395.0, + "step": 1630 + }, + { + "epoch": 0.8342710997442455, + "grad_norm": 9.367307662963867, + "learning_rate": 1e-06, + "loss": 0.3544, + "mean_token_accuracy": 0.8561090230941772, + "num_tokens": 5244749.0, + "step": 1631 + }, + { + "epoch": 0.8347826086956521, + "grad_norm": 10.043229103088379, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8732118606567383, + "num_tokens": 5247480.0, + "step": 1632 + }, + { + "epoch": 0.8352941176470589, + "grad_norm": 8.533258438110352, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8621900081634521, + "num_tokens": 5251901.0, + "step": 1633 + }, + { + "epoch": 0.8358056265984655, + "grad_norm": 8.087579727172852, + "learning_rate": 1e-06, + "loss": 0.301, + "mean_token_accuracy": 0.8920685052871704, + "num_tokens": 5255989.0, + "step": 1634 + }, + { + "epoch": 0.8363171355498721, + "grad_norm": 12.329662322998047, + "learning_rate": 1e-06, + "loss": 0.4013, + "mean_token_accuracy": 0.8662786483764648, + "num_tokens": 5259343.0, + "step": 1635 + }, + { + "epoch": 0.8368286445012788, + "grad_norm": 9.158223152160645, + "learning_rate": 1e-06, + "loss": 0.3193, + "mean_token_accuracy": 0.8762412667274475, + "num_tokens": 5262107.0, + "step": 1636 + }, + { + "epoch": 0.8373401534526854, + "grad_norm": 11.404151916503906, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8613559007644653, + "num_tokens": 5264858.0, + "step": 1637 + }, + { + "epoch": 0.837851662404092, + "grad_norm": 8.166488647460938, + "learning_rate": 1e-06, + "loss": 0.2745, + "mean_token_accuracy": 0.8946831822395325, + "num_tokens": 5267495.0, + "step": 1638 + }, + { + "epoch": 0.8383631713554988, + "grad_norm": 9.862654685974121, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.855745255947113, + "num_tokens": 5271336.0, + "step": 1639 + }, + { + "epoch": 0.8388746803069054, + "grad_norm": 8.949222564697266, + "learning_rate": 1e-06, + "loss": 0.2436, + "mean_token_accuracy": 0.9022504091262817, + "num_tokens": 5274242.0, + "step": 1640 + }, + { + "epoch": 0.839386189258312, + "grad_norm": 9.227609634399414, + "learning_rate": 1e-06, + "loss": 0.275, + "mean_token_accuracy": 0.9017477035522461, + "num_tokens": 5276901.0, + "step": 1641 + }, + { + "epoch": 0.8398976982097187, + "grad_norm": 11.339672088623047, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8392603397369385, + "num_tokens": 5279524.0, + "step": 1642 + }, + { + "epoch": 0.8404092071611253, + "grad_norm": 10.626178741455078, + "learning_rate": 1e-06, + "loss": 0.365, + "mean_token_accuracy": 0.8590599298477173, + "num_tokens": 5282724.0, + "step": 1643 + }, + { + "epoch": 0.8409207161125319, + "grad_norm": 10.495753288269043, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8645463585853577, + "num_tokens": 5285247.0, + "step": 1644 + }, + { + "epoch": 0.8414322250639387, + "grad_norm": 8.281630516052246, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8799906969070435, + "num_tokens": 5288659.0, + "step": 1645 + }, + { + "epoch": 0.8419437340153453, + "grad_norm": 9.435977935791016, + "learning_rate": 1e-06, + "loss": 0.3173, + "mean_token_accuracy": 0.8933417797088623, + "num_tokens": 5291612.0, + "step": 1646 + }, + { + "epoch": 0.8424552429667519, + "grad_norm": 9.31412124633789, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8654493689537048, + "num_tokens": 5295006.0, + "step": 1647 + }, + { + "epoch": 0.8429667519181586, + "grad_norm": 9.295581817626953, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.858381986618042, + "num_tokens": 5298449.0, + "step": 1648 + }, + { + "epoch": 0.8434782608695652, + "grad_norm": 10.868557929992676, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8613688349723816, + "num_tokens": 5301356.0, + "step": 1649 + }, + { + "epoch": 0.8439897698209718, + "grad_norm": 10.91551399230957, + "learning_rate": 1e-06, + "loss": 0.3906, + "mean_token_accuracy": 0.8543702960014343, + "num_tokens": 5304284.0, + "step": 1650 + }, + { + "epoch": 0.8445012787723786, + "grad_norm": 6.003626823425293, + "learning_rate": 1e-06, + "loss": 0.2335, + "mean_token_accuracy": 0.8945273756980896, + "num_tokens": 5308375.0, + "step": 1651 + }, + { + "epoch": 0.8450127877237852, + "grad_norm": 9.92884349822998, + "learning_rate": 1e-06, + "loss": 0.3936, + "mean_token_accuracy": 0.8397674560546875, + "num_tokens": 5311736.0, + "step": 1652 + }, + { + "epoch": 0.8455242966751918, + "grad_norm": 8.175894737243652, + "learning_rate": 1e-06, + "loss": 0.2547, + "mean_token_accuracy": 0.8944237232208252, + "num_tokens": 5314491.0, + "step": 1653 + }, + { + "epoch": 0.8460358056265984, + "grad_norm": 7.174847602844238, + "learning_rate": 1e-06, + "loss": 0.2185, + "mean_token_accuracy": 0.8885598182678223, + "num_tokens": 5317707.0, + "step": 1654 + }, + { + "epoch": 0.8465473145780051, + "grad_norm": 9.161316871643066, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.880072295665741, + "num_tokens": 5320926.0, + "step": 1655 + }, + { + "epoch": 0.8470588235294118, + "grad_norm": 10.402179718017578, + "learning_rate": 1e-06, + "loss": 0.404, + "mean_token_accuracy": 0.8544130325317383, + "num_tokens": 5324146.0, + "step": 1656 + }, + { + "epoch": 0.8475703324808184, + "grad_norm": 9.969283103942871, + "learning_rate": 1e-06, + "loss": 0.3356, + "mean_token_accuracy": 0.8923952579498291, + "num_tokens": 5327087.0, + "step": 1657 + }, + { + "epoch": 0.8480818414322251, + "grad_norm": 8.790184020996094, + "learning_rate": 1e-06, + "loss": 0.3198, + "mean_token_accuracy": 0.8852670192718506, + "num_tokens": 5330719.0, + "step": 1658 + }, + { + "epoch": 0.8485933503836317, + "grad_norm": 9.248987197875977, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.9014105796813965, + "num_tokens": 5333874.0, + "step": 1659 + }, + { + "epoch": 0.8491048593350383, + "grad_norm": 9.493138313293457, + "learning_rate": 1e-06, + "loss": 0.3583, + "mean_token_accuracy": 0.8768458366394043, + "num_tokens": 5336413.0, + "step": 1660 + }, + { + "epoch": 0.849616368286445, + "grad_norm": 9.672404289245605, + "learning_rate": 1e-06, + "loss": 0.3183, + "mean_token_accuracy": 0.8747866153717041, + "num_tokens": 5339673.0, + "step": 1661 + }, + { + "epoch": 0.8501278772378517, + "grad_norm": 12.667340278625488, + "learning_rate": 1e-06, + "loss": 0.3759, + "mean_token_accuracy": 0.8618285655975342, + "num_tokens": 5342542.0, + "step": 1662 + }, + { + "epoch": 0.8506393861892583, + "grad_norm": 12.908563613891602, + "learning_rate": 1e-06, + "loss": 0.3826, + "mean_token_accuracy": 0.8725765943527222, + "num_tokens": 5345251.0, + "step": 1663 + }, + { + "epoch": 0.851150895140665, + "grad_norm": 12.308781623840332, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8846691846847534, + "num_tokens": 5348247.0, + "step": 1664 + }, + { + "epoch": 0.8516624040920716, + "grad_norm": 9.591017723083496, + "learning_rate": 1e-06, + "loss": 0.3249, + "mean_token_accuracy": 0.8780287504196167, + "num_tokens": 5351692.0, + "step": 1665 + }, + { + "epoch": 0.8521739130434782, + "grad_norm": 6.341416358947754, + "learning_rate": 1e-06, + "loss": 0.1952, + "mean_token_accuracy": 0.9151962995529175, + "num_tokens": 5354552.0, + "step": 1666 + }, + { + "epoch": 0.852685421994885, + "grad_norm": 12.207740783691406, + "learning_rate": 1e-06, + "loss": 0.4282, + "mean_token_accuracy": 0.8335827589035034, + "num_tokens": 5358108.0, + "step": 1667 + }, + { + "epoch": 0.8531969309462916, + "grad_norm": 8.921407699584961, + "learning_rate": 1e-06, + "loss": 0.2791, + "mean_token_accuracy": 0.8702735304832458, + "num_tokens": 5361621.0, + "step": 1668 + }, + { + "epoch": 0.8537084398976982, + "grad_norm": 13.131113052368164, + "learning_rate": 1e-06, + "loss": 0.4167, + "mean_token_accuracy": 0.8332965970039368, + "num_tokens": 5365012.0, + "step": 1669 + }, + { + "epoch": 0.8542199488491049, + "grad_norm": 10.240674018859863, + "learning_rate": 1e-06, + "loss": 0.4468, + "mean_token_accuracy": 0.8527196645736694, + "num_tokens": 5368379.0, + "step": 1670 + }, + { + "epoch": 0.8547314578005115, + "grad_norm": 15.952637672424316, + "learning_rate": 1e-06, + "loss": 0.4114, + "mean_token_accuracy": 0.8713852167129517, + "num_tokens": 5371375.0, + "step": 1671 + }, + { + "epoch": 0.8552429667519181, + "grad_norm": 11.946005821228027, + "learning_rate": 1e-06, + "loss": 0.3431, + "mean_token_accuracy": 0.8688996434211731, + "num_tokens": 5374777.0, + "step": 1672 + }, + { + "epoch": 0.8557544757033249, + "grad_norm": 9.270288467407227, + "learning_rate": 1e-06, + "loss": 0.4216, + "mean_token_accuracy": 0.8411046266555786, + "num_tokens": 5378440.0, + "step": 1673 + }, + { + "epoch": 0.8562659846547315, + "grad_norm": 10.018656730651855, + "learning_rate": 1e-06, + "loss": 0.3128, + "mean_token_accuracy": 0.8875820636749268, + "num_tokens": 5381960.0, + "step": 1674 + }, + { + "epoch": 0.8567774936061381, + "grad_norm": 10.250947952270508, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8835318684577942, + "num_tokens": 5385120.0, + "step": 1675 + }, + { + "epoch": 0.8572890025575448, + "grad_norm": 8.748244285583496, + "learning_rate": 1e-06, + "loss": 0.3282, + "mean_token_accuracy": 0.8850246667861938, + "num_tokens": 5389124.0, + "step": 1676 + }, + { + "epoch": 0.8578005115089514, + "grad_norm": 10.452244758605957, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.8942792415618896, + "num_tokens": 5392785.0, + "step": 1677 + }, + { + "epoch": 0.858312020460358, + "grad_norm": 13.666722297668457, + "learning_rate": 1e-06, + "loss": 0.4416, + "mean_token_accuracy": 0.8368538618087769, + "num_tokens": 5395582.0, + "step": 1678 + }, + { + "epoch": 0.8588235294117647, + "grad_norm": 7.159384727478027, + "learning_rate": 1e-06, + "loss": 0.2236, + "mean_token_accuracy": 0.8664146065711975, + "num_tokens": 5401300.0, + "step": 1679 + }, + { + "epoch": 0.8593350383631714, + "grad_norm": 11.707398414611816, + "learning_rate": 1e-06, + "loss": 0.3667, + "mean_token_accuracy": 0.8865926265716553, + "num_tokens": 5404686.0, + "step": 1680 + }, + { + "epoch": 0.859846547314578, + "grad_norm": 9.979896545410156, + "learning_rate": 1e-06, + "loss": 0.3528, + "mean_token_accuracy": 0.8882976174354553, + "num_tokens": 5407451.0, + "step": 1681 + }, + { + "epoch": 0.8603580562659846, + "grad_norm": 9.782352447509766, + "learning_rate": 1e-06, + "loss": 0.3404, + "mean_token_accuracy": 0.871674656867981, + "num_tokens": 5410188.0, + "step": 1682 + }, + { + "epoch": 0.8608695652173913, + "grad_norm": 10.3337984085083, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.9102768898010254, + "num_tokens": 5412838.0, + "step": 1683 + }, + { + "epoch": 0.8613810741687979, + "grad_norm": 10.153091430664062, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8684147000312805, + "num_tokens": 5415676.0, + "step": 1684 + }, + { + "epoch": 0.8618925831202046, + "grad_norm": 11.157464027404785, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.8620736598968506, + "num_tokens": 5419035.0, + "step": 1685 + }, + { + "epoch": 0.8624040920716113, + "grad_norm": 10.373400688171387, + "learning_rate": 1e-06, + "loss": 0.3618, + "mean_token_accuracy": 0.8825751543045044, + "num_tokens": 5422028.0, + "step": 1686 + }, + { + "epoch": 0.8629156010230179, + "grad_norm": 10.091876029968262, + "learning_rate": 1e-06, + "loss": 0.3328, + "mean_token_accuracy": 0.8955710530281067, + "num_tokens": 5425475.0, + "step": 1687 + }, + { + "epoch": 0.8634271099744245, + "grad_norm": 8.12048053741455, + "learning_rate": 1e-06, + "loss": 0.2641, + "mean_token_accuracy": 0.9128739833831787, + "num_tokens": 5428537.0, + "step": 1688 + }, + { + "epoch": 0.8639386189258312, + "grad_norm": 10.647255897521973, + "learning_rate": 1e-06, + "loss": 0.3675, + "mean_token_accuracy": 0.8530564308166504, + "num_tokens": 5433085.0, + "step": 1689 + }, + { + "epoch": 0.8644501278772379, + "grad_norm": 10.591727256774902, + "learning_rate": 1e-06, + "loss": 0.4196, + "mean_token_accuracy": 0.8399131298065186, + "num_tokens": 5435728.0, + "step": 1690 + }, + { + "epoch": 0.8649616368286445, + "grad_norm": 13.287940979003906, + "learning_rate": 1e-06, + "loss": 0.4499, + "mean_token_accuracy": 0.8255155086517334, + "num_tokens": 5438773.0, + "step": 1691 + }, + { + "epoch": 0.8654731457800512, + "grad_norm": 5.807607650756836, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.9059714078903198, + "num_tokens": 5443325.0, + "step": 1692 + }, + { + "epoch": 0.8659846547314578, + "grad_norm": 8.856586456298828, + "learning_rate": 1e-06, + "loss": 0.3007, + "mean_token_accuracy": 0.858931839466095, + "num_tokens": 5446225.0, + "step": 1693 + }, + { + "epoch": 0.8664961636828644, + "grad_norm": 9.457186698913574, + "learning_rate": 1e-06, + "loss": 0.3562, + "mean_token_accuracy": 0.84660404920578, + "num_tokens": 5449243.0, + "step": 1694 + }, + { + "epoch": 0.8670076726342711, + "grad_norm": 9.718971252441406, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8864067792892456, + "num_tokens": 5452758.0, + "step": 1695 + }, + { + "epoch": 0.8675191815856778, + "grad_norm": 8.184743881225586, + "learning_rate": 1e-06, + "loss": 0.2804, + "mean_token_accuracy": 0.877143919467926, + "num_tokens": 5456373.0, + "step": 1696 + }, + { + "epoch": 0.8680306905370844, + "grad_norm": 11.87070369720459, + "learning_rate": 1e-06, + "loss": 0.3892, + "mean_token_accuracy": 0.8516175746917725, + "num_tokens": 5459284.0, + "step": 1697 + }, + { + "epoch": 0.8685421994884911, + "grad_norm": 10.18520450592041, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8565394282341003, + "num_tokens": 5462785.0, + "step": 1698 + }, + { + "epoch": 0.8690537084398977, + "grad_norm": 11.26785659790039, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8400020003318787, + "num_tokens": 5465611.0, + "step": 1699 + }, + { + "epoch": 0.8695652173913043, + "grad_norm": 13.135306358337402, + "learning_rate": 1e-06, + "loss": 0.4385, + "mean_token_accuracy": 0.8506631851196289, + "num_tokens": 5469089.0, + "step": 1700 + }, + { + "epoch": 0.870076726342711, + "grad_norm": 9.349228858947754, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.8813060522079468, + "num_tokens": 5472592.0, + "step": 1701 + }, + { + "epoch": 0.8705882352941177, + "grad_norm": 9.523401260375977, + "learning_rate": 1e-06, + "loss": 0.3078, + "mean_token_accuracy": 0.8521884083747864, + "num_tokens": 5475748.0, + "step": 1702 + }, + { + "epoch": 0.8710997442455243, + "grad_norm": 11.052667617797852, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.8534029722213745, + "num_tokens": 5479957.0, + "step": 1703 + }, + { + "epoch": 0.871611253196931, + "grad_norm": 7.423989772796631, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.891018271446228, + "num_tokens": 5483996.0, + "step": 1704 + }, + { + "epoch": 0.8721227621483376, + "grad_norm": 7.793636322021484, + "learning_rate": 1e-06, + "loss": 0.3032, + "mean_token_accuracy": 0.887912392616272, + "num_tokens": 5487648.0, + "step": 1705 + }, + { + "epoch": 0.8726342710997442, + "grad_norm": 10.3742036819458, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8581774234771729, + "num_tokens": 5491033.0, + "step": 1706 + }, + { + "epoch": 0.8731457800511508, + "grad_norm": 11.31786823272705, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.8764630556106567, + "num_tokens": 5493429.0, + "step": 1707 + }, + { + "epoch": 0.8736572890025576, + "grad_norm": 11.787618637084961, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8614195585250854, + "num_tokens": 5496578.0, + "step": 1708 + }, + { + "epoch": 0.8741687979539642, + "grad_norm": 12.497101783752441, + "learning_rate": 1e-06, + "loss": 0.2826, + "mean_token_accuracy": 0.8930184841156006, + "num_tokens": 5499380.0, + "step": 1709 + }, + { + "epoch": 0.8746803069053708, + "grad_norm": 10.673199653625488, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8846899271011353, + "num_tokens": 5502784.0, + "step": 1710 + }, + { + "epoch": 0.8751918158567775, + "grad_norm": 9.303808212280273, + "learning_rate": 1e-06, + "loss": 0.3162, + "mean_token_accuracy": 0.895602822303772, + "num_tokens": 5506402.0, + "step": 1711 + }, + { + "epoch": 0.8757033248081841, + "grad_norm": 10.53343391418457, + "learning_rate": 1e-06, + "loss": 0.3611, + "mean_token_accuracy": 0.8657284379005432, + "num_tokens": 5509660.0, + "step": 1712 + }, + { + "epoch": 0.8762148337595907, + "grad_norm": 9.144359588623047, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.8993679285049438, + "num_tokens": 5512825.0, + "step": 1713 + }, + { + "epoch": 0.8767263427109975, + "grad_norm": 12.82458209991455, + "learning_rate": 1e-06, + "loss": 0.3778, + "mean_token_accuracy": 0.8922197222709656, + "num_tokens": 5516902.0, + "step": 1714 + }, + { + "epoch": 0.8772378516624041, + "grad_norm": 9.3959379196167, + "learning_rate": 1e-06, + "loss": 0.3284, + "mean_token_accuracy": 0.8913158774375916, + "num_tokens": 5519185.0, + "step": 1715 + }, + { + "epoch": 0.8777493606138107, + "grad_norm": 12.102991104125977, + "learning_rate": 1e-06, + "loss": 0.3492, + "mean_token_accuracy": 0.8387947082519531, + "num_tokens": 5522550.0, + "step": 1716 + }, + { + "epoch": 0.8782608695652174, + "grad_norm": 13.266404151916504, + "learning_rate": 1e-06, + "loss": 0.4267, + "mean_token_accuracy": 0.8366205096244812, + "num_tokens": 5525544.0, + "step": 1717 + }, + { + "epoch": 0.878772378516624, + "grad_norm": 9.892602920532227, + "learning_rate": 1e-06, + "loss": 0.4835, + "mean_token_accuracy": 0.8578935861587524, + "num_tokens": 5528739.0, + "step": 1718 + }, + { + "epoch": 0.8792838874680307, + "grad_norm": 9.667169570922852, + "learning_rate": 1e-06, + "loss": 0.3271, + "mean_token_accuracy": 0.8952462077140808, + "num_tokens": 5531000.0, + "step": 1719 + }, + { + "epoch": 0.8797953964194374, + "grad_norm": 10.223864555358887, + "learning_rate": 1e-06, + "loss": 0.3876, + "mean_token_accuracy": 0.8524680137634277, + "num_tokens": 5534672.0, + "step": 1720 + }, + { + "epoch": 0.880306905370844, + "grad_norm": 9.668170928955078, + "learning_rate": 1e-06, + "loss": 0.3332, + "mean_token_accuracy": 0.8832581043243408, + "num_tokens": 5537354.0, + "step": 1721 + }, + { + "epoch": 0.8808184143222506, + "grad_norm": 9.875805854797363, + "learning_rate": 1e-06, + "loss": 0.2929, + "mean_token_accuracy": 0.8958301544189453, + "num_tokens": 5540116.0, + "step": 1722 + }, + { + "epoch": 0.8813299232736573, + "grad_norm": 7.1014227867126465, + "learning_rate": 1e-06, + "loss": 0.2568, + "mean_token_accuracy": 0.8909897804260254, + "num_tokens": 5543452.0, + "step": 1723 + }, + { + "epoch": 0.881841432225064, + "grad_norm": 9.702300071716309, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.882194995880127, + "num_tokens": 5546558.0, + "step": 1724 + }, + { + "epoch": 0.8823529411764706, + "grad_norm": 11.465450286865234, + "learning_rate": 1e-06, + "loss": 0.3345, + "mean_token_accuracy": 0.8742806911468506, + "num_tokens": 5549463.0, + "step": 1725 + }, + { + "epoch": 0.8828644501278773, + "grad_norm": 8.20936393737793, + "learning_rate": 1e-06, + "loss": 0.283, + "mean_token_accuracy": 0.8890825510025024, + "num_tokens": 5552403.0, + "step": 1726 + }, + { + "epoch": 0.8833759590792839, + "grad_norm": 9.95260238647461, + "learning_rate": 1e-06, + "loss": 0.349, + "mean_token_accuracy": 0.8763831853866577, + "num_tokens": 5555735.0, + "step": 1727 + }, + { + "epoch": 0.8838874680306905, + "grad_norm": 10.080280303955078, + "learning_rate": 1e-06, + "loss": 0.3861, + "mean_token_accuracy": 0.8586724996566772, + "num_tokens": 5558158.0, + "step": 1728 + }, + { + "epoch": 0.8843989769820972, + "grad_norm": 6.19307804107666, + "learning_rate": 1e-06, + "loss": 0.1758, + "mean_token_accuracy": 0.9003782272338867, + "num_tokens": 5561372.0, + "step": 1729 + }, + { + "epoch": 0.8849104859335039, + "grad_norm": 8.034289360046387, + "learning_rate": 1e-06, + "loss": 0.2186, + "mean_token_accuracy": 0.9173023700714111, + "num_tokens": 5564984.0, + "step": 1730 + }, + { + "epoch": 0.8854219948849105, + "grad_norm": 9.678297996520996, + "learning_rate": 1e-06, + "loss": 0.3317, + "mean_token_accuracy": 0.865941047668457, + "num_tokens": 5568131.0, + "step": 1731 + }, + { + "epoch": 0.8859335038363171, + "grad_norm": 10.006068229675293, + "learning_rate": 1e-06, + "loss": 0.2525, + "mean_token_accuracy": 0.8876792192459106, + "num_tokens": 5571136.0, + "step": 1732 + }, + { + "epoch": 0.8864450127877238, + "grad_norm": 9.5525484085083, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8683269023895264, + "num_tokens": 5574450.0, + "step": 1733 + }, + { + "epoch": 0.8869565217391304, + "grad_norm": 12.097278594970703, + "learning_rate": 1e-06, + "loss": 0.3859, + "mean_token_accuracy": 0.844780445098877, + "num_tokens": 5576747.0, + "step": 1734 + }, + { + "epoch": 0.887468030690537, + "grad_norm": 9.353001594543457, + "learning_rate": 1e-06, + "loss": 0.2454, + "mean_token_accuracy": 0.9012380242347717, + "num_tokens": 5579782.0, + "step": 1735 + }, + { + "epoch": 0.8879795396419438, + "grad_norm": 12.643240928649902, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.864344596862793, + "num_tokens": 5582467.0, + "step": 1736 + }, + { + "epoch": 0.8884910485933504, + "grad_norm": 12.836359977722168, + "learning_rate": 1e-06, + "loss": 0.4035, + "mean_token_accuracy": 0.8434275984764099, + "num_tokens": 5585518.0, + "step": 1737 + }, + { + "epoch": 0.889002557544757, + "grad_norm": 9.465705871582031, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8451176881790161, + "num_tokens": 5588015.0, + "step": 1738 + }, + { + "epoch": 0.8895140664961637, + "grad_norm": 9.93173599243164, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.876177966594696, + "num_tokens": 5590810.0, + "step": 1739 + }, + { + "epoch": 0.8900255754475703, + "grad_norm": 14.122694969177246, + "learning_rate": 1e-06, + "loss": 0.4941, + "mean_token_accuracy": 0.8388926982879639, + "num_tokens": 5594250.0, + "step": 1740 + }, + { + "epoch": 0.8905370843989769, + "grad_norm": 8.645501136779785, + "learning_rate": 1e-06, + "loss": 0.2795, + "mean_token_accuracy": 0.8875266909599304, + "num_tokens": 5597209.0, + "step": 1741 + }, + { + "epoch": 0.8910485933503837, + "grad_norm": 12.824755668640137, + "learning_rate": 1e-06, + "loss": 0.4929, + "mean_token_accuracy": 0.829745888710022, + "num_tokens": 5600371.0, + "step": 1742 + }, + { + "epoch": 0.8915601023017903, + "grad_norm": 10.831865310668945, + "learning_rate": 1e-06, + "loss": 0.3642, + "mean_token_accuracy": 0.8607924580574036, + "num_tokens": 5603858.0, + "step": 1743 + }, + { + "epoch": 0.8920716112531969, + "grad_norm": 8.46545696258545, + "learning_rate": 1e-06, + "loss": 0.2985, + "mean_token_accuracy": 0.84892737865448, + "num_tokens": 5606990.0, + "step": 1744 + }, + { + "epoch": 0.8925831202046036, + "grad_norm": 8.457815170288086, + "learning_rate": 1e-06, + "loss": 0.3304, + "mean_token_accuracy": 0.8722027540206909, + "num_tokens": 5610740.0, + "step": 1745 + }, + { + "epoch": 0.8930946291560102, + "grad_norm": 8.862600326538086, + "learning_rate": 1e-06, + "loss": 0.3381, + "mean_token_accuracy": 0.8874211311340332, + "num_tokens": 5614711.0, + "step": 1746 + }, + { + "epoch": 0.8936061381074168, + "grad_norm": 7.527688503265381, + "learning_rate": 1e-06, + "loss": 0.262, + "mean_token_accuracy": 0.8948119282722473, + "num_tokens": 5618867.0, + "step": 1747 + }, + { + "epoch": 0.8941176470588236, + "grad_norm": 9.759631156921387, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8901472091674805, + "num_tokens": 5622312.0, + "step": 1748 + }, + { + "epoch": 0.8946291560102302, + "grad_norm": 10.00052261352539, + "learning_rate": 1e-06, + "loss": 0.2802, + "mean_token_accuracy": 0.8756320476531982, + "num_tokens": 5625582.0, + "step": 1749 + }, + { + "epoch": 0.8951406649616368, + "grad_norm": 9.998783111572266, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8761058449745178, + "num_tokens": 5629144.0, + "step": 1750 + }, + { + "epoch": 0.8956521739130435, + "grad_norm": 12.592811584472656, + "learning_rate": 1e-06, + "loss": 0.3984, + "mean_token_accuracy": 0.8352102637290955, + "num_tokens": 5632757.0, + "step": 1751 + }, + { + "epoch": 0.8961636828644501, + "grad_norm": 14.171846389770508, + "learning_rate": 1e-06, + "loss": 0.4252, + "mean_token_accuracy": 0.8681344389915466, + "num_tokens": 5635522.0, + "step": 1752 + }, + { + "epoch": 0.8966751918158568, + "grad_norm": 16.775177001953125, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8944215774536133, + "num_tokens": 5638475.0, + "step": 1753 + }, + { + "epoch": 0.8971867007672635, + "grad_norm": 12.823132514953613, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.847888171672821, + "num_tokens": 5641217.0, + "step": 1754 + }, + { + "epoch": 0.8976982097186701, + "grad_norm": 10.183679580688477, + "learning_rate": 1e-06, + "loss": 0.4346, + "mean_token_accuracy": 0.8395867347717285, + "num_tokens": 5645274.0, + "step": 1755 + }, + { + "epoch": 0.8982097186700767, + "grad_norm": 14.283907890319824, + "learning_rate": 1e-06, + "loss": 0.3772, + "mean_token_accuracy": 0.8664829134941101, + "num_tokens": 5648228.0, + "step": 1756 + }, + { + "epoch": 0.8987212276214833, + "grad_norm": 8.096670150756836, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8497451543807983, + "num_tokens": 5652150.0, + "step": 1757 + }, + { + "epoch": 0.89923273657289, + "grad_norm": 8.250391960144043, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.862657904624939, + "num_tokens": 5656471.0, + "step": 1758 + }, + { + "epoch": 0.8997442455242967, + "grad_norm": 10.840344429016113, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.9037787914276123, + "num_tokens": 5659107.0, + "step": 1759 + }, + { + "epoch": 0.9002557544757033, + "grad_norm": 8.840492248535156, + "learning_rate": 1e-06, + "loss": 0.3435, + "mean_token_accuracy": 0.8563918471336365, + "num_tokens": 5662730.0, + "step": 1760 + }, + { + "epoch": 0.90076726342711, + "grad_norm": 7.5944719314575195, + "learning_rate": 1e-06, + "loss": 0.2551, + "mean_token_accuracy": 0.8977339267730713, + "num_tokens": 5666376.0, + "step": 1761 + }, + { + "epoch": 0.9012787723785166, + "grad_norm": 6.41674280166626, + "learning_rate": 1e-06, + "loss": 0.2802, + "mean_token_accuracy": 0.8914106488227844, + "num_tokens": 5670161.0, + "step": 1762 + }, + { + "epoch": 0.9017902813299232, + "grad_norm": 9.682319641113281, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8479565382003784, + "num_tokens": 5672832.0, + "step": 1763 + }, + { + "epoch": 0.90230179028133, + "grad_norm": 10.942481994628906, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8988257050514221, + "num_tokens": 5676107.0, + "step": 1764 + }, + { + "epoch": 0.9028132992327366, + "grad_norm": 11.416784286499023, + "learning_rate": 1e-06, + "loss": 0.4521, + "mean_token_accuracy": 0.8475826382637024, + "num_tokens": 5680430.0, + "step": 1765 + }, + { + "epoch": 0.9033248081841432, + "grad_norm": 11.35175895690918, + "learning_rate": 1e-06, + "loss": 0.4316, + "mean_token_accuracy": 0.8447217345237732, + "num_tokens": 5683933.0, + "step": 1766 + }, + { + "epoch": 0.9038363171355499, + "grad_norm": 5.932085037231445, + "learning_rate": 1e-06, + "loss": 0.295, + "mean_token_accuracy": 0.8742663860321045, + "num_tokens": 5688588.0, + "step": 1767 + }, + { + "epoch": 0.9043478260869565, + "grad_norm": 7.247668743133545, + "learning_rate": 1e-06, + "loss": 0.2566, + "mean_token_accuracy": 0.8804409503936768, + "num_tokens": 5691344.0, + "step": 1768 + }, + { + "epoch": 0.9048593350383631, + "grad_norm": 12.248641014099121, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8740463852882385, + "num_tokens": 5694257.0, + "step": 1769 + }, + { + "epoch": 0.9053708439897699, + "grad_norm": 9.841583251953125, + "learning_rate": 1e-06, + "loss": 0.3649, + "mean_token_accuracy": 0.8444823026657104, + "num_tokens": 5697721.0, + "step": 1770 + }, + { + "epoch": 0.9058823529411765, + "grad_norm": 7.686307430267334, + "learning_rate": 1e-06, + "loss": 0.2194, + "mean_token_accuracy": 0.9063351154327393, + "num_tokens": 5700869.0, + "step": 1771 + }, + { + "epoch": 0.9063938618925831, + "grad_norm": 10.027108192443848, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8844228982925415, + "num_tokens": 5704123.0, + "step": 1772 + }, + { + "epoch": 0.9069053708439898, + "grad_norm": 10.574183464050293, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8817062973976135, + "num_tokens": 5706919.0, + "step": 1773 + }, + { + "epoch": 0.9074168797953964, + "grad_norm": 10.864557266235352, + "learning_rate": 1e-06, + "loss": 0.3722, + "mean_token_accuracy": 0.8650789260864258, + "num_tokens": 5709886.0, + "step": 1774 + }, + { + "epoch": 0.907928388746803, + "grad_norm": 11.827249526977539, + "learning_rate": 1e-06, + "loss": 0.2562, + "mean_token_accuracy": 0.8908913135528564, + "num_tokens": 5712933.0, + "step": 1775 + }, + { + "epoch": 0.9084398976982098, + "grad_norm": 9.374771118164062, + "learning_rate": 1e-06, + "loss": 0.2805, + "mean_token_accuracy": 0.8874462842941284, + "num_tokens": 5715359.0, + "step": 1776 + }, + { + "epoch": 0.9089514066496164, + "grad_norm": 9.336636543273926, + "learning_rate": 1e-06, + "loss": 0.297, + "mean_token_accuracy": 0.8893956542015076, + "num_tokens": 5717812.0, + "step": 1777 + }, + { + "epoch": 0.909462915601023, + "grad_norm": 8.20764446258545, + "learning_rate": 1e-06, + "loss": 0.3209, + "mean_token_accuracy": 0.8707597851753235, + "num_tokens": 5721563.0, + "step": 1778 + }, + { + "epoch": 0.9099744245524297, + "grad_norm": 10.425188064575195, + "learning_rate": 1e-06, + "loss": 0.3158, + "mean_token_accuracy": 0.8859663009643555, + "num_tokens": 5724554.0, + "step": 1779 + }, + { + "epoch": 0.9104859335038363, + "grad_norm": 12.733518600463867, + "learning_rate": 1e-06, + "loss": 0.4053, + "mean_token_accuracy": 0.8498404026031494, + "num_tokens": 5727109.0, + "step": 1780 + }, + { + "epoch": 0.910997442455243, + "grad_norm": 6.773622035980225, + "learning_rate": 1e-06, + "loss": 0.3182, + "mean_token_accuracy": 0.8556085824966431, + "num_tokens": 5730681.0, + "step": 1781 + }, + { + "epoch": 0.9115089514066497, + "grad_norm": 6.755671501159668, + "learning_rate": 1e-06, + "loss": 0.2756, + "mean_token_accuracy": 0.884344756603241, + "num_tokens": 5734652.0, + "step": 1782 + }, + { + "epoch": 0.9120204603580563, + "grad_norm": 12.066892623901367, + "learning_rate": 1e-06, + "loss": 0.4448, + "mean_token_accuracy": 0.8219987154006958, + "num_tokens": 5737701.0, + "step": 1783 + }, + { + "epoch": 0.9125319693094629, + "grad_norm": 10.278913497924805, + "learning_rate": 1e-06, + "loss": 0.3168, + "mean_token_accuracy": 0.8825331926345825, + "num_tokens": 5741425.0, + "step": 1784 + }, + { + "epoch": 0.9130434782608695, + "grad_norm": 7.819056987762451, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8749319911003113, + "num_tokens": 5745310.0, + "step": 1785 + }, + { + "epoch": 0.9135549872122762, + "grad_norm": 19.440631866455078, + "learning_rate": 1e-06, + "loss": 0.448, + "mean_token_accuracy": 0.8707516193389893, + "num_tokens": 5748510.0, + "step": 1786 + }, + { + "epoch": 0.9140664961636829, + "grad_norm": 7.4809088706970215, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.9265083074569702, + "num_tokens": 5752256.0, + "step": 1787 + }, + { + "epoch": 0.9145780051150895, + "grad_norm": 11.878425598144531, + "learning_rate": 1e-06, + "loss": 0.4552, + "mean_token_accuracy": 0.8302189707756042, + "num_tokens": 5755443.0, + "step": 1788 + }, + { + "epoch": 0.9150895140664962, + "grad_norm": 9.312557220458984, + "learning_rate": 1e-06, + "loss": 0.3734, + "mean_token_accuracy": 0.8853450417518616, + "num_tokens": 5758361.0, + "step": 1789 + }, + { + "epoch": 0.9156010230179028, + "grad_norm": 9.41576099395752, + "learning_rate": 1e-06, + "loss": 0.3657, + "mean_token_accuracy": 0.8862948417663574, + "num_tokens": 5761569.0, + "step": 1790 + }, + { + "epoch": 0.9161125319693094, + "grad_norm": 8.612565040588379, + "learning_rate": 1e-06, + "loss": 0.298, + "mean_token_accuracy": 0.8665047883987427, + "num_tokens": 5764882.0, + "step": 1791 + }, + { + "epoch": 0.9166240409207161, + "grad_norm": 9.275410652160645, + "learning_rate": 1e-06, + "loss": 0.3223, + "mean_token_accuracy": 0.8713959455490112, + "num_tokens": 5768396.0, + "step": 1792 + }, + { + "epoch": 0.9171355498721228, + "grad_norm": 9.500862121582031, + "learning_rate": 1e-06, + "loss": 0.3117, + "mean_token_accuracy": 0.8821506500244141, + "num_tokens": 5771711.0, + "step": 1793 + }, + { + "epoch": 0.9176470588235294, + "grad_norm": 6.617624282836914, + "learning_rate": 1e-06, + "loss": 0.2677, + "mean_token_accuracy": 0.882436990737915, + "num_tokens": 5776593.0, + "step": 1794 + }, + { + "epoch": 0.9181585677749361, + "grad_norm": 8.467758178710938, + "learning_rate": 1e-06, + "loss": 0.2972, + "mean_token_accuracy": 0.8966899514198303, + "num_tokens": 5779469.0, + "step": 1795 + }, + { + "epoch": 0.9186700767263427, + "grad_norm": 14.047616004943848, + "learning_rate": 1e-06, + "loss": 0.4681, + "mean_token_accuracy": 0.829250156879425, + "num_tokens": 5782658.0, + "step": 1796 + }, + { + "epoch": 0.9191815856777493, + "grad_norm": 12.568936347961426, + "learning_rate": 1e-06, + "loss": 0.3849, + "mean_token_accuracy": 0.8685415387153625, + "num_tokens": 5786355.0, + "step": 1797 + }, + { + "epoch": 0.919693094629156, + "grad_norm": 10.599090576171875, + "learning_rate": 1e-06, + "loss": 0.3948, + "mean_token_accuracy": 0.8508914709091187, + "num_tokens": 5789573.0, + "step": 1798 + }, + { + "epoch": 0.9202046035805627, + "grad_norm": 7.6911540031433105, + "learning_rate": 1e-06, + "loss": 0.347, + "mean_token_accuracy": 0.8689992427825928, + "num_tokens": 5793263.0, + "step": 1799 + }, + { + "epoch": 0.9207161125319693, + "grad_norm": 11.160712242126465, + "learning_rate": 1e-06, + "loss": 0.3316, + "mean_token_accuracy": 0.87493896484375, + "num_tokens": 5796860.0, + "step": 1800 + }, + { + "epoch": 0.921227621483376, + "grad_norm": 8.149566650390625, + "learning_rate": 1e-06, + "loss": 0.3091, + "mean_token_accuracy": 0.8464152812957764, + "num_tokens": 5800876.0, + "step": 1801 + }, + { + "epoch": 0.9217391304347826, + "grad_norm": 9.999375343322754, + "learning_rate": 1e-06, + "loss": 0.2864, + "mean_token_accuracy": 0.8852084875106812, + "num_tokens": 5803606.0, + "step": 1802 + }, + { + "epoch": 0.9222506393861892, + "grad_norm": 20.766775131225586, + "learning_rate": 1e-06, + "loss": 0.4755, + "mean_token_accuracy": 0.8572753667831421, + "num_tokens": 5806670.0, + "step": 1803 + }, + { + "epoch": 0.922762148337596, + "grad_norm": 9.014578819274902, + "learning_rate": 1e-06, + "loss": 0.3904, + "mean_token_accuracy": 0.8668480515480042, + "num_tokens": 5810509.0, + "step": 1804 + }, + { + "epoch": 0.9232736572890026, + "grad_norm": 11.905348777770996, + "learning_rate": 1e-06, + "loss": 0.4503, + "mean_token_accuracy": 0.8385705947875977, + "num_tokens": 5813489.0, + "step": 1805 + }, + { + "epoch": 0.9237851662404092, + "grad_norm": 12.910139083862305, + "learning_rate": 1e-06, + "loss": 0.4111, + "mean_token_accuracy": 0.8398308753967285, + "num_tokens": 5816382.0, + "step": 1806 + }, + { + "epoch": 0.9242966751918159, + "grad_norm": 10.446874618530273, + "learning_rate": 1e-06, + "loss": 0.4048, + "mean_token_accuracy": 0.8380569219589233, + "num_tokens": 5819389.0, + "step": 1807 + }, + { + "epoch": 0.9248081841432225, + "grad_norm": 9.332135200500488, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8798348903656006, + "num_tokens": 5823126.0, + "step": 1808 + }, + { + "epoch": 0.9253196930946291, + "grad_norm": 9.5265474319458, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8772762417793274, + "num_tokens": 5827236.0, + "step": 1809 + }, + { + "epoch": 0.9258312020460358, + "grad_norm": 12.544051170349121, + "learning_rate": 1e-06, + "loss": 0.4338, + "mean_token_accuracy": 0.8509196043014526, + "num_tokens": 5830355.0, + "step": 1810 + }, + { + "epoch": 0.9263427109974425, + "grad_norm": 9.353307723999023, + "learning_rate": 1e-06, + "loss": 0.3242, + "mean_token_accuracy": 0.855860710144043, + "num_tokens": 5833916.0, + "step": 1811 + }, + { + "epoch": 0.9268542199488491, + "grad_norm": 9.53299331665039, + "learning_rate": 1e-06, + "loss": 0.3458, + "mean_token_accuracy": 0.8690820932388306, + "num_tokens": 5836978.0, + "step": 1812 + }, + { + "epoch": 0.9273657289002557, + "grad_norm": 7.786085605621338, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8857384324073792, + "num_tokens": 5840636.0, + "step": 1813 + }, + { + "epoch": 0.9278772378516624, + "grad_norm": 10.322925567626953, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8866874575614929, + "num_tokens": 5843831.0, + "step": 1814 + }, + { + "epoch": 0.928388746803069, + "grad_norm": 9.261771202087402, + "learning_rate": 1e-06, + "loss": 0.3347, + "mean_token_accuracy": 0.8659929633140564, + "num_tokens": 5847162.0, + "step": 1815 + }, + { + "epoch": 0.9289002557544757, + "grad_norm": 7.217800617218018, + "learning_rate": 1e-06, + "loss": 0.3154, + "mean_token_accuracy": 0.8641607761383057, + "num_tokens": 5850617.0, + "step": 1816 + }, + { + "epoch": 0.9294117647058824, + "grad_norm": 9.269378662109375, + "learning_rate": 1e-06, + "loss": 0.3754, + "mean_token_accuracy": 0.8508214950561523, + "num_tokens": 5853678.0, + "step": 1817 + }, + { + "epoch": 0.929923273657289, + "grad_norm": 11.130687713623047, + "learning_rate": 1e-06, + "loss": 0.3755, + "mean_token_accuracy": 0.8609741926193237, + "num_tokens": 5856993.0, + "step": 1818 + }, + { + "epoch": 0.9304347826086956, + "grad_norm": 10.659587860107422, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8621212244033813, + "num_tokens": 5860247.0, + "step": 1819 + }, + { + "epoch": 0.9309462915601023, + "grad_norm": 16.598846435546875, + "learning_rate": 1e-06, + "loss": 0.3924, + "mean_token_accuracy": 0.835124135017395, + "num_tokens": 5863363.0, + "step": 1820 + }, + { + "epoch": 0.931457800511509, + "grad_norm": 10.510907173156738, + "learning_rate": 1e-06, + "loss": 0.3413, + "mean_token_accuracy": 0.8699283003807068, + "num_tokens": 5866736.0, + "step": 1821 + }, + { + "epoch": 0.9319693094629156, + "grad_norm": 8.262035369873047, + "learning_rate": 1e-06, + "loss": 0.2445, + "mean_token_accuracy": 0.9040182828903198, + "num_tokens": 5870080.0, + "step": 1822 + }, + { + "epoch": 0.9324808184143223, + "grad_norm": 9.447933197021484, + "learning_rate": 1e-06, + "loss": 0.382, + "mean_token_accuracy": 0.8699007630348206, + "num_tokens": 5873828.0, + "step": 1823 + }, + { + "epoch": 0.9329923273657289, + "grad_norm": 10.582279205322266, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8635668754577637, + "num_tokens": 5876588.0, + "step": 1824 + }, + { + "epoch": 0.9335038363171355, + "grad_norm": 10.590688705444336, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8650941848754883, + "num_tokens": 5880040.0, + "step": 1825 + }, + { + "epoch": 0.9340153452685422, + "grad_norm": 8.609936714172363, + "learning_rate": 1e-06, + "loss": 0.3461, + "mean_token_accuracy": 0.8518325090408325, + "num_tokens": 5883914.0, + "step": 1826 + }, + { + "epoch": 0.9345268542199489, + "grad_norm": 11.33944320678711, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8562108278274536, + "num_tokens": 5886889.0, + "step": 1827 + }, + { + "epoch": 0.9350383631713555, + "grad_norm": 8.455419540405273, + "learning_rate": 1e-06, + "loss": 0.2683, + "mean_token_accuracy": 0.8955143690109253, + "num_tokens": 5889436.0, + "step": 1828 + }, + { + "epoch": 0.9355498721227622, + "grad_norm": 7.195250034332275, + "learning_rate": 1e-06, + "loss": 0.2858, + "mean_token_accuracy": 0.8835214376449585, + "num_tokens": 5893208.0, + "step": 1829 + }, + { + "epoch": 0.9360613810741688, + "grad_norm": 13.725746154785156, + "learning_rate": 1e-06, + "loss": 0.4471, + "mean_token_accuracy": 0.8473601341247559, + "num_tokens": 5896149.0, + "step": 1830 + }, + { + "epoch": 0.9365728900255754, + "grad_norm": 7.854837417602539, + "learning_rate": 1e-06, + "loss": 0.2576, + "mean_token_accuracy": 0.8831301331520081, + "num_tokens": 5899057.0, + "step": 1831 + }, + { + "epoch": 0.9370843989769821, + "grad_norm": 11.049175262451172, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.8555527329444885, + "num_tokens": 5902319.0, + "step": 1832 + }, + { + "epoch": 0.9375959079283888, + "grad_norm": 11.149850845336914, + "learning_rate": 1e-06, + "loss": 0.396, + "mean_token_accuracy": 0.8325660228729248, + "num_tokens": 5904532.0, + "step": 1833 + }, + { + "epoch": 0.9381074168797954, + "grad_norm": 10.26293659210205, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8794773817062378, + "num_tokens": 5907702.0, + "step": 1834 + }, + { + "epoch": 0.9386189258312021, + "grad_norm": 10.798257827758789, + "learning_rate": 1e-06, + "loss": 0.3393, + "mean_token_accuracy": 0.8686633706092834, + "num_tokens": 5910635.0, + "step": 1835 + }, + { + "epoch": 0.9391304347826087, + "grad_norm": 11.536276817321777, + "learning_rate": 1e-06, + "loss": 0.3176, + "mean_token_accuracy": 0.9042043685913086, + "num_tokens": 5913277.0, + "step": 1836 + }, + { + "epoch": 0.9396419437340153, + "grad_norm": 8.857481956481934, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8466681241989136, + "num_tokens": 5916450.0, + "step": 1837 + }, + { + "epoch": 0.9401534526854219, + "grad_norm": 12.865599632263184, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8528176546096802, + "num_tokens": 5919170.0, + "step": 1838 + }, + { + "epoch": 0.9406649616368287, + "grad_norm": 9.078664779663086, + "learning_rate": 1e-06, + "loss": 0.3835, + "mean_token_accuracy": 0.8576120138168335, + "num_tokens": 5923563.0, + "step": 1839 + }, + { + "epoch": 0.9411764705882353, + "grad_norm": 10.859594345092773, + "learning_rate": 1e-06, + "loss": 0.4171, + "mean_token_accuracy": 0.8358932733535767, + "num_tokens": 5926907.0, + "step": 1840 + }, + { + "epoch": 0.9416879795396419, + "grad_norm": 10.736958503723145, + "learning_rate": 1e-06, + "loss": 0.2291, + "mean_token_accuracy": 0.9143545627593994, + "num_tokens": 5930173.0, + "step": 1841 + }, + { + "epoch": 0.9421994884910486, + "grad_norm": 9.910636901855469, + "learning_rate": 1e-06, + "loss": 0.3467, + "mean_token_accuracy": 0.8703409433364868, + "num_tokens": 5933649.0, + "step": 1842 + }, + { + "epoch": 0.9427109974424552, + "grad_norm": 9.29403305053711, + "learning_rate": 1e-06, + "loss": 0.2513, + "mean_token_accuracy": 0.9081666469573975, + "num_tokens": 5937131.0, + "step": 1843 + }, + { + "epoch": 0.9432225063938618, + "grad_norm": 10.216670036315918, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8876056671142578, + "num_tokens": 5939780.0, + "step": 1844 + }, + { + "epoch": 0.9437340153452686, + "grad_norm": 7.959906578063965, + "learning_rate": 1e-06, + "loss": 0.3301, + "mean_token_accuracy": 0.865768551826477, + "num_tokens": 5943390.0, + "step": 1845 + }, + { + "epoch": 0.9442455242966752, + "grad_norm": 11.729002952575684, + "learning_rate": 1e-06, + "loss": 0.3697, + "mean_token_accuracy": 0.8697785139083862, + "num_tokens": 5947499.0, + "step": 1846 + }, + { + "epoch": 0.9447570332480818, + "grad_norm": 12.032901763916016, + "learning_rate": 1e-06, + "loss": 0.4124, + "mean_token_accuracy": 0.8531156778335571, + "num_tokens": 5950304.0, + "step": 1847 + }, + { + "epoch": 0.9452685421994885, + "grad_norm": 10.77314567565918, + "learning_rate": 1e-06, + "loss": 0.3283, + "mean_token_accuracy": 0.8732913732528687, + "num_tokens": 5952867.0, + "step": 1848 + }, + { + "epoch": 0.9457800511508951, + "grad_norm": 10.157402992248535, + "learning_rate": 1e-06, + "loss": 0.3535, + "mean_token_accuracy": 0.8439590334892273, + "num_tokens": 5955985.0, + "step": 1849 + }, + { + "epoch": 0.9462915601023018, + "grad_norm": 12.82271671295166, + "learning_rate": 1e-06, + "loss": 0.4461, + "mean_token_accuracy": 0.823830246925354, + "num_tokens": 5958467.0, + "step": 1850 + }, + { + "epoch": 0.9468030690537085, + "grad_norm": 8.1602201461792, + "learning_rate": 1e-06, + "loss": 0.2943, + "mean_token_accuracy": 0.8728857636451721, + "num_tokens": 5961881.0, + "step": 1851 + }, + { + "epoch": 0.9473145780051151, + "grad_norm": 11.250657081604004, + "learning_rate": 1e-06, + "loss": 0.338, + "mean_token_accuracy": 0.896099865436554, + "num_tokens": 5965139.0, + "step": 1852 + }, + { + "epoch": 0.9478260869565217, + "grad_norm": 11.094274520874023, + "learning_rate": 1e-06, + "loss": 0.3291, + "mean_token_accuracy": 0.8868443965911865, + "num_tokens": 5968299.0, + "step": 1853 + }, + { + "epoch": 0.9483375959079284, + "grad_norm": 12.64639663696289, + "learning_rate": 1e-06, + "loss": 0.4073, + "mean_token_accuracy": 0.8567849397659302, + "num_tokens": 5971049.0, + "step": 1854 + }, + { + "epoch": 0.948849104859335, + "grad_norm": 7.367737293243408, + "learning_rate": 1e-06, + "loss": 0.2074, + "mean_token_accuracy": 0.9020362496376038, + "num_tokens": 5974078.0, + "step": 1855 + }, + { + "epoch": 0.9493606138107417, + "grad_norm": 10.110246658325195, + "learning_rate": 1e-06, + "loss": 0.3405, + "mean_token_accuracy": 0.8807647824287415, + "num_tokens": 5977589.0, + "step": 1856 + }, + { + "epoch": 0.9498721227621484, + "grad_norm": 10.872359275817871, + "learning_rate": 1e-06, + "loss": 0.3423, + "mean_token_accuracy": 0.8751095533370972, + "num_tokens": 5979973.0, + "step": 1857 + }, + { + "epoch": 0.950383631713555, + "grad_norm": 8.371319770812988, + "learning_rate": 1e-06, + "loss": 0.2667, + "mean_token_accuracy": 0.9067526459693909, + "num_tokens": 5983177.0, + "step": 1858 + }, + { + "epoch": 0.9508951406649616, + "grad_norm": 8.673552513122559, + "learning_rate": 1e-06, + "loss": 0.2498, + "mean_token_accuracy": 0.9133299589157104, + "num_tokens": 5986592.0, + "step": 1859 + }, + { + "epoch": 0.9514066496163683, + "grad_norm": 11.206194877624512, + "learning_rate": 1e-06, + "loss": 0.2738, + "mean_token_accuracy": 0.8817435503005981, + "num_tokens": 5990065.0, + "step": 1860 + }, + { + "epoch": 0.951918158567775, + "grad_norm": 9.106558799743652, + "learning_rate": 1e-06, + "loss": 0.2868, + "mean_token_accuracy": 0.89422607421875, + "num_tokens": 5993106.0, + "step": 1861 + }, + { + "epoch": 0.9524296675191816, + "grad_norm": 9.369205474853516, + "learning_rate": 1e-06, + "loss": 0.2794, + "mean_token_accuracy": 0.8900784254074097, + "num_tokens": 5995529.0, + "step": 1862 + }, + { + "epoch": 0.9529411764705882, + "grad_norm": 14.74465560913086, + "learning_rate": 1e-06, + "loss": 0.4486, + "mean_token_accuracy": 0.8279302716255188, + "num_tokens": 5998200.0, + "step": 1863 + }, + { + "epoch": 0.9534526854219949, + "grad_norm": 15.580414772033691, + "learning_rate": 1e-06, + "loss": 0.3949, + "mean_token_accuracy": 0.857196569442749, + "num_tokens": 6001646.0, + "step": 1864 + }, + { + "epoch": 0.9539641943734015, + "grad_norm": 13.029714584350586, + "learning_rate": 1e-06, + "loss": 0.3893, + "mean_token_accuracy": 0.8760090470314026, + "num_tokens": 6005248.0, + "step": 1865 + }, + { + "epoch": 0.9544757033248081, + "grad_norm": 8.922656059265137, + "learning_rate": 1e-06, + "loss": 0.304, + "mean_token_accuracy": 0.843093752861023, + "num_tokens": 6008225.0, + "step": 1866 + }, + { + "epoch": 0.9549872122762149, + "grad_norm": 6.335081100463867, + "learning_rate": 1e-06, + "loss": 0.228, + "mean_token_accuracy": 0.9082845449447632, + "num_tokens": 6011047.0, + "step": 1867 + }, + { + "epoch": 0.9554987212276215, + "grad_norm": 14.27310562133789, + "learning_rate": 1e-06, + "loss": 0.3799, + "mean_token_accuracy": 0.8446473479270935, + "num_tokens": 6014000.0, + "step": 1868 + }, + { + "epoch": 0.9560102301790281, + "grad_norm": 10.24742317199707, + "learning_rate": 1e-06, + "loss": 0.3941, + "mean_token_accuracy": 0.840658962726593, + "num_tokens": 6016682.0, + "step": 1869 + }, + { + "epoch": 0.9565217391304348, + "grad_norm": 9.891044616699219, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8716636896133423, + "num_tokens": 6019454.0, + "step": 1870 + }, + { + "epoch": 0.9570332480818414, + "grad_norm": 12.028022766113281, + "learning_rate": 1e-06, + "loss": 0.4045, + "mean_token_accuracy": 0.859347939491272, + "num_tokens": 6022902.0, + "step": 1871 + }, + { + "epoch": 0.957544757033248, + "grad_norm": 10.50909423828125, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.879418134689331, + "num_tokens": 6026434.0, + "step": 1872 + }, + { + "epoch": 0.9580562659846548, + "grad_norm": 10.133185386657715, + "learning_rate": 1e-06, + "loss": 0.3389, + "mean_token_accuracy": 0.871963620185852, + "num_tokens": 6029593.0, + "step": 1873 + }, + { + "epoch": 0.9585677749360614, + "grad_norm": 11.870383262634277, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.899768590927124, + "num_tokens": 6032402.0, + "step": 1874 + }, + { + "epoch": 0.959079283887468, + "grad_norm": 10.093192100524902, + "learning_rate": 1e-06, + "loss": 0.3088, + "mean_token_accuracy": 0.8775118589401245, + "num_tokens": 6035070.0, + "step": 1875 + }, + { + "epoch": 0.9595907928388747, + "grad_norm": 8.748804092407227, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8817700147628784, + "num_tokens": 6039190.0, + "step": 1876 + }, + { + "epoch": 0.9601023017902813, + "grad_norm": 11.062827110290527, + "learning_rate": 1e-06, + "loss": 0.3928, + "mean_token_accuracy": 0.8622124791145325, + "num_tokens": 6041997.0, + "step": 1877 + }, + { + "epoch": 0.960613810741688, + "grad_norm": 7.335335731506348, + "learning_rate": 1e-06, + "loss": 0.2866, + "mean_token_accuracy": 0.9138604402542114, + "num_tokens": 6045823.0, + "step": 1878 + }, + { + "epoch": 0.9611253196930947, + "grad_norm": 10.870229721069336, + "learning_rate": 1e-06, + "loss": 0.313, + "mean_token_accuracy": 0.8938180804252625, + "num_tokens": 6048875.0, + "step": 1879 + }, + { + "epoch": 0.9616368286445013, + "grad_norm": 9.419452667236328, + "learning_rate": 1e-06, + "loss": 0.3285, + "mean_token_accuracy": 0.8719357252120972, + "num_tokens": 6052681.0, + "step": 1880 + }, + { + "epoch": 0.9621483375959079, + "grad_norm": 8.554157257080078, + "learning_rate": 1e-06, + "loss": 0.2762, + "mean_token_accuracy": 0.878696858882904, + "num_tokens": 6055423.0, + "step": 1881 + }, + { + "epoch": 0.9626598465473146, + "grad_norm": 11.2485933303833, + "learning_rate": 1e-06, + "loss": 0.4258, + "mean_token_accuracy": 0.832088828086853, + "num_tokens": 6058338.0, + "step": 1882 + }, + { + "epoch": 0.9631713554987212, + "grad_norm": 6.360633850097656, + "learning_rate": 1e-06, + "loss": 0.2545, + "mean_token_accuracy": 0.8820041418075562, + "num_tokens": 6063544.0, + "step": 1883 + }, + { + "epoch": 0.9636828644501279, + "grad_norm": 10.276885032653809, + "learning_rate": 1e-06, + "loss": 0.3005, + "mean_token_accuracy": 0.9022059440612793, + "num_tokens": 6066693.0, + "step": 1884 + }, + { + "epoch": 0.9641943734015346, + "grad_norm": 8.57137680053711, + "learning_rate": 1e-06, + "loss": 0.2877, + "mean_token_accuracy": 0.8357293605804443, + "num_tokens": 6069196.0, + "step": 1885 + }, + { + "epoch": 0.9647058823529412, + "grad_norm": 11.043481826782227, + "learning_rate": 1e-06, + "loss": 0.3832, + "mean_token_accuracy": 0.8655958771705627, + "num_tokens": 6072532.0, + "step": 1886 + }, + { + "epoch": 0.9652173913043478, + "grad_norm": 13.092544555664062, + "learning_rate": 1e-06, + "loss": 0.3654, + "mean_token_accuracy": 0.8550058603286743, + "num_tokens": 6075303.0, + "step": 1887 + }, + { + "epoch": 0.9657289002557545, + "grad_norm": 8.75853157043457, + "learning_rate": 1e-06, + "loss": 0.3669, + "mean_token_accuracy": 0.8359799385070801, + "num_tokens": 6079035.0, + "step": 1888 + }, + { + "epoch": 0.9662404092071611, + "grad_norm": 12.671667098999023, + "learning_rate": 1e-06, + "loss": 0.3575, + "mean_token_accuracy": 0.8691971898078918, + "num_tokens": 6081942.0, + "step": 1889 + }, + { + "epoch": 0.9667519181585678, + "grad_norm": 13.049062728881836, + "learning_rate": 1e-06, + "loss": 0.3118, + "mean_token_accuracy": 0.8543379306793213, + "num_tokens": 6084794.0, + "step": 1890 + }, + { + "epoch": 0.9672634271099744, + "grad_norm": 10.941046714782715, + "learning_rate": 1e-06, + "loss": 0.3793, + "mean_token_accuracy": 0.8763275146484375, + "num_tokens": 6087724.0, + "step": 1891 + }, + { + "epoch": 0.9677749360613811, + "grad_norm": 11.254379272460938, + "learning_rate": 1e-06, + "loss": 0.4374, + "mean_token_accuracy": 0.8461354970932007, + "num_tokens": 6090689.0, + "step": 1892 + }, + { + "epoch": 0.9682864450127877, + "grad_norm": 10.04345989227295, + "learning_rate": 1e-06, + "loss": 0.3212, + "mean_token_accuracy": 0.8668617010116577, + "num_tokens": 6093374.0, + "step": 1893 + }, + { + "epoch": 0.9687979539641943, + "grad_norm": 9.12835693359375, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8887135982513428, + "num_tokens": 6096655.0, + "step": 1894 + }, + { + "epoch": 0.969309462915601, + "grad_norm": 8.95710277557373, + "learning_rate": 1e-06, + "loss": 0.2788, + "mean_token_accuracy": 0.8980638980865479, + "num_tokens": 6099577.0, + "step": 1895 + }, + { + "epoch": 0.9698209718670077, + "grad_norm": 9.348088264465332, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8649005889892578, + "num_tokens": 6102250.0, + "step": 1896 + }, + { + "epoch": 0.9703324808184143, + "grad_norm": 14.387263298034668, + "learning_rate": 1e-06, + "loss": 0.4192, + "mean_token_accuracy": 0.8477216362953186, + "num_tokens": 6104637.0, + "step": 1897 + }, + { + "epoch": 0.970843989769821, + "grad_norm": 8.81554889678955, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8806307315826416, + "num_tokens": 6107178.0, + "step": 1898 + }, + { + "epoch": 0.9713554987212276, + "grad_norm": 11.015532493591309, + "learning_rate": 1e-06, + "loss": 0.4149, + "mean_token_accuracy": 0.8416322469711304, + "num_tokens": 6110016.0, + "step": 1899 + }, + { + "epoch": 0.9718670076726342, + "grad_norm": 11.135212898254395, + "learning_rate": 1e-06, + "loss": 0.4064, + "mean_token_accuracy": 0.827875554561615, + "num_tokens": 6112786.0, + "step": 1900 + }, + { + "epoch": 0.972378516624041, + "grad_norm": 9.50338077545166, + "learning_rate": 1e-06, + "loss": 0.3368, + "mean_token_accuracy": 0.8568511009216309, + "num_tokens": 6116355.0, + "step": 1901 + }, + { + "epoch": 0.9728900255754476, + "grad_norm": 7.5643744468688965, + "learning_rate": 1e-06, + "loss": 0.2867, + "mean_token_accuracy": 0.8893576264381409, + "num_tokens": 6120551.0, + "step": 1902 + }, + { + "epoch": 0.9734015345268542, + "grad_norm": 10.998876571655273, + "learning_rate": 1e-06, + "loss": 0.4327, + "mean_token_accuracy": 0.8189389705657959, + "num_tokens": 6123585.0, + "step": 1903 + }, + { + "epoch": 0.9739130434782609, + "grad_norm": 10.874146461486816, + "learning_rate": 1e-06, + "loss": 0.3608, + "mean_token_accuracy": 0.8695531487464905, + "num_tokens": 6126945.0, + "step": 1904 + }, + { + "epoch": 0.9744245524296675, + "grad_norm": 8.766685485839844, + "learning_rate": 1e-06, + "loss": 0.2847, + "mean_token_accuracy": 0.859684407711029, + "num_tokens": 6129304.0, + "step": 1905 + }, + { + "epoch": 0.9749360613810741, + "grad_norm": 12.50619888305664, + "learning_rate": 1e-06, + "loss": 0.3988, + "mean_token_accuracy": 0.8545893430709839, + "num_tokens": 6132357.0, + "step": 1906 + }, + { + "epoch": 0.9754475703324809, + "grad_norm": 11.358783721923828, + "learning_rate": 1e-06, + "loss": 0.4071, + "mean_token_accuracy": 0.843694806098938, + "num_tokens": 6135387.0, + "step": 1907 + }, + { + "epoch": 0.9759590792838875, + "grad_norm": 10.30732536315918, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8599261045455933, + "num_tokens": 6138892.0, + "step": 1908 + }, + { + "epoch": 0.9764705882352941, + "grad_norm": 10.720654487609863, + "learning_rate": 1e-06, + "loss": 0.299, + "mean_token_accuracy": 0.8658676147460938, + "num_tokens": 6142913.0, + "step": 1909 + }, + { + "epoch": 0.9769820971867008, + "grad_norm": 9.958632469177246, + "learning_rate": 1e-06, + "loss": 0.3434, + "mean_token_accuracy": 0.8564116954803467, + "num_tokens": 6145479.0, + "step": 1910 + }, + { + "epoch": 0.9774936061381074, + "grad_norm": 7.507368087768555, + "learning_rate": 1e-06, + "loss": 0.2799, + "mean_token_accuracy": 0.8674569129943848, + "num_tokens": 6149365.0, + "step": 1911 + }, + { + "epoch": 0.978005115089514, + "grad_norm": 8.633092880249023, + "learning_rate": 1e-06, + "loss": 0.2951, + "mean_token_accuracy": 0.8899548649787903, + "num_tokens": 6152019.0, + "step": 1912 + }, + { + "epoch": 0.9785166240409208, + "grad_norm": 9.633544921875, + "learning_rate": 1e-06, + "loss": 0.3873, + "mean_token_accuracy": 0.836529016494751, + "num_tokens": 6155454.0, + "step": 1913 + }, + { + "epoch": 0.9790281329923274, + "grad_norm": 10.423551559448242, + "learning_rate": 1e-06, + "loss": 0.3569, + "mean_token_accuracy": 0.8528612852096558, + "num_tokens": 6158484.0, + "step": 1914 + }, + { + "epoch": 0.979539641943734, + "grad_norm": 12.000800132751465, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8820412158966064, + "num_tokens": 6161070.0, + "step": 1915 + }, + { + "epoch": 0.9800511508951406, + "grad_norm": 10.663427352905273, + "learning_rate": 1e-06, + "loss": 0.3664, + "mean_token_accuracy": 0.8823215961456299, + "num_tokens": 6164557.0, + "step": 1916 + }, + { + "epoch": 0.9805626598465473, + "grad_norm": 14.494248390197754, + "learning_rate": 1e-06, + "loss": 0.3455, + "mean_token_accuracy": 0.8756532073020935, + "num_tokens": 6167970.0, + "step": 1917 + }, + { + "epoch": 0.981074168797954, + "grad_norm": 12.98196792602539, + "learning_rate": 1e-06, + "loss": 0.5, + "mean_token_accuracy": 0.8296828269958496, + "num_tokens": 6170723.0, + "step": 1918 + }, + { + "epoch": 0.9815856777493606, + "grad_norm": 9.919306755065918, + "learning_rate": 1e-06, + "loss": 0.4401, + "mean_token_accuracy": 0.8384247422218323, + "num_tokens": 6174038.0, + "step": 1919 + }, + { + "epoch": 0.9820971867007673, + "grad_norm": 13.009435653686523, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8624880909919739, + "num_tokens": 6177372.0, + "step": 1920 + }, + { + "epoch": 0.9826086956521739, + "grad_norm": 10.406901359558105, + "learning_rate": 1e-06, + "loss": 0.2753, + "mean_token_accuracy": 0.8849393129348755, + "num_tokens": 6180587.0, + "step": 1921 + }, + { + "epoch": 0.9831202046035805, + "grad_norm": 8.727790832519531, + "learning_rate": 1e-06, + "loss": 0.3172, + "mean_token_accuracy": 0.8783478140830994, + "num_tokens": 6183604.0, + "step": 1922 + }, + { + "epoch": 0.9836317135549872, + "grad_norm": 9.282896995544434, + "learning_rate": 1e-06, + "loss": 0.3443, + "mean_token_accuracy": 0.8696402311325073, + "num_tokens": 6186748.0, + "step": 1923 + }, + { + "epoch": 0.9841432225063939, + "grad_norm": 7.464517116546631, + "learning_rate": 1e-06, + "loss": 0.2866, + "mean_token_accuracy": 0.867503821849823, + "num_tokens": 6190084.0, + "step": 1924 + }, + { + "epoch": 0.9846547314578005, + "grad_norm": 12.772773742675781, + "learning_rate": 1e-06, + "loss": 0.3362, + "mean_token_accuracy": 0.881097674369812, + "num_tokens": 6192767.0, + "step": 1925 + }, + { + "epoch": 0.9851662404092072, + "grad_norm": 12.875429153442383, + "learning_rate": 1e-06, + "loss": 0.4337, + "mean_token_accuracy": 0.8438615798950195, + "num_tokens": 6195819.0, + "step": 1926 + }, + { + "epoch": 0.9856777493606138, + "grad_norm": 7.067956924438477, + "learning_rate": 1e-06, + "loss": 0.2662, + "mean_token_accuracy": 0.8819020986557007, + "num_tokens": 6199655.0, + "step": 1927 + }, + { + "epoch": 0.9861892583120204, + "grad_norm": 7.998693943023682, + "learning_rate": 1e-06, + "loss": 0.2476, + "mean_token_accuracy": 0.9064714312553406, + "num_tokens": 6203260.0, + "step": 1928 + }, + { + "epoch": 0.9867007672634271, + "grad_norm": 9.855828285217285, + "learning_rate": 1e-06, + "loss": 0.3509, + "mean_token_accuracy": 0.851188063621521, + "num_tokens": 6205944.0, + "step": 1929 + }, + { + "epoch": 0.9872122762148338, + "grad_norm": 11.063214302062988, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8484962582588196, + "num_tokens": 6209289.0, + "step": 1930 + }, + { + "epoch": 0.9877237851662404, + "grad_norm": 10.095337867736816, + "learning_rate": 1e-06, + "loss": 0.3542, + "mean_token_accuracy": 0.851567268371582, + "num_tokens": 6212604.0, + "step": 1931 + }, + { + "epoch": 0.9882352941176471, + "grad_norm": 12.27108097076416, + "learning_rate": 1e-06, + "loss": 0.3821, + "mean_token_accuracy": 0.8698005676269531, + "num_tokens": 6215938.0, + "step": 1932 + }, + { + "epoch": 0.9887468030690537, + "grad_norm": 12.034805297851562, + "learning_rate": 1e-06, + "loss": 0.4063, + "mean_token_accuracy": 0.8396046161651611, + "num_tokens": 6219354.0, + "step": 1933 + }, + { + "epoch": 0.9892583120204603, + "grad_norm": 10.233430862426758, + "learning_rate": 1e-06, + "loss": 0.3889, + "mean_token_accuracy": 0.8727390170097351, + "num_tokens": 6222644.0, + "step": 1934 + }, + { + "epoch": 0.989769820971867, + "grad_norm": 10.444085121154785, + "learning_rate": 1e-06, + "loss": 0.3998, + "mean_token_accuracy": 0.8353595733642578, + "num_tokens": 6225948.0, + "step": 1935 + }, + { + "epoch": 0.9902813299232737, + "grad_norm": 12.590639114379883, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8727457523345947, + "num_tokens": 6228755.0, + "step": 1936 + }, + { + "epoch": 0.9907928388746803, + "grad_norm": 6.2985687255859375, + "learning_rate": 1e-06, + "loss": 0.2242, + "mean_token_accuracy": 0.8986254334449768, + "num_tokens": 6233354.0, + "step": 1937 + }, + { + "epoch": 0.991304347826087, + "grad_norm": 11.22882080078125, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8748458623886108, + "num_tokens": 6235928.0, + "step": 1938 + }, + { + "epoch": 0.9918158567774936, + "grad_norm": 8.491560935974121, + "learning_rate": 1e-06, + "loss": 0.4059, + "mean_token_accuracy": 0.8668158650398254, + "num_tokens": 6239636.0, + "step": 1939 + }, + { + "epoch": 0.9923273657289002, + "grad_norm": 10.948592185974121, + "learning_rate": 1e-06, + "loss": 0.4328, + "mean_token_accuracy": 0.8415216207504272, + "num_tokens": 6242498.0, + "step": 1940 + }, + { + "epoch": 0.992838874680307, + "grad_norm": 11.122300148010254, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8974547386169434, + "num_tokens": 6245716.0, + "step": 1941 + }, + { + "epoch": 0.9933503836317136, + "grad_norm": 9.468517303466797, + "learning_rate": 1e-06, + "loss": 0.4055, + "mean_token_accuracy": 0.8589158058166504, + "num_tokens": 6248741.0, + "step": 1942 + }, + { + "epoch": 0.9938618925831202, + "grad_norm": 9.069449424743652, + "learning_rate": 1e-06, + "loss": 0.2638, + "mean_token_accuracy": 0.8688387870788574, + "num_tokens": 6251848.0, + "step": 1943 + }, + { + "epoch": 0.9943734015345268, + "grad_norm": 9.870692253112793, + "learning_rate": 1e-06, + "loss": 0.2703, + "mean_token_accuracy": 0.893984317779541, + "num_tokens": 6254844.0, + "step": 1944 + }, + { + "epoch": 0.9948849104859335, + "grad_norm": 9.740588188171387, + "learning_rate": 1e-06, + "loss": 0.405, + "mean_token_accuracy": 0.8437222242355347, + "num_tokens": 6259369.0, + "step": 1945 + }, + { + "epoch": 0.9953964194373401, + "grad_norm": 9.99333667755127, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8794070482254028, + "num_tokens": 6262237.0, + "step": 1946 + }, + { + "epoch": 0.9959079283887468, + "grad_norm": 10.752159118652344, + "learning_rate": 1e-06, + "loss": 0.3732, + "mean_token_accuracy": 0.8533077239990234, + "num_tokens": 6265503.0, + "step": 1947 + }, + { + "epoch": 0.9964194373401535, + "grad_norm": 11.073052406311035, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8696871399879456, + "num_tokens": 6268295.0, + "step": 1948 + }, + { + "epoch": 0.9969309462915601, + "grad_norm": 10.025158882141113, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.8585880398750305, + "num_tokens": 6271738.0, + "step": 1949 + }, + { + "epoch": 0.9974424552429667, + "grad_norm": 8.958059310913086, + "learning_rate": 1e-06, + "loss": 0.3376, + "mean_token_accuracy": 0.8894600868225098, + "num_tokens": 6274270.0, + "step": 1950 + }, + { + "epoch": 0.9979539641943734, + "grad_norm": 11.882721900939941, + "learning_rate": 1e-06, + "loss": 0.4566, + "mean_token_accuracy": 0.8085463047027588, + "num_tokens": 6277396.0, + "step": 1951 + }, + { + "epoch": 0.99846547314578, + "grad_norm": 10.540449142456055, + "learning_rate": 1e-06, + "loss": 0.4062, + "mean_token_accuracy": 0.8485376834869385, + "num_tokens": 6280025.0, + "step": 1952 + }, + { + "epoch": 0.9989769820971867, + "grad_norm": 11.552967071533203, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.9009994268417358, + "num_tokens": 6283125.0, + "step": 1953 + }, + { + "epoch": 0.9994884910485934, + "grad_norm": 11.177425384521484, + "learning_rate": 1e-06, + "loss": 0.3884, + "mean_token_accuracy": 0.8571267127990723, + "num_tokens": 6286200.0, + "step": 1954 + }, + { + "epoch": 1.0, + "grad_norm": 10.099602699279785, + "learning_rate": 1e-06, + "loss": 0.3514, + "mean_token_accuracy": 0.8216556310653687, + "num_tokens": 6287581.0, + "step": 1955 + }, + { + "epoch": 1.0005115089514067, + "grad_norm": 12.946795463562012, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.8727253675460815, + "num_tokens": 6291186.0, + "step": 1956 + }, + { + "epoch": 1.0010230179028132, + "grad_norm": 6.192601680755615, + "learning_rate": 1e-06, + "loss": 0.201, + "mean_token_accuracy": 0.9370733499526978, + "num_tokens": 6294982.0, + "step": 1957 + }, + { + "epoch": 1.00153452685422, + "grad_norm": 8.876642227172852, + "learning_rate": 1e-06, + "loss": 0.2598, + "mean_token_accuracy": 0.8859929442405701, + "num_tokens": 6297517.0, + "step": 1958 + }, + { + "epoch": 1.0020460358056267, + "grad_norm": 5.864957809448242, + "learning_rate": 1e-06, + "loss": 0.1809, + "mean_token_accuracy": 0.8982748985290527, + "num_tokens": 6300780.0, + "step": 1959 + }, + { + "epoch": 1.0025575447570332, + "grad_norm": 10.472183227539062, + "learning_rate": 1e-06, + "loss": 0.3043, + "mean_token_accuracy": 0.888367772102356, + "num_tokens": 6303431.0, + "step": 1960 + }, + { + "epoch": 1.00306905370844, + "grad_norm": 10.376982688903809, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8998605012893677, + "num_tokens": 6306998.0, + "step": 1961 + }, + { + "epoch": 1.0035805626598466, + "grad_norm": 8.649822235107422, + "learning_rate": 1e-06, + "loss": 0.2461, + "mean_token_accuracy": 0.9031010866165161, + "num_tokens": 6309810.0, + "step": 1962 + }, + { + "epoch": 1.0040920716112531, + "grad_norm": 8.934394836425781, + "learning_rate": 1e-06, + "loss": 0.2964, + "mean_token_accuracy": 0.8890820741653442, + "num_tokens": 6312745.0, + "step": 1963 + }, + { + "epoch": 1.0046035805626599, + "grad_norm": 8.31836986541748, + "learning_rate": 1e-06, + "loss": 0.2079, + "mean_token_accuracy": 0.9114376902580261, + "num_tokens": 6316062.0, + "step": 1964 + }, + { + "epoch": 1.0051150895140666, + "grad_norm": 9.570793151855469, + "learning_rate": 1e-06, + "loss": 0.3422, + "mean_token_accuracy": 0.8848227262496948, + "num_tokens": 6319187.0, + "step": 1965 + }, + { + "epoch": 1.005626598465473, + "grad_norm": 5.8979172706604, + "learning_rate": 1e-06, + "loss": 0.1966, + "mean_token_accuracy": 0.8944156765937805, + "num_tokens": 6323449.0, + "step": 1966 + }, + { + "epoch": 1.0061381074168798, + "grad_norm": 2.473165273666382, + "learning_rate": 1e-06, + "loss": 0.0687, + "mean_token_accuracy": 0.934076726436615, + "num_tokens": 6328632.0, + "step": 1967 + }, + { + "epoch": 1.0066496163682865, + "grad_norm": 7.392898082733154, + "learning_rate": 1e-06, + "loss": 0.1745, + "mean_token_accuracy": 0.944297194480896, + "num_tokens": 6332088.0, + "step": 1968 + }, + { + "epoch": 1.007161125319693, + "grad_norm": 10.207306861877441, + "learning_rate": 1e-06, + "loss": 0.2799, + "mean_token_accuracy": 0.9062924981117249, + "num_tokens": 6334818.0, + "step": 1969 + }, + { + "epoch": 1.0076726342710998, + "grad_norm": 6.160454273223877, + "learning_rate": 1e-06, + "loss": 0.1684, + "mean_token_accuracy": 0.9046338796615601, + "num_tokens": 6339545.0, + "step": 1970 + }, + { + "epoch": 1.0081841432225065, + "grad_norm": 6.8325581550598145, + "learning_rate": 1e-06, + "loss": 0.2119, + "mean_token_accuracy": 0.9159064292907715, + "num_tokens": 6342655.0, + "step": 1971 + }, + { + "epoch": 1.008695652173913, + "grad_norm": 9.196486473083496, + "learning_rate": 1e-06, + "loss": 0.2177, + "mean_token_accuracy": 0.8957911729812622, + "num_tokens": 6345496.0, + "step": 1972 + }, + { + "epoch": 1.0092071611253197, + "grad_norm": 8.688499450683594, + "learning_rate": 1e-06, + "loss": 0.2147, + "mean_token_accuracy": 0.9133082628250122, + "num_tokens": 6349078.0, + "step": 1973 + }, + { + "epoch": 1.0097186700767264, + "grad_norm": 6.1312456130981445, + "learning_rate": 1e-06, + "loss": 0.1654, + "mean_token_accuracy": 0.9389306306838989, + "num_tokens": 6352805.0, + "step": 1974 + }, + { + "epoch": 1.010230179028133, + "grad_norm": 6.702034950256348, + "learning_rate": 1e-06, + "loss": 0.1265, + "mean_token_accuracy": 0.9115538001060486, + "num_tokens": 6356057.0, + "step": 1975 + }, + { + "epoch": 1.0107416879795397, + "grad_norm": 9.037252426147461, + "learning_rate": 1e-06, + "loss": 0.243, + "mean_token_accuracy": 0.9017506837844849, + "num_tokens": 6359137.0, + "step": 1976 + }, + { + "epoch": 1.0112531969309464, + "grad_norm": 10.714866638183594, + "learning_rate": 1e-06, + "loss": 0.1819, + "mean_token_accuracy": 0.9165876507759094, + "num_tokens": 6362163.0, + "step": 1977 + }, + { + "epoch": 1.011764705882353, + "grad_norm": 12.624104499816895, + "learning_rate": 1e-06, + "loss": 0.3603, + "mean_token_accuracy": 0.8748760223388672, + "num_tokens": 6364696.0, + "step": 1978 + }, + { + "epoch": 1.0122762148337596, + "grad_norm": 11.684576988220215, + "learning_rate": 1e-06, + "loss": 0.2352, + "mean_token_accuracy": 0.904522180557251, + "num_tokens": 6367637.0, + "step": 1979 + }, + { + "epoch": 1.0127877237851663, + "grad_norm": 5.953277587890625, + "learning_rate": 1e-06, + "loss": 0.1656, + "mean_token_accuracy": 0.9346135854721069, + "num_tokens": 6371490.0, + "step": 1980 + }, + { + "epoch": 1.0132992327365729, + "grad_norm": 14.105256080627441, + "learning_rate": 1e-06, + "loss": 0.3717, + "mean_token_accuracy": 0.8472762703895569, + "num_tokens": 6374500.0, + "step": 1981 + }, + { + "epoch": 1.0138107416879796, + "grad_norm": 7.867595195770264, + "learning_rate": 1e-06, + "loss": 0.1977, + "mean_token_accuracy": 0.9061039686203003, + "num_tokens": 6377966.0, + "step": 1982 + }, + { + "epoch": 1.014322250639386, + "grad_norm": 12.618179321289062, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8878450393676758, + "num_tokens": 6380468.0, + "step": 1983 + }, + { + "epoch": 1.0148337595907928, + "grad_norm": 11.71037483215332, + "learning_rate": 1e-06, + "loss": 0.2581, + "mean_token_accuracy": 0.9090365171432495, + "num_tokens": 6383173.0, + "step": 1984 + }, + { + "epoch": 1.0153452685421995, + "grad_norm": 12.881159782409668, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.8984578847885132, + "num_tokens": 6386038.0, + "step": 1985 + }, + { + "epoch": 1.015856777493606, + "grad_norm": 9.39144229888916, + "learning_rate": 1e-06, + "loss": 0.1761, + "mean_token_accuracy": 0.9033627510070801, + "num_tokens": 6389290.0, + "step": 1986 + }, + { + "epoch": 1.0163682864450128, + "grad_norm": 7.9691057205200195, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.8723887205123901, + "num_tokens": 6392955.0, + "step": 1987 + }, + { + "epoch": 1.0168797953964195, + "grad_norm": 7.764408588409424, + "learning_rate": 1e-06, + "loss": 0.1877, + "mean_token_accuracy": 0.9121403694152832, + "num_tokens": 6396232.0, + "step": 1988 + }, + { + "epoch": 1.017391304347826, + "grad_norm": 9.55738639831543, + "learning_rate": 1e-06, + "loss": 0.2696, + "mean_token_accuracy": 0.9049888253211975, + "num_tokens": 6400026.0, + "step": 1989 + }, + { + "epoch": 1.0179028132992327, + "grad_norm": 4.946193218231201, + "learning_rate": 1e-06, + "loss": 0.1399, + "mean_token_accuracy": 0.9347535371780396, + "num_tokens": 6404229.0, + "step": 1990 + }, + { + "epoch": 1.0184143222506394, + "grad_norm": 7.995392799377441, + "learning_rate": 1e-06, + "loss": 0.2242, + "mean_token_accuracy": 0.89801025390625, + "num_tokens": 6407111.0, + "step": 1991 + }, + { + "epoch": 1.018925831202046, + "grad_norm": 7.1131110191345215, + "learning_rate": 1e-06, + "loss": 0.1556, + "mean_token_accuracy": 0.9347720146179199, + "num_tokens": 6410476.0, + "step": 1992 + }, + { + "epoch": 1.0194373401534527, + "grad_norm": 7.999195098876953, + "learning_rate": 1e-06, + "loss": 0.1871, + "mean_token_accuracy": 0.9029157161712646, + "num_tokens": 6413121.0, + "step": 1993 + }, + { + "epoch": 1.0199488491048594, + "grad_norm": 9.228401184082031, + "learning_rate": 1e-06, + "loss": 0.2039, + "mean_token_accuracy": 0.916806161403656, + "num_tokens": 6416174.0, + "step": 1994 + }, + { + "epoch": 1.020460358056266, + "grad_norm": 6.55952262878418, + "learning_rate": 1e-06, + "loss": 0.1506, + "mean_token_accuracy": 0.9472317695617676, + "num_tokens": 6419716.0, + "step": 1995 + }, + { + "epoch": 1.0209718670076726, + "grad_norm": 9.9708833694458, + "learning_rate": 1e-06, + "loss": 0.2401, + "mean_token_accuracy": 0.9035178422927856, + "num_tokens": 6423082.0, + "step": 1996 + }, + { + "epoch": 1.0214833759590793, + "grad_norm": 6.394842624664307, + "learning_rate": 1e-06, + "loss": 0.1497, + "mean_token_accuracy": 0.9320316314697266, + "num_tokens": 6427045.0, + "step": 1997 + }, + { + "epoch": 1.0219948849104858, + "grad_norm": 7.496931076049805, + "learning_rate": 1e-06, + "loss": 0.1139, + "mean_token_accuracy": 0.9166667461395264, + "num_tokens": 6431749.0, + "step": 1998 + }, + { + "epoch": 1.0225063938618926, + "grad_norm": 7.3062310218811035, + "learning_rate": 1e-06, + "loss": 0.1736, + "mean_token_accuracy": 0.9319337010383606, + "num_tokens": 6435297.0, + "step": 1999 + }, + { + "epoch": 1.0230179028132993, + "grad_norm": 8.752823829650879, + "learning_rate": 1e-06, + "loss": 0.219, + "mean_token_accuracy": 0.8895165920257568, + "num_tokens": 6438213.0, + "step": 2000 + }, + { + "epoch": 1.0235294117647058, + "grad_norm": 13.58824634552002, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.9044070243835449, + "num_tokens": 6440872.0, + "step": 2001 + }, + { + "epoch": 1.0240409207161125, + "grad_norm": 8.670300483703613, + "learning_rate": 1e-06, + "loss": 0.1896, + "mean_token_accuracy": 0.9180177450180054, + "num_tokens": 6444531.0, + "step": 2002 + }, + { + "epoch": 1.0245524296675192, + "grad_norm": 7.262824535369873, + "learning_rate": 1e-06, + "loss": 0.1728, + "mean_token_accuracy": 0.9159639477729797, + "num_tokens": 6449129.0, + "step": 2003 + }, + { + "epoch": 1.0250639386189258, + "grad_norm": 11.13241958618164, + "learning_rate": 1e-06, + "loss": 0.2399, + "mean_token_accuracy": 0.8968622088432312, + "num_tokens": 6453110.0, + "step": 2004 + }, + { + "epoch": 1.0255754475703325, + "grad_norm": 8.184669494628906, + "learning_rate": 1e-06, + "loss": 0.2032, + "mean_token_accuracy": 0.9223646521568298, + "num_tokens": 6456120.0, + "step": 2005 + }, + { + "epoch": 1.0260869565217392, + "grad_norm": 9.21471881866455, + "learning_rate": 1e-06, + "loss": 0.2358, + "mean_token_accuracy": 0.8625186681747437, + "num_tokens": 6460029.0, + "step": 2006 + }, + { + "epoch": 1.0265984654731457, + "grad_norm": 7.360350131988525, + "learning_rate": 1e-06, + "loss": 0.1689, + "mean_token_accuracy": 0.8932004570960999, + "num_tokens": 6463411.0, + "step": 2007 + }, + { + "epoch": 1.0271099744245524, + "grad_norm": 16.154407501220703, + "learning_rate": 1e-06, + "loss": 0.2242, + "mean_token_accuracy": 0.8811615705490112, + "num_tokens": 6466450.0, + "step": 2008 + }, + { + "epoch": 1.0276214833759592, + "grad_norm": 8.609838485717773, + "learning_rate": 1e-06, + "loss": 0.1769, + "mean_token_accuracy": 0.9233922958374023, + "num_tokens": 6470511.0, + "step": 2009 + }, + { + "epoch": 1.0281329923273657, + "grad_norm": 11.289992332458496, + "learning_rate": 1e-06, + "loss": 0.2807, + "mean_token_accuracy": 0.8890948295593262, + "num_tokens": 6473517.0, + "step": 2010 + }, + { + "epoch": 1.0286445012787724, + "grad_norm": 16.058837890625, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8548891544342041, + "num_tokens": 6476132.0, + "step": 2011 + }, + { + "epoch": 1.029156010230179, + "grad_norm": 10.802562713623047, + "learning_rate": 1e-06, + "loss": 0.1982, + "mean_token_accuracy": 0.9300517439842224, + "num_tokens": 6479038.0, + "step": 2012 + }, + { + "epoch": 1.0296675191815856, + "grad_norm": 12.963872909545898, + "learning_rate": 1e-06, + "loss": 0.2089, + "mean_token_accuracy": 0.9053371548652649, + "num_tokens": 6481737.0, + "step": 2013 + }, + { + "epoch": 1.0301790281329923, + "grad_norm": 10.089676856994629, + "learning_rate": 1e-06, + "loss": 0.2399, + "mean_token_accuracy": 0.9189918041229248, + "num_tokens": 6484763.0, + "step": 2014 + }, + { + "epoch": 1.030690537084399, + "grad_norm": 13.686384201049805, + "learning_rate": 1e-06, + "loss": 0.2708, + "mean_token_accuracy": 0.9016717672348022, + "num_tokens": 6488734.0, + "step": 2015 + }, + { + "epoch": 1.0312020460358056, + "grad_norm": 12.584836959838867, + "learning_rate": 1e-06, + "loss": 0.2221, + "mean_token_accuracy": 0.9245135188102722, + "num_tokens": 6491794.0, + "step": 2016 + }, + { + "epoch": 1.0317135549872123, + "grad_norm": 7.287815570831299, + "learning_rate": 1e-06, + "loss": 0.1422, + "mean_token_accuracy": 0.9348129034042358, + "num_tokens": 6495451.0, + "step": 2017 + }, + { + "epoch": 1.032225063938619, + "grad_norm": 12.459138870239258, + "learning_rate": 1e-06, + "loss": 0.262, + "mean_token_accuracy": 0.9012028574943542, + "num_tokens": 6498787.0, + "step": 2018 + }, + { + "epoch": 1.0327365728900255, + "grad_norm": 9.559507369995117, + "learning_rate": 1e-06, + "loss": 0.2452, + "mean_token_accuracy": 0.9125533103942871, + "num_tokens": 6502595.0, + "step": 2019 + }, + { + "epoch": 1.0332480818414322, + "grad_norm": 9.056572914123535, + "learning_rate": 1e-06, + "loss": 0.2447, + "mean_token_accuracy": 0.9001255035400391, + "num_tokens": 6505472.0, + "step": 2020 + }, + { + "epoch": 1.033759590792839, + "grad_norm": 13.847321510314941, + "learning_rate": 1e-06, + "loss": 0.2685, + "mean_token_accuracy": 0.9105544090270996, + "num_tokens": 6507973.0, + "step": 2021 + }, + { + "epoch": 1.0342710997442455, + "grad_norm": 8.016133308410645, + "learning_rate": 1e-06, + "loss": 0.1711, + "mean_token_accuracy": 0.9254741668701172, + "num_tokens": 6510712.0, + "step": 2022 + }, + { + "epoch": 1.0347826086956522, + "grad_norm": 11.29806137084961, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.9047673940658569, + "num_tokens": 6513621.0, + "step": 2023 + }, + { + "epoch": 1.035294117647059, + "grad_norm": 9.570524215698242, + "learning_rate": 1e-06, + "loss": 0.1662, + "mean_token_accuracy": 0.9187249541282654, + "num_tokens": 6516989.0, + "step": 2024 + }, + { + "epoch": 1.0358056265984654, + "grad_norm": 10.686314582824707, + "learning_rate": 1e-06, + "loss": 0.2668, + "mean_token_accuracy": 0.8844625949859619, + "num_tokens": 6520142.0, + "step": 2025 + }, + { + "epoch": 1.0363171355498721, + "grad_norm": 11.014719009399414, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8591126799583435, + "num_tokens": 6523351.0, + "step": 2026 + }, + { + "epoch": 1.0368286445012789, + "grad_norm": 8.918785095214844, + "learning_rate": 1e-06, + "loss": 0.2294, + "mean_token_accuracy": 0.8948628306388855, + "num_tokens": 6526714.0, + "step": 2027 + }, + { + "epoch": 1.0373401534526854, + "grad_norm": 12.855202674865723, + "learning_rate": 1e-06, + "loss": 0.3416, + "mean_token_accuracy": 0.8564804792404175, + "num_tokens": 6529724.0, + "step": 2028 + }, + { + "epoch": 1.037851662404092, + "grad_norm": 8.690150260925293, + "learning_rate": 1e-06, + "loss": 0.1892, + "mean_token_accuracy": 0.9087094068527222, + "num_tokens": 6533006.0, + "step": 2029 + }, + { + "epoch": 1.0383631713554988, + "grad_norm": 5.229085922241211, + "learning_rate": 1e-06, + "loss": 0.1328, + "mean_token_accuracy": 0.9295992851257324, + "num_tokens": 6536150.0, + "step": 2030 + }, + { + "epoch": 1.0388746803069053, + "grad_norm": 7.8369879722595215, + "learning_rate": 1e-06, + "loss": 0.2335, + "mean_token_accuracy": 0.9149670600891113, + "num_tokens": 6539561.0, + "step": 2031 + }, + { + "epoch": 1.039386189258312, + "grad_norm": 28.13499641418457, + "learning_rate": 1e-06, + "loss": 0.2007, + "mean_token_accuracy": 0.9274671673774719, + "num_tokens": 6542693.0, + "step": 2032 + }, + { + "epoch": 1.0398976982097188, + "grad_norm": 6.536016941070557, + "learning_rate": 1e-06, + "loss": 0.1635, + "mean_token_accuracy": 0.9298107624053955, + "num_tokens": 6546442.0, + "step": 2033 + }, + { + "epoch": 1.0404092071611253, + "grad_norm": 10.021181106567383, + "learning_rate": 1e-06, + "loss": 0.2591, + "mean_token_accuracy": 0.9057794213294983, + "num_tokens": 6550286.0, + "step": 2034 + }, + { + "epoch": 1.040920716112532, + "grad_norm": 8.222187995910645, + "learning_rate": 1e-06, + "loss": 0.2033, + "mean_token_accuracy": 0.9324749708175659, + "num_tokens": 6553440.0, + "step": 2035 + }, + { + "epoch": 1.0414322250639385, + "grad_norm": 10.216361045837402, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.8867111206054688, + "num_tokens": 6556487.0, + "step": 2036 + }, + { + "epoch": 1.0419437340153452, + "grad_norm": 7.6159443855285645, + "learning_rate": 1e-06, + "loss": 0.1879, + "mean_token_accuracy": 0.9216346740722656, + "num_tokens": 6560186.0, + "step": 2037 + }, + { + "epoch": 1.042455242966752, + "grad_norm": 8.485763549804688, + "learning_rate": 1e-06, + "loss": 0.201, + "mean_token_accuracy": 0.9272310733795166, + "num_tokens": 6563751.0, + "step": 2038 + }, + { + "epoch": 1.0429667519181585, + "grad_norm": 8.709568977355957, + "learning_rate": 1e-06, + "loss": 0.1862, + "mean_token_accuracy": 0.9169557094573975, + "num_tokens": 6566548.0, + "step": 2039 + }, + { + "epoch": 1.0434782608695652, + "grad_norm": 15.496163368225098, + "learning_rate": 1e-06, + "loss": 0.3197, + "mean_token_accuracy": 0.8914110660552979, + "num_tokens": 6569674.0, + "step": 2040 + }, + { + "epoch": 1.043989769820972, + "grad_norm": 11.49150276184082, + "learning_rate": 1e-06, + "loss": 0.2088, + "mean_token_accuracy": 0.9324663877487183, + "num_tokens": 6572501.0, + "step": 2041 + }, + { + "epoch": 1.0445012787723784, + "grad_norm": 5.530673980712891, + "learning_rate": 1e-06, + "loss": 0.138, + "mean_token_accuracy": 0.9541738033294678, + "num_tokens": 6575636.0, + "step": 2042 + }, + { + "epoch": 1.0450127877237851, + "grad_norm": 10.270037651062012, + "learning_rate": 1e-06, + "loss": 0.2669, + "mean_token_accuracy": 0.8762193322181702, + "num_tokens": 6578122.0, + "step": 2043 + }, + { + "epoch": 1.0455242966751919, + "grad_norm": 12.504125595092773, + "learning_rate": 1e-06, + "loss": 0.2455, + "mean_token_accuracy": 0.9223916530609131, + "num_tokens": 6581653.0, + "step": 2044 + }, + { + "epoch": 1.0460358056265984, + "grad_norm": 14.55234146118164, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8826515674591064, + "num_tokens": 6584797.0, + "step": 2045 + }, + { + "epoch": 1.046547314578005, + "grad_norm": 13.326356887817383, + "learning_rate": 1e-06, + "loss": 0.2151, + "mean_token_accuracy": 0.9185922145843506, + "num_tokens": 6588288.0, + "step": 2046 + }, + { + "epoch": 1.0470588235294118, + "grad_norm": 2.9745407104492188, + "learning_rate": 1e-06, + "loss": 0.0813, + "mean_token_accuracy": 0.9514995813369751, + "num_tokens": 6592790.0, + "step": 2047 + }, + { + "epoch": 1.0475703324808183, + "grad_norm": 4.363402843475342, + "learning_rate": 1e-06, + "loss": 0.1066, + "mean_token_accuracy": 0.9229443669319153, + "num_tokens": 6597886.0, + "step": 2048 + }, + { + "epoch": 1.048081841432225, + "grad_norm": 7.2108941078186035, + "learning_rate": 1e-06, + "loss": 0.1688, + "mean_token_accuracy": 0.9411910772323608, + "num_tokens": 6601200.0, + "step": 2049 + }, + { + "epoch": 1.0485933503836318, + "grad_norm": 12.141940116882324, + "learning_rate": 1e-06, + "loss": 0.2216, + "mean_token_accuracy": 0.9262883067131042, + "num_tokens": 6604358.0, + "step": 2050 + }, + { + "epoch": 1.0491048593350383, + "grad_norm": 18.39200782775879, + "learning_rate": 1e-06, + "loss": 0.2683, + "mean_token_accuracy": 0.9331657290458679, + "num_tokens": 6607678.0, + "step": 2051 + }, + { + "epoch": 1.049616368286445, + "grad_norm": 6.152890205383301, + "learning_rate": 1e-06, + "loss": 0.1336, + "mean_token_accuracy": 0.9337465763092041, + "num_tokens": 6611857.0, + "step": 2052 + }, + { + "epoch": 1.0501278772378517, + "grad_norm": 9.311521530151367, + "learning_rate": 1e-06, + "loss": 0.1752, + "mean_token_accuracy": 0.9312773942947388, + "num_tokens": 6615586.0, + "step": 2053 + }, + { + "epoch": 1.0506393861892582, + "grad_norm": 12.86143970489502, + "learning_rate": 1e-06, + "loss": 0.2316, + "mean_token_accuracy": 0.8885865807533264, + "num_tokens": 6617892.0, + "step": 2054 + }, + { + "epoch": 1.051150895140665, + "grad_norm": 11.524815559387207, + "learning_rate": 1e-06, + "loss": 0.2569, + "mean_token_accuracy": 0.8742471933364868, + "num_tokens": 6620743.0, + "step": 2055 + }, + { + "epoch": 1.0516624040920717, + "grad_norm": 11.04532527923584, + "learning_rate": 1e-06, + "loss": 0.3252, + "mean_token_accuracy": 0.8431333303451538, + "num_tokens": 6623701.0, + "step": 2056 + }, + { + "epoch": 1.0521739130434782, + "grad_norm": 10.255529403686523, + "learning_rate": 1e-06, + "loss": 0.246, + "mean_token_accuracy": 0.8763452172279358, + "num_tokens": 6626741.0, + "step": 2057 + }, + { + "epoch": 1.052685421994885, + "grad_norm": 10.830379486083984, + "learning_rate": 1e-06, + "loss": 0.2574, + "mean_token_accuracy": 0.9107431173324585, + "num_tokens": 6629978.0, + "step": 2058 + }, + { + "epoch": 1.0531969309462916, + "grad_norm": 10.432997703552246, + "learning_rate": 1e-06, + "loss": 0.2343, + "mean_token_accuracy": 0.9157902002334595, + "num_tokens": 6633203.0, + "step": 2059 + }, + { + "epoch": 1.0537084398976981, + "grad_norm": 12.1557035446167, + "learning_rate": 1e-06, + "loss": 0.2463, + "mean_token_accuracy": 0.8939677476882935, + "num_tokens": 6636607.0, + "step": 2060 + }, + { + "epoch": 1.0542199488491049, + "grad_norm": 10.318500518798828, + "learning_rate": 1e-06, + "loss": 0.3011, + "mean_token_accuracy": 0.8755599856376648, + "num_tokens": 6639285.0, + "step": 2061 + }, + { + "epoch": 1.0547314578005116, + "grad_norm": 20.269052505493164, + "learning_rate": 1e-06, + "loss": 0.3233, + "mean_token_accuracy": 0.8773162961006165, + "num_tokens": 6643287.0, + "step": 2062 + }, + { + "epoch": 1.055242966751918, + "grad_norm": 16.074600219726562, + "learning_rate": 1e-06, + "loss": 0.2658, + "mean_token_accuracy": 0.8874098062515259, + "num_tokens": 6646776.0, + "step": 2063 + }, + { + "epoch": 1.0557544757033248, + "grad_norm": 12.912442207336426, + "learning_rate": 1e-06, + "loss": 0.2411, + "mean_token_accuracy": 0.9021714925765991, + "num_tokens": 6649581.0, + "step": 2064 + }, + { + "epoch": 1.0562659846547315, + "grad_norm": 12.160850524902344, + "learning_rate": 1e-06, + "loss": 0.2696, + "mean_token_accuracy": 0.8922759294509888, + "num_tokens": 6653589.0, + "step": 2065 + }, + { + "epoch": 1.056777493606138, + "grad_norm": 10.617358207702637, + "learning_rate": 1e-06, + "loss": 0.2285, + "mean_token_accuracy": 0.9118250608444214, + "num_tokens": 6656519.0, + "step": 2066 + }, + { + "epoch": 1.0572890025575448, + "grad_norm": 17.645519256591797, + "learning_rate": 1e-06, + "loss": 0.2632, + "mean_token_accuracy": 0.9063986539840698, + "num_tokens": 6659392.0, + "step": 2067 + }, + { + "epoch": 1.0578005115089515, + "grad_norm": 11.590937614440918, + "learning_rate": 1e-06, + "loss": 0.3089, + "mean_token_accuracy": 0.8933702707290649, + "num_tokens": 6661706.0, + "step": 2068 + }, + { + "epoch": 1.058312020460358, + "grad_norm": 8.383427619934082, + "learning_rate": 1e-06, + "loss": 0.1566, + "mean_token_accuracy": 0.927408754825592, + "num_tokens": 6664883.0, + "step": 2069 + }, + { + "epoch": 1.0588235294117647, + "grad_norm": 12.651789665222168, + "learning_rate": 1e-06, + "loss": 0.2127, + "mean_token_accuracy": 0.9261206388473511, + "num_tokens": 6667561.0, + "step": 2070 + }, + { + "epoch": 1.0593350383631714, + "grad_norm": 7.786194801330566, + "learning_rate": 1e-06, + "loss": 0.2097, + "mean_token_accuracy": 0.867415726184845, + "num_tokens": 6670949.0, + "step": 2071 + }, + { + "epoch": 1.059846547314578, + "grad_norm": 9.361618041992188, + "learning_rate": 1e-06, + "loss": 0.2057, + "mean_token_accuracy": 0.9048292636871338, + "num_tokens": 6674539.0, + "step": 2072 + }, + { + "epoch": 1.0603580562659847, + "grad_norm": 11.623002052307129, + "learning_rate": 1e-06, + "loss": 0.3178, + "mean_token_accuracy": 0.8740578889846802, + "num_tokens": 6677163.0, + "step": 2073 + }, + { + "epoch": 1.0608695652173914, + "grad_norm": 7.314454555511475, + "learning_rate": 1e-06, + "loss": 0.1831, + "mean_token_accuracy": 0.9235742092132568, + "num_tokens": 6681541.0, + "step": 2074 + }, + { + "epoch": 1.061381074168798, + "grad_norm": 10.31442928314209, + "learning_rate": 1e-06, + "loss": 0.242, + "mean_token_accuracy": 0.908051073551178, + "num_tokens": 6684765.0, + "step": 2075 + }, + { + "epoch": 1.0618925831202046, + "grad_norm": 9.782906532287598, + "learning_rate": 1e-06, + "loss": 0.2237, + "mean_token_accuracy": 0.9231007695198059, + "num_tokens": 6687276.0, + "step": 2076 + }, + { + "epoch": 1.0624040920716114, + "grad_norm": 9.431356430053711, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.8971902132034302, + "num_tokens": 6690589.0, + "step": 2077 + }, + { + "epoch": 1.0629156010230179, + "grad_norm": 8.70428466796875, + "learning_rate": 1e-06, + "loss": 0.2441, + "mean_token_accuracy": 0.8827615976333618, + "num_tokens": 6693333.0, + "step": 2078 + }, + { + "epoch": 1.0634271099744246, + "grad_norm": 11.001129150390625, + "learning_rate": 1e-06, + "loss": 0.2887, + "mean_token_accuracy": 0.8956708908081055, + "num_tokens": 6695518.0, + "step": 2079 + }, + { + "epoch": 1.0639386189258313, + "grad_norm": 10.012603759765625, + "learning_rate": 1e-06, + "loss": 0.2678, + "mean_token_accuracy": 0.8807495832443237, + "num_tokens": 6698527.0, + "step": 2080 + }, + { + "epoch": 1.0644501278772378, + "grad_norm": 15.307744026184082, + "learning_rate": 1e-06, + "loss": 0.3482, + "mean_token_accuracy": 0.8784973621368408, + "num_tokens": 6702087.0, + "step": 2081 + }, + { + "epoch": 1.0649616368286445, + "grad_norm": 7.77947473526001, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.9111495614051819, + "num_tokens": 6705361.0, + "step": 2082 + }, + { + "epoch": 1.065473145780051, + "grad_norm": 8.23793888092041, + "learning_rate": 1e-06, + "loss": 0.2599, + "mean_token_accuracy": 0.8875932693481445, + "num_tokens": 6709583.0, + "step": 2083 + }, + { + "epoch": 1.0659846547314578, + "grad_norm": 10.798311233520508, + "learning_rate": 1e-06, + "loss": 0.2848, + "mean_token_accuracy": 0.8991026282310486, + "num_tokens": 6712030.0, + "step": 2084 + }, + { + "epoch": 1.0664961636828645, + "grad_norm": 10.701519966125488, + "learning_rate": 1e-06, + "loss": 0.261, + "mean_token_accuracy": 0.9027113318443298, + "num_tokens": 6715567.0, + "step": 2085 + }, + { + "epoch": 1.0670076726342712, + "grad_norm": 11.342215538024902, + "learning_rate": 1e-06, + "loss": 0.2498, + "mean_token_accuracy": 0.8885114192962646, + "num_tokens": 6718325.0, + "step": 2086 + }, + { + "epoch": 1.0675191815856777, + "grad_norm": 5.401583194732666, + "learning_rate": 1e-06, + "loss": 0.1081, + "mean_token_accuracy": 0.9363837838172913, + "num_tokens": 6722015.0, + "step": 2087 + }, + { + "epoch": 1.0680306905370844, + "grad_norm": 11.210810661315918, + "learning_rate": 1e-06, + "loss": 0.1555, + "mean_token_accuracy": 0.9173449277877808, + "num_tokens": 6725282.0, + "step": 2088 + }, + { + "epoch": 1.068542199488491, + "grad_norm": 6.89104700088501, + "learning_rate": 1e-06, + "loss": 0.1755, + "mean_token_accuracy": 0.9347108006477356, + "num_tokens": 6729996.0, + "step": 2089 + }, + { + "epoch": 1.0690537084398977, + "grad_norm": 9.431035041809082, + "learning_rate": 1e-06, + "loss": 0.188, + "mean_token_accuracy": 0.9236336350440979, + "num_tokens": 6733474.0, + "step": 2090 + }, + { + "epoch": 1.0695652173913044, + "grad_norm": 12.664275169372559, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.897087037563324, + "num_tokens": 6735914.0, + "step": 2091 + }, + { + "epoch": 1.0700767263427111, + "grad_norm": 8.90561294555664, + "learning_rate": 1e-06, + "loss": 0.2297, + "mean_token_accuracy": 0.9207029342651367, + "num_tokens": 6739050.0, + "step": 2092 + }, + { + "epoch": 1.0705882352941176, + "grad_norm": 13.278166770935059, + "learning_rate": 1e-06, + "loss": 0.2823, + "mean_token_accuracy": 0.886242151260376, + "num_tokens": 6741568.0, + "step": 2093 + }, + { + "epoch": 1.0710997442455243, + "grad_norm": 10.151549339294434, + "learning_rate": 1e-06, + "loss": 0.2409, + "mean_token_accuracy": 0.882265567779541, + "num_tokens": 6744385.0, + "step": 2094 + }, + { + "epoch": 1.0716112531969308, + "grad_norm": 12.924657821655273, + "learning_rate": 1e-06, + "loss": 0.3401, + "mean_token_accuracy": 0.8506594300270081, + "num_tokens": 6747042.0, + "step": 2095 + }, + { + "epoch": 1.0721227621483376, + "grad_norm": 12.357630729675293, + "learning_rate": 1e-06, + "loss": 0.2514, + "mean_token_accuracy": 0.9115992784500122, + "num_tokens": 6749860.0, + "step": 2096 + }, + { + "epoch": 1.0726342710997443, + "grad_norm": 11.000275611877441, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.9015095233917236, + "num_tokens": 6752966.0, + "step": 2097 + }, + { + "epoch": 1.0731457800511508, + "grad_norm": 11.294280052185059, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8682204484939575, + "num_tokens": 6755971.0, + "step": 2098 + }, + { + "epoch": 1.0736572890025575, + "grad_norm": 8.445996284484863, + "learning_rate": 1e-06, + "loss": 0.1673, + "mean_token_accuracy": 0.9405840039253235, + "num_tokens": 6759418.0, + "step": 2099 + }, + { + "epoch": 1.0741687979539642, + "grad_norm": 10.094491004943848, + "learning_rate": 1e-06, + "loss": 0.2194, + "mean_token_accuracy": 0.8988722562789917, + "num_tokens": 6762955.0, + "step": 2100 + }, + { + "epoch": 1.0746803069053708, + "grad_norm": 11.572063446044922, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8942611217498779, + "num_tokens": 6766565.0, + "step": 2101 + }, + { + "epoch": 1.0751918158567775, + "grad_norm": 7.548501968383789, + "learning_rate": 1e-06, + "loss": 0.1822, + "mean_token_accuracy": 0.9258499145507812, + "num_tokens": 6770285.0, + "step": 2102 + }, + { + "epoch": 1.0757033248081842, + "grad_norm": 15.54690933227539, + "learning_rate": 1e-06, + "loss": 0.33, + "mean_token_accuracy": 0.878604531288147, + "num_tokens": 6773185.0, + "step": 2103 + }, + { + "epoch": 1.0762148337595907, + "grad_norm": 9.071660995483398, + "learning_rate": 1e-06, + "loss": 0.2628, + "mean_token_accuracy": 0.9018183350563049, + "num_tokens": 6775897.0, + "step": 2104 + }, + { + "epoch": 1.0767263427109974, + "grad_norm": 10.25224494934082, + "learning_rate": 1e-06, + "loss": 0.2664, + "mean_token_accuracy": 0.9010364413261414, + "num_tokens": 6778801.0, + "step": 2105 + }, + { + "epoch": 1.0772378516624042, + "grad_norm": 10.377336502075195, + "learning_rate": 1e-06, + "loss": 0.2848, + "mean_token_accuracy": 0.8749977946281433, + "num_tokens": 6781224.0, + "step": 2106 + }, + { + "epoch": 1.0777493606138107, + "grad_norm": 12.00639533996582, + "learning_rate": 1e-06, + "loss": 0.2017, + "mean_token_accuracy": 0.9142764806747437, + "num_tokens": 6784499.0, + "step": 2107 + }, + { + "epoch": 1.0782608695652174, + "grad_norm": 12.651263236999512, + "learning_rate": 1e-06, + "loss": 0.3548, + "mean_token_accuracy": 0.8539076447486877, + "num_tokens": 6787068.0, + "step": 2108 + }, + { + "epoch": 1.078772378516624, + "grad_norm": 7.261185169219971, + "learning_rate": 1e-06, + "loss": 0.1534, + "mean_token_accuracy": 0.9454278945922852, + "num_tokens": 6790017.0, + "step": 2109 + }, + { + "epoch": 1.0792838874680306, + "grad_norm": 8.94935417175293, + "learning_rate": 1e-06, + "loss": 0.2001, + "mean_token_accuracy": 0.8969190120697021, + "num_tokens": 6793092.0, + "step": 2110 + }, + { + "epoch": 1.0797953964194373, + "grad_norm": 9.14257526397705, + "learning_rate": 1e-06, + "loss": 0.2005, + "mean_token_accuracy": 0.8975969552993774, + "num_tokens": 6795840.0, + "step": 2111 + }, + { + "epoch": 1.080306905370844, + "grad_norm": 8.920404434204102, + "learning_rate": 1e-06, + "loss": 0.2121, + "mean_token_accuracy": 0.9052276611328125, + "num_tokens": 6799015.0, + "step": 2112 + }, + { + "epoch": 1.0808184143222506, + "grad_norm": 7.568833827972412, + "learning_rate": 1e-06, + "loss": 0.1823, + "mean_token_accuracy": 0.9173954725265503, + "num_tokens": 6801966.0, + "step": 2113 + }, + { + "epoch": 1.0813299232736573, + "grad_norm": 10.158663749694824, + "learning_rate": 1e-06, + "loss": 0.2144, + "mean_token_accuracy": 0.9135432839393616, + "num_tokens": 6805286.0, + "step": 2114 + }, + { + "epoch": 1.081841432225064, + "grad_norm": 8.76334285736084, + "learning_rate": 1e-06, + "loss": 0.256, + "mean_token_accuracy": 0.8995426893234253, + "num_tokens": 6808341.0, + "step": 2115 + }, + { + "epoch": 1.0823529411764705, + "grad_norm": 7.389133453369141, + "learning_rate": 1e-06, + "loss": 0.2012, + "mean_token_accuracy": 0.8667203783988953, + "num_tokens": 6811485.0, + "step": 2116 + }, + { + "epoch": 1.0828644501278772, + "grad_norm": 12.964170455932617, + "learning_rate": 1e-06, + "loss": 0.3008, + "mean_token_accuracy": 0.8863992094993591, + "num_tokens": 6815383.0, + "step": 2117 + }, + { + "epoch": 1.083375959079284, + "grad_norm": 10.670763969421387, + "learning_rate": 1e-06, + "loss": 0.2694, + "mean_token_accuracy": 0.8999520540237427, + "num_tokens": 6818767.0, + "step": 2118 + }, + { + "epoch": 1.0838874680306905, + "grad_norm": 11.44982624053955, + "learning_rate": 1e-06, + "loss": 0.2695, + "mean_token_accuracy": 0.8888975381851196, + "num_tokens": 6822323.0, + "step": 2119 + }, + { + "epoch": 1.0843989769820972, + "grad_norm": 7.362820625305176, + "learning_rate": 1e-06, + "loss": 0.1574, + "mean_token_accuracy": 0.9398925304412842, + "num_tokens": 6825460.0, + "step": 2120 + }, + { + "epoch": 1.084910485933504, + "grad_norm": 16.659887313842773, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.9057125449180603, + "num_tokens": 6828019.0, + "step": 2121 + }, + { + "epoch": 1.0854219948849104, + "grad_norm": 12.025336265563965, + "learning_rate": 1e-06, + "loss": 0.2314, + "mean_token_accuracy": 0.8899120688438416, + "num_tokens": 6830864.0, + "step": 2122 + }, + { + "epoch": 1.0859335038363171, + "grad_norm": 9.709813117980957, + "learning_rate": 1e-06, + "loss": 0.2276, + "mean_token_accuracy": 0.8702185153961182, + "num_tokens": 6834733.0, + "step": 2123 + }, + { + "epoch": 1.0864450127877239, + "grad_norm": 9.989523887634277, + "learning_rate": 1e-06, + "loss": 0.2331, + "mean_token_accuracy": 0.8943054676055908, + "num_tokens": 6837760.0, + "step": 2124 + }, + { + "epoch": 1.0869565217391304, + "grad_norm": 8.943840980529785, + "learning_rate": 1e-06, + "loss": 0.1871, + "mean_token_accuracy": 0.925469160079956, + "num_tokens": 6841339.0, + "step": 2125 + }, + { + "epoch": 1.087468030690537, + "grad_norm": 9.286567687988281, + "learning_rate": 1e-06, + "loss": 0.1462, + "mean_token_accuracy": 0.938710629940033, + "num_tokens": 6845007.0, + "step": 2126 + }, + { + "epoch": 1.0879795396419438, + "grad_norm": 10.317068099975586, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8878107070922852, + "num_tokens": 6847945.0, + "step": 2127 + }, + { + "epoch": 1.0884910485933503, + "grad_norm": 8.341854095458984, + "learning_rate": 1e-06, + "loss": 0.1773, + "mean_token_accuracy": 0.9127386808395386, + "num_tokens": 6850550.0, + "step": 2128 + }, + { + "epoch": 1.089002557544757, + "grad_norm": 11.504859924316406, + "learning_rate": 1e-06, + "loss": 0.3073, + "mean_token_accuracy": 0.8613033294677734, + "num_tokens": 6853377.0, + "step": 2129 + }, + { + "epoch": 1.0895140664961638, + "grad_norm": 18.018394470214844, + "learning_rate": 1e-06, + "loss": 0.3419, + "mean_token_accuracy": 0.8741469383239746, + "num_tokens": 6857374.0, + "step": 2130 + }, + { + "epoch": 1.0900255754475703, + "grad_norm": 11.40362548828125, + "learning_rate": 1e-06, + "loss": 0.2616, + "mean_token_accuracy": 0.9042646884918213, + "num_tokens": 6860215.0, + "step": 2131 + }, + { + "epoch": 1.090537084398977, + "grad_norm": 7.488648891448975, + "learning_rate": 1e-06, + "loss": 0.1815, + "mean_token_accuracy": 0.9015781879425049, + "num_tokens": 6863377.0, + "step": 2132 + }, + { + "epoch": 1.0910485933503837, + "grad_norm": 11.109898567199707, + "learning_rate": 1e-06, + "loss": 0.3102, + "mean_token_accuracy": 0.8958115577697754, + "num_tokens": 6866697.0, + "step": 2133 + }, + { + "epoch": 1.0915601023017902, + "grad_norm": 10.50112247467041, + "learning_rate": 1e-06, + "loss": 0.2606, + "mean_token_accuracy": 0.8908811211585999, + "num_tokens": 6869552.0, + "step": 2134 + }, + { + "epoch": 1.092071611253197, + "grad_norm": 9.52924919128418, + "learning_rate": 1e-06, + "loss": 0.1492, + "mean_token_accuracy": 0.9335397481918335, + "num_tokens": 6872873.0, + "step": 2135 + }, + { + "epoch": 1.0925831202046035, + "grad_norm": 10.917764663696289, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8935797810554504, + "num_tokens": 6876188.0, + "step": 2136 + }, + { + "epoch": 1.0930946291560102, + "grad_norm": 11.402142524719238, + "learning_rate": 1e-06, + "loss": 0.2203, + "mean_token_accuracy": 0.9218116402626038, + "num_tokens": 6879336.0, + "step": 2137 + }, + { + "epoch": 1.093606138107417, + "grad_norm": 9.737712860107422, + "learning_rate": 1e-06, + "loss": 0.2641, + "mean_token_accuracy": 0.8980284929275513, + "num_tokens": 6881894.0, + "step": 2138 + }, + { + "epoch": 1.0941176470588236, + "grad_norm": 11.521879196166992, + "learning_rate": 1e-06, + "loss": 0.2487, + "mean_token_accuracy": 0.8948795795440674, + "num_tokens": 6885001.0, + "step": 2139 + }, + { + "epoch": 1.0946291560102301, + "grad_norm": 11.982460021972656, + "learning_rate": 1e-06, + "loss": 0.2466, + "mean_token_accuracy": 0.9148514866828918, + "num_tokens": 6888055.0, + "step": 2140 + }, + { + "epoch": 1.0951406649616369, + "grad_norm": 10.948339462280273, + "learning_rate": 1e-06, + "loss": 0.2355, + "mean_token_accuracy": 0.9063748717308044, + "num_tokens": 6891136.0, + "step": 2141 + }, + { + "epoch": 1.0956521739130434, + "grad_norm": 14.180554389953613, + "learning_rate": 1e-06, + "loss": 0.2638, + "mean_token_accuracy": 0.8912028074264526, + "num_tokens": 6893457.0, + "step": 2142 + }, + { + "epoch": 1.09616368286445, + "grad_norm": 11.968879699707031, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8425794839859009, + "num_tokens": 6896159.0, + "step": 2143 + }, + { + "epoch": 1.0966751918158568, + "grad_norm": 11.927451133728027, + "learning_rate": 1e-06, + "loss": 0.1808, + "mean_token_accuracy": 0.9250073432922363, + "num_tokens": 6898795.0, + "step": 2144 + }, + { + "epoch": 1.0971867007672633, + "grad_norm": 9.313322067260742, + "learning_rate": 1e-06, + "loss": 0.2577, + "mean_token_accuracy": 0.9094603061676025, + "num_tokens": 6902010.0, + "step": 2145 + }, + { + "epoch": 1.09769820971867, + "grad_norm": 9.714174270629883, + "learning_rate": 1e-06, + "loss": 0.2182, + "mean_token_accuracy": 0.9062336683273315, + "num_tokens": 6904756.0, + "step": 2146 + }, + { + "epoch": 1.0982097186700768, + "grad_norm": 9.40809154510498, + "learning_rate": 1e-06, + "loss": 0.2302, + "mean_token_accuracy": 0.9146444201469421, + "num_tokens": 6907483.0, + "step": 2147 + }, + { + "epoch": 1.0987212276214833, + "grad_norm": 8.729114532470703, + "learning_rate": 1e-06, + "loss": 0.2332, + "mean_token_accuracy": 0.9011156558990479, + "num_tokens": 6909798.0, + "step": 2148 + }, + { + "epoch": 1.09923273657289, + "grad_norm": 9.993867874145508, + "learning_rate": 1e-06, + "loss": 0.2319, + "mean_token_accuracy": 0.8958847522735596, + "num_tokens": 6913742.0, + "step": 2149 + }, + { + "epoch": 1.0997442455242967, + "grad_norm": 12.671309471130371, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.9010518193244934, + "num_tokens": 6916680.0, + "step": 2150 + }, + { + "epoch": 1.1002557544757032, + "grad_norm": 11.859639167785645, + "learning_rate": 1e-06, + "loss": 0.2611, + "mean_token_accuracy": 0.8855000734329224, + "num_tokens": 6919766.0, + "step": 2151 + }, + { + "epoch": 1.10076726342711, + "grad_norm": 10.367818832397461, + "learning_rate": 1e-06, + "loss": 0.2423, + "mean_token_accuracy": 0.9049134254455566, + "num_tokens": 6922818.0, + "step": 2152 + }, + { + "epoch": 1.1012787723785167, + "grad_norm": 12.123175621032715, + "learning_rate": 1e-06, + "loss": 0.2385, + "mean_token_accuracy": 0.9096941947937012, + "num_tokens": 6925889.0, + "step": 2153 + }, + { + "epoch": 1.1017902813299232, + "grad_norm": 11.665446281433105, + "learning_rate": 1e-06, + "loss": 0.3218, + "mean_token_accuracy": 0.8628591895103455, + "num_tokens": 6929281.0, + "step": 2154 + }, + { + "epoch": 1.10230179028133, + "grad_norm": 7.594274520874023, + "learning_rate": 1e-06, + "loss": 0.1856, + "mean_token_accuracy": 0.9025778770446777, + "num_tokens": 6932830.0, + "step": 2155 + }, + { + "epoch": 1.1028132992327366, + "grad_norm": 6.93482780456543, + "learning_rate": 1e-06, + "loss": 0.1636, + "mean_token_accuracy": 0.8982844352722168, + "num_tokens": 6935944.0, + "step": 2156 + }, + { + "epoch": 1.1033248081841431, + "grad_norm": 9.710680961608887, + "learning_rate": 1e-06, + "loss": 0.1761, + "mean_token_accuracy": 0.895551323890686, + "num_tokens": 6938917.0, + "step": 2157 + }, + { + "epoch": 1.1038363171355499, + "grad_norm": 10.239545822143555, + "learning_rate": 1e-06, + "loss": 0.225, + "mean_token_accuracy": 0.909196674823761, + "num_tokens": 6942214.0, + "step": 2158 + }, + { + "epoch": 1.1043478260869566, + "grad_norm": 9.68403434753418, + "learning_rate": 1e-06, + "loss": 0.244, + "mean_token_accuracy": 0.9034098386764526, + "num_tokens": 6946231.0, + "step": 2159 + }, + { + "epoch": 1.104859335038363, + "grad_norm": 14.125535011291504, + "learning_rate": 1e-06, + "loss": 0.2769, + "mean_token_accuracy": 0.9101502895355225, + "num_tokens": 6949347.0, + "step": 2160 + }, + { + "epoch": 1.1053708439897698, + "grad_norm": 10.293949127197266, + "learning_rate": 1e-06, + "loss": 0.2072, + "mean_token_accuracy": 0.9175481200218201, + "num_tokens": 6952847.0, + "step": 2161 + }, + { + "epoch": 1.1058823529411765, + "grad_norm": 10.058213233947754, + "learning_rate": 1e-06, + "loss": 0.2099, + "mean_token_accuracy": 0.9226607084274292, + "num_tokens": 6955547.0, + "step": 2162 + }, + { + "epoch": 1.106393861892583, + "grad_norm": 8.147599220275879, + "learning_rate": 1e-06, + "loss": 0.1807, + "mean_token_accuracy": 0.9226847887039185, + "num_tokens": 6958752.0, + "step": 2163 + }, + { + "epoch": 1.1069053708439898, + "grad_norm": 7.398456573486328, + "learning_rate": 1e-06, + "loss": 0.1362, + "mean_token_accuracy": 0.8978420495986938, + "num_tokens": 6962770.0, + "step": 2164 + }, + { + "epoch": 1.1074168797953965, + "grad_norm": 14.994071960449219, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.868838369846344, + "num_tokens": 6965625.0, + "step": 2165 + }, + { + "epoch": 1.107928388746803, + "grad_norm": 10.1029634475708, + "learning_rate": 1e-06, + "loss": 0.2821, + "mean_token_accuracy": 0.8935680389404297, + "num_tokens": 6968651.0, + "step": 2166 + }, + { + "epoch": 1.1084398976982097, + "grad_norm": 9.740897178649902, + "learning_rate": 1e-06, + "loss": 0.1835, + "mean_token_accuracy": 0.9098591804504395, + "num_tokens": 6972372.0, + "step": 2167 + }, + { + "epoch": 1.1089514066496164, + "grad_norm": 11.398907661437988, + "learning_rate": 1e-06, + "loss": 0.224, + "mean_token_accuracy": 0.8998673558235168, + "num_tokens": 6976254.0, + "step": 2168 + }, + { + "epoch": 1.109462915601023, + "grad_norm": 7.591893672943115, + "learning_rate": 1e-06, + "loss": 0.1701, + "mean_token_accuracy": 0.9155486226081848, + "num_tokens": 6979773.0, + "step": 2169 + }, + { + "epoch": 1.1099744245524297, + "grad_norm": 6.704920768737793, + "learning_rate": 1e-06, + "loss": 0.2036, + "mean_token_accuracy": 0.911674439907074, + "num_tokens": 6982188.0, + "step": 2170 + }, + { + "epoch": 1.1104859335038364, + "grad_norm": 14.896906852722168, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8543835282325745, + "num_tokens": 6985452.0, + "step": 2171 + }, + { + "epoch": 1.110997442455243, + "grad_norm": 10.423784255981445, + "learning_rate": 1e-06, + "loss": 0.2116, + "mean_token_accuracy": 0.8688236474990845, + "num_tokens": 6988613.0, + "step": 2172 + }, + { + "epoch": 1.1115089514066496, + "grad_norm": 8.68607234954834, + "learning_rate": 1e-06, + "loss": 0.2531, + "mean_token_accuracy": 0.8929252624511719, + "num_tokens": 6992193.0, + "step": 2173 + }, + { + "epoch": 1.1120204603580564, + "grad_norm": 10.454185485839844, + "learning_rate": 1e-06, + "loss": 0.2687, + "mean_token_accuracy": 0.8876752257347107, + "num_tokens": 6995480.0, + "step": 2174 + }, + { + "epoch": 1.1125319693094629, + "grad_norm": 9.458806991577148, + "learning_rate": 1e-06, + "loss": 0.2621, + "mean_token_accuracy": 0.8991872072219849, + "num_tokens": 6998147.0, + "step": 2175 + }, + { + "epoch": 1.1130434782608696, + "grad_norm": 9.870410919189453, + "learning_rate": 1e-06, + "loss": 0.2612, + "mean_token_accuracy": 0.8867816925048828, + "num_tokens": 7001554.0, + "step": 2176 + }, + { + "epoch": 1.1135549872122763, + "grad_norm": 10.648259162902832, + "learning_rate": 1e-06, + "loss": 0.239, + "mean_token_accuracy": 0.9097261428833008, + "num_tokens": 7005214.0, + "step": 2177 + }, + { + "epoch": 1.1140664961636828, + "grad_norm": 11.083989143371582, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8630902767181396, + "num_tokens": 7008883.0, + "step": 2178 + }, + { + "epoch": 1.1145780051150895, + "grad_norm": 6.87490177154541, + "learning_rate": 1e-06, + "loss": 0.1966, + "mean_token_accuracy": 0.9170922040939331, + "num_tokens": 7011563.0, + "step": 2179 + }, + { + "epoch": 1.1150895140664963, + "grad_norm": 8.983192443847656, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9170784950256348, + "num_tokens": 7014605.0, + "step": 2180 + }, + { + "epoch": 1.1156010230179028, + "grad_norm": 11.61516284942627, + "learning_rate": 1e-06, + "loss": 0.2679, + "mean_token_accuracy": 0.8739153146743774, + "num_tokens": 7017327.0, + "step": 2181 + }, + { + "epoch": 1.1161125319693095, + "grad_norm": 11.410086631774902, + "learning_rate": 1e-06, + "loss": 0.2547, + "mean_token_accuracy": 0.8968445062637329, + "num_tokens": 7019614.0, + "step": 2182 + }, + { + "epoch": 1.1166240409207162, + "grad_norm": 8.471768379211426, + "learning_rate": 1e-06, + "loss": 0.2501, + "mean_token_accuracy": 0.908902108669281, + "num_tokens": 7023197.0, + "step": 2183 + }, + { + "epoch": 1.1171355498721227, + "grad_norm": 8.693395614624023, + "learning_rate": 1e-06, + "loss": 0.2232, + "mean_token_accuracy": 0.9077087640762329, + "num_tokens": 7025954.0, + "step": 2184 + }, + { + "epoch": 1.1176470588235294, + "grad_norm": 10.378610610961914, + "learning_rate": 1e-06, + "loss": 0.2761, + "mean_token_accuracy": 0.9104347229003906, + "num_tokens": 7028827.0, + "step": 2185 + }, + { + "epoch": 1.1181585677749362, + "grad_norm": 10.277524948120117, + "learning_rate": 1e-06, + "loss": 0.2511, + "mean_token_accuracy": 0.8919599652290344, + "num_tokens": 7031676.0, + "step": 2186 + }, + { + "epoch": 1.1186700767263427, + "grad_norm": 10.464884757995605, + "learning_rate": 1e-06, + "loss": 0.2478, + "mean_token_accuracy": 0.9113698601722717, + "num_tokens": 7034587.0, + "step": 2187 + }, + { + "epoch": 1.1191815856777494, + "grad_norm": 12.082450866699219, + "learning_rate": 1e-06, + "loss": 0.2491, + "mean_token_accuracy": 0.9057352542877197, + "num_tokens": 7037224.0, + "step": 2188 + }, + { + "epoch": 1.119693094629156, + "grad_norm": 17.93623161315918, + "learning_rate": 1e-06, + "loss": 0.2947, + "mean_token_accuracy": 0.8757959008216858, + "num_tokens": 7040402.0, + "step": 2189 + }, + { + "epoch": 1.1202046035805626, + "grad_norm": 10.067076683044434, + "learning_rate": 1e-06, + "loss": 0.2699, + "mean_token_accuracy": 0.8881385326385498, + "num_tokens": 7043350.0, + "step": 2190 + }, + { + "epoch": 1.1207161125319693, + "grad_norm": 6.395715236663818, + "learning_rate": 1e-06, + "loss": 0.1816, + "mean_token_accuracy": 0.9214979410171509, + "num_tokens": 7046201.0, + "step": 2191 + }, + { + "epoch": 1.121227621483376, + "grad_norm": 8.651055335998535, + "learning_rate": 1e-06, + "loss": 0.2034, + "mean_token_accuracy": 0.8938620090484619, + "num_tokens": 7049801.0, + "step": 2192 + }, + { + "epoch": 1.1217391304347826, + "grad_norm": 11.159689903259277, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.8851997256278992, + "num_tokens": 7052980.0, + "step": 2193 + }, + { + "epoch": 1.1222506393861893, + "grad_norm": 5.518963813781738, + "learning_rate": 1e-06, + "loss": 0.1309, + "mean_token_accuracy": 0.9298708438873291, + "num_tokens": 7056384.0, + "step": 2194 + }, + { + "epoch": 1.1227621483375958, + "grad_norm": 10.532037734985352, + "learning_rate": 1e-06, + "loss": 0.2214, + "mean_token_accuracy": 0.9183001518249512, + "num_tokens": 7059667.0, + "step": 2195 + }, + { + "epoch": 1.1232736572890025, + "grad_norm": 8.093231201171875, + "learning_rate": 1e-06, + "loss": 0.1722, + "mean_token_accuracy": 0.9367104172706604, + "num_tokens": 7062625.0, + "step": 2196 + }, + { + "epoch": 1.1237851662404093, + "grad_norm": 13.824202537536621, + "learning_rate": 1e-06, + "loss": 0.3247, + "mean_token_accuracy": 0.8696835041046143, + "num_tokens": 7065601.0, + "step": 2197 + }, + { + "epoch": 1.1242966751918158, + "grad_norm": 8.840738296508789, + "learning_rate": 1e-06, + "loss": 0.2052, + "mean_token_accuracy": 0.9148761034011841, + "num_tokens": 7068072.0, + "step": 2198 + }, + { + "epoch": 1.1248081841432225, + "grad_norm": 8.736235618591309, + "learning_rate": 1e-06, + "loss": 0.1657, + "mean_token_accuracy": 0.9297216534614563, + "num_tokens": 7071861.0, + "step": 2199 + }, + { + "epoch": 1.1253196930946292, + "grad_norm": 8.654455184936523, + "learning_rate": 1e-06, + "loss": 0.2097, + "mean_token_accuracy": 0.9111760854721069, + "num_tokens": 7075081.0, + "step": 2200 + }, + { + "epoch": 1.1258312020460357, + "grad_norm": 10.193924903869629, + "learning_rate": 1e-06, + "loss": 0.2659, + "mean_token_accuracy": 0.8943027853965759, + "num_tokens": 7078423.0, + "step": 2201 + }, + { + "epoch": 1.1263427109974424, + "grad_norm": 18.680208206176758, + "learning_rate": 1e-06, + "loss": 0.2577, + "mean_token_accuracy": 0.9019107818603516, + "num_tokens": 7082040.0, + "step": 2202 + }, + { + "epoch": 1.1268542199488492, + "grad_norm": 8.822380065917969, + "learning_rate": 1e-06, + "loss": 0.1596, + "mean_token_accuracy": 0.918121337890625, + "num_tokens": 7085493.0, + "step": 2203 + }, + { + "epoch": 1.1273657289002557, + "grad_norm": 9.148045539855957, + "learning_rate": 1e-06, + "loss": 0.216, + "mean_token_accuracy": 0.8951320052146912, + "num_tokens": 7088469.0, + "step": 2204 + }, + { + "epoch": 1.1278772378516624, + "grad_norm": 10.412572860717773, + "learning_rate": 1e-06, + "loss": 0.2286, + "mean_token_accuracy": 0.906450629234314, + "num_tokens": 7092651.0, + "step": 2205 + }, + { + "epoch": 1.128388746803069, + "grad_norm": 8.110722541809082, + "learning_rate": 1e-06, + "loss": 0.1776, + "mean_token_accuracy": 0.9122579097747803, + "num_tokens": 7095650.0, + "step": 2206 + }, + { + "epoch": 1.1289002557544756, + "grad_norm": 6.776087760925293, + "learning_rate": 1e-06, + "loss": 0.1609, + "mean_token_accuracy": 0.9062701463699341, + "num_tokens": 7099405.0, + "step": 2207 + }, + { + "epoch": 1.1294117647058823, + "grad_norm": 11.753105163574219, + "learning_rate": 1e-06, + "loss": 0.2625, + "mean_token_accuracy": 0.8886975049972534, + "num_tokens": 7102010.0, + "step": 2208 + }, + { + "epoch": 1.129923273657289, + "grad_norm": 11.09388542175293, + "learning_rate": 1e-06, + "loss": 0.2985, + "mean_token_accuracy": 0.8969377279281616, + "num_tokens": 7105358.0, + "step": 2209 + }, + { + "epoch": 1.1304347826086956, + "grad_norm": 11.158631324768066, + "learning_rate": 1e-06, + "loss": 0.2526, + "mean_token_accuracy": 0.8901106715202332, + "num_tokens": 7109687.0, + "step": 2210 + }, + { + "epoch": 1.1309462915601023, + "grad_norm": 8.196617126464844, + "learning_rate": 1e-06, + "loss": 0.1472, + "mean_token_accuracy": 0.937610387802124, + "num_tokens": 7113926.0, + "step": 2211 + }, + { + "epoch": 1.131457800511509, + "grad_norm": 10.463065147399902, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.8652879595756531, + "num_tokens": 7116798.0, + "step": 2212 + }, + { + "epoch": 1.1319693094629155, + "grad_norm": 15.74439811706543, + "learning_rate": 1e-06, + "loss": 0.322, + "mean_token_accuracy": 0.8825240731239319, + "num_tokens": 7120583.0, + "step": 2213 + }, + { + "epoch": 1.1324808184143222, + "grad_norm": 10.719107627868652, + "learning_rate": 1e-06, + "loss": 0.2591, + "mean_token_accuracy": 0.9126416444778442, + "num_tokens": 7123680.0, + "step": 2214 + }, + { + "epoch": 1.132992327365729, + "grad_norm": 15.57072925567627, + "learning_rate": 1e-06, + "loss": 0.3727, + "mean_token_accuracy": 0.861985981464386, + "num_tokens": 7127012.0, + "step": 2215 + }, + { + "epoch": 1.1335038363171355, + "grad_norm": 10.768899917602539, + "learning_rate": 1e-06, + "loss": 0.2135, + "mean_token_accuracy": 0.9344347715377808, + "num_tokens": 7131137.0, + "step": 2216 + }, + { + "epoch": 1.1340153452685422, + "grad_norm": 7.441157817840576, + "learning_rate": 1e-06, + "loss": 0.191, + "mean_token_accuracy": 0.9127061367034912, + "num_tokens": 7134448.0, + "step": 2217 + }, + { + "epoch": 1.134526854219949, + "grad_norm": 7.743613243103027, + "learning_rate": 1e-06, + "loss": 0.1744, + "mean_token_accuracy": 0.9251848459243774, + "num_tokens": 7138300.0, + "step": 2218 + }, + { + "epoch": 1.1350383631713554, + "grad_norm": 9.557419776916504, + "learning_rate": 1e-06, + "loss": 0.209, + "mean_token_accuracy": 0.9137699604034424, + "num_tokens": 7140953.0, + "step": 2219 + }, + { + "epoch": 1.1355498721227621, + "grad_norm": 10.573720932006836, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.889086902141571, + "num_tokens": 7143484.0, + "step": 2220 + }, + { + "epoch": 1.1360613810741689, + "grad_norm": 11.527999877929688, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8696370124816895, + "num_tokens": 7146293.0, + "step": 2221 + }, + { + "epoch": 1.1365728900255754, + "grad_norm": 11.142582893371582, + "learning_rate": 1e-06, + "loss": 0.282, + "mean_token_accuracy": 0.875360906124115, + "num_tokens": 7149510.0, + "step": 2222 + }, + { + "epoch": 1.137084398976982, + "grad_norm": 7.383630752563477, + "learning_rate": 1e-06, + "loss": 0.1952, + "mean_token_accuracy": 0.9095562696456909, + "num_tokens": 7152716.0, + "step": 2223 + }, + { + "epoch": 1.1375959079283888, + "grad_norm": 7.881943225860596, + "learning_rate": 1e-06, + "loss": 0.2249, + "mean_token_accuracy": 0.9053328633308411, + "num_tokens": 7156087.0, + "step": 2224 + }, + { + "epoch": 1.1381074168797953, + "grad_norm": 15.485722541809082, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.897774875164032, + "num_tokens": 7159447.0, + "step": 2225 + }, + { + "epoch": 1.138618925831202, + "grad_norm": 9.101622581481934, + "learning_rate": 1e-06, + "loss": 0.2693, + "mean_token_accuracy": 0.8605761528015137, + "num_tokens": 7162093.0, + "step": 2226 + }, + { + "epoch": 1.1391304347826088, + "grad_norm": 8.914535522460938, + "learning_rate": 1e-06, + "loss": 0.2707, + "mean_token_accuracy": 0.8860595226287842, + "num_tokens": 7164839.0, + "step": 2227 + }, + { + "epoch": 1.1396419437340153, + "grad_norm": 9.755510330200195, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.9002564549446106, + "num_tokens": 7168261.0, + "step": 2228 + }, + { + "epoch": 1.140153452685422, + "grad_norm": 11.964424133300781, + "learning_rate": 1e-06, + "loss": 0.3471, + "mean_token_accuracy": 0.8734836578369141, + "num_tokens": 7170991.0, + "step": 2229 + }, + { + "epoch": 1.1406649616368287, + "grad_norm": 11.50944995880127, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8896127939224243, + "num_tokens": 7174286.0, + "step": 2230 + }, + { + "epoch": 1.1411764705882352, + "grad_norm": 8.13188648223877, + "learning_rate": 1e-06, + "loss": 0.2607, + "mean_token_accuracy": 0.8819520473480225, + "num_tokens": 7176808.0, + "step": 2231 + }, + { + "epoch": 1.141687979539642, + "grad_norm": 10.460338592529297, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.8734570145606995, + "num_tokens": 7180486.0, + "step": 2232 + }, + { + "epoch": 1.1421994884910487, + "grad_norm": 10.273818016052246, + "learning_rate": 1e-06, + "loss": 0.249, + "mean_token_accuracy": 0.874618649482727, + "num_tokens": 7183594.0, + "step": 2233 + }, + { + "epoch": 1.1427109974424552, + "grad_norm": 9.358654975891113, + "learning_rate": 1e-06, + "loss": 0.2018, + "mean_token_accuracy": 0.9253371953964233, + "num_tokens": 7186634.0, + "step": 2234 + }, + { + "epoch": 1.143222506393862, + "grad_norm": 5.083751201629639, + "learning_rate": 1e-06, + "loss": 0.1223, + "mean_token_accuracy": 0.9274636507034302, + "num_tokens": 7190335.0, + "step": 2235 + }, + { + "epoch": 1.1437340153452684, + "grad_norm": 12.024280548095703, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.8818497061729431, + "num_tokens": 7193758.0, + "step": 2236 + }, + { + "epoch": 1.1442455242966751, + "grad_norm": 9.915191650390625, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8672480583190918, + "num_tokens": 7196737.0, + "step": 2237 + }, + { + "epoch": 1.1447570332480819, + "grad_norm": 8.451594352722168, + "learning_rate": 1e-06, + "loss": 0.2437, + "mean_token_accuracy": 0.9062439203262329, + "num_tokens": 7199375.0, + "step": 2238 + }, + { + "epoch": 1.1452685421994886, + "grad_norm": 10.076199531555176, + "learning_rate": 1e-06, + "loss": 0.206, + "mean_token_accuracy": 0.9206429123878479, + "num_tokens": 7202145.0, + "step": 2239 + }, + { + "epoch": 1.145780051150895, + "grad_norm": 9.698476791381836, + "learning_rate": 1e-06, + "loss": 0.1965, + "mean_token_accuracy": 0.9210574626922607, + "num_tokens": 7205797.0, + "step": 2240 + }, + { + "epoch": 1.1462915601023018, + "grad_norm": 14.476765632629395, + "learning_rate": 1e-06, + "loss": 0.2518, + "mean_token_accuracy": 0.9162307381629944, + "num_tokens": 7208536.0, + "step": 2241 + }, + { + "epoch": 1.1468030690537083, + "grad_norm": 9.40636157989502, + "learning_rate": 1e-06, + "loss": 0.2222, + "mean_token_accuracy": 0.9087151885032654, + "num_tokens": 7211640.0, + "step": 2242 + }, + { + "epoch": 1.147314578005115, + "grad_norm": 11.81423282623291, + "learning_rate": 1e-06, + "loss": 0.2486, + "mean_token_accuracy": 0.9009236097335815, + "num_tokens": 7214532.0, + "step": 2243 + }, + { + "epoch": 1.1478260869565218, + "grad_norm": 10.715665817260742, + "learning_rate": 1e-06, + "loss": 0.2267, + "mean_token_accuracy": 0.9254713654518127, + "num_tokens": 7218072.0, + "step": 2244 + }, + { + "epoch": 1.1483375959079285, + "grad_norm": 6.6856255531311035, + "learning_rate": 1e-06, + "loss": 0.1676, + "mean_token_accuracy": 0.934845507144928, + "num_tokens": 7222325.0, + "step": 2245 + }, + { + "epoch": 1.148849104859335, + "grad_norm": 10.246898651123047, + "learning_rate": 1e-06, + "loss": 0.2349, + "mean_token_accuracy": 0.9174743890762329, + "num_tokens": 7225574.0, + "step": 2246 + }, + { + "epoch": 1.1493606138107417, + "grad_norm": 8.606731414794922, + "learning_rate": 1e-06, + "loss": 0.2241, + "mean_token_accuracy": 0.9303894639015198, + "num_tokens": 7228320.0, + "step": 2247 + }, + { + "epoch": 1.1498721227621482, + "grad_norm": 9.883976936340332, + "learning_rate": 1e-06, + "loss": 0.2671, + "mean_token_accuracy": 0.9000351428985596, + "num_tokens": 7231947.0, + "step": 2248 + }, + { + "epoch": 1.150383631713555, + "grad_norm": 7.924631118774414, + "learning_rate": 1e-06, + "loss": 0.2046, + "mean_token_accuracy": 0.8979977369308472, + "num_tokens": 7235068.0, + "step": 2249 + }, + { + "epoch": 1.1508951406649617, + "grad_norm": 12.56478214263916, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.9083164930343628, + "num_tokens": 7238350.0, + "step": 2250 + }, + { + "epoch": 1.1514066496163684, + "grad_norm": 15.274283409118652, + "learning_rate": 1e-06, + "loss": 0.1977, + "mean_token_accuracy": 0.9100494980812073, + "num_tokens": 7241645.0, + "step": 2251 + }, + { + "epoch": 1.151918158567775, + "grad_norm": 8.614984512329102, + "learning_rate": 1e-06, + "loss": 0.198, + "mean_token_accuracy": 0.9222941398620605, + "num_tokens": 7244770.0, + "step": 2252 + }, + { + "epoch": 1.1524296675191816, + "grad_norm": 6.090347766876221, + "learning_rate": 1e-06, + "loss": 0.1424, + "mean_token_accuracy": 0.9287472367286682, + "num_tokens": 7248484.0, + "step": 2253 + }, + { + "epoch": 1.1529411764705881, + "grad_norm": 9.479400634765625, + "learning_rate": 1e-06, + "loss": 0.2542, + "mean_token_accuracy": 0.8905287384986877, + "num_tokens": 7250880.0, + "step": 2254 + }, + { + "epoch": 1.1534526854219949, + "grad_norm": 11.54339599609375, + "learning_rate": 1e-06, + "loss": 0.2663, + "mean_token_accuracy": 0.8780152201652527, + "num_tokens": 7254214.0, + "step": 2255 + }, + { + "epoch": 1.1539641943734016, + "grad_norm": 12.150045394897461, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8903452157974243, + "num_tokens": 7257229.0, + "step": 2256 + }, + { + "epoch": 1.154475703324808, + "grad_norm": 6.807754993438721, + "learning_rate": 1e-06, + "loss": 0.1571, + "mean_token_accuracy": 0.9372312426567078, + "num_tokens": 7260649.0, + "step": 2257 + }, + { + "epoch": 1.1549872122762148, + "grad_norm": 10.31013298034668, + "learning_rate": 1e-06, + "loss": 0.1969, + "mean_token_accuracy": 0.9259955883026123, + "num_tokens": 7264216.0, + "step": 2258 + }, + { + "epoch": 1.1554987212276215, + "grad_norm": 11.307927131652832, + "learning_rate": 1e-06, + "loss": 0.3421, + "mean_token_accuracy": 0.8720974922180176, + "num_tokens": 7267351.0, + "step": 2259 + }, + { + "epoch": 1.156010230179028, + "grad_norm": 10.879768371582031, + "learning_rate": 1e-06, + "loss": 0.3139, + "mean_token_accuracy": 0.8752263784408569, + "num_tokens": 7269793.0, + "step": 2260 + }, + { + "epoch": 1.1565217391304348, + "grad_norm": 10.00703239440918, + "learning_rate": 1e-06, + "loss": 0.2126, + "mean_token_accuracy": 0.9273735284805298, + "num_tokens": 7272943.0, + "step": 2261 + }, + { + "epoch": 1.1570332480818415, + "grad_norm": 9.97009563446045, + "learning_rate": 1e-06, + "loss": 0.2389, + "mean_token_accuracy": 0.9030206203460693, + "num_tokens": 7276242.0, + "step": 2262 + }, + { + "epoch": 1.157544757033248, + "grad_norm": 9.487983703613281, + "learning_rate": 1e-06, + "loss": 0.1938, + "mean_token_accuracy": 0.9281735420227051, + "num_tokens": 7279378.0, + "step": 2263 + }, + { + "epoch": 1.1580562659846547, + "grad_norm": 5.779391765594482, + "learning_rate": 1e-06, + "loss": 0.1219, + "mean_token_accuracy": 0.9459250569343567, + "num_tokens": 7283189.0, + "step": 2264 + }, + { + "epoch": 1.1585677749360614, + "grad_norm": 9.229057312011719, + "learning_rate": 1e-06, + "loss": 0.1989, + "mean_token_accuracy": 0.9167025089263916, + "num_tokens": 7287018.0, + "step": 2265 + }, + { + "epoch": 1.159079283887468, + "grad_norm": 9.422408103942871, + "learning_rate": 1e-06, + "loss": 0.246, + "mean_token_accuracy": 0.9026352167129517, + "num_tokens": 7289683.0, + "step": 2266 + }, + { + "epoch": 1.1595907928388747, + "grad_norm": 8.128993034362793, + "learning_rate": 1e-06, + "loss": 0.183, + "mean_token_accuracy": 0.9215973019599915, + "num_tokens": 7293303.0, + "step": 2267 + }, + { + "epoch": 1.1601023017902814, + "grad_norm": 6.570508003234863, + "learning_rate": 1e-06, + "loss": 0.1277, + "mean_token_accuracy": 0.9451005458831787, + "num_tokens": 7296667.0, + "step": 2268 + }, + { + "epoch": 1.160613810741688, + "grad_norm": 9.003545761108398, + "learning_rate": 1e-06, + "loss": 0.1933, + "mean_token_accuracy": 0.9150851964950562, + "num_tokens": 7299518.0, + "step": 2269 + }, + { + "epoch": 1.1611253196930946, + "grad_norm": 9.602745056152344, + "learning_rate": 1e-06, + "loss": 0.1948, + "mean_token_accuracy": 0.934184730052948, + "num_tokens": 7303529.0, + "step": 2270 + }, + { + "epoch": 1.1616368286445014, + "grad_norm": 7.843327522277832, + "learning_rate": 1e-06, + "loss": 0.1951, + "mean_token_accuracy": 0.8984816074371338, + "num_tokens": 7306992.0, + "step": 2271 + }, + { + "epoch": 1.1621483375959079, + "grad_norm": 12.803255081176758, + "learning_rate": 1e-06, + "loss": 0.3384, + "mean_token_accuracy": 0.8646969795227051, + "num_tokens": 7309284.0, + "step": 2272 + }, + { + "epoch": 1.1626598465473146, + "grad_norm": 8.032770156860352, + "learning_rate": 1e-06, + "loss": 0.1673, + "mean_token_accuracy": 0.934913158416748, + "num_tokens": 7312763.0, + "step": 2273 + }, + { + "epoch": 1.1631713554987213, + "grad_norm": 10.543634414672852, + "learning_rate": 1e-06, + "loss": 0.2755, + "mean_token_accuracy": 0.8900793790817261, + "num_tokens": 7315799.0, + "step": 2274 + }, + { + "epoch": 1.1636828644501278, + "grad_norm": 15.229647636413574, + "learning_rate": 1e-06, + "loss": 0.3121, + "mean_token_accuracy": 0.8773176670074463, + "num_tokens": 7318677.0, + "step": 2275 + }, + { + "epoch": 1.1641943734015345, + "grad_norm": 7.012701511383057, + "learning_rate": 1e-06, + "loss": 0.1684, + "mean_token_accuracy": 0.92997807264328, + "num_tokens": 7322410.0, + "step": 2276 + }, + { + "epoch": 1.1647058823529413, + "grad_norm": 10.997614860534668, + "learning_rate": 1e-06, + "loss": 0.2412, + "mean_token_accuracy": 0.9027017951011658, + "num_tokens": 7325664.0, + "step": 2277 + }, + { + "epoch": 1.1652173913043478, + "grad_norm": 11.710683822631836, + "learning_rate": 1e-06, + "loss": 0.2857, + "mean_token_accuracy": 0.8967065811157227, + "num_tokens": 7328804.0, + "step": 2278 + }, + { + "epoch": 1.1657289002557545, + "grad_norm": 10.967050552368164, + "learning_rate": 1e-06, + "loss": 0.2825, + "mean_token_accuracy": 0.8942763805389404, + "num_tokens": 7331942.0, + "step": 2279 + }, + { + "epoch": 1.1662404092071612, + "grad_norm": 10.43457317352295, + "learning_rate": 1e-06, + "loss": 0.2593, + "mean_token_accuracy": 0.9112716913223267, + "num_tokens": 7334785.0, + "step": 2280 + }, + { + "epoch": 1.1667519181585677, + "grad_norm": 14.696212768554688, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.8929820656776428, + "num_tokens": 7337509.0, + "step": 2281 + }, + { + "epoch": 1.1672634271099744, + "grad_norm": 10.002018928527832, + "learning_rate": 1e-06, + "loss": 0.249, + "mean_token_accuracy": 0.9145799875259399, + "num_tokens": 7340669.0, + "step": 2282 + }, + { + "epoch": 1.1677749360613812, + "grad_norm": 10.641711235046387, + "learning_rate": 1e-06, + "loss": 0.2844, + "mean_token_accuracy": 0.886335551738739, + "num_tokens": 7343984.0, + "step": 2283 + }, + { + "epoch": 1.1682864450127877, + "grad_norm": 9.825963973999023, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8789708614349365, + "num_tokens": 7346964.0, + "step": 2284 + }, + { + "epoch": 1.1687979539641944, + "grad_norm": 12.641743659973145, + "learning_rate": 1e-06, + "loss": 0.2171, + "mean_token_accuracy": 0.9150124192237854, + "num_tokens": 7350130.0, + "step": 2285 + }, + { + "epoch": 1.1693094629156011, + "grad_norm": 14.324407577514648, + "learning_rate": 1e-06, + "loss": 0.3126, + "mean_token_accuracy": 0.8526054620742798, + "num_tokens": 7353395.0, + "step": 2286 + }, + { + "epoch": 1.1698209718670076, + "grad_norm": 10.402459144592285, + "learning_rate": 1e-06, + "loss": 0.3204, + "mean_token_accuracy": 0.8664726614952087, + "num_tokens": 7356064.0, + "step": 2287 + }, + { + "epoch": 1.1703324808184143, + "grad_norm": 12.148517608642578, + "learning_rate": 1e-06, + "loss": 0.2626, + "mean_token_accuracy": 0.8834758996963501, + "num_tokens": 7359229.0, + "step": 2288 + }, + { + "epoch": 1.1708439897698208, + "grad_norm": 8.175779342651367, + "learning_rate": 1e-06, + "loss": 0.2323, + "mean_token_accuracy": 0.9148679971694946, + "num_tokens": 7362147.0, + "step": 2289 + }, + { + "epoch": 1.1713554987212276, + "grad_norm": 10.948620796203613, + "learning_rate": 1e-06, + "loss": 0.264, + "mean_token_accuracy": 0.894093930721283, + "num_tokens": 7364908.0, + "step": 2290 + }, + { + "epoch": 1.1718670076726343, + "grad_norm": 10.299412727355957, + "learning_rate": 1e-06, + "loss": 0.2817, + "mean_token_accuracy": 0.8876246213912964, + "num_tokens": 7367305.0, + "step": 2291 + }, + { + "epoch": 1.172378516624041, + "grad_norm": 6.201018810272217, + "learning_rate": 1e-06, + "loss": 0.1842, + "mean_token_accuracy": 0.9289469122886658, + "num_tokens": 7371091.0, + "step": 2292 + }, + { + "epoch": 1.1728900255754475, + "grad_norm": 10.40210247039795, + "learning_rate": 1e-06, + "loss": 0.2094, + "mean_token_accuracy": 0.9063732028007507, + "num_tokens": 7374155.0, + "step": 2293 + }, + { + "epoch": 1.1734015345268543, + "grad_norm": 10.68731689453125, + "learning_rate": 1e-06, + "loss": 0.2389, + "mean_token_accuracy": 0.9016037583351135, + "num_tokens": 7377739.0, + "step": 2294 + }, + { + "epoch": 1.1739130434782608, + "grad_norm": 13.778437614440918, + "learning_rate": 1e-06, + "loss": 0.229, + "mean_token_accuracy": 0.9042640924453735, + "num_tokens": 7380349.0, + "step": 2295 + }, + { + "epoch": 1.1744245524296675, + "grad_norm": 9.428450584411621, + "learning_rate": 1e-06, + "loss": 0.1816, + "mean_token_accuracy": 0.9271889925003052, + "num_tokens": 7383749.0, + "step": 2296 + }, + { + "epoch": 1.1749360613810742, + "grad_norm": 14.102363586425781, + "learning_rate": 1e-06, + "loss": 0.3269, + "mean_token_accuracy": 0.9188183546066284, + "num_tokens": 7387294.0, + "step": 2297 + }, + { + "epoch": 1.175447570332481, + "grad_norm": 10.868465423583984, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.8654447793960571, + "num_tokens": 7390393.0, + "step": 2298 + }, + { + "epoch": 1.1759590792838874, + "grad_norm": 10.227408409118652, + "learning_rate": 1e-06, + "loss": 0.2647, + "mean_token_accuracy": 0.8934793472290039, + "num_tokens": 7392896.0, + "step": 2299 + }, + { + "epoch": 1.1764705882352942, + "grad_norm": 8.750523567199707, + "learning_rate": 1e-06, + "loss": 0.1681, + "mean_token_accuracy": 0.8995080590248108, + "num_tokens": 7395912.0, + "step": 2300 + }, + { + "epoch": 1.1769820971867007, + "grad_norm": 10.249576568603516, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.899311900138855, + "num_tokens": 7398736.0, + "step": 2301 + }, + { + "epoch": 1.1774936061381074, + "grad_norm": 10.514887809753418, + "learning_rate": 1e-06, + "loss": 0.2331, + "mean_token_accuracy": 0.8952968716621399, + "num_tokens": 7401457.0, + "step": 2302 + }, + { + "epoch": 1.178005115089514, + "grad_norm": 8.722318649291992, + "learning_rate": 1e-06, + "loss": 0.2207, + "mean_token_accuracy": 0.9097669124603271, + "num_tokens": 7404920.0, + "step": 2303 + }, + { + "epoch": 1.1785166240409208, + "grad_norm": 8.603432655334473, + "learning_rate": 1e-06, + "loss": 0.1879, + "mean_token_accuracy": 0.8921282291412354, + "num_tokens": 7408609.0, + "step": 2304 + }, + { + "epoch": 1.1790281329923273, + "grad_norm": 5.783303260803223, + "learning_rate": 1e-06, + "loss": 0.1169, + "mean_token_accuracy": 0.9247517585754395, + "num_tokens": 7411995.0, + "step": 2305 + }, + { + "epoch": 1.179539641943734, + "grad_norm": 10.577475547790527, + "learning_rate": 1e-06, + "loss": 0.2475, + "mean_token_accuracy": 0.903266429901123, + "num_tokens": 7415186.0, + "step": 2306 + }, + { + "epoch": 1.1800511508951406, + "grad_norm": 6.866580486297607, + "learning_rate": 1e-06, + "loss": 0.1489, + "mean_token_accuracy": 0.9270710945129395, + "num_tokens": 7419840.0, + "step": 2307 + }, + { + "epoch": 1.1805626598465473, + "grad_norm": 7.82284688949585, + "learning_rate": 1e-06, + "loss": 0.1844, + "mean_token_accuracy": 0.9346861243247986, + "num_tokens": 7422459.0, + "step": 2308 + }, + { + "epoch": 1.181074168797954, + "grad_norm": 10.665806770324707, + "learning_rate": 1e-06, + "loss": 0.2282, + "mean_token_accuracy": 0.9020093083381653, + "num_tokens": 7425154.0, + "step": 2309 + }, + { + "epoch": 1.1815856777493605, + "grad_norm": 6.140814304351807, + "learning_rate": 1e-06, + "loss": 0.163, + "mean_token_accuracy": 0.9342641830444336, + "num_tokens": 7429114.0, + "step": 2310 + }, + { + "epoch": 1.1820971867007672, + "grad_norm": 10.108406066894531, + "learning_rate": 1e-06, + "loss": 0.2178, + "mean_token_accuracy": 0.8904531002044678, + "num_tokens": 7432984.0, + "step": 2311 + }, + { + "epoch": 1.182608695652174, + "grad_norm": 6.111667156219482, + "learning_rate": 1e-06, + "loss": 0.1633, + "mean_token_accuracy": 0.9319311380386353, + "num_tokens": 7436384.0, + "step": 2312 + }, + { + "epoch": 1.1831202046035805, + "grad_norm": 6.50772762298584, + "learning_rate": 1e-06, + "loss": 0.1438, + "mean_token_accuracy": 0.9395736455917358, + "num_tokens": 7440203.0, + "step": 2313 + }, + { + "epoch": 1.1836317135549872, + "grad_norm": 8.479541778564453, + "learning_rate": 1e-06, + "loss": 0.2239, + "mean_token_accuracy": 0.8935945630073547, + "num_tokens": 7443360.0, + "step": 2314 + }, + { + "epoch": 1.184143222506394, + "grad_norm": 10.597058296203613, + "learning_rate": 1e-06, + "loss": 0.2392, + "mean_token_accuracy": 0.891452968120575, + "num_tokens": 7446102.0, + "step": 2315 + }, + { + "epoch": 1.1846547314578004, + "grad_norm": 8.241751670837402, + "learning_rate": 1e-06, + "loss": 0.1728, + "mean_token_accuracy": 0.9173241257667542, + "num_tokens": 7448586.0, + "step": 2316 + }, + { + "epoch": 1.1851662404092072, + "grad_norm": 13.269973754882812, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8944816589355469, + "num_tokens": 7451549.0, + "step": 2317 + }, + { + "epoch": 1.1856777493606139, + "grad_norm": 9.878308296203613, + "learning_rate": 1e-06, + "loss": 0.2359, + "mean_token_accuracy": 0.9010183215141296, + "num_tokens": 7454129.0, + "step": 2318 + }, + { + "epoch": 1.1861892583120204, + "grad_norm": 11.356729507446289, + "learning_rate": 1e-06, + "loss": 0.2456, + "mean_token_accuracy": 0.9135680198669434, + "num_tokens": 7457139.0, + "step": 2319 + }, + { + "epoch": 1.186700767263427, + "grad_norm": 9.09647274017334, + "learning_rate": 1e-06, + "loss": 0.2247, + "mean_token_accuracy": 0.8752237558364868, + "num_tokens": 7460345.0, + "step": 2320 + }, + { + "epoch": 1.1872122762148338, + "grad_norm": 30.375797271728516, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8825308084487915, + "num_tokens": 7463270.0, + "step": 2321 + }, + { + "epoch": 1.1877237851662403, + "grad_norm": 11.469158172607422, + "learning_rate": 1e-06, + "loss": 0.2065, + "mean_token_accuracy": 0.8942556977272034, + "num_tokens": 7466653.0, + "step": 2322 + }, + { + "epoch": 1.188235294117647, + "grad_norm": 11.228367805480957, + "learning_rate": 1e-06, + "loss": 0.2761, + "mean_token_accuracy": 0.8916431665420532, + "num_tokens": 7469280.0, + "step": 2323 + }, + { + "epoch": 1.1887468030690538, + "grad_norm": 8.862345695495605, + "learning_rate": 1e-06, + "loss": 0.1934, + "mean_token_accuracy": 0.913243293762207, + "num_tokens": 7472927.0, + "step": 2324 + }, + { + "epoch": 1.1892583120204603, + "grad_norm": 10.44587230682373, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.9037233591079712, + "num_tokens": 7476077.0, + "step": 2325 + }, + { + "epoch": 1.189769820971867, + "grad_norm": 10.522994041442871, + "learning_rate": 1e-06, + "loss": 0.1959, + "mean_token_accuracy": 0.911578893661499, + "num_tokens": 7479465.0, + "step": 2326 + }, + { + "epoch": 1.1902813299232737, + "grad_norm": 7.147512435913086, + "learning_rate": 1e-06, + "loss": 0.1356, + "mean_token_accuracy": 0.9464237093925476, + "num_tokens": 7483657.0, + "step": 2327 + }, + { + "epoch": 1.1907928388746802, + "grad_norm": 7.53249454498291, + "learning_rate": 1e-06, + "loss": 0.1746, + "mean_token_accuracy": 0.8877224326133728, + "num_tokens": 7487123.0, + "step": 2328 + }, + { + "epoch": 1.191304347826087, + "grad_norm": 7.211447238922119, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.9051289558410645, + "num_tokens": 7489951.0, + "step": 2329 + }, + { + "epoch": 1.1918158567774937, + "grad_norm": 11.794792175292969, + "learning_rate": 1e-06, + "loss": 0.3153, + "mean_token_accuracy": 0.8858539462089539, + "num_tokens": 7493128.0, + "step": 2330 + }, + { + "epoch": 1.1923273657289002, + "grad_norm": 11.626081466674805, + "learning_rate": 1e-06, + "loss": 0.275, + "mean_token_accuracy": 0.8984275460243225, + "num_tokens": 7496249.0, + "step": 2331 + }, + { + "epoch": 1.192838874680307, + "grad_norm": 10.21019172668457, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8896880149841309, + "num_tokens": 7499182.0, + "step": 2332 + }, + { + "epoch": 1.1933503836317136, + "grad_norm": 7.799311637878418, + "learning_rate": 1e-06, + "loss": 0.1746, + "mean_token_accuracy": 0.9187701940536499, + "num_tokens": 7503352.0, + "step": 2333 + }, + { + "epoch": 1.1938618925831201, + "grad_norm": 9.464651107788086, + "learning_rate": 1e-06, + "loss": 0.2331, + "mean_token_accuracy": 0.9100872278213501, + "num_tokens": 7506370.0, + "step": 2334 + }, + { + "epoch": 1.1943734015345269, + "grad_norm": 8.752459526062012, + "learning_rate": 1e-06, + "loss": 0.2204, + "mean_token_accuracy": 0.9027407765388489, + "num_tokens": 7508880.0, + "step": 2335 + }, + { + "epoch": 1.1948849104859336, + "grad_norm": 8.235797882080078, + "learning_rate": 1e-06, + "loss": 0.2848, + "mean_token_accuracy": 0.9055810570716858, + "num_tokens": 7512455.0, + "step": 2336 + }, + { + "epoch": 1.19539641943734, + "grad_norm": 10.575154304504395, + "learning_rate": 1e-06, + "loss": 0.2626, + "mean_token_accuracy": 0.8969382047653198, + "num_tokens": 7516058.0, + "step": 2337 + }, + { + "epoch": 1.1959079283887468, + "grad_norm": 10.463808059692383, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.89722740650177, + "num_tokens": 7519053.0, + "step": 2338 + }, + { + "epoch": 1.1964194373401535, + "grad_norm": 10.708813667297363, + "learning_rate": 1e-06, + "loss": 0.2182, + "mean_token_accuracy": 0.9291764497756958, + "num_tokens": 7522671.0, + "step": 2339 + }, + { + "epoch": 1.19693094629156, + "grad_norm": 9.841679573059082, + "learning_rate": 1e-06, + "loss": 0.2239, + "mean_token_accuracy": 0.9135267734527588, + "num_tokens": 7525664.0, + "step": 2340 + }, + { + "epoch": 1.1974424552429668, + "grad_norm": 8.3696870803833, + "learning_rate": 1e-06, + "loss": 0.2225, + "mean_token_accuracy": 0.9103610515594482, + "num_tokens": 7529098.0, + "step": 2341 + }, + { + "epoch": 1.1979539641943733, + "grad_norm": 7.886106014251709, + "learning_rate": 1e-06, + "loss": 0.2128, + "mean_token_accuracy": 0.9166443347930908, + "num_tokens": 7532157.0, + "step": 2342 + }, + { + "epoch": 1.19846547314578, + "grad_norm": 20.893205642700195, + "learning_rate": 1e-06, + "loss": 0.2723, + "mean_token_accuracy": 0.905353844165802, + "num_tokens": 7534832.0, + "step": 2343 + }, + { + "epoch": 1.1989769820971867, + "grad_norm": 14.570858001708984, + "learning_rate": 1e-06, + "loss": 0.2626, + "mean_token_accuracy": 0.9044578671455383, + "num_tokens": 7537310.0, + "step": 2344 + }, + { + "epoch": 1.1994884910485935, + "grad_norm": 11.090411186218262, + "learning_rate": 1e-06, + "loss": 0.2379, + "mean_token_accuracy": 0.8947941064834595, + "num_tokens": 7540241.0, + "step": 2345 + }, + { + "epoch": 1.2, + "grad_norm": 9.674985885620117, + "learning_rate": 1e-06, + "loss": 0.1859, + "mean_token_accuracy": 0.9280813336372375, + "num_tokens": 7543509.0, + "step": 2346 + }, + { + "epoch": 1.2005115089514067, + "grad_norm": 9.151853561401367, + "learning_rate": 1e-06, + "loss": 0.199, + "mean_token_accuracy": 0.9286362528800964, + "num_tokens": 7547420.0, + "step": 2347 + }, + { + "epoch": 1.2010230179028132, + "grad_norm": 11.509537696838379, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8987661004066467, + "num_tokens": 7550494.0, + "step": 2348 + }, + { + "epoch": 1.20153452685422, + "grad_norm": 8.364404678344727, + "learning_rate": 1e-06, + "loss": 0.2182, + "mean_token_accuracy": 0.9215635061264038, + "num_tokens": 7552996.0, + "step": 2349 + }, + { + "epoch": 1.2020460358056266, + "grad_norm": 8.15385627746582, + "learning_rate": 1e-06, + "loss": 0.1722, + "mean_token_accuracy": 0.9122105836868286, + "num_tokens": 7556951.0, + "step": 2350 + }, + { + "epoch": 1.2025575447570334, + "grad_norm": 10.159811973571777, + "learning_rate": 1e-06, + "loss": 0.2822, + "mean_token_accuracy": 0.8819750547409058, + "num_tokens": 7560302.0, + "step": 2351 + }, + { + "epoch": 1.2030690537084399, + "grad_norm": 10.380499839782715, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8888737559318542, + "num_tokens": 7563169.0, + "step": 2352 + }, + { + "epoch": 1.2035805626598466, + "grad_norm": 6.694550514221191, + "learning_rate": 1e-06, + "loss": 0.1922, + "mean_token_accuracy": 0.9332709312438965, + "num_tokens": 7567452.0, + "step": 2353 + }, + { + "epoch": 1.204092071611253, + "grad_norm": 12.224334716796875, + "learning_rate": 1e-06, + "loss": 0.3305, + "mean_token_accuracy": 0.8942030072212219, + "num_tokens": 7570477.0, + "step": 2354 + }, + { + "epoch": 1.2046035805626598, + "grad_norm": 8.678828239440918, + "learning_rate": 1e-06, + "loss": 0.2063, + "mean_token_accuracy": 0.9323636293411255, + "num_tokens": 7573769.0, + "step": 2355 + }, + { + "epoch": 1.2051150895140665, + "grad_norm": 10.944123268127441, + "learning_rate": 1e-06, + "loss": 0.264, + "mean_token_accuracy": 0.9206808805465698, + "num_tokens": 7576251.0, + "step": 2356 + }, + { + "epoch": 1.2056265984654733, + "grad_norm": 10.228043556213379, + "learning_rate": 1e-06, + "loss": 0.273, + "mean_token_accuracy": 0.8776668310165405, + "num_tokens": 7579708.0, + "step": 2357 + }, + { + "epoch": 1.2061381074168798, + "grad_norm": 7.355031967163086, + "learning_rate": 1e-06, + "loss": 0.1731, + "mean_token_accuracy": 0.932651162147522, + "num_tokens": 7583490.0, + "step": 2358 + }, + { + "epoch": 1.2066496163682865, + "grad_norm": 13.602950096130371, + "learning_rate": 1e-06, + "loss": 0.2297, + "mean_token_accuracy": 0.9117079973220825, + "num_tokens": 7586294.0, + "step": 2359 + }, + { + "epoch": 1.207161125319693, + "grad_norm": 9.263099670410156, + "learning_rate": 1e-06, + "loss": 0.2678, + "mean_token_accuracy": 0.8856875896453857, + "num_tokens": 7588910.0, + "step": 2360 + }, + { + "epoch": 1.2076726342710997, + "grad_norm": 6.875961780548096, + "learning_rate": 1e-06, + "loss": 0.1876, + "mean_token_accuracy": 0.9150086641311646, + "num_tokens": 7592496.0, + "step": 2361 + }, + { + "epoch": 1.2081841432225064, + "grad_norm": 9.853730201721191, + "learning_rate": 1e-06, + "loss": 0.3085, + "mean_token_accuracy": 0.883721113204956, + "num_tokens": 7595764.0, + "step": 2362 + }, + { + "epoch": 1.208695652173913, + "grad_norm": 7.882992744445801, + "learning_rate": 1e-06, + "loss": 0.1181, + "mean_token_accuracy": 0.9371020793914795, + "num_tokens": 7599968.0, + "step": 2363 + }, + { + "epoch": 1.2092071611253197, + "grad_norm": 6.243054389953613, + "learning_rate": 1e-06, + "loss": 0.166, + "mean_token_accuracy": 0.9207566976547241, + "num_tokens": 7604443.0, + "step": 2364 + }, + { + "epoch": 1.2097186700767264, + "grad_norm": 10.288790702819824, + "learning_rate": 1e-06, + "loss": 0.2121, + "mean_token_accuracy": 0.9238802194595337, + "num_tokens": 7606886.0, + "step": 2365 + }, + { + "epoch": 1.210230179028133, + "grad_norm": 9.880804061889648, + "learning_rate": 1e-06, + "loss": 0.2829, + "mean_token_accuracy": 0.904741108417511, + "num_tokens": 7610182.0, + "step": 2366 + }, + { + "epoch": 1.2107416879795396, + "grad_norm": 7.746084690093994, + "learning_rate": 1e-06, + "loss": 0.2724, + "mean_token_accuracy": 0.8624510765075684, + "num_tokens": 7613590.0, + "step": 2367 + }, + { + "epoch": 1.2112531969309464, + "grad_norm": 8.165873527526855, + "learning_rate": 1e-06, + "loss": 0.2492, + "mean_token_accuracy": 0.8920614123344421, + "num_tokens": 7617422.0, + "step": 2368 + }, + { + "epoch": 1.2117647058823529, + "grad_norm": 5.209260940551758, + "learning_rate": 1e-06, + "loss": 0.1315, + "mean_token_accuracy": 0.9371291399002075, + "num_tokens": 7620346.0, + "step": 2369 + }, + { + "epoch": 1.2122762148337596, + "grad_norm": 10.824814796447754, + "learning_rate": 1e-06, + "loss": 0.3053, + "mean_token_accuracy": 0.8815687298774719, + "num_tokens": 7623132.0, + "step": 2370 + }, + { + "epoch": 1.2127877237851663, + "grad_norm": 14.180096626281738, + "learning_rate": 1e-06, + "loss": 0.2746, + "mean_token_accuracy": 0.8980230689048767, + "num_tokens": 7626459.0, + "step": 2371 + }, + { + "epoch": 1.2132992327365728, + "grad_norm": 10.357406616210938, + "learning_rate": 1e-06, + "loss": 0.2689, + "mean_token_accuracy": 0.877007246017456, + "num_tokens": 7629523.0, + "step": 2372 + }, + { + "epoch": 1.2138107416879795, + "grad_norm": 8.409415245056152, + "learning_rate": 1e-06, + "loss": 0.2077, + "mean_token_accuracy": 0.9222427010536194, + "num_tokens": 7633148.0, + "step": 2373 + }, + { + "epoch": 1.2143222506393863, + "grad_norm": 7.615177631378174, + "learning_rate": 1e-06, + "loss": 0.1754, + "mean_token_accuracy": 0.9389593601226807, + "num_tokens": 7636144.0, + "step": 2374 + }, + { + "epoch": 1.2148337595907928, + "grad_norm": 4.479589939117432, + "learning_rate": 1e-06, + "loss": 0.1226, + "mean_token_accuracy": 0.8901768326759338, + "num_tokens": 7639918.0, + "step": 2375 + }, + { + "epoch": 1.2153452685421995, + "grad_norm": 10.817500114440918, + "learning_rate": 1e-06, + "loss": 0.2722, + "mean_token_accuracy": 0.8771656155586243, + "num_tokens": 7643145.0, + "step": 2376 + }, + { + "epoch": 1.2158567774936062, + "grad_norm": 15.576383590698242, + "learning_rate": 1e-06, + "loss": 0.2447, + "mean_token_accuracy": 0.8947656154632568, + "num_tokens": 7645979.0, + "step": 2377 + }, + { + "epoch": 1.2163682864450127, + "grad_norm": 11.267574310302734, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.8646629452705383, + "num_tokens": 7648479.0, + "step": 2378 + }, + { + "epoch": 1.2168797953964194, + "grad_norm": 12.813312530517578, + "learning_rate": 1e-06, + "loss": 0.2395, + "mean_token_accuracy": 0.927165150642395, + "num_tokens": 7651698.0, + "step": 2379 + }, + { + "epoch": 1.2173913043478262, + "grad_norm": 10.61540412902832, + "learning_rate": 1e-06, + "loss": 0.247, + "mean_token_accuracy": 0.8879525661468506, + "num_tokens": 7654758.0, + "step": 2380 + }, + { + "epoch": 1.2179028132992327, + "grad_norm": 8.349333763122559, + "learning_rate": 1e-06, + "loss": 0.1876, + "mean_token_accuracy": 0.9269155263900757, + "num_tokens": 7658745.0, + "step": 2381 + }, + { + "epoch": 1.2184143222506394, + "grad_norm": 9.90965747833252, + "learning_rate": 1e-06, + "loss": 0.209, + "mean_token_accuracy": 0.9161934852600098, + "num_tokens": 7661068.0, + "step": 2382 + }, + { + "epoch": 1.2189258312020461, + "grad_norm": 10.91044807434082, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8658170700073242, + "num_tokens": 7663479.0, + "step": 2383 + }, + { + "epoch": 1.2194373401534526, + "grad_norm": 13.430299758911133, + "learning_rate": 1e-06, + "loss": 0.3238, + "mean_token_accuracy": 0.8674119710922241, + "num_tokens": 7665780.0, + "step": 2384 + }, + { + "epoch": 1.2199488491048593, + "grad_norm": 13.390008926391602, + "learning_rate": 1e-06, + "loss": 0.2119, + "mean_token_accuracy": 0.9313534498214722, + "num_tokens": 7668433.0, + "step": 2385 + }, + { + "epoch": 1.220460358056266, + "grad_norm": 10.68946647644043, + "learning_rate": 1e-06, + "loss": 0.2639, + "mean_token_accuracy": 0.8954988718032837, + "num_tokens": 7671072.0, + "step": 2386 + }, + { + "epoch": 1.2209718670076726, + "grad_norm": 11.468865394592285, + "learning_rate": 1e-06, + "loss": 0.2469, + "mean_token_accuracy": 0.9113237857818604, + "num_tokens": 7674417.0, + "step": 2387 + }, + { + "epoch": 1.2214833759590793, + "grad_norm": 10.143746376037598, + "learning_rate": 1e-06, + "loss": 0.2319, + "mean_token_accuracy": 0.9015309810638428, + "num_tokens": 7677440.0, + "step": 2388 + }, + { + "epoch": 1.221994884910486, + "grad_norm": 9.708734512329102, + "learning_rate": 1e-06, + "loss": 0.2402, + "mean_token_accuracy": 0.923632025718689, + "num_tokens": 7680422.0, + "step": 2389 + }, + { + "epoch": 1.2225063938618925, + "grad_norm": 11.135278701782227, + "learning_rate": 1e-06, + "loss": 0.2414, + "mean_token_accuracy": 0.8977274298667908, + "num_tokens": 7683496.0, + "step": 2390 + }, + { + "epoch": 1.2230179028132993, + "grad_norm": 13.346844673156738, + "learning_rate": 1e-06, + "loss": 0.3374, + "mean_token_accuracy": 0.8493053913116455, + "num_tokens": 7686713.0, + "step": 2391 + }, + { + "epoch": 1.223529411764706, + "grad_norm": 12.2607421875, + "learning_rate": 1e-06, + "loss": 0.3311, + "mean_token_accuracy": 0.8785520195960999, + "num_tokens": 7689367.0, + "step": 2392 + }, + { + "epoch": 1.2240409207161125, + "grad_norm": 6.516062259674072, + "learning_rate": 1e-06, + "loss": 0.1751, + "mean_token_accuracy": 0.8898175954818726, + "num_tokens": 7692521.0, + "step": 2393 + }, + { + "epoch": 1.2245524296675192, + "grad_norm": 10.362821578979492, + "learning_rate": 1e-06, + "loss": 0.247, + "mean_token_accuracy": 0.8887643814086914, + "num_tokens": 7695192.0, + "step": 2394 + }, + { + "epoch": 1.2250639386189257, + "grad_norm": 11.872576713562012, + "learning_rate": 1e-06, + "loss": 0.2683, + "mean_token_accuracy": 0.8895737528800964, + "num_tokens": 7698583.0, + "step": 2395 + }, + { + "epoch": 1.2255754475703324, + "grad_norm": 6.939410209655762, + "learning_rate": 1e-06, + "loss": 0.1915, + "mean_token_accuracy": 0.8958267569541931, + "num_tokens": 7702441.0, + "step": 2396 + }, + { + "epoch": 1.2260869565217392, + "grad_norm": 5.900575160980225, + "learning_rate": 1e-06, + "loss": 0.1553, + "mean_token_accuracy": 0.8970444798469543, + "num_tokens": 7706463.0, + "step": 2397 + }, + { + "epoch": 1.2265984654731459, + "grad_norm": 8.665765762329102, + "learning_rate": 1e-06, + "loss": 0.251, + "mean_token_accuracy": 0.8828216195106506, + "num_tokens": 7708948.0, + "step": 2398 + }, + { + "epoch": 1.2271099744245524, + "grad_norm": 5.7349853515625, + "learning_rate": 1e-06, + "loss": 0.1605, + "mean_token_accuracy": 0.9168531894683838, + "num_tokens": 7712543.0, + "step": 2399 + }, + { + "epoch": 1.227621483375959, + "grad_norm": 11.195204734802246, + "learning_rate": 1e-06, + "loss": 0.2389, + "mean_token_accuracy": 0.911304771900177, + "num_tokens": 7715257.0, + "step": 2400 + }, + { + "epoch": 1.2281329923273656, + "grad_norm": 10.962186813354492, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8832653164863586, + "num_tokens": 7718290.0, + "step": 2401 + }, + { + "epoch": 1.2286445012787723, + "grad_norm": 6.758986473083496, + "learning_rate": 1e-06, + "loss": 0.1427, + "mean_token_accuracy": 0.9292382001876831, + "num_tokens": 7721581.0, + "step": 2402 + }, + { + "epoch": 1.229156010230179, + "grad_norm": 9.670469284057617, + "learning_rate": 1e-06, + "loss": 0.2356, + "mean_token_accuracy": 0.9088340997695923, + "num_tokens": 7724470.0, + "step": 2403 + }, + { + "epoch": 1.2296675191815858, + "grad_norm": 9.62004566192627, + "learning_rate": 1e-06, + "loss": 0.2667, + "mean_token_accuracy": 0.8764674067497253, + "num_tokens": 7727456.0, + "step": 2404 + }, + { + "epoch": 1.2301790281329923, + "grad_norm": 12.087051391601562, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.8991427421569824, + "num_tokens": 7730952.0, + "step": 2405 + }, + { + "epoch": 1.230690537084399, + "grad_norm": 12.774473190307617, + "learning_rate": 1e-06, + "loss": 0.154, + "mean_token_accuracy": 0.9399009943008423, + "num_tokens": 7733983.0, + "step": 2406 + }, + { + "epoch": 1.2312020460358055, + "grad_norm": 7.885640621185303, + "learning_rate": 1e-06, + "loss": 0.1601, + "mean_token_accuracy": 0.9340598583221436, + "num_tokens": 7736857.0, + "step": 2407 + }, + { + "epoch": 1.2317135549872122, + "grad_norm": 10.139996528625488, + "learning_rate": 1e-06, + "loss": 0.3045, + "mean_token_accuracy": 0.8921017646789551, + "num_tokens": 7739791.0, + "step": 2408 + }, + { + "epoch": 1.232225063938619, + "grad_norm": 8.685599327087402, + "learning_rate": 1e-06, + "loss": 0.2732, + "mean_token_accuracy": 0.9010617733001709, + "num_tokens": 7742629.0, + "step": 2409 + }, + { + "epoch": 1.2327365728900257, + "grad_norm": 11.226583480834961, + "learning_rate": 1e-06, + "loss": 0.252, + "mean_token_accuracy": 0.9030073285102844, + "num_tokens": 7745329.0, + "step": 2410 + }, + { + "epoch": 1.2332480818414322, + "grad_norm": 15.51103687286377, + "learning_rate": 1e-06, + "loss": 0.2583, + "mean_token_accuracy": 0.8970853090286255, + "num_tokens": 7748880.0, + "step": 2411 + }, + { + "epoch": 1.233759590792839, + "grad_norm": 10.704931259155273, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8836535215377808, + "num_tokens": 7751703.0, + "step": 2412 + }, + { + "epoch": 1.2342710997442454, + "grad_norm": 9.782238006591797, + "learning_rate": 1e-06, + "loss": 0.1926, + "mean_token_accuracy": 0.9210023283958435, + "num_tokens": 7754573.0, + "step": 2413 + }, + { + "epoch": 1.2347826086956522, + "grad_norm": 11.140286445617676, + "learning_rate": 1e-06, + "loss": 0.2634, + "mean_token_accuracy": 0.9053783416748047, + "num_tokens": 7757113.0, + "step": 2414 + }, + { + "epoch": 1.2352941176470589, + "grad_norm": 9.31492805480957, + "learning_rate": 1e-06, + "loss": 0.227, + "mean_token_accuracy": 0.91717529296875, + "num_tokens": 7759828.0, + "step": 2415 + }, + { + "epoch": 1.2358056265984654, + "grad_norm": 7.406130313873291, + "learning_rate": 1e-06, + "loss": 0.187, + "mean_token_accuracy": 0.9105051159858704, + "num_tokens": 7764461.0, + "step": 2416 + }, + { + "epoch": 1.236317135549872, + "grad_norm": 11.253067970275879, + "learning_rate": 1e-06, + "loss": 0.3129, + "mean_token_accuracy": 0.8905633687973022, + "num_tokens": 7767854.0, + "step": 2417 + }, + { + "epoch": 1.2368286445012788, + "grad_norm": 10.684154510498047, + "learning_rate": 1e-06, + "loss": 0.2038, + "mean_token_accuracy": 0.9270439743995667, + "num_tokens": 7771250.0, + "step": 2418 + }, + { + "epoch": 1.2373401534526853, + "grad_norm": 11.038979530334473, + "learning_rate": 1e-06, + "loss": 0.2711, + "mean_token_accuracy": 0.9044871926307678, + "num_tokens": 7774978.0, + "step": 2419 + }, + { + "epoch": 1.237851662404092, + "grad_norm": 9.063117980957031, + "learning_rate": 1e-06, + "loss": 0.1804, + "mean_token_accuracy": 0.9032435417175293, + "num_tokens": 7778155.0, + "step": 2420 + }, + { + "epoch": 1.2383631713554988, + "grad_norm": 10.839597702026367, + "learning_rate": 1e-06, + "loss": 0.2989, + "mean_token_accuracy": 0.9094196557998657, + "num_tokens": 7780891.0, + "step": 2421 + }, + { + "epoch": 1.2388746803069053, + "grad_norm": 10.859537124633789, + "learning_rate": 1e-06, + "loss": 0.2185, + "mean_token_accuracy": 0.9106225967407227, + "num_tokens": 7784024.0, + "step": 2422 + }, + { + "epoch": 1.239386189258312, + "grad_norm": 8.96334171295166, + "learning_rate": 1e-06, + "loss": 0.2505, + "mean_token_accuracy": 0.8897750973701477, + "num_tokens": 7786622.0, + "step": 2423 + }, + { + "epoch": 1.2398976982097187, + "grad_norm": 8.7139253616333, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.8938214182853699, + "num_tokens": 7789781.0, + "step": 2424 + }, + { + "epoch": 1.2404092071611252, + "grad_norm": 6.976244926452637, + "learning_rate": 1e-06, + "loss": 0.1577, + "mean_token_accuracy": 0.9415236711502075, + "num_tokens": 7793278.0, + "step": 2425 + }, + { + "epoch": 1.240920716112532, + "grad_norm": 13.17209243774414, + "learning_rate": 1e-06, + "loss": 0.3065, + "mean_token_accuracy": 0.8821002244949341, + "num_tokens": 7796325.0, + "step": 2426 + }, + { + "epoch": 1.2414322250639387, + "grad_norm": 9.579154014587402, + "learning_rate": 1e-06, + "loss": 0.2263, + "mean_token_accuracy": 0.9115524888038635, + "num_tokens": 7799171.0, + "step": 2427 + }, + { + "epoch": 1.2419437340153452, + "grad_norm": 10.722174644470215, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8907182216644287, + "num_tokens": 7802170.0, + "step": 2428 + }, + { + "epoch": 1.242455242966752, + "grad_norm": 12.015296936035156, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.9015496969223022, + "num_tokens": 7804672.0, + "step": 2429 + }, + { + "epoch": 1.2429667519181586, + "grad_norm": 9.13812255859375, + "learning_rate": 1e-06, + "loss": 0.1952, + "mean_token_accuracy": 0.9192427396774292, + "num_tokens": 7808042.0, + "step": 2430 + }, + { + "epoch": 1.2434782608695651, + "grad_norm": 9.92416000366211, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.9047596454620361, + "num_tokens": 7811325.0, + "step": 2431 + }, + { + "epoch": 1.2439897698209719, + "grad_norm": 9.291655540466309, + "learning_rate": 1e-06, + "loss": 0.1611, + "mean_token_accuracy": 0.937720537185669, + "num_tokens": 7814598.0, + "step": 2432 + }, + { + "epoch": 1.2445012787723786, + "grad_norm": 6.031901836395264, + "learning_rate": 1e-06, + "loss": 0.1802, + "mean_token_accuracy": 0.9357547760009766, + "num_tokens": 7817616.0, + "step": 2433 + }, + { + "epoch": 1.245012787723785, + "grad_norm": 9.16755485534668, + "learning_rate": 1e-06, + "loss": 0.184, + "mean_token_accuracy": 0.9121342897415161, + "num_tokens": 7822418.0, + "step": 2434 + }, + { + "epoch": 1.2455242966751918, + "grad_norm": 10.626504898071289, + "learning_rate": 1e-06, + "loss": 0.2594, + "mean_token_accuracy": 0.8939999341964722, + "num_tokens": 7825561.0, + "step": 2435 + }, + { + "epoch": 1.2460358056265985, + "grad_norm": 9.615360260009766, + "learning_rate": 1e-06, + "loss": 0.2243, + "mean_token_accuracy": 0.9128599762916565, + "num_tokens": 7829234.0, + "step": 2436 + }, + { + "epoch": 1.246547314578005, + "grad_norm": 13.588842391967773, + "learning_rate": 1e-06, + "loss": 0.2923, + "mean_token_accuracy": 0.9010221362113953, + "num_tokens": 7831784.0, + "step": 2437 + }, + { + "epoch": 1.2470588235294118, + "grad_norm": 10.215218544006348, + "learning_rate": 1e-06, + "loss": 0.3013, + "mean_token_accuracy": 0.8887909650802612, + "num_tokens": 7835646.0, + "step": 2438 + }, + { + "epoch": 1.2475703324808185, + "grad_norm": 13.627972602844238, + "learning_rate": 1e-06, + "loss": 0.3314, + "mean_token_accuracy": 0.8670408129692078, + "num_tokens": 7838344.0, + "step": 2439 + }, + { + "epoch": 1.248081841432225, + "grad_norm": 9.2189359664917, + "learning_rate": 1e-06, + "loss": 0.2555, + "mean_token_accuracy": 0.8858668804168701, + "num_tokens": 7841764.0, + "step": 2440 + }, + { + "epoch": 1.2485933503836317, + "grad_norm": 12.51242733001709, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.8901302814483643, + "num_tokens": 7844850.0, + "step": 2441 + }, + { + "epoch": 1.2491048593350385, + "grad_norm": 10.402539253234863, + "learning_rate": 1e-06, + "loss": 0.2566, + "mean_token_accuracy": 0.8691179156303406, + "num_tokens": 7848267.0, + "step": 2442 + }, + { + "epoch": 1.249616368286445, + "grad_norm": 10.665450096130371, + "learning_rate": 1e-06, + "loss": 0.2515, + "mean_token_accuracy": 0.8983856439590454, + "num_tokens": 7850849.0, + "step": 2443 + }, + { + "epoch": 1.2501278772378517, + "grad_norm": 9.927290916442871, + "learning_rate": 1e-06, + "loss": 0.2146, + "mean_token_accuracy": 0.9139714241027832, + "num_tokens": 7853613.0, + "step": 2444 + }, + { + "epoch": 1.2506393861892584, + "grad_norm": 14.770565032958984, + "learning_rate": 1e-06, + "loss": 0.3037, + "mean_token_accuracy": 0.8879669308662415, + "num_tokens": 7856883.0, + "step": 2445 + }, + { + "epoch": 1.251150895140665, + "grad_norm": 9.583887100219727, + "learning_rate": 1e-06, + "loss": 0.2309, + "mean_token_accuracy": 0.9080090522766113, + "num_tokens": 7859486.0, + "step": 2446 + }, + { + "epoch": 1.2516624040920716, + "grad_norm": 8.064517974853516, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9096459150314331, + "num_tokens": 7863474.0, + "step": 2447 + }, + { + "epoch": 1.2521739130434781, + "grad_norm": 8.445737838745117, + "learning_rate": 1e-06, + "loss": 0.2282, + "mean_token_accuracy": 0.8989612460136414, + "num_tokens": 7866094.0, + "step": 2448 + }, + { + "epoch": 1.2526854219948849, + "grad_norm": 6.152837753295898, + "learning_rate": 1e-06, + "loss": 0.1549, + "mean_token_accuracy": 0.9052224159240723, + "num_tokens": 7871065.0, + "step": 2449 + }, + { + "epoch": 1.2531969309462916, + "grad_norm": 9.845537185668945, + "learning_rate": 1e-06, + "loss": 0.258, + "mean_token_accuracy": 0.9034159779548645, + "num_tokens": 7873929.0, + "step": 2450 + }, + { + "epoch": 1.2537084398976983, + "grad_norm": 10.05599308013916, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.8630140423774719, + "num_tokens": 7877679.0, + "step": 2451 + }, + { + "epoch": 1.2542199488491048, + "grad_norm": 9.97265625, + "learning_rate": 1e-06, + "loss": 0.2765, + "mean_token_accuracy": 0.8919030427932739, + "num_tokens": 7881004.0, + "step": 2452 + }, + { + "epoch": 1.2547314578005115, + "grad_norm": 7.676721572875977, + "learning_rate": 1e-06, + "loss": 0.1999, + "mean_token_accuracy": 0.9027707576751709, + "num_tokens": 7884880.0, + "step": 2453 + }, + { + "epoch": 1.255242966751918, + "grad_norm": 6.158743858337402, + "learning_rate": 1e-06, + "loss": 0.1593, + "mean_token_accuracy": 0.9349721670150757, + "num_tokens": 7888882.0, + "step": 2454 + }, + { + "epoch": 1.2557544757033248, + "grad_norm": 7.1248698234558105, + "learning_rate": 1e-06, + "loss": 0.1932, + "mean_token_accuracy": 0.9159289598464966, + "num_tokens": 7892285.0, + "step": 2455 + }, + { + "epoch": 1.2562659846547315, + "grad_norm": 9.060072898864746, + "learning_rate": 1e-06, + "loss": 0.1827, + "mean_token_accuracy": 0.9198334217071533, + "num_tokens": 7895280.0, + "step": 2456 + }, + { + "epoch": 1.2567774936061382, + "grad_norm": 7.235004425048828, + "learning_rate": 1e-06, + "loss": 0.2176, + "mean_token_accuracy": 0.9134711027145386, + "num_tokens": 7898463.0, + "step": 2457 + }, + { + "epoch": 1.2572890025575447, + "grad_norm": 6.1268181800842285, + "learning_rate": 1e-06, + "loss": 0.1767, + "mean_token_accuracy": 0.9137348532676697, + "num_tokens": 7902201.0, + "step": 2458 + }, + { + "epoch": 1.2578005115089514, + "grad_norm": 9.568827629089355, + "learning_rate": 1e-06, + "loss": 0.2404, + "mean_token_accuracy": 0.90834641456604, + "num_tokens": 7904773.0, + "step": 2459 + }, + { + "epoch": 1.258312020460358, + "grad_norm": 9.406166076660156, + "learning_rate": 1e-06, + "loss": 0.2532, + "mean_token_accuracy": 0.912297248840332, + "num_tokens": 7908387.0, + "step": 2460 + }, + { + "epoch": 1.2588235294117647, + "grad_norm": 9.27165412902832, + "learning_rate": 1e-06, + "loss": 0.2347, + "mean_token_accuracy": 0.9104753732681274, + "num_tokens": 7911274.0, + "step": 2461 + }, + { + "epoch": 1.2593350383631714, + "grad_norm": 7.360652923583984, + "learning_rate": 1e-06, + "loss": 0.2152, + "mean_token_accuracy": 0.906417727470398, + "num_tokens": 7913978.0, + "step": 2462 + }, + { + "epoch": 1.2598465473145781, + "grad_norm": 9.328874588012695, + "learning_rate": 1e-06, + "loss": 0.2429, + "mean_token_accuracy": 0.9044013023376465, + "num_tokens": 7917414.0, + "step": 2463 + }, + { + "epoch": 1.2603580562659846, + "grad_norm": 11.6745023727417, + "learning_rate": 1e-06, + "loss": 0.2913, + "mean_token_accuracy": 0.8855386972427368, + "num_tokens": 7920393.0, + "step": 2464 + }, + { + "epoch": 1.2608695652173914, + "grad_norm": 6.776053428649902, + "learning_rate": 1e-06, + "loss": 0.1673, + "mean_token_accuracy": 0.9140942692756653, + "num_tokens": 7923729.0, + "step": 2465 + }, + { + "epoch": 1.2613810741687979, + "grad_norm": 8.653167724609375, + "learning_rate": 1e-06, + "loss": 0.25, + "mean_token_accuracy": 0.9017761945724487, + "num_tokens": 7926678.0, + "step": 2466 + }, + { + "epoch": 1.2618925831202046, + "grad_norm": 10.874027252197266, + "learning_rate": 1e-06, + "loss": 0.2385, + "mean_token_accuracy": 0.8729538917541504, + "num_tokens": 7929846.0, + "step": 2467 + }, + { + "epoch": 1.2624040920716113, + "grad_norm": 8.149084091186523, + "learning_rate": 1e-06, + "loss": 0.1778, + "mean_token_accuracy": 0.921485424041748, + "num_tokens": 7934518.0, + "step": 2468 + }, + { + "epoch": 1.262915601023018, + "grad_norm": 7.578909397125244, + "learning_rate": 1e-06, + "loss": 0.1808, + "mean_token_accuracy": 0.9403023719787598, + "num_tokens": 7938677.0, + "step": 2469 + }, + { + "epoch": 1.2634271099744245, + "grad_norm": 9.995565414428711, + "learning_rate": 1e-06, + "loss": 0.2988, + "mean_token_accuracy": 0.8897225856781006, + "num_tokens": 7941426.0, + "step": 2470 + }, + { + "epoch": 1.2639386189258313, + "grad_norm": 12.826891899108887, + "learning_rate": 1e-06, + "loss": 0.2021, + "mean_token_accuracy": 0.9299858808517456, + "num_tokens": 7944281.0, + "step": 2471 + }, + { + "epoch": 1.2644501278772378, + "grad_norm": 10.351372718811035, + "learning_rate": 1e-06, + "loss": 0.2809, + "mean_token_accuracy": 0.8946292400360107, + "num_tokens": 7947754.0, + "step": 2472 + }, + { + "epoch": 1.2649616368286445, + "grad_norm": 9.837038040161133, + "learning_rate": 1e-06, + "loss": 0.2154, + "mean_token_accuracy": 0.916118323802948, + "num_tokens": 7951313.0, + "step": 2473 + }, + { + "epoch": 1.2654731457800512, + "grad_norm": 13.161717414855957, + "learning_rate": 1e-06, + "loss": 0.2199, + "mean_token_accuracy": 0.9085960388183594, + "num_tokens": 7954147.0, + "step": 2474 + }, + { + "epoch": 1.265984654731458, + "grad_norm": 11.735905647277832, + "learning_rate": 1e-06, + "loss": 0.2276, + "mean_token_accuracy": 0.8934736251831055, + "num_tokens": 7957501.0, + "step": 2475 + }, + { + "epoch": 1.2664961636828644, + "grad_norm": 10.730567932128906, + "learning_rate": 1e-06, + "loss": 0.2187, + "mean_token_accuracy": 0.9134904146194458, + "num_tokens": 7960700.0, + "step": 2476 + }, + { + "epoch": 1.2670076726342712, + "grad_norm": 8.768452644348145, + "learning_rate": 1e-06, + "loss": 0.2105, + "mean_token_accuracy": 0.9003841876983643, + "num_tokens": 7964242.0, + "step": 2477 + }, + { + "epoch": 1.2675191815856777, + "grad_norm": 7.38411808013916, + "learning_rate": 1e-06, + "loss": 0.2108, + "mean_token_accuracy": 0.9046842455863953, + "num_tokens": 7967458.0, + "step": 2478 + }, + { + "epoch": 1.2680306905370844, + "grad_norm": 9.631558418273926, + "learning_rate": 1e-06, + "loss": 0.2462, + "mean_token_accuracy": 0.8999325633049011, + "num_tokens": 7970955.0, + "step": 2479 + }, + { + "epoch": 1.2685421994884911, + "grad_norm": 8.701212882995605, + "learning_rate": 1e-06, + "loss": 0.194, + "mean_token_accuracy": 0.9023979902267456, + "num_tokens": 7975474.0, + "step": 2480 + }, + { + "epoch": 1.2690537084398976, + "grad_norm": 8.671873092651367, + "learning_rate": 1e-06, + "loss": 0.1802, + "mean_token_accuracy": 0.9418807625770569, + "num_tokens": 7978594.0, + "step": 2481 + }, + { + "epoch": 1.2695652173913043, + "grad_norm": 9.651960372924805, + "learning_rate": 1e-06, + "loss": 0.2137, + "mean_token_accuracy": 0.9083234071731567, + "num_tokens": 7982556.0, + "step": 2482 + }, + { + "epoch": 1.270076726342711, + "grad_norm": 9.47386360168457, + "learning_rate": 1e-06, + "loss": 0.1935, + "mean_token_accuracy": 0.9086822867393494, + "num_tokens": 7985975.0, + "step": 2483 + }, + { + "epoch": 1.2705882352941176, + "grad_norm": 14.074112892150879, + "learning_rate": 1e-06, + "loss": 0.2783, + "mean_token_accuracy": 0.8965648412704468, + "num_tokens": 7989162.0, + "step": 2484 + }, + { + "epoch": 1.2710997442455243, + "grad_norm": 12.042901992797852, + "learning_rate": 1e-06, + "loss": 0.3383, + "mean_token_accuracy": 0.8718320727348328, + "num_tokens": 7991955.0, + "step": 2485 + }, + { + "epoch": 1.271611253196931, + "grad_norm": 10.951269149780273, + "learning_rate": 1e-06, + "loss": 0.2904, + "mean_token_accuracy": 0.89207923412323, + "num_tokens": 7995160.0, + "step": 2486 + }, + { + "epoch": 1.2721227621483375, + "grad_norm": 10.521699905395508, + "learning_rate": 1e-06, + "loss": 0.2066, + "mean_token_accuracy": 0.8756783604621887, + "num_tokens": 7999280.0, + "step": 2487 + }, + { + "epoch": 1.2726342710997443, + "grad_norm": 11.791793823242188, + "learning_rate": 1e-06, + "loss": 0.3671, + "mean_token_accuracy": 0.8616831302642822, + "num_tokens": 8002144.0, + "step": 2488 + }, + { + "epoch": 1.273145780051151, + "grad_norm": 11.527076721191406, + "learning_rate": 1e-06, + "loss": 0.2977, + "mean_token_accuracy": 0.9034454822540283, + "num_tokens": 8005293.0, + "step": 2489 + }, + { + "epoch": 1.2736572890025575, + "grad_norm": 11.646149635314941, + "learning_rate": 1e-06, + "loss": 0.2599, + "mean_token_accuracy": 0.8971637487411499, + "num_tokens": 8007682.0, + "step": 2490 + }, + { + "epoch": 1.2741687979539642, + "grad_norm": 13.846641540527344, + "learning_rate": 1e-06, + "loss": 0.2741, + "mean_token_accuracy": 0.8976921439170837, + "num_tokens": 8010972.0, + "step": 2491 + }, + { + "epoch": 1.274680306905371, + "grad_norm": 9.049361228942871, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.8911446332931519, + "num_tokens": 8013545.0, + "step": 2492 + }, + { + "epoch": 1.2751918158567774, + "grad_norm": 7.810546398162842, + "learning_rate": 1e-06, + "loss": 0.1942, + "mean_token_accuracy": 0.9004279971122742, + "num_tokens": 8016871.0, + "step": 2493 + }, + { + "epoch": 1.2757033248081842, + "grad_norm": 11.685001373291016, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.9086683392524719, + "num_tokens": 8020172.0, + "step": 2494 + }, + { + "epoch": 1.2762148337595907, + "grad_norm": 8.487569808959961, + "learning_rate": 1e-06, + "loss": 0.2328, + "mean_token_accuracy": 0.9038727283477783, + "num_tokens": 8023455.0, + "step": 2495 + }, + { + "epoch": 1.2767263427109974, + "grad_norm": 8.746512413024902, + "learning_rate": 1e-06, + "loss": 0.2104, + "mean_token_accuracy": 0.9264565706253052, + "num_tokens": 8027299.0, + "step": 2496 + }, + { + "epoch": 1.2772378516624041, + "grad_norm": 8.809109687805176, + "learning_rate": 1e-06, + "loss": 0.2557, + "mean_token_accuracy": 0.9017099142074585, + "num_tokens": 8030348.0, + "step": 2497 + }, + { + "epoch": 1.2777493606138108, + "grad_norm": 10.882429122924805, + "learning_rate": 1e-06, + "loss": 0.2737, + "mean_token_accuracy": 0.8883862495422363, + "num_tokens": 8033133.0, + "step": 2498 + }, + { + "epoch": 1.2782608695652173, + "grad_norm": 10.634249687194824, + "learning_rate": 1e-06, + "loss": 0.2744, + "mean_token_accuracy": 0.9009536504745483, + "num_tokens": 8035831.0, + "step": 2499 + }, + { + "epoch": 1.278772378516624, + "grad_norm": 7.0483856201171875, + "learning_rate": 1e-06, + "loss": 0.1851, + "mean_token_accuracy": 0.9150853157043457, + "num_tokens": 8039522.0, + "step": 2500 + }, + { + "epoch": 1.2792838874680306, + "grad_norm": 10.873405456542969, + "learning_rate": 1e-06, + "loss": 0.273, + "mean_token_accuracy": 0.8967517614364624, + "num_tokens": 8041795.0, + "step": 2501 + }, + { + "epoch": 1.2797953964194373, + "grad_norm": 8.974239349365234, + "learning_rate": 1e-06, + "loss": 0.2862, + "mean_token_accuracy": 0.8815082311630249, + "num_tokens": 8044478.0, + "step": 2502 + }, + { + "epoch": 1.280306905370844, + "grad_norm": 12.24012565612793, + "learning_rate": 1e-06, + "loss": 0.27, + "mean_token_accuracy": 0.8853340744972229, + "num_tokens": 8047348.0, + "step": 2503 + }, + { + "epoch": 1.2808184143222507, + "grad_norm": 8.58014965057373, + "learning_rate": 1e-06, + "loss": 0.2684, + "mean_token_accuracy": 0.898651123046875, + "num_tokens": 8050364.0, + "step": 2504 + }, + { + "epoch": 1.2813299232736572, + "grad_norm": 5.082701683044434, + "learning_rate": 1e-06, + "loss": 0.1455, + "mean_token_accuracy": 0.8935897946357727, + "num_tokens": 8054251.0, + "step": 2505 + }, + { + "epoch": 1.281841432225064, + "grad_norm": 5.246040344238281, + "learning_rate": 1e-06, + "loss": 0.1772, + "mean_token_accuracy": 0.9234027862548828, + "num_tokens": 8058569.0, + "step": 2506 + }, + { + "epoch": 1.2823529411764705, + "grad_norm": 9.590325355529785, + "learning_rate": 1e-06, + "loss": 0.2346, + "mean_token_accuracy": 0.9018070697784424, + "num_tokens": 8061497.0, + "step": 2507 + }, + { + "epoch": 1.2828644501278772, + "grad_norm": 7.713584899902344, + "learning_rate": 1e-06, + "loss": 0.1354, + "mean_token_accuracy": 0.9187432527542114, + "num_tokens": 8064482.0, + "step": 2508 + }, + { + "epoch": 1.283375959079284, + "grad_norm": 12.179485321044922, + "learning_rate": 1e-06, + "loss": 0.2746, + "mean_token_accuracy": 0.8821240067481995, + "num_tokens": 8067835.0, + "step": 2509 + }, + { + "epoch": 1.2838874680306906, + "grad_norm": 7.991079330444336, + "learning_rate": 1e-06, + "loss": 0.1278, + "mean_token_accuracy": 0.9373586177825928, + "num_tokens": 8071241.0, + "step": 2510 + }, + { + "epoch": 1.2843989769820972, + "grad_norm": 15.628573417663574, + "learning_rate": 1e-06, + "loss": 0.2068, + "mean_token_accuracy": 0.9245911240577698, + "num_tokens": 8075597.0, + "step": 2511 + }, + { + "epoch": 1.2849104859335039, + "grad_norm": 8.44843578338623, + "learning_rate": 1e-06, + "loss": 0.1976, + "mean_token_accuracy": 0.8926385641098022, + "num_tokens": 8078462.0, + "step": 2512 + }, + { + "epoch": 1.2854219948849104, + "grad_norm": 8.790046691894531, + "learning_rate": 1e-06, + "loss": 0.2376, + "mean_token_accuracy": 0.892796516418457, + "num_tokens": 8082345.0, + "step": 2513 + }, + { + "epoch": 1.285933503836317, + "grad_norm": 8.79268741607666, + "learning_rate": 1e-06, + "loss": 0.174, + "mean_token_accuracy": 0.9409554600715637, + "num_tokens": 8085529.0, + "step": 2514 + }, + { + "epoch": 1.2864450127877238, + "grad_norm": 10.806342124938965, + "learning_rate": 1e-06, + "loss": 0.2445, + "mean_token_accuracy": 0.8861964344978333, + "num_tokens": 8089475.0, + "step": 2515 + }, + { + "epoch": 1.2869565217391306, + "grad_norm": 9.151985168457031, + "learning_rate": 1e-06, + "loss": 0.2474, + "mean_token_accuracy": 0.9035661220550537, + "num_tokens": 8092088.0, + "step": 2516 + }, + { + "epoch": 1.287468030690537, + "grad_norm": 4.340137481689453, + "learning_rate": 1e-06, + "loss": 0.1126, + "mean_token_accuracy": 0.9171856641769409, + "num_tokens": 8095681.0, + "step": 2517 + }, + { + "epoch": 1.2879795396419438, + "grad_norm": 11.158088684082031, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.9047209620475769, + "num_tokens": 8099003.0, + "step": 2518 + }, + { + "epoch": 1.2884910485933503, + "grad_norm": 9.935681343078613, + "learning_rate": 1e-06, + "loss": 0.2104, + "mean_token_accuracy": 0.9191575050354004, + "num_tokens": 8103076.0, + "step": 2519 + }, + { + "epoch": 1.289002557544757, + "grad_norm": 8.420722961425781, + "learning_rate": 1e-06, + "loss": 0.2001, + "mean_token_accuracy": 0.9021108150482178, + "num_tokens": 8106428.0, + "step": 2520 + }, + { + "epoch": 1.2895140664961637, + "grad_norm": 7.715994358062744, + "learning_rate": 1e-06, + "loss": 0.193, + "mean_token_accuracy": 0.927747905254364, + "num_tokens": 8110525.0, + "step": 2521 + }, + { + "epoch": 1.2900255754475705, + "grad_norm": 10.094199180603027, + "learning_rate": 1e-06, + "loss": 0.2093, + "mean_token_accuracy": 0.8990477323532104, + "num_tokens": 8113240.0, + "step": 2522 + }, + { + "epoch": 1.290537084398977, + "grad_norm": 14.619002342224121, + "learning_rate": 1e-06, + "loss": 0.2955, + "mean_token_accuracy": 0.8860317468643188, + "num_tokens": 8116074.0, + "step": 2523 + }, + { + "epoch": 1.2910485933503837, + "grad_norm": 14.986299514770508, + "learning_rate": 1e-06, + "loss": 0.2976, + "mean_token_accuracy": 0.8883541226387024, + "num_tokens": 8118924.0, + "step": 2524 + }, + { + "epoch": 1.2915601023017902, + "grad_norm": 11.995738983154297, + "learning_rate": 1e-06, + "loss": 0.2644, + "mean_token_accuracy": 0.8894813060760498, + "num_tokens": 8121617.0, + "step": 2525 + }, + { + "epoch": 1.292071611253197, + "grad_norm": 9.273662567138672, + "learning_rate": 1e-06, + "loss": 0.1971, + "mean_token_accuracy": 0.9023976922035217, + "num_tokens": 8124182.0, + "step": 2526 + }, + { + "epoch": 1.2925831202046036, + "grad_norm": 7.104301929473877, + "learning_rate": 1e-06, + "loss": 0.1972, + "mean_token_accuracy": 0.9250998497009277, + "num_tokens": 8127644.0, + "step": 2527 + }, + { + "epoch": 1.2930946291560104, + "grad_norm": 11.901294708251953, + "learning_rate": 1e-06, + "loss": 0.2318, + "mean_token_accuracy": 0.9031376838684082, + "num_tokens": 8130323.0, + "step": 2528 + }, + { + "epoch": 1.2936061381074169, + "grad_norm": 17.764514923095703, + "learning_rate": 1e-06, + "loss": 0.3062, + "mean_token_accuracy": 0.8823012113571167, + "num_tokens": 8132720.0, + "step": 2529 + }, + { + "epoch": 1.2941176470588236, + "grad_norm": 12.926644325256348, + "learning_rate": 1e-06, + "loss": 0.2487, + "mean_token_accuracy": 0.8961057662963867, + "num_tokens": 8135227.0, + "step": 2530 + }, + { + "epoch": 1.29462915601023, + "grad_norm": 8.731025695800781, + "learning_rate": 1e-06, + "loss": 0.2174, + "mean_token_accuracy": 0.9245085716247559, + "num_tokens": 8138160.0, + "step": 2531 + }, + { + "epoch": 1.2951406649616368, + "grad_norm": 10.137456893920898, + "learning_rate": 1e-06, + "loss": 0.297, + "mean_token_accuracy": 0.8980768918991089, + "num_tokens": 8141384.0, + "step": 2532 + }, + { + "epoch": 1.2956521739130435, + "grad_norm": 8.469741821289062, + "learning_rate": 1e-06, + "loss": 0.1971, + "mean_token_accuracy": 0.9097927212715149, + "num_tokens": 8144306.0, + "step": 2533 + }, + { + "epoch": 1.29616368286445, + "grad_norm": 9.227808952331543, + "learning_rate": 1e-06, + "loss": 0.2415, + "mean_token_accuracy": 0.8988843560218811, + "num_tokens": 8146835.0, + "step": 2534 + }, + { + "epoch": 1.2966751918158568, + "grad_norm": 5.159260272979736, + "learning_rate": 1e-06, + "loss": 0.1546, + "mean_token_accuracy": 0.9214944839477539, + "num_tokens": 8150628.0, + "step": 2535 + }, + { + "epoch": 1.2971867007672635, + "grad_norm": 7.997294902801514, + "learning_rate": 1e-06, + "loss": 0.1591, + "mean_token_accuracy": 0.9260174036026001, + "num_tokens": 8153754.0, + "step": 2536 + }, + { + "epoch": 1.29769820971867, + "grad_norm": 15.857327461242676, + "learning_rate": 1e-06, + "loss": 0.323, + "mean_token_accuracy": 0.871071994304657, + "num_tokens": 8156823.0, + "step": 2537 + }, + { + "epoch": 1.2982097186700767, + "grad_norm": 8.091849327087402, + "learning_rate": 1e-06, + "loss": 0.2212, + "mean_token_accuracy": 0.8822873830795288, + "num_tokens": 8160751.0, + "step": 2538 + }, + { + "epoch": 1.2987212276214835, + "grad_norm": 8.25487995147705, + "learning_rate": 1e-06, + "loss": 0.2384, + "mean_token_accuracy": 0.8814765810966492, + "num_tokens": 8163980.0, + "step": 2539 + }, + { + "epoch": 1.29923273657289, + "grad_norm": 9.867486000061035, + "learning_rate": 1e-06, + "loss": 0.2863, + "mean_token_accuracy": 0.8801731467247009, + "num_tokens": 8167055.0, + "step": 2540 + }, + { + "epoch": 1.2997442455242967, + "grad_norm": 9.603473663330078, + "learning_rate": 1e-06, + "loss": 0.2321, + "mean_token_accuracy": 0.9167886972427368, + "num_tokens": 8169611.0, + "step": 2541 + }, + { + "epoch": 1.3002557544757034, + "grad_norm": 4.671981334686279, + "learning_rate": 1e-06, + "loss": 0.1138, + "mean_token_accuracy": 0.9393048286437988, + "num_tokens": 8173876.0, + "step": 2542 + }, + { + "epoch": 1.30076726342711, + "grad_norm": 12.575286865234375, + "learning_rate": 1e-06, + "loss": 0.2966, + "mean_token_accuracy": 0.8838611245155334, + "num_tokens": 8176803.0, + "step": 2543 + }, + { + "epoch": 1.3012787723785166, + "grad_norm": 11.2921724319458, + "learning_rate": 1e-06, + "loss": 0.2399, + "mean_token_accuracy": 0.8955235481262207, + "num_tokens": 8180394.0, + "step": 2544 + }, + { + "epoch": 1.3017902813299234, + "grad_norm": 9.942792892456055, + "learning_rate": 1e-06, + "loss": 0.2411, + "mean_token_accuracy": 0.8937733173370361, + "num_tokens": 8183268.0, + "step": 2545 + }, + { + "epoch": 1.3023017902813299, + "grad_norm": 11.569206237792969, + "learning_rate": 1e-06, + "loss": 0.2677, + "mean_token_accuracy": 0.903581440448761, + "num_tokens": 8186442.0, + "step": 2546 + }, + { + "epoch": 1.3028132992327366, + "grad_norm": 12.132904052734375, + "learning_rate": 1e-06, + "loss": 0.2414, + "mean_token_accuracy": 0.9298357367515564, + "num_tokens": 8189834.0, + "step": 2547 + }, + { + "epoch": 1.303324808184143, + "grad_norm": 9.53484058380127, + "learning_rate": 1e-06, + "loss": 0.2449, + "mean_token_accuracy": 0.9024474620819092, + "num_tokens": 8192710.0, + "step": 2548 + }, + { + "epoch": 1.3038363171355498, + "grad_norm": 9.668228149414062, + "learning_rate": 1e-06, + "loss": 0.2071, + "mean_token_accuracy": 0.9145469665527344, + "num_tokens": 8195727.0, + "step": 2549 + }, + { + "epoch": 1.3043478260869565, + "grad_norm": 9.234206199645996, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.8990169763565063, + "num_tokens": 8199103.0, + "step": 2550 + }, + { + "epoch": 1.3048593350383633, + "grad_norm": 8.74771785736084, + "learning_rate": 1e-06, + "loss": 0.1853, + "mean_token_accuracy": 0.9344924688339233, + "num_tokens": 8202430.0, + "step": 2551 + }, + { + "epoch": 1.3053708439897698, + "grad_norm": 10.081318855285645, + "learning_rate": 1e-06, + "loss": 0.2528, + "mean_token_accuracy": 0.9010133147239685, + "num_tokens": 8205404.0, + "step": 2552 + }, + { + "epoch": 1.3058823529411765, + "grad_norm": 5.893933296203613, + "learning_rate": 1e-06, + "loss": 0.1263, + "mean_token_accuracy": 0.9321616291999817, + "num_tokens": 8208804.0, + "step": 2553 + }, + { + "epoch": 1.306393861892583, + "grad_norm": 11.282991409301758, + "learning_rate": 1e-06, + "loss": 0.3592, + "mean_token_accuracy": 0.8599017858505249, + "num_tokens": 8212465.0, + "step": 2554 + }, + { + "epoch": 1.3069053708439897, + "grad_norm": 8.109036445617676, + "learning_rate": 1e-06, + "loss": 0.185, + "mean_token_accuracy": 0.9281240105628967, + "num_tokens": 8214941.0, + "step": 2555 + }, + { + "epoch": 1.3074168797953964, + "grad_norm": 10.238203048706055, + "learning_rate": 1e-06, + "loss": 0.2104, + "mean_token_accuracy": 0.9171148538589478, + "num_tokens": 8218303.0, + "step": 2556 + }, + { + "epoch": 1.3079283887468032, + "grad_norm": 7.201146602630615, + "learning_rate": 1e-06, + "loss": 0.1634, + "mean_token_accuracy": 0.909917414188385, + "num_tokens": 8222231.0, + "step": 2557 + }, + { + "epoch": 1.3084398976982097, + "grad_norm": 9.590178489685059, + "learning_rate": 1e-06, + "loss": 0.231, + "mean_token_accuracy": 0.880340039730072, + "num_tokens": 8226091.0, + "step": 2558 + }, + { + "epoch": 1.3089514066496164, + "grad_norm": 6.96278715133667, + "learning_rate": 1e-06, + "loss": 0.1798, + "mean_token_accuracy": 0.9149190783500671, + "num_tokens": 8230214.0, + "step": 2559 + }, + { + "epoch": 1.309462915601023, + "grad_norm": 10.012982368469238, + "learning_rate": 1e-06, + "loss": 0.2414, + "mean_token_accuracy": 0.8975480794906616, + "num_tokens": 8232844.0, + "step": 2560 + }, + { + "epoch": 1.3099744245524296, + "grad_norm": 11.145002365112305, + "learning_rate": 1e-06, + "loss": 0.2655, + "mean_token_accuracy": 0.8942848443984985, + "num_tokens": 8236025.0, + "step": 2561 + }, + { + "epoch": 1.3104859335038364, + "grad_norm": 7.650847911834717, + "learning_rate": 1e-06, + "loss": 0.2092, + "mean_token_accuracy": 0.9144128561019897, + "num_tokens": 8239960.0, + "step": 2562 + }, + { + "epoch": 1.310997442455243, + "grad_norm": 15.070687294006348, + "learning_rate": 1e-06, + "loss": 0.2461, + "mean_token_accuracy": 0.8990862965583801, + "num_tokens": 8242930.0, + "step": 2563 + }, + { + "epoch": 1.3115089514066496, + "grad_norm": 11.917089462280273, + "learning_rate": 1e-06, + "loss": 0.2745, + "mean_token_accuracy": 0.8996566534042358, + "num_tokens": 8245750.0, + "step": 2564 + }, + { + "epoch": 1.3120204603580563, + "grad_norm": 13.317955017089844, + "learning_rate": 1e-06, + "loss": 0.3438, + "mean_token_accuracy": 0.8716052770614624, + "num_tokens": 8248475.0, + "step": 2565 + }, + { + "epoch": 1.3125319693094628, + "grad_norm": 10.511192321777344, + "learning_rate": 1e-06, + "loss": 0.2007, + "mean_token_accuracy": 0.9067864418029785, + "num_tokens": 8252496.0, + "step": 2566 + }, + { + "epoch": 1.3130434782608695, + "grad_norm": 8.873238563537598, + "learning_rate": 1e-06, + "loss": 0.1874, + "mean_token_accuracy": 0.9283908605575562, + "num_tokens": 8255350.0, + "step": 2567 + }, + { + "epoch": 1.3135549872122763, + "grad_norm": 10.054226875305176, + "learning_rate": 1e-06, + "loss": 0.2707, + "mean_token_accuracy": 0.8978056907653809, + "num_tokens": 8258044.0, + "step": 2568 + }, + { + "epoch": 1.314066496163683, + "grad_norm": 11.719409942626953, + "learning_rate": 1e-06, + "loss": 0.2035, + "mean_token_accuracy": 0.9164992570877075, + "num_tokens": 8261081.0, + "step": 2569 + }, + { + "epoch": 1.3145780051150895, + "grad_norm": 7.443212509155273, + "learning_rate": 1e-06, + "loss": 0.1902, + "mean_token_accuracy": 0.9165651798248291, + "num_tokens": 8264421.0, + "step": 2570 + }, + { + "epoch": 1.3150895140664962, + "grad_norm": 8.755417823791504, + "learning_rate": 1e-06, + "loss": 0.2045, + "mean_token_accuracy": 0.9262794852256775, + "num_tokens": 8267142.0, + "step": 2571 + }, + { + "epoch": 1.3156010230179027, + "grad_norm": 7.317359447479248, + "learning_rate": 1e-06, + "loss": 0.1356, + "mean_token_accuracy": 0.9421349167823792, + "num_tokens": 8270684.0, + "step": 2572 + }, + { + "epoch": 1.3161125319693094, + "grad_norm": 6.569379806518555, + "learning_rate": 1e-06, + "loss": 0.2029, + "mean_token_accuracy": 0.9082022905349731, + "num_tokens": 8274278.0, + "step": 2573 + }, + { + "epoch": 1.3166240409207162, + "grad_norm": 9.995498657226562, + "learning_rate": 1e-06, + "loss": 0.3014, + "mean_token_accuracy": 0.8741519451141357, + "num_tokens": 8277711.0, + "step": 2574 + }, + { + "epoch": 1.317135549872123, + "grad_norm": 11.544756889343262, + "learning_rate": 1e-06, + "loss": 0.2422, + "mean_token_accuracy": 0.9032183885574341, + "num_tokens": 8280411.0, + "step": 2575 + }, + { + "epoch": 1.3176470588235294, + "grad_norm": 10.396629333496094, + "learning_rate": 1e-06, + "loss": 0.2857, + "mean_token_accuracy": 0.8682359457015991, + "num_tokens": 8282810.0, + "step": 2576 + }, + { + "epoch": 1.3181585677749361, + "grad_norm": 6.80357551574707, + "learning_rate": 1e-06, + "loss": 0.1643, + "mean_token_accuracy": 0.8991316556930542, + "num_tokens": 8286053.0, + "step": 2577 + }, + { + "epoch": 1.3186700767263426, + "grad_norm": 9.666768074035645, + "learning_rate": 1e-06, + "loss": 0.2571, + "mean_token_accuracy": 0.8875732421875, + "num_tokens": 8289055.0, + "step": 2578 + }, + { + "epoch": 1.3191815856777493, + "grad_norm": 8.17360782623291, + "learning_rate": 1e-06, + "loss": 0.1965, + "mean_token_accuracy": 0.9024648666381836, + "num_tokens": 8292624.0, + "step": 2579 + }, + { + "epoch": 1.319693094629156, + "grad_norm": 7.135474681854248, + "learning_rate": 1e-06, + "loss": 0.2048, + "mean_token_accuracy": 0.9307152628898621, + "num_tokens": 8296267.0, + "step": 2580 + }, + { + "epoch": 1.3202046035805628, + "grad_norm": 11.841385841369629, + "learning_rate": 1e-06, + "loss": 0.3255, + "mean_token_accuracy": 0.8824602365493774, + "num_tokens": 8299319.0, + "step": 2581 + }, + { + "epoch": 1.3207161125319693, + "grad_norm": 10.390336990356445, + "learning_rate": 1e-06, + "loss": 0.2407, + "mean_token_accuracy": 0.9113785028457642, + "num_tokens": 8303306.0, + "step": 2582 + }, + { + "epoch": 1.321227621483376, + "grad_norm": 10.509865760803223, + "learning_rate": 1e-06, + "loss": 0.2375, + "mean_token_accuracy": 0.9135936498641968, + "num_tokens": 8306218.0, + "step": 2583 + }, + { + "epoch": 1.3217391304347825, + "grad_norm": 9.248383522033691, + "learning_rate": 1e-06, + "loss": 0.2375, + "mean_token_accuracy": 0.9226138591766357, + "num_tokens": 8308963.0, + "step": 2584 + }, + { + "epoch": 1.3222506393861893, + "grad_norm": 14.575706481933594, + "learning_rate": 1e-06, + "loss": 0.4283, + "mean_token_accuracy": 0.8543977737426758, + "num_tokens": 8312012.0, + "step": 2585 + }, + { + "epoch": 1.322762148337596, + "grad_norm": 7.160074710845947, + "learning_rate": 1e-06, + "loss": 0.2043, + "mean_token_accuracy": 0.8951680064201355, + "num_tokens": 8314705.0, + "step": 2586 + }, + { + "epoch": 1.3232736572890025, + "grad_norm": 9.528254508972168, + "learning_rate": 1e-06, + "loss": 0.2309, + "mean_token_accuracy": 0.8940316438674927, + "num_tokens": 8317998.0, + "step": 2587 + }, + { + "epoch": 1.3237851662404092, + "grad_norm": 9.798362731933594, + "learning_rate": 1e-06, + "loss": 0.2309, + "mean_token_accuracy": 0.9100013971328735, + "num_tokens": 8320627.0, + "step": 2588 + }, + { + "epoch": 1.324296675191816, + "grad_norm": 11.460164070129395, + "learning_rate": 1e-06, + "loss": 0.306, + "mean_token_accuracy": 0.878216028213501, + "num_tokens": 8323812.0, + "step": 2589 + }, + { + "epoch": 1.3248081841432224, + "grad_norm": 9.556857109069824, + "learning_rate": 1e-06, + "loss": 0.1923, + "mean_token_accuracy": 0.9185407161712646, + "num_tokens": 8327243.0, + "step": 2590 + }, + { + "epoch": 1.3253196930946292, + "grad_norm": 13.796842575073242, + "learning_rate": 1e-06, + "loss": 0.2455, + "mean_token_accuracy": 0.9198888540267944, + "num_tokens": 8330334.0, + "step": 2591 + }, + { + "epoch": 1.3258312020460359, + "grad_norm": 10.814268112182617, + "learning_rate": 1e-06, + "loss": 0.3294, + "mean_token_accuracy": 0.8878090381622314, + "num_tokens": 8333244.0, + "step": 2592 + }, + { + "epoch": 1.3263427109974424, + "grad_norm": 9.302190780639648, + "learning_rate": 1e-06, + "loss": 0.2496, + "mean_token_accuracy": 0.9055168032646179, + "num_tokens": 8336285.0, + "step": 2593 + }, + { + "epoch": 1.3268542199488491, + "grad_norm": 6.287835121154785, + "learning_rate": 1e-06, + "loss": 0.2016, + "mean_token_accuracy": 0.894889771938324, + "num_tokens": 8339776.0, + "step": 2594 + }, + { + "epoch": 1.3273657289002558, + "grad_norm": 11.033446311950684, + "learning_rate": 1e-06, + "loss": 0.3559, + "mean_token_accuracy": 0.8604770302772522, + "num_tokens": 8342418.0, + "step": 2595 + }, + { + "epoch": 1.3278772378516623, + "grad_norm": 6.6986846923828125, + "learning_rate": 1e-06, + "loss": 0.1771, + "mean_token_accuracy": 0.9383216500282288, + "num_tokens": 8345996.0, + "step": 2596 + }, + { + "epoch": 1.328388746803069, + "grad_norm": 9.3150634765625, + "learning_rate": 1e-06, + "loss": 0.2758, + "mean_token_accuracy": 0.8846895098686218, + "num_tokens": 8348750.0, + "step": 2597 + }, + { + "epoch": 1.3289002557544758, + "grad_norm": 11.978989601135254, + "learning_rate": 1e-06, + "loss": 0.2102, + "mean_token_accuracy": 0.9047770500183105, + "num_tokens": 8351574.0, + "step": 2598 + }, + { + "epoch": 1.3294117647058823, + "grad_norm": 9.146011352539062, + "learning_rate": 1e-06, + "loss": 0.1802, + "mean_token_accuracy": 0.9099382758140564, + "num_tokens": 8355725.0, + "step": 2599 + }, + { + "epoch": 1.329923273657289, + "grad_norm": 7.793693542480469, + "learning_rate": 1e-06, + "loss": 0.1885, + "mean_token_accuracy": 0.9185223579406738, + "num_tokens": 8359377.0, + "step": 2600 + }, + { + "epoch": 1.3304347826086955, + "grad_norm": 7.516993045806885, + "learning_rate": 1e-06, + "loss": 0.1625, + "mean_token_accuracy": 0.9168404340744019, + "num_tokens": 8362901.0, + "step": 2601 + }, + { + "epoch": 1.3309462915601022, + "grad_norm": 10.579753875732422, + "learning_rate": 1e-06, + "loss": 0.2523, + "mean_token_accuracy": 0.9021778702735901, + "num_tokens": 8365675.0, + "step": 2602 + }, + { + "epoch": 1.331457800511509, + "grad_norm": 14.58873462677002, + "learning_rate": 1e-06, + "loss": 0.2122, + "mean_token_accuracy": 0.9320859909057617, + "num_tokens": 8368267.0, + "step": 2603 + }, + { + "epoch": 1.3319693094629157, + "grad_norm": 11.630057334899902, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.8733430504798889, + "num_tokens": 8371474.0, + "step": 2604 + }, + { + "epoch": 1.3324808184143222, + "grad_norm": 5.467602729797363, + "learning_rate": 1e-06, + "loss": 0.1208, + "mean_token_accuracy": 0.9180538058280945, + "num_tokens": 8376146.0, + "step": 2605 + }, + { + "epoch": 1.332992327365729, + "grad_norm": 9.225869178771973, + "learning_rate": 1e-06, + "loss": 0.1667, + "mean_token_accuracy": 0.9128890037536621, + "num_tokens": 8380586.0, + "step": 2606 + }, + { + "epoch": 1.3335038363171354, + "grad_norm": 14.155524253845215, + "learning_rate": 1e-06, + "loss": 0.2802, + "mean_token_accuracy": 0.8976949453353882, + "num_tokens": 8384626.0, + "step": 2607 + }, + { + "epoch": 1.3340153452685422, + "grad_norm": 6.4874653816223145, + "learning_rate": 1e-06, + "loss": 0.1787, + "mean_token_accuracy": 0.9094250798225403, + "num_tokens": 8388545.0, + "step": 2608 + }, + { + "epoch": 1.3345268542199489, + "grad_norm": 8.98349380493164, + "learning_rate": 1e-06, + "loss": 0.1909, + "mean_token_accuracy": 0.9171416163444519, + "num_tokens": 8392339.0, + "step": 2609 + }, + { + "epoch": 1.3350383631713556, + "grad_norm": 7.844724655151367, + "learning_rate": 1e-06, + "loss": 0.1706, + "mean_token_accuracy": 0.9058558940887451, + "num_tokens": 8396088.0, + "step": 2610 + }, + { + "epoch": 1.335549872122762, + "grad_norm": 8.627533912658691, + "learning_rate": 1e-06, + "loss": 0.2046, + "mean_token_accuracy": 0.9109455943107605, + "num_tokens": 8399351.0, + "step": 2611 + }, + { + "epoch": 1.3360613810741688, + "grad_norm": 7.614812850952148, + "learning_rate": 1e-06, + "loss": 0.1694, + "mean_token_accuracy": 0.909098744392395, + "num_tokens": 8402499.0, + "step": 2612 + }, + { + "epoch": 1.3365728900255753, + "grad_norm": 14.927414894104004, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.9146877527236938, + "num_tokens": 8405749.0, + "step": 2613 + }, + { + "epoch": 1.337084398976982, + "grad_norm": 10.883075714111328, + "learning_rate": 1e-06, + "loss": 0.2944, + "mean_token_accuracy": 0.8688514232635498, + "num_tokens": 8409253.0, + "step": 2614 + }, + { + "epoch": 1.3375959079283888, + "grad_norm": 8.206954002380371, + "learning_rate": 1e-06, + "loss": 0.2329, + "mean_token_accuracy": 0.9143577814102173, + "num_tokens": 8412163.0, + "step": 2615 + }, + { + "epoch": 1.3381074168797955, + "grad_norm": 7.496329307556152, + "learning_rate": 1e-06, + "loss": 0.2085, + "mean_token_accuracy": 0.8522863984107971, + "num_tokens": 8416112.0, + "step": 2616 + }, + { + "epoch": 1.338618925831202, + "grad_norm": 15.816980361938477, + "learning_rate": 1e-06, + "loss": 0.3581, + "mean_token_accuracy": 0.8762684464454651, + "num_tokens": 8419332.0, + "step": 2617 + }, + { + "epoch": 1.3391304347826087, + "grad_norm": 9.821800231933594, + "learning_rate": 1e-06, + "loss": 0.2112, + "mean_token_accuracy": 0.9242023825645447, + "num_tokens": 8421836.0, + "step": 2618 + }, + { + "epoch": 1.3396419437340152, + "grad_norm": 11.134312629699707, + "learning_rate": 1e-06, + "loss": 0.1959, + "mean_token_accuracy": 0.9269711375236511, + "num_tokens": 8425179.0, + "step": 2619 + }, + { + "epoch": 1.340153452685422, + "grad_norm": 8.064162254333496, + "learning_rate": 1e-06, + "loss": 0.181, + "mean_token_accuracy": 0.8970611691474915, + "num_tokens": 8428744.0, + "step": 2620 + }, + { + "epoch": 1.3406649616368287, + "grad_norm": 9.486059188842773, + "learning_rate": 1e-06, + "loss": 0.2389, + "mean_token_accuracy": 0.9212443828582764, + "num_tokens": 8432227.0, + "step": 2621 + }, + { + "epoch": 1.3411764705882354, + "grad_norm": 11.15693187713623, + "learning_rate": 1e-06, + "loss": 0.3497, + "mean_token_accuracy": 0.8678015470504761, + "num_tokens": 8435423.0, + "step": 2622 + }, + { + "epoch": 1.341687979539642, + "grad_norm": 5.42397403717041, + "learning_rate": 1e-06, + "loss": 0.1244, + "mean_token_accuracy": 0.9221982359886169, + "num_tokens": 8438694.0, + "step": 2623 + }, + { + "epoch": 1.3421994884910486, + "grad_norm": 8.924321174621582, + "learning_rate": 1e-06, + "loss": 0.1969, + "mean_token_accuracy": 0.9007896184921265, + "num_tokens": 8441771.0, + "step": 2624 + }, + { + "epoch": 1.3427109974424551, + "grad_norm": 8.316510200500488, + "learning_rate": 1e-06, + "loss": 0.1271, + "mean_token_accuracy": 0.9386913776397705, + "num_tokens": 8446149.0, + "step": 2625 + }, + { + "epoch": 1.3432225063938619, + "grad_norm": 9.916954040527344, + "learning_rate": 1e-06, + "loss": 0.2094, + "mean_token_accuracy": 0.8843207955360413, + "num_tokens": 8449174.0, + "step": 2626 + }, + { + "epoch": 1.3437340153452686, + "grad_norm": 9.423904418945312, + "learning_rate": 1e-06, + "loss": 0.1498, + "mean_token_accuracy": 0.9310312867164612, + "num_tokens": 8452658.0, + "step": 2627 + }, + { + "epoch": 1.3442455242966753, + "grad_norm": 15.855687141418457, + "learning_rate": 1e-06, + "loss": 0.2534, + "mean_token_accuracy": 0.9309476613998413, + "num_tokens": 8455711.0, + "step": 2628 + }, + { + "epoch": 1.3447570332480818, + "grad_norm": 11.494719505310059, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.9099302291870117, + "num_tokens": 8459318.0, + "step": 2629 + }, + { + "epoch": 1.3452685421994885, + "grad_norm": 7.303058624267578, + "learning_rate": 1e-06, + "loss": 0.1751, + "mean_token_accuracy": 0.9048219919204712, + "num_tokens": 8462976.0, + "step": 2630 + }, + { + "epoch": 1.345780051150895, + "grad_norm": 12.824055671691895, + "learning_rate": 1e-06, + "loss": 0.2922, + "mean_token_accuracy": 0.8795217275619507, + "num_tokens": 8466948.0, + "step": 2631 + }, + { + "epoch": 1.3462915601023018, + "grad_norm": 13.811022758483887, + "learning_rate": 1e-06, + "loss": 0.3907, + "mean_token_accuracy": 0.8796792030334473, + "num_tokens": 8469835.0, + "step": 2632 + }, + { + "epoch": 1.3468030690537085, + "grad_norm": 9.128850936889648, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.8934276103973389, + "num_tokens": 8473103.0, + "step": 2633 + }, + { + "epoch": 1.3473145780051152, + "grad_norm": 11.856691360473633, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8773627281188965, + "num_tokens": 8475522.0, + "step": 2634 + }, + { + "epoch": 1.3478260869565217, + "grad_norm": 18.911638259887695, + "learning_rate": 1e-06, + "loss": 0.295, + "mean_token_accuracy": 0.8863832950592041, + "num_tokens": 8478374.0, + "step": 2635 + }, + { + "epoch": 1.3483375959079285, + "grad_norm": 11.905740737915039, + "learning_rate": 1e-06, + "loss": 0.3006, + "mean_token_accuracy": 0.8694403767585754, + "num_tokens": 8481475.0, + "step": 2636 + }, + { + "epoch": 1.348849104859335, + "grad_norm": 9.810194969177246, + "learning_rate": 1e-06, + "loss": 0.2585, + "mean_token_accuracy": 0.8838869333267212, + "num_tokens": 8484656.0, + "step": 2637 + }, + { + "epoch": 1.3493606138107417, + "grad_norm": 11.128705978393555, + "learning_rate": 1e-06, + "loss": 0.2661, + "mean_token_accuracy": 0.8829017877578735, + "num_tokens": 8487513.0, + "step": 2638 + }, + { + "epoch": 1.3498721227621484, + "grad_norm": 9.80784797668457, + "learning_rate": 1e-06, + "loss": 0.2808, + "mean_token_accuracy": 0.9189819097518921, + "num_tokens": 8490533.0, + "step": 2639 + }, + { + "epoch": 1.350383631713555, + "grad_norm": 10.300896644592285, + "learning_rate": 1e-06, + "loss": 0.327, + "mean_token_accuracy": 0.8626703023910522, + "num_tokens": 8493286.0, + "step": 2640 + }, + { + "epoch": 1.3508951406649616, + "grad_norm": 8.968619346618652, + "learning_rate": 1e-06, + "loss": 0.2008, + "mean_token_accuracy": 0.9211738109588623, + "num_tokens": 8497088.0, + "step": 2641 + }, + { + "epoch": 1.3514066496163684, + "grad_norm": 6.322421073913574, + "learning_rate": 1e-06, + "loss": 0.1725, + "mean_token_accuracy": 0.9399903416633606, + "num_tokens": 8500699.0, + "step": 2642 + }, + { + "epoch": 1.3519181585677749, + "grad_norm": 9.09371280670166, + "learning_rate": 1e-06, + "loss": 0.1997, + "mean_token_accuracy": 0.9084452986717224, + "num_tokens": 8503832.0, + "step": 2643 + }, + { + "epoch": 1.3524296675191816, + "grad_norm": 10.790209770202637, + "learning_rate": 1e-06, + "loss": 0.2252, + "mean_token_accuracy": 0.8909125924110413, + "num_tokens": 8506585.0, + "step": 2644 + }, + { + "epoch": 1.3529411764705883, + "grad_norm": 12.124738693237305, + "learning_rate": 1e-06, + "loss": 0.2707, + "mean_token_accuracy": 0.8931481838226318, + "num_tokens": 8509959.0, + "step": 2645 + }, + { + "epoch": 1.3534526854219948, + "grad_norm": 8.16139030456543, + "learning_rate": 1e-06, + "loss": 0.212, + "mean_token_accuracy": 0.9252384901046753, + "num_tokens": 8512969.0, + "step": 2646 + }, + { + "epoch": 1.3539641943734015, + "grad_norm": 10.265334129333496, + "learning_rate": 1e-06, + "loss": 0.2856, + "mean_token_accuracy": 0.8986542820930481, + "num_tokens": 8516094.0, + "step": 2647 + }, + { + "epoch": 1.3544757033248083, + "grad_norm": 7.012290954589844, + "learning_rate": 1e-06, + "loss": 0.1484, + "mean_token_accuracy": 0.9240108728408813, + "num_tokens": 8519969.0, + "step": 2648 + }, + { + "epoch": 1.3549872122762148, + "grad_norm": 10.884221076965332, + "learning_rate": 1e-06, + "loss": 0.2725, + "mean_token_accuracy": 0.8952309489250183, + "num_tokens": 8523041.0, + "step": 2649 + }, + { + "epoch": 1.3554987212276215, + "grad_norm": 15.477529525756836, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.9085657596588135, + "num_tokens": 8525670.0, + "step": 2650 + }, + { + "epoch": 1.3560102301790282, + "grad_norm": 8.587037086486816, + "learning_rate": 1e-06, + "loss": 0.2298, + "mean_token_accuracy": 0.8907398581504822, + "num_tokens": 8529583.0, + "step": 2651 + }, + { + "epoch": 1.3565217391304347, + "grad_norm": 9.687176704406738, + "learning_rate": 1e-06, + "loss": 0.2752, + "mean_token_accuracy": 0.8666069507598877, + "num_tokens": 8532544.0, + "step": 2652 + }, + { + "epoch": 1.3570332480818414, + "grad_norm": 11.2189302444458, + "learning_rate": 1e-06, + "loss": 0.2794, + "mean_token_accuracy": 0.8955722451210022, + "num_tokens": 8536109.0, + "step": 2653 + }, + { + "epoch": 1.357544757033248, + "grad_norm": 8.188714027404785, + "learning_rate": 1e-06, + "loss": 0.2149, + "mean_token_accuracy": 0.9252948760986328, + "num_tokens": 8540038.0, + "step": 2654 + }, + { + "epoch": 1.3580562659846547, + "grad_norm": 11.311461448669434, + "learning_rate": 1e-06, + "loss": 0.2637, + "mean_token_accuracy": 0.878476619720459, + "num_tokens": 8542818.0, + "step": 2655 + }, + { + "epoch": 1.3585677749360614, + "grad_norm": 9.302752494812012, + "learning_rate": 1e-06, + "loss": 0.2607, + "mean_token_accuracy": 0.8893982172012329, + "num_tokens": 8546423.0, + "step": 2656 + }, + { + "epoch": 1.3590792838874681, + "grad_norm": 9.787663459777832, + "learning_rate": 1e-06, + "loss": 0.1855, + "mean_token_accuracy": 0.9170368313789368, + "num_tokens": 8550211.0, + "step": 2657 + }, + { + "epoch": 1.3595907928388746, + "grad_norm": 6.193150043487549, + "learning_rate": 1e-06, + "loss": 0.1491, + "mean_token_accuracy": 0.9300994873046875, + "num_tokens": 8553913.0, + "step": 2658 + }, + { + "epoch": 1.3601023017902814, + "grad_norm": 11.430700302124023, + "learning_rate": 1e-06, + "loss": 0.3003, + "mean_token_accuracy": 0.894626259803772, + "num_tokens": 8556726.0, + "step": 2659 + }, + { + "epoch": 1.3606138107416879, + "grad_norm": 10.691494941711426, + "learning_rate": 1e-06, + "loss": 0.2553, + "mean_token_accuracy": 0.9118005633354187, + "num_tokens": 8559805.0, + "step": 2660 + }, + { + "epoch": 1.3611253196930946, + "grad_norm": 7.076081275939941, + "learning_rate": 1e-06, + "loss": 0.2047, + "mean_token_accuracy": 0.9139398336410522, + "num_tokens": 8562815.0, + "step": 2661 + }, + { + "epoch": 1.3616368286445013, + "grad_norm": 7.894775867462158, + "learning_rate": 1e-06, + "loss": 0.2097, + "mean_token_accuracy": 0.9171464443206787, + "num_tokens": 8566490.0, + "step": 2662 + }, + { + "epoch": 1.362148337595908, + "grad_norm": 10.489535331726074, + "learning_rate": 1e-06, + "loss": 0.2313, + "mean_token_accuracy": 0.8964427709579468, + "num_tokens": 8569683.0, + "step": 2663 + }, + { + "epoch": 1.3626598465473145, + "grad_norm": 6.953126430511475, + "learning_rate": 1e-06, + "loss": 0.1667, + "mean_token_accuracy": 0.9295298457145691, + "num_tokens": 8572724.0, + "step": 2664 + }, + { + "epoch": 1.3631713554987213, + "grad_norm": 11.5058012008667, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.8934129476547241, + "num_tokens": 8575866.0, + "step": 2665 + }, + { + "epoch": 1.3636828644501278, + "grad_norm": 4.67785120010376, + "learning_rate": 1e-06, + "loss": 0.1385, + "mean_token_accuracy": 0.9033175706863403, + "num_tokens": 8579939.0, + "step": 2666 + }, + { + "epoch": 1.3641943734015345, + "grad_norm": 7.2175679206848145, + "learning_rate": 1e-06, + "loss": 0.2085, + "mean_token_accuracy": 0.9139742851257324, + "num_tokens": 8583020.0, + "step": 2667 + }, + { + "epoch": 1.3647058823529412, + "grad_norm": 11.161226272583008, + "learning_rate": 1e-06, + "loss": 0.3041, + "mean_token_accuracy": 0.8787931203842163, + "num_tokens": 8585622.0, + "step": 2668 + }, + { + "epoch": 1.365217391304348, + "grad_norm": 6.797589302062988, + "learning_rate": 1e-06, + "loss": 0.1633, + "mean_token_accuracy": 0.9237881302833557, + "num_tokens": 8588226.0, + "step": 2669 + }, + { + "epoch": 1.3657289002557544, + "grad_norm": 10.840399742126465, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.8932566046714783, + "num_tokens": 8591503.0, + "step": 2670 + }, + { + "epoch": 1.3662404092071612, + "grad_norm": 10.009895324707031, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.904582142829895, + "num_tokens": 8595389.0, + "step": 2671 + }, + { + "epoch": 1.3667519181585677, + "grad_norm": 7.468221664428711, + "learning_rate": 1e-06, + "loss": 0.1873, + "mean_token_accuracy": 0.8936406970024109, + "num_tokens": 8599571.0, + "step": 2672 + }, + { + "epoch": 1.3672634271099744, + "grad_norm": 5.885266304016113, + "learning_rate": 1e-06, + "loss": 0.1289, + "mean_token_accuracy": 0.961573600769043, + "num_tokens": 8603027.0, + "step": 2673 + }, + { + "epoch": 1.3677749360613811, + "grad_norm": 11.189764976501465, + "learning_rate": 1e-06, + "loss": 0.2598, + "mean_token_accuracy": 0.8944615125656128, + "num_tokens": 8605953.0, + "step": 2674 + }, + { + "epoch": 1.3682864450127878, + "grad_norm": 6.834293842315674, + "learning_rate": 1e-06, + "loss": 0.1486, + "mean_token_accuracy": 0.9382761120796204, + "num_tokens": 8610205.0, + "step": 2675 + }, + { + "epoch": 1.3687979539641943, + "grad_norm": 12.26466178894043, + "learning_rate": 1e-06, + "loss": 0.294, + "mean_token_accuracy": 0.8844531774520874, + "num_tokens": 8613443.0, + "step": 2676 + }, + { + "epoch": 1.369309462915601, + "grad_norm": 9.5924072265625, + "learning_rate": 1e-06, + "loss": 0.2163, + "mean_token_accuracy": 0.9040603637695312, + "num_tokens": 8616454.0, + "step": 2677 + }, + { + "epoch": 1.3698209718670076, + "grad_norm": 11.945526123046875, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.8942030668258667, + "num_tokens": 8619444.0, + "step": 2678 + }, + { + "epoch": 1.3703324808184143, + "grad_norm": 9.81358528137207, + "learning_rate": 1e-06, + "loss": 0.287, + "mean_token_accuracy": 0.8823139667510986, + "num_tokens": 8622303.0, + "step": 2679 + }, + { + "epoch": 1.370843989769821, + "grad_norm": 8.731855392456055, + "learning_rate": 1e-06, + "loss": 0.1832, + "mean_token_accuracy": 0.9185656309127808, + "num_tokens": 8625362.0, + "step": 2680 + }, + { + "epoch": 1.3713554987212278, + "grad_norm": 6.344738960266113, + "learning_rate": 1e-06, + "loss": 0.1457, + "mean_token_accuracy": 0.9169124960899353, + "num_tokens": 8629246.0, + "step": 2681 + }, + { + "epoch": 1.3718670076726343, + "grad_norm": 12.464205741882324, + "learning_rate": 1e-06, + "loss": 0.3549, + "mean_token_accuracy": 0.8641268014907837, + "num_tokens": 8632671.0, + "step": 2682 + }, + { + "epoch": 1.372378516624041, + "grad_norm": 9.944014549255371, + "learning_rate": 1e-06, + "loss": 0.2556, + "mean_token_accuracy": 0.900776207447052, + "num_tokens": 8635662.0, + "step": 2683 + }, + { + "epoch": 1.3728900255754475, + "grad_norm": 11.488277435302734, + "learning_rate": 1e-06, + "loss": 0.2395, + "mean_token_accuracy": 0.9134902954101562, + "num_tokens": 8638100.0, + "step": 2684 + }, + { + "epoch": 1.3734015345268542, + "grad_norm": 9.296186447143555, + "learning_rate": 1e-06, + "loss": 0.2213, + "mean_token_accuracy": 0.8970261216163635, + "num_tokens": 8642188.0, + "step": 2685 + }, + { + "epoch": 1.373913043478261, + "grad_norm": 12.041662216186523, + "learning_rate": 1e-06, + "loss": 0.3151, + "mean_token_accuracy": 0.8757628798484802, + "num_tokens": 8645165.0, + "step": 2686 + }, + { + "epoch": 1.3744245524296677, + "grad_norm": 10.701714515686035, + "learning_rate": 1e-06, + "loss": 0.1886, + "mean_token_accuracy": 0.9329000115394592, + "num_tokens": 8648482.0, + "step": 2687 + }, + { + "epoch": 1.3749360613810742, + "grad_norm": 11.624438285827637, + "learning_rate": 1e-06, + "loss": 0.2187, + "mean_token_accuracy": 0.9147021770477295, + "num_tokens": 8652002.0, + "step": 2688 + }, + { + "epoch": 1.3754475703324809, + "grad_norm": 9.967523574829102, + "learning_rate": 1e-06, + "loss": 0.2111, + "mean_token_accuracy": 0.9148613214492798, + "num_tokens": 8654727.0, + "step": 2689 + }, + { + "epoch": 1.3759590792838874, + "grad_norm": 3.6880807876586914, + "learning_rate": 1e-06, + "loss": 0.0828, + "mean_token_accuracy": 0.9230068325996399, + "num_tokens": 8659898.0, + "step": 2690 + }, + { + "epoch": 1.3764705882352941, + "grad_norm": 10.661142349243164, + "learning_rate": 1e-06, + "loss": 0.2102, + "mean_token_accuracy": 0.9171688556671143, + "num_tokens": 8663181.0, + "step": 2691 + }, + { + "epoch": 1.3769820971867008, + "grad_norm": 9.725395202636719, + "learning_rate": 1e-06, + "loss": 0.2051, + "mean_token_accuracy": 0.8928872346878052, + "num_tokens": 8665832.0, + "step": 2692 + }, + { + "epoch": 1.3774936061381073, + "grad_norm": 6.9602131843566895, + "learning_rate": 1e-06, + "loss": 0.1863, + "mean_token_accuracy": 0.9098195433616638, + "num_tokens": 8669847.0, + "step": 2693 + }, + { + "epoch": 1.378005115089514, + "grad_norm": 9.80439281463623, + "learning_rate": 1e-06, + "loss": 0.2452, + "mean_token_accuracy": 0.8809831142425537, + "num_tokens": 8673267.0, + "step": 2694 + }, + { + "epoch": 1.3785166240409208, + "grad_norm": 9.684041976928711, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.909351110458374, + "num_tokens": 8676489.0, + "step": 2695 + }, + { + "epoch": 1.3790281329923273, + "grad_norm": 7.473410606384277, + "learning_rate": 1e-06, + "loss": 0.1947, + "mean_token_accuracy": 0.9197259545326233, + "num_tokens": 8680439.0, + "step": 2696 + }, + { + "epoch": 1.379539641943734, + "grad_norm": 10.959285736083984, + "learning_rate": 1e-06, + "loss": 0.2016, + "mean_token_accuracy": 0.9109476804733276, + "num_tokens": 8682784.0, + "step": 2697 + }, + { + "epoch": 1.3800511508951407, + "grad_norm": 8.704439163208008, + "learning_rate": 1e-06, + "loss": 0.2266, + "mean_token_accuracy": 0.9059333205223083, + "num_tokens": 8686102.0, + "step": 2698 + }, + { + "epoch": 1.3805626598465472, + "grad_norm": 13.973176956176758, + "learning_rate": 1e-06, + "loss": 0.2619, + "mean_token_accuracy": 0.9162961840629578, + "num_tokens": 8688784.0, + "step": 2699 + }, + { + "epoch": 1.381074168797954, + "grad_norm": 11.213776588439941, + "learning_rate": 1e-06, + "loss": 0.2574, + "mean_token_accuracy": 0.8972117900848389, + "num_tokens": 8692027.0, + "step": 2700 + }, + { + "epoch": 1.3815856777493607, + "grad_norm": 14.288193702697754, + "learning_rate": 1e-06, + "loss": 0.2983, + "mean_token_accuracy": 0.9122514724731445, + "num_tokens": 8695005.0, + "step": 2701 + }, + { + "epoch": 1.3820971867007672, + "grad_norm": 8.321462631225586, + "learning_rate": 1e-06, + "loss": 0.2396, + "mean_token_accuracy": 0.8851733207702637, + "num_tokens": 8697603.0, + "step": 2702 + }, + { + "epoch": 1.382608695652174, + "grad_norm": 14.291135787963867, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8914332389831543, + "num_tokens": 8701609.0, + "step": 2703 + }, + { + "epoch": 1.3831202046035806, + "grad_norm": 7.9024786949157715, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.9227774143218994, + "num_tokens": 8705036.0, + "step": 2704 + }, + { + "epoch": 1.3836317135549872, + "grad_norm": 11.385885238647461, + "learning_rate": 1e-06, + "loss": 0.2763, + "mean_token_accuracy": 0.8759843707084656, + "num_tokens": 8707728.0, + "step": 2705 + }, + { + "epoch": 1.3841432225063939, + "grad_norm": 9.846577644348145, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8614906072616577, + "num_tokens": 8710800.0, + "step": 2706 + }, + { + "epoch": 1.3846547314578004, + "grad_norm": 10.7793607711792, + "learning_rate": 1e-06, + "loss": 0.2533, + "mean_token_accuracy": 0.8934228420257568, + "num_tokens": 8713429.0, + "step": 2707 + }, + { + "epoch": 1.385166240409207, + "grad_norm": 13.362187385559082, + "learning_rate": 1e-06, + "loss": 0.2743, + "mean_token_accuracy": 0.895746111869812, + "num_tokens": 8716384.0, + "step": 2708 + }, + { + "epoch": 1.3856777493606138, + "grad_norm": 7.920930862426758, + "learning_rate": 1e-06, + "loss": 0.2134, + "mean_token_accuracy": 0.9345698952674866, + "num_tokens": 8719143.0, + "step": 2709 + }, + { + "epoch": 1.3861892583120206, + "grad_norm": 8.330885887145996, + "learning_rate": 1e-06, + "loss": 0.2152, + "mean_token_accuracy": 0.8948838710784912, + "num_tokens": 8722434.0, + "step": 2710 + }, + { + "epoch": 1.386700767263427, + "grad_norm": 10.841970443725586, + "learning_rate": 1e-06, + "loss": 0.308, + "mean_token_accuracy": 0.8909254670143127, + "num_tokens": 8725803.0, + "step": 2711 + }, + { + "epoch": 1.3872122762148338, + "grad_norm": 8.021007537841797, + "learning_rate": 1e-06, + "loss": 0.2239, + "mean_token_accuracy": 0.9038336277008057, + "num_tokens": 8729074.0, + "step": 2712 + }, + { + "epoch": 1.3877237851662403, + "grad_norm": 7.8609137535095215, + "learning_rate": 1e-06, + "loss": 0.1545, + "mean_token_accuracy": 0.9474061727523804, + "num_tokens": 8731912.0, + "step": 2713 + }, + { + "epoch": 1.388235294117647, + "grad_norm": 12.179813385009766, + "learning_rate": 1e-06, + "loss": 0.2482, + "mean_token_accuracy": 0.9030788540840149, + "num_tokens": 8734527.0, + "step": 2714 + }, + { + "epoch": 1.3887468030690537, + "grad_norm": 10.762114524841309, + "learning_rate": 1e-06, + "loss": 0.2389, + "mean_token_accuracy": 0.917340874671936, + "num_tokens": 8737070.0, + "step": 2715 + }, + { + "epoch": 1.3892583120204605, + "grad_norm": 8.83244800567627, + "learning_rate": 1e-06, + "loss": 0.2428, + "mean_token_accuracy": 0.881449818611145, + "num_tokens": 8740126.0, + "step": 2716 + }, + { + "epoch": 1.389769820971867, + "grad_norm": 10.64786434173584, + "learning_rate": 1e-06, + "loss": 0.3108, + "mean_token_accuracy": 0.8856152892112732, + "num_tokens": 8743297.0, + "step": 2717 + }, + { + "epoch": 1.3902813299232737, + "grad_norm": 12.25816822052002, + "learning_rate": 1e-06, + "loss": 0.3353, + "mean_token_accuracy": 0.8642745018005371, + "num_tokens": 8746468.0, + "step": 2718 + }, + { + "epoch": 1.3907928388746802, + "grad_norm": 8.408622741699219, + "learning_rate": 1e-06, + "loss": 0.1777, + "mean_token_accuracy": 0.9078578948974609, + "num_tokens": 8750036.0, + "step": 2719 + }, + { + "epoch": 1.391304347826087, + "grad_norm": 10.65027141571045, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8838843107223511, + "num_tokens": 8753533.0, + "step": 2720 + }, + { + "epoch": 1.3918158567774936, + "grad_norm": 8.568170547485352, + "learning_rate": 1e-06, + "loss": 0.1885, + "mean_token_accuracy": 0.9223646521568298, + "num_tokens": 8756929.0, + "step": 2721 + }, + { + "epoch": 1.3923273657289004, + "grad_norm": 8.362929344177246, + "learning_rate": 1e-06, + "loss": 0.1927, + "mean_token_accuracy": 0.9054615497589111, + "num_tokens": 8760329.0, + "step": 2722 + }, + { + "epoch": 1.3928388746803069, + "grad_norm": 11.048633575439453, + "learning_rate": 1e-06, + "loss": 0.2837, + "mean_token_accuracy": 0.8947738409042358, + "num_tokens": 8763818.0, + "step": 2723 + }, + { + "epoch": 1.3933503836317136, + "grad_norm": 9.053523063659668, + "learning_rate": 1e-06, + "loss": 0.2252, + "mean_token_accuracy": 0.906471848487854, + "num_tokens": 8767064.0, + "step": 2724 + }, + { + "epoch": 1.39386189258312, + "grad_norm": 7.258176326751709, + "learning_rate": 1e-06, + "loss": 0.1315, + "mean_token_accuracy": 0.9075444936752319, + "num_tokens": 8770253.0, + "step": 2725 + }, + { + "epoch": 1.3943734015345268, + "grad_norm": 12.148516654968262, + "learning_rate": 1e-06, + "loss": 0.277, + "mean_token_accuracy": 0.8873158693313599, + "num_tokens": 8773666.0, + "step": 2726 + }, + { + "epoch": 1.3948849104859335, + "grad_norm": 11.213422775268555, + "learning_rate": 1e-06, + "loss": 0.3373, + "mean_token_accuracy": 0.8553203344345093, + "num_tokens": 8776471.0, + "step": 2727 + }, + { + "epoch": 1.3953964194373403, + "grad_norm": 10.130282402038574, + "learning_rate": 1e-06, + "loss": 0.2642, + "mean_token_accuracy": 0.915960431098938, + "num_tokens": 8779848.0, + "step": 2728 + }, + { + "epoch": 1.3959079283887468, + "grad_norm": 8.17841911315918, + "learning_rate": 1e-06, + "loss": 0.2162, + "mean_token_accuracy": 0.9071304798126221, + "num_tokens": 8782477.0, + "step": 2729 + }, + { + "epoch": 1.3964194373401535, + "grad_norm": 9.134471893310547, + "learning_rate": 1e-06, + "loss": 0.2331, + "mean_token_accuracy": 0.9121652841567993, + "num_tokens": 8785616.0, + "step": 2730 + }, + { + "epoch": 1.39693094629156, + "grad_norm": 9.614777565002441, + "learning_rate": 1e-06, + "loss": 0.2604, + "mean_token_accuracy": 0.8760092258453369, + "num_tokens": 8788916.0, + "step": 2731 + }, + { + "epoch": 1.3974424552429667, + "grad_norm": 6.9807305335998535, + "learning_rate": 1e-06, + "loss": 0.1871, + "mean_token_accuracy": 0.9165738224983215, + "num_tokens": 8792307.0, + "step": 2732 + }, + { + "epoch": 1.3979539641943735, + "grad_norm": 7.519774436950684, + "learning_rate": 1e-06, + "loss": 0.1664, + "mean_token_accuracy": 0.9030117392539978, + "num_tokens": 8795928.0, + "step": 2733 + }, + { + "epoch": 1.3984654731457802, + "grad_norm": 9.232126235961914, + "learning_rate": 1e-06, + "loss": 0.2191, + "mean_token_accuracy": 0.8966174125671387, + "num_tokens": 8799539.0, + "step": 2734 + }, + { + "epoch": 1.3989769820971867, + "grad_norm": 10.255256652832031, + "learning_rate": 1e-06, + "loss": 0.2336, + "mean_token_accuracy": 0.9061307907104492, + "num_tokens": 8803112.0, + "step": 2735 + }, + { + "epoch": 1.3994884910485934, + "grad_norm": 11.790825843811035, + "learning_rate": 1e-06, + "loss": 0.3079, + "mean_token_accuracy": 0.8914765119552612, + "num_tokens": 8805856.0, + "step": 2736 + }, + { + "epoch": 1.4, + "grad_norm": 6.448992729187012, + "learning_rate": 1e-06, + "loss": 0.1979, + "mean_token_accuracy": 0.9016945362091064, + "num_tokens": 8809385.0, + "step": 2737 + }, + { + "epoch": 1.4005115089514066, + "grad_norm": 13.681587219238281, + "learning_rate": 1e-06, + "loss": 0.2668, + "mean_token_accuracy": 0.9024173021316528, + "num_tokens": 8812368.0, + "step": 2738 + }, + { + "epoch": 1.4010230179028134, + "grad_norm": 12.010673522949219, + "learning_rate": 1e-06, + "loss": 0.2831, + "mean_token_accuracy": 0.9013009071350098, + "num_tokens": 8814565.0, + "step": 2739 + }, + { + "epoch": 1.40153452685422, + "grad_norm": 9.025914192199707, + "learning_rate": 1e-06, + "loss": 0.2027, + "mean_token_accuracy": 0.8965970277786255, + "num_tokens": 8818382.0, + "step": 2740 + }, + { + "epoch": 1.4020460358056266, + "grad_norm": 21.254972457885742, + "learning_rate": 1e-06, + "loss": 0.3992, + "mean_token_accuracy": 0.8669582009315491, + "num_tokens": 8821301.0, + "step": 2741 + }, + { + "epoch": 1.4025575447570333, + "grad_norm": 10.09659481048584, + "learning_rate": 1e-06, + "loss": 0.2785, + "mean_token_accuracy": 0.8819496631622314, + "num_tokens": 8824590.0, + "step": 2742 + }, + { + "epoch": 1.4030690537084398, + "grad_norm": 12.48638916015625, + "learning_rate": 1e-06, + "loss": 0.3206, + "mean_token_accuracy": 0.8584789037704468, + "num_tokens": 8827288.0, + "step": 2743 + }, + { + "epoch": 1.4035805626598465, + "grad_norm": 9.844518661499023, + "learning_rate": 1e-06, + "loss": 0.2625, + "mean_token_accuracy": 0.8987595438957214, + "num_tokens": 8830905.0, + "step": 2744 + }, + { + "epoch": 1.4040920716112533, + "grad_norm": 7.709944248199463, + "learning_rate": 1e-06, + "loss": 0.1702, + "mean_token_accuracy": 0.9293867349624634, + "num_tokens": 8834772.0, + "step": 2745 + }, + { + "epoch": 1.4046035805626598, + "grad_norm": 9.901899337768555, + "learning_rate": 1e-06, + "loss": 0.2519, + "mean_token_accuracy": 0.8848921656608582, + "num_tokens": 8838453.0, + "step": 2746 + }, + { + "epoch": 1.4051150895140665, + "grad_norm": 7.157768726348877, + "learning_rate": 1e-06, + "loss": 0.1653, + "mean_token_accuracy": 0.9137787818908691, + "num_tokens": 8842074.0, + "step": 2747 + }, + { + "epoch": 1.4056265984654732, + "grad_norm": 7.153957843780518, + "learning_rate": 1e-06, + "loss": 0.1943, + "mean_token_accuracy": 0.8888072371482849, + "num_tokens": 8845403.0, + "step": 2748 + }, + { + "epoch": 1.4061381074168797, + "grad_norm": 13.351516723632812, + "learning_rate": 1e-06, + "loss": 0.3111, + "mean_token_accuracy": 0.8939145803451538, + "num_tokens": 8848623.0, + "step": 2749 + }, + { + "epoch": 1.4066496163682864, + "grad_norm": 10.659113883972168, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8918051719665527, + "num_tokens": 8851940.0, + "step": 2750 + }, + { + "epoch": 1.4071611253196932, + "grad_norm": 5.952152252197266, + "learning_rate": 1e-06, + "loss": 0.1003, + "mean_token_accuracy": 0.9366258978843689, + "num_tokens": 8855655.0, + "step": 2751 + }, + { + "epoch": 1.4076726342710997, + "grad_norm": 6.97130823135376, + "learning_rate": 1e-06, + "loss": 0.1574, + "mean_token_accuracy": 0.9436094164848328, + "num_tokens": 8858367.0, + "step": 2752 + }, + { + "epoch": 1.4081841432225064, + "grad_norm": 10.010527610778809, + "learning_rate": 1e-06, + "loss": 0.2114, + "mean_token_accuracy": 0.92664635181427, + "num_tokens": 8861721.0, + "step": 2753 + }, + { + "epoch": 1.4086956521739131, + "grad_norm": 9.520807266235352, + "learning_rate": 1e-06, + "loss": 0.2938, + "mean_token_accuracy": 0.8825892210006714, + "num_tokens": 8864975.0, + "step": 2754 + }, + { + "epoch": 1.4092071611253196, + "grad_norm": 10.255073547363281, + "learning_rate": 1e-06, + "loss": 0.3156, + "mean_token_accuracy": 0.8664449453353882, + "num_tokens": 8868017.0, + "step": 2755 + }, + { + "epoch": 1.4097186700767264, + "grad_norm": 10.560738563537598, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.8949314951896667, + "num_tokens": 8870515.0, + "step": 2756 + }, + { + "epoch": 1.410230179028133, + "grad_norm": 10.056426048278809, + "learning_rate": 1e-06, + "loss": 0.2388, + "mean_token_accuracy": 0.9036006927490234, + "num_tokens": 8873448.0, + "step": 2757 + }, + { + "epoch": 1.4107416879795396, + "grad_norm": 11.6856107711792, + "learning_rate": 1e-06, + "loss": 0.2749, + "mean_token_accuracy": 0.9013816118240356, + "num_tokens": 8876824.0, + "step": 2758 + }, + { + "epoch": 1.4112531969309463, + "grad_norm": 6.322577476501465, + "learning_rate": 1e-06, + "loss": 0.1605, + "mean_token_accuracy": 0.9310145974159241, + "num_tokens": 8879492.0, + "step": 2759 + }, + { + "epoch": 1.4117647058823528, + "grad_norm": 8.238568305969238, + "learning_rate": 1e-06, + "loss": 0.1316, + "mean_token_accuracy": 0.9504259824752808, + "num_tokens": 8882820.0, + "step": 2760 + }, + { + "epoch": 1.4122762148337595, + "grad_norm": 9.211566925048828, + "learning_rate": 1e-06, + "loss": 0.3131, + "mean_token_accuracy": 0.8913086652755737, + "num_tokens": 8886054.0, + "step": 2761 + }, + { + "epoch": 1.4127877237851663, + "grad_norm": 6.62302827835083, + "learning_rate": 1e-06, + "loss": 0.1637, + "mean_token_accuracy": 0.9384276866912842, + "num_tokens": 8890087.0, + "step": 2762 + }, + { + "epoch": 1.413299232736573, + "grad_norm": 9.221421241760254, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.8812136650085449, + "num_tokens": 8892457.0, + "step": 2763 + }, + { + "epoch": 1.4138107416879795, + "grad_norm": 15.06576919555664, + "learning_rate": 1e-06, + "loss": 0.255, + "mean_token_accuracy": 0.9033163785934448, + "num_tokens": 8895379.0, + "step": 2764 + }, + { + "epoch": 1.4143222506393862, + "grad_norm": 8.146806716918945, + "learning_rate": 1e-06, + "loss": 0.2283, + "mean_token_accuracy": 0.921387791633606, + "num_tokens": 8898354.0, + "step": 2765 + }, + { + "epoch": 1.4148337595907927, + "grad_norm": 8.775912284851074, + "learning_rate": 1e-06, + "loss": 0.2077, + "mean_token_accuracy": 0.892532467842102, + "num_tokens": 8902233.0, + "step": 2766 + }, + { + "epoch": 1.4153452685421994, + "grad_norm": 11.746432304382324, + "learning_rate": 1e-06, + "loss": 0.317, + "mean_token_accuracy": 0.865138590335846, + "num_tokens": 8905218.0, + "step": 2767 + }, + { + "epoch": 1.4158567774936062, + "grad_norm": 8.21529483795166, + "learning_rate": 1e-06, + "loss": 0.2115, + "mean_token_accuracy": 0.9192121028900146, + "num_tokens": 8909463.0, + "step": 2768 + }, + { + "epoch": 1.416368286445013, + "grad_norm": 11.867725372314453, + "learning_rate": 1e-06, + "loss": 0.226, + "mean_token_accuracy": 0.921469509601593, + "num_tokens": 8912911.0, + "step": 2769 + }, + { + "epoch": 1.4168797953964194, + "grad_norm": 11.195525169372559, + "learning_rate": 1e-06, + "loss": 0.24, + "mean_token_accuracy": 0.9075682163238525, + "num_tokens": 8915835.0, + "step": 2770 + }, + { + "epoch": 1.4173913043478261, + "grad_norm": 7.430218696594238, + "learning_rate": 1e-06, + "loss": 0.1713, + "mean_token_accuracy": 0.9178112149238586, + "num_tokens": 8919543.0, + "step": 2771 + }, + { + "epoch": 1.4179028132992326, + "grad_norm": 7.796029090881348, + "learning_rate": 1e-06, + "loss": 0.2212, + "mean_token_accuracy": 0.9090335369110107, + "num_tokens": 8921869.0, + "step": 2772 + }, + { + "epoch": 1.4184143222506393, + "grad_norm": 9.627340316772461, + "learning_rate": 1e-06, + "loss": 0.2491, + "mean_token_accuracy": 0.9148342609405518, + "num_tokens": 8924275.0, + "step": 2773 + }, + { + "epoch": 1.418925831202046, + "grad_norm": 9.412725448608398, + "learning_rate": 1e-06, + "loss": 0.2575, + "mean_token_accuracy": 0.8979384899139404, + "num_tokens": 8927267.0, + "step": 2774 + }, + { + "epoch": 1.4194373401534528, + "grad_norm": 9.885931968688965, + "learning_rate": 1e-06, + "loss": 0.2654, + "mean_token_accuracy": 0.8580262064933777, + "num_tokens": 8930442.0, + "step": 2775 + }, + { + "epoch": 1.4199488491048593, + "grad_norm": 10.269948959350586, + "learning_rate": 1e-06, + "loss": 0.1996, + "mean_token_accuracy": 0.909332275390625, + "num_tokens": 8934224.0, + "step": 2776 + }, + { + "epoch": 1.420460358056266, + "grad_norm": 10.622692108154297, + "learning_rate": 1e-06, + "loss": 0.2862, + "mean_token_accuracy": 0.892867922782898, + "num_tokens": 8937790.0, + "step": 2777 + }, + { + "epoch": 1.4209718670076725, + "grad_norm": 12.444400787353516, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8750519752502441, + "num_tokens": 8940778.0, + "step": 2778 + }, + { + "epoch": 1.4214833759590793, + "grad_norm": 10.421502113342285, + "learning_rate": 1e-06, + "loss": 0.2375, + "mean_token_accuracy": 0.8978806734085083, + "num_tokens": 8943569.0, + "step": 2779 + }, + { + "epoch": 1.421994884910486, + "grad_norm": 6.84736967086792, + "learning_rate": 1e-06, + "loss": 0.1986, + "mean_token_accuracy": 0.9064544439315796, + "num_tokens": 8946845.0, + "step": 2780 + }, + { + "epoch": 1.4225063938618927, + "grad_norm": 10.402934074401855, + "learning_rate": 1e-06, + "loss": 0.2169, + "mean_token_accuracy": 0.9202649593353271, + "num_tokens": 8949971.0, + "step": 2781 + }, + { + "epoch": 1.4230179028132992, + "grad_norm": 9.424779891967773, + "learning_rate": 1e-06, + "loss": 0.249, + "mean_token_accuracy": 0.9051821231842041, + "num_tokens": 8952584.0, + "step": 2782 + }, + { + "epoch": 1.423529411764706, + "grad_norm": 12.162043571472168, + "learning_rate": 1e-06, + "loss": 0.3489, + "mean_token_accuracy": 0.8394032716751099, + "num_tokens": 8956278.0, + "step": 2783 + }, + { + "epoch": 1.4240409207161124, + "grad_norm": 9.804418563842773, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.9101330041885376, + "num_tokens": 8959658.0, + "step": 2784 + }, + { + "epoch": 1.4245524296675192, + "grad_norm": 10.837544441223145, + "learning_rate": 1e-06, + "loss": 0.2069, + "mean_token_accuracy": 0.9215554594993591, + "num_tokens": 8962883.0, + "step": 2785 + }, + { + "epoch": 1.4250639386189259, + "grad_norm": 8.180221557617188, + "learning_rate": 1e-06, + "loss": 0.1952, + "mean_token_accuracy": 0.9098141193389893, + "num_tokens": 8966753.0, + "step": 2786 + }, + { + "epoch": 1.4255754475703326, + "grad_norm": 12.421046257019043, + "learning_rate": 1e-06, + "loss": 0.3169, + "mean_token_accuracy": 0.8899506330490112, + "num_tokens": 8969728.0, + "step": 2787 + }, + { + "epoch": 1.4260869565217391, + "grad_norm": 11.14476490020752, + "learning_rate": 1e-06, + "loss": 0.326, + "mean_token_accuracy": 0.9024215936660767, + "num_tokens": 8972303.0, + "step": 2788 + }, + { + "epoch": 1.4265984654731458, + "grad_norm": 10.709599494934082, + "learning_rate": 1e-06, + "loss": 0.2619, + "mean_token_accuracy": 0.9044545888900757, + "num_tokens": 8974970.0, + "step": 2789 + }, + { + "epoch": 1.4271099744245523, + "grad_norm": 9.519298553466797, + "learning_rate": 1e-06, + "loss": 0.2534, + "mean_token_accuracy": 0.9064401388168335, + "num_tokens": 8978406.0, + "step": 2790 + }, + { + "epoch": 1.427621483375959, + "grad_norm": 7.509286403656006, + "learning_rate": 1e-06, + "loss": 0.1702, + "mean_token_accuracy": 0.9491022825241089, + "num_tokens": 8981875.0, + "step": 2791 + }, + { + "epoch": 1.4281329923273658, + "grad_norm": 14.584221839904785, + "learning_rate": 1e-06, + "loss": 0.3688, + "mean_token_accuracy": 0.8761628866195679, + "num_tokens": 8985022.0, + "step": 2792 + }, + { + "epoch": 1.4286445012787725, + "grad_norm": 13.11989974975586, + "learning_rate": 1e-06, + "loss": 0.3329, + "mean_token_accuracy": 0.8601464033126831, + "num_tokens": 8987762.0, + "step": 2793 + }, + { + "epoch": 1.429156010230179, + "grad_norm": 9.428189277648926, + "learning_rate": 1e-06, + "loss": 0.255, + "mean_token_accuracy": 0.8735226392745972, + "num_tokens": 8990968.0, + "step": 2794 + }, + { + "epoch": 1.4296675191815857, + "grad_norm": 6.257648944854736, + "learning_rate": 1e-06, + "loss": 0.1573, + "mean_token_accuracy": 0.9203839302062988, + "num_tokens": 8994072.0, + "step": 2795 + }, + { + "epoch": 1.4301790281329922, + "grad_norm": 8.29046630859375, + "learning_rate": 1e-06, + "loss": 0.226, + "mean_token_accuracy": 0.9102915525436401, + "num_tokens": 8997214.0, + "step": 2796 + }, + { + "epoch": 1.430690537084399, + "grad_norm": 11.293846130371094, + "learning_rate": 1e-06, + "loss": 0.2865, + "mean_token_accuracy": 0.8950104713439941, + "num_tokens": 8999876.0, + "step": 2797 + }, + { + "epoch": 1.4312020460358057, + "grad_norm": 9.171022415161133, + "learning_rate": 1e-06, + "loss": 0.1822, + "mean_token_accuracy": 0.9272085428237915, + "num_tokens": 9003311.0, + "step": 2798 + }, + { + "epoch": 1.4317135549872122, + "grad_norm": 8.750227928161621, + "learning_rate": 1e-06, + "loss": 0.223, + "mean_token_accuracy": 0.9059319496154785, + "num_tokens": 9006399.0, + "step": 2799 + }, + { + "epoch": 1.432225063938619, + "grad_norm": 4.947676181793213, + "learning_rate": 1e-06, + "loss": 0.153, + "mean_token_accuracy": 0.9025083184242249, + "num_tokens": 9009904.0, + "step": 2800 + }, + { + "epoch": 1.4327365728900257, + "grad_norm": 11.752762794494629, + "learning_rate": 1e-06, + "loss": 0.2037, + "mean_token_accuracy": 0.8992236256599426, + "num_tokens": 9013273.0, + "step": 2801 + }, + { + "epoch": 1.4332480818414322, + "grad_norm": 12.948911666870117, + "learning_rate": 1e-06, + "loss": 0.3464, + "mean_token_accuracy": 0.8629941344261169, + "num_tokens": 9016886.0, + "step": 2802 + }, + { + "epoch": 1.4337595907928389, + "grad_norm": 9.061339378356934, + "learning_rate": 1e-06, + "loss": 0.1912, + "mean_token_accuracy": 0.8955203294754028, + "num_tokens": 9020794.0, + "step": 2803 + }, + { + "epoch": 1.4342710997442456, + "grad_norm": 8.940227508544922, + "learning_rate": 1e-06, + "loss": 0.1974, + "mean_token_accuracy": 0.9335438013076782, + "num_tokens": 9023645.0, + "step": 2804 + }, + { + "epoch": 1.434782608695652, + "grad_norm": 5.8720574378967285, + "learning_rate": 1e-06, + "loss": 0.1199, + "mean_token_accuracy": 0.9357694387435913, + "num_tokens": 9027839.0, + "step": 2805 + }, + { + "epoch": 1.4352941176470588, + "grad_norm": 10.615172386169434, + "learning_rate": 1e-06, + "loss": 0.2338, + "mean_token_accuracy": 0.8900771141052246, + "num_tokens": 9031027.0, + "step": 2806 + }, + { + "epoch": 1.4358056265984656, + "grad_norm": 11.108299255371094, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8894472718238831, + "num_tokens": 9034126.0, + "step": 2807 + }, + { + "epoch": 1.436317135549872, + "grad_norm": 11.8408203125, + "learning_rate": 1e-06, + "loss": 0.3107, + "mean_token_accuracy": 0.8943365216255188, + "num_tokens": 9037150.0, + "step": 2808 + }, + { + "epoch": 1.4368286445012788, + "grad_norm": 10.66466236114502, + "learning_rate": 1e-06, + "loss": 0.2494, + "mean_token_accuracy": 0.9209878444671631, + "num_tokens": 9040541.0, + "step": 2809 + }, + { + "epoch": 1.4373401534526855, + "grad_norm": 10.94638442993164, + "learning_rate": 1e-06, + "loss": 0.2522, + "mean_token_accuracy": 0.9057635068893433, + "num_tokens": 9043690.0, + "step": 2810 + }, + { + "epoch": 1.437851662404092, + "grad_norm": 12.08336353302002, + "learning_rate": 1e-06, + "loss": 0.3152, + "mean_token_accuracy": 0.8925125598907471, + "num_tokens": 9046524.0, + "step": 2811 + }, + { + "epoch": 1.4383631713554987, + "grad_norm": 9.965158462524414, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.8776451349258423, + "num_tokens": 9049927.0, + "step": 2812 + }, + { + "epoch": 1.4388746803069052, + "grad_norm": 7.971109867095947, + "learning_rate": 1e-06, + "loss": 0.1875, + "mean_token_accuracy": 0.934290885925293, + "num_tokens": 9053356.0, + "step": 2813 + }, + { + "epoch": 1.439386189258312, + "grad_norm": 11.47408390045166, + "learning_rate": 1e-06, + "loss": 0.2766, + "mean_token_accuracy": 0.9056003093719482, + "num_tokens": 9055922.0, + "step": 2814 + }, + { + "epoch": 1.4398976982097187, + "grad_norm": 13.410865783691406, + "learning_rate": 1e-06, + "loss": 0.3463, + "mean_token_accuracy": 0.8657436370849609, + "num_tokens": 9058955.0, + "step": 2815 + }, + { + "epoch": 1.4404092071611254, + "grad_norm": 12.984672546386719, + "learning_rate": 1e-06, + "loss": 0.2557, + "mean_token_accuracy": 0.8943055868148804, + "num_tokens": 9061353.0, + "step": 2816 + }, + { + "epoch": 1.440920716112532, + "grad_norm": 9.0365571975708, + "learning_rate": 1e-06, + "loss": 0.2321, + "mean_token_accuracy": 0.9013487696647644, + "num_tokens": 9064487.0, + "step": 2817 + }, + { + "epoch": 1.4414322250639386, + "grad_norm": 7.408951759338379, + "learning_rate": 1e-06, + "loss": 0.2023, + "mean_token_accuracy": 0.9162425994873047, + "num_tokens": 9068223.0, + "step": 2818 + }, + { + "epoch": 1.4419437340153451, + "grad_norm": 8.96444034576416, + "learning_rate": 1e-06, + "loss": 0.2238, + "mean_token_accuracy": 0.9024150967597961, + "num_tokens": 9071484.0, + "step": 2819 + }, + { + "epoch": 1.4424552429667519, + "grad_norm": 7.569674968719482, + "learning_rate": 1e-06, + "loss": 0.2366, + "mean_token_accuracy": 0.8968055248260498, + "num_tokens": 9075465.0, + "step": 2820 + }, + { + "epoch": 1.4429667519181586, + "grad_norm": 10.163558959960938, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.8767213225364685, + "num_tokens": 9078168.0, + "step": 2821 + }, + { + "epoch": 1.4434782608695653, + "grad_norm": 11.203125, + "learning_rate": 1e-06, + "loss": 0.2592, + "mean_token_accuracy": 0.8983762860298157, + "num_tokens": 9081043.0, + "step": 2822 + }, + { + "epoch": 1.4439897698209718, + "grad_norm": 10.254469871520996, + "learning_rate": 1e-06, + "loss": 0.25, + "mean_token_accuracy": 0.9182907342910767, + "num_tokens": 9084360.0, + "step": 2823 + }, + { + "epoch": 1.4445012787723785, + "grad_norm": 7.268659591674805, + "learning_rate": 1e-06, + "loss": 0.1509, + "mean_token_accuracy": 0.9278776049613953, + "num_tokens": 9088462.0, + "step": 2824 + }, + { + "epoch": 1.445012787723785, + "grad_norm": 7.277657985687256, + "learning_rate": 1e-06, + "loss": 0.1942, + "mean_token_accuracy": 0.9188940525054932, + "num_tokens": 9091672.0, + "step": 2825 + }, + { + "epoch": 1.4455242966751918, + "grad_norm": 13.012499809265137, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8715893626213074, + "num_tokens": 9095030.0, + "step": 2826 + }, + { + "epoch": 1.4460358056265985, + "grad_norm": 15.095736503601074, + "learning_rate": 1e-06, + "loss": 0.3175, + "mean_token_accuracy": 0.9039100408554077, + "num_tokens": 9098552.0, + "step": 2827 + }, + { + "epoch": 1.4465473145780052, + "grad_norm": 11.475841522216797, + "learning_rate": 1e-06, + "loss": 0.3099, + "mean_token_accuracy": 0.9068273901939392, + "num_tokens": 9102135.0, + "step": 2828 + }, + { + "epoch": 1.4470588235294117, + "grad_norm": 13.38649845123291, + "learning_rate": 1e-06, + "loss": 0.2369, + "mean_token_accuracy": 0.9030575752258301, + "num_tokens": 9105555.0, + "step": 2829 + }, + { + "epoch": 1.4475703324808185, + "grad_norm": 14.717737197875977, + "learning_rate": 1e-06, + "loss": 0.3938, + "mean_token_accuracy": 0.8867205381393433, + "num_tokens": 9109587.0, + "step": 2830 + }, + { + "epoch": 1.448081841432225, + "grad_norm": 14.103325843811035, + "learning_rate": 1e-06, + "loss": 0.2902, + "mean_token_accuracy": 0.9024202823638916, + "num_tokens": 9112265.0, + "step": 2831 + }, + { + "epoch": 1.4485933503836317, + "grad_norm": 9.399691581726074, + "learning_rate": 1e-06, + "loss": 0.1823, + "mean_token_accuracy": 0.943244218826294, + "num_tokens": 9115912.0, + "step": 2832 + }, + { + "epoch": 1.4491048593350384, + "grad_norm": 7.640307903289795, + "learning_rate": 1e-06, + "loss": 0.213, + "mean_token_accuracy": 0.8910161256790161, + "num_tokens": 9118891.0, + "step": 2833 + }, + { + "epoch": 1.4496163682864451, + "grad_norm": 8.752928733825684, + "learning_rate": 1e-06, + "loss": 0.2629, + "mean_token_accuracy": 0.9020704030990601, + "num_tokens": 9121971.0, + "step": 2834 + }, + { + "epoch": 1.4501278772378516, + "grad_norm": 8.688077926635742, + "learning_rate": 1e-06, + "loss": 0.2514, + "mean_token_accuracy": 0.9150344729423523, + "num_tokens": 9125592.0, + "step": 2835 + }, + { + "epoch": 1.4506393861892584, + "grad_norm": 8.449121475219727, + "learning_rate": 1e-06, + "loss": 0.2148, + "mean_token_accuracy": 0.9288302659988403, + "num_tokens": 9128004.0, + "step": 2836 + }, + { + "epoch": 1.4511508951406649, + "grad_norm": 9.790771484375, + "learning_rate": 1e-06, + "loss": 0.2515, + "mean_token_accuracy": 0.8929462432861328, + "num_tokens": 9131997.0, + "step": 2837 + }, + { + "epoch": 1.4516624040920716, + "grad_norm": 8.838275909423828, + "learning_rate": 1e-06, + "loss": 0.2651, + "mean_token_accuracy": 0.8832376599311829, + "num_tokens": 9135155.0, + "step": 2838 + }, + { + "epoch": 1.4521739130434783, + "grad_norm": 5.3308000564575195, + "learning_rate": 1e-06, + "loss": 0.1606, + "mean_token_accuracy": 0.9273087382316589, + "num_tokens": 9138260.0, + "step": 2839 + }, + { + "epoch": 1.452685421994885, + "grad_norm": 11.600654602050781, + "learning_rate": 1e-06, + "loss": 0.3017, + "mean_token_accuracy": 0.889359712600708, + "num_tokens": 9140978.0, + "step": 2840 + }, + { + "epoch": 1.4531969309462915, + "grad_norm": 9.153985977172852, + "learning_rate": 1e-06, + "loss": 0.3493, + "mean_token_accuracy": 0.849410355091095, + "num_tokens": 9144709.0, + "step": 2841 + }, + { + "epoch": 1.4537084398976983, + "grad_norm": 5.5438947677612305, + "learning_rate": 1e-06, + "loss": 0.1629, + "mean_token_accuracy": 0.9240304231643677, + "num_tokens": 9148765.0, + "step": 2842 + }, + { + "epoch": 1.4542199488491048, + "grad_norm": 9.963811874389648, + "learning_rate": 1e-06, + "loss": 0.2033, + "mean_token_accuracy": 0.9265329837799072, + "num_tokens": 9151614.0, + "step": 2843 + }, + { + "epoch": 1.4547314578005115, + "grad_norm": 10.357684135437012, + "learning_rate": 1e-06, + "loss": 0.2809, + "mean_token_accuracy": 0.9111241102218628, + "num_tokens": 9154511.0, + "step": 2844 + }, + { + "epoch": 1.4552429667519182, + "grad_norm": 7.516500473022461, + "learning_rate": 1e-06, + "loss": 0.2261, + "mean_token_accuracy": 0.9079405069351196, + "num_tokens": 9157898.0, + "step": 2845 + }, + { + "epoch": 1.455754475703325, + "grad_norm": 16.889925003051758, + "learning_rate": 1e-06, + "loss": 0.381, + "mean_token_accuracy": 0.8681473731994629, + "num_tokens": 9160701.0, + "step": 2846 + }, + { + "epoch": 1.4562659846547314, + "grad_norm": 6.805542469024658, + "learning_rate": 1e-06, + "loss": 0.2031, + "mean_token_accuracy": 0.924920916557312, + "num_tokens": 9164038.0, + "step": 2847 + }, + { + "epoch": 1.4567774936061382, + "grad_norm": 8.8577880859375, + "learning_rate": 1e-06, + "loss": 0.2445, + "mean_token_accuracy": 0.916351318359375, + "num_tokens": 9166868.0, + "step": 2848 + }, + { + "epoch": 1.4572890025575447, + "grad_norm": 12.000168800354004, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8823493719100952, + "num_tokens": 9169594.0, + "step": 2849 + }, + { + "epoch": 1.4578005115089514, + "grad_norm": 9.867874145507812, + "learning_rate": 1e-06, + "loss": 0.2312, + "mean_token_accuracy": 0.908149242401123, + "num_tokens": 9172288.0, + "step": 2850 + }, + { + "epoch": 1.4583120204603581, + "grad_norm": 13.05994987487793, + "learning_rate": 1e-06, + "loss": 0.3276, + "mean_token_accuracy": 0.8671819567680359, + "num_tokens": 9174840.0, + "step": 2851 + }, + { + "epoch": 1.4588235294117646, + "grad_norm": 7.663493633270264, + "learning_rate": 1e-06, + "loss": 0.2274, + "mean_token_accuracy": 0.8858671188354492, + "num_tokens": 9177815.0, + "step": 2852 + }, + { + "epoch": 1.4593350383631714, + "grad_norm": 6.872354507446289, + "learning_rate": 1e-06, + "loss": 0.2057, + "mean_token_accuracy": 0.8955850005149841, + "num_tokens": 9181954.0, + "step": 2853 + }, + { + "epoch": 1.459846547314578, + "grad_norm": 6.824888229370117, + "learning_rate": 1e-06, + "loss": 0.1881, + "mean_token_accuracy": 0.8967519998550415, + "num_tokens": 9185173.0, + "step": 2854 + }, + { + "epoch": 1.4603580562659846, + "grad_norm": 9.021980285644531, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8826198577880859, + "num_tokens": 9187594.0, + "step": 2855 + }, + { + "epoch": 1.4608695652173913, + "grad_norm": 9.742558479309082, + "learning_rate": 1e-06, + "loss": 0.2217, + "mean_token_accuracy": 0.8816282749176025, + "num_tokens": 9190426.0, + "step": 2856 + }, + { + "epoch": 1.461381074168798, + "grad_norm": 7.477635860443115, + "learning_rate": 1e-06, + "loss": 0.153, + "mean_token_accuracy": 0.9195733666419983, + "num_tokens": 9194061.0, + "step": 2857 + }, + { + "epoch": 1.4618925831202045, + "grad_norm": 9.783860206604004, + "learning_rate": 1e-06, + "loss": 0.2554, + "mean_token_accuracy": 0.9048619270324707, + "num_tokens": 9197630.0, + "step": 2858 + }, + { + "epoch": 1.4624040920716113, + "grad_norm": 7.474545955657959, + "learning_rate": 1e-06, + "loss": 0.1977, + "mean_token_accuracy": 0.8873283863067627, + "num_tokens": 9201565.0, + "step": 2859 + }, + { + "epoch": 1.4629156010230178, + "grad_norm": 7.820626258850098, + "learning_rate": 1e-06, + "loss": 0.2457, + "mean_token_accuracy": 0.9103886485099792, + "num_tokens": 9204552.0, + "step": 2860 + }, + { + "epoch": 1.4634271099744245, + "grad_norm": 8.684306144714355, + "learning_rate": 1e-06, + "loss": 0.1773, + "mean_token_accuracy": 0.9053130745887756, + "num_tokens": 9207470.0, + "step": 2861 + }, + { + "epoch": 1.4639386189258312, + "grad_norm": 9.956319808959961, + "learning_rate": 1e-06, + "loss": 0.2557, + "mean_token_accuracy": 0.887955904006958, + "num_tokens": 9210059.0, + "step": 2862 + }, + { + "epoch": 1.464450127877238, + "grad_norm": 7.712399005889893, + "learning_rate": 1e-06, + "loss": 0.1575, + "mean_token_accuracy": 0.928320586681366, + "num_tokens": 9213431.0, + "step": 2863 + }, + { + "epoch": 1.4649616368286444, + "grad_norm": 11.056105613708496, + "learning_rate": 1e-06, + "loss": 0.1942, + "mean_token_accuracy": 0.8917320966720581, + "num_tokens": 9216578.0, + "step": 2864 + }, + { + "epoch": 1.4654731457800512, + "grad_norm": 9.219926834106445, + "learning_rate": 1e-06, + "loss": 0.2341, + "mean_token_accuracy": 0.9015806913375854, + "num_tokens": 9219276.0, + "step": 2865 + }, + { + "epoch": 1.4659846547314577, + "grad_norm": 6.963474750518799, + "learning_rate": 1e-06, + "loss": 0.1762, + "mean_token_accuracy": 0.9298791885375977, + "num_tokens": 9222182.0, + "step": 2866 + }, + { + "epoch": 1.4664961636828644, + "grad_norm": 14.097882270812988, + "learning_rate": 1e-06, + "loss": 0.2535, + "mean_token_accuracy": 0.8683000802993774, + "num_tokens": 9224904.0, + "step": 2867 + }, + { + "epoch": 1.4670076726342711, + "grad_norm": 12.143342971801758, + "learning_rate": 1e-06, + "loss": 0.2832, + "mean_token_accuracy": 0.8736855983734131, + "num_tokens": 9229221.0, + "step": 2868 + }, + { + "epoch": 1.4675191815856778, + "grad_norm": 10.034734725952148, + "learning_rate": 1e-06, + "loss": 0.2844, + "mean_token_accuracy": 0.8834766149520874, + "num_tokens": 9231923.0, + "step": 2869 + }, + { + "epoch": 1.4680306905370843, + "grad_norm": 17.00454330444336, + "learning_rate": 1e-06, + "loss": 0.3551, + "mean_token_accuracy": 0.8979614973068237, + "num_tokens": 9234517.0, + "step": 2870 + }, + { + "epoch": 1.468542199488491, + "grad_norm": 9.529804229736328, + "learning_rate": 1e-06, + "loss": 0.1997, + "mean_token_accuracy": 0.92452073097229, + "num_tokens": 9237914.0, + "step": 2871 + }, + { + "epoch": 1.4690537084398976, + "grad_norm": 8.583544731140137, + "learning_rate": 1e-06, + "loss": 0.2435, + "mean_token_accuracy": 0.898933470249176, + "num_tokens": 9241019.0, + "step": 2872 + }, + { + "epoch": 1.4695652173913043, + "grad_norm": 8.568554878234863, + "learning_rate": 1e-06, + "loss": 0.2598, + "mean_token_accuracy": 0.8979400396347046, + "num_tokens": 9243621.0, + "step": 2873 + }, + { + "epoch": 1.470076726342711, + "grad_norm": 8.389328002929688, + "learning_rate": 1e-06, + "loss": 0.2106, + "mean_token_accuracy": 0.9147839546203613, + "num_tokens": 9246722.0, + "step": 2874 + }, + { + "epoch": 1.4705882352941178, + "grad_norm": 7.027134895324707, + "learning_rate": 1e-06, + "loss": 0.1949, + "mean_token_accuracy": 0.9286389350891113, + "num_tokens": 9249643.0, + "step": 2875 + }, + { + "epoch": 1.4710997442455243, + "grad_norm": 8.626124382019043, + "learning_rate": 1e-06, + "loss": 0.1884, + "mean_token_accuracy": 0.8946614861488342, + "num_tokens": 9252837.0, + "step": 2876 + }, + { + "epoch": 1.471611253196931, + "grad_norm": 7.631728649139404, + "learning_rate": 1e-06, + "loss": 0.1637, + "mean_token_accuracy": 0.92271888256073, + "num_tokens": 9255950.0, + "step": 2877 + }, + { + "epoch": 1.4721227621483375, + "grad_norm": 9.829291343688965, + "learning_rate": 1e-06, + "loss": 0.1967, + "mean_token_accuracy": 0.9058042764663696, + "num_tokens": 9258869.0, + "step": 2878 + }, + { + "epoch": 1.4726342710997442, + "grad_norm": 11.901947975158691, + "learning_rate": 1e-06, + "loss": 0.243, + "mean_token_accuracy": 0.9060786366462708, + "num_tokens": 9262019.0, + "step": 2879 + }, + { + "epoch": 1.473145780051151, + "grad_norm": 10.173417091369629, + "learning_rate": 1e-06, + "loss": 0.2623, + "mean_token_accuracy": 0.8861643671989441, + "num_tokens": 9264853.0, + "step": 2880 + }, + { + "epoch": 1.4736572890025577, + "grad_norm": 11.181746482849121, + "learning_rate": 1e-06, + "loss": 0.3501, + "mean_token_accuracy": 0.8847341537475586, + "num_tokens": 9267367.0, + "step": 2881 + }, + { + "epoch": 1.4741687979539642, + "grad_norm": 14.126924514770508, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.8780984878540039, + "num_tokens": 9270738.0, + "step": 2882 + }, + { + "epoch": 1.4746803069053709, + "grad_norm": 11.228874206542969, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.9053456783294678, + "num_tokens": 9273803.0, + "step": 2883 + }, + { + "epoch": 1.4751918158567774, + "grad_norm": 12.00344467163086, + "learning_rate": 1e-06, + "loss": 0.2441, + "mean_token_accuracy": 0.8836091160774231, + "num_tokens": 9276683.0, + "step": 2884 + }, + { + "epoch": 1.4757033248081841, + "grad_norm": 12.989521026611328, + "learning_rate": 1e-06, + "loss": 0.241, + "mean_token_accuracy": 0.9011324644088745, + "num_tokens": 9280758.0, + "step": 2885 + }, + { + "epoch": 1.4762148337595908, + "grad_norm": 8.198751449584961, + "learning_rate": 1e-06, + "loss": 0.1676, + "mean_token_accuracy": 0.9306782484054565, + "num_tokens": 9283627.0, + "step": 2886 + }, + { + "epoch": 1.4767263427109976, + "grad_norm": 14.133607864379883, + "learning_rate": 1e-06, + "loss": 0.325, + "mean_token_accuracy": 0.8767179250717163, + "num_tokens": 9286145.0, + "step": 2887 + }, + { + "epoch": 1.477237851662404, + "grad_norm": 10.89964771270752, + "learning_rate": 1e-06, + "loss": 0.2919, + "mean_token_accuracy": 0.8755694031715393, + "num_tokens": 9289548.0, + "step": 2888 + }, + { + "epoch": 1.4777493606138108, + "grad_norm": 15.049982070922852, + "learning_rate": 1e-06, + "loss": 0.2857, + "mean_token_accuracy": 0.9024688005447388, + "num_tokens": 9293042.0, + "step": 2889 + }, + { + "epoch": 1.4782608695652173, + "grad_norm": 8.550556182861328, + "learning_rate": 1e-06, + "loss": 0.1834, + "mean_token_accuracy": 0.9040482640266418, + "num_tokens": 9296754.0, + "step": 2890 + }, + { + "epoch": 1.478772378516624, + "grad_norm": 11.225390434265137, + "learning_rate": 1e-06, + "loss": 0.2758, + "mean_token_accuracy": 0.8934040069580078, + "num_tokens": 9299661.0, + "step": 2891 + }, + { + "epoch": 1.4792838874680307, + "grad_norm": 10.093518257141113, + "learning_rate": 1e-06, + "loss": 0.2665, + "mean_token_accuracy": 0.9147257804870605, + "num_tokens": 9302389.0, + "step": 2892 + }, + { + "epoch": 1.4797953964194375, + "grad_norm": 14.245892524719238, + "learning_rate": 1e-06, + "loss": 0.3813, + "mean_token_accuracy": 0.8520196676254272, + "num_tokens": 9304860.0, + "step": 2893 + }, + { + "epoch": 1.480306905370844, + "grad_norm": 14.295648574829102, + "learning_rate": 1e-06, + "loss": 0.4141, + "mean_token_accuracy": 0.8558720350265503, + "num_tokens": 9307368.0, + "step": 2894 + }, + { + "epoch": 1.4808184143222507, + "grad_norm": 12.931587219238281, + "learning_rate": 1e-06, + "loss": 0.3806, + "mean_token_accuracy": 0.8778461813926697, + "num_tokens": 9310133.0, + "step": 2895 + }, + { + "epoch": 1.4813299232736572, + "grad_norm": 10.648883819580078, + "learning_rate": 1e-06, + "loss": 0.2441, + "mean_token_accuracy": 0.9059284925460815, + "num_tokens": 9312948.0, + "step": 2896 + }, + { + "epoch": 1.481841432225064, + "grad_norm": 8.427512168884277, + "learning_rate": 1e-06, + "loss": 0.2385, + "mean_token_accuracy": 0.9135655164718628, + "num_tokens": 9316880.0, + "step": 2897 + }, + { + "epoch": 1.4823529411764707, + "grad_norm": 8.508673667907715, + "learning_rate": 1e-06, + "loss": 0.2599, + "mean_token_accuracy": 0.8959820866584778, + "num_tokens": 9319874.0, + "step": 2898 + }, + { + "epoch": 1.4828644501278774, + "grad_norm": 11.05754566192627, + "learning_rate": 1e-06, + "loss": 0.2231, + "mean_token_accuracy": 0.9349981546401978, + "num_tokens": 9322488.0, + "step": 2899 + }, + { + "epoch": 1.4833759590792839, + "grad_norm": 9.28853988647461, + "learning_rate": 1e-06, + "loss": 0.2371, + "mean_token_accuracy": 0.890076756477356, + "num_tokens": 9326023.0, + "step": 2900 + }, + { + "epoch": 1.4838874680306906, + "grad_norm": 9.181792259216309, + "learning_rate": 1e-06, + "loss": 0.2408, + "mean_token_accuracy": 0.9206840991973877, + "num_tokens": 9329165.0, + "step": 2901 + }, + { + "epoch": 1.484398976982097, + "grad_norm": 8.075638771057129, + "learning_rate": 1e-06, + "loss": 0.2426, + "mean_token_accuracy": 0.8928764462471008, + "num_tokens": 9332721.0, + "step": 2902 + }, + { + "epoch": 1.4849104859335038, + "grad_norm": 8.191912651062012, + "learning_rate": 1e-06, + "loss": 0.2138, + "mean_token_accuracy": 0.9301005601882935, + "num_tokens": 9335635.0, + "step": 2903 + }, + { + "epoch": 1.4854219948849106, + "grad_norm": 8.380122184753418, + "learning_rate": 1e-06, + "loss": 0.1731, + "mean_token_accuracy": 0.9285497665405273, + "num_tokens": 9338811.0, + "step": 2904 + }, + { + "epoch": 1.485933503836317, + "grad_norm": 7.542749404907227, + "learning_rate": 1e-06, + "loss": 0.1351, + "mean_token_accuracy": 0.9416476488113403, + "num_tokens": 9342712.0, + "step": 2905 + }, + { + "epoch": 1.4864450127877238, + "grad_norm": 8.579122543334961, + "learning_rate": 1e-06, + "loss": 0.2639, + "mean_token_accuracy": 0.8823269009590149, + "num_tokens": 9345460.0, + "step": 2906 + }, + { + "epoch": 1.4869565217391305, + "grad_norm": 10.597126960754395, + "learning_rate": 1e-06, + "loss": 0.269, + "mean_token_accuracy": 0.8878358602523804, + "num_tokens": 9348623.0, + "step": 2907 + }, + { + "epoch": 1.487468030690537, + "grad_norm": 6.245246410369873, + "learning_rate": 1e-06, + "loss": 0.1483, + "mean_token_accuracy": 0.8820489048957825, + "num_tokens": 9352777.0, + "step": 2908 + }, + { + "epoch": 1.4879795396419437, + "grad_norm": 7.815675735473633, + "learning_rate": 1e-06, + "loss": 0.1809, + "mean_token_accuracy": 0.9249047636985779, + "num_tokens": 9355389.0, + "step": 2909 + }, + { + "epoch": 1.4884910485933505, + "grad_norm": 7.027017116546631, + "learning_rate": 1e-06, + "loss": 0.1866, + "mean_token_accuracy": 0.8839118480682373, + "num_tokens": 9358493.0, + "step": 2910 + }, + { + "epoch": 1.489002557544757, + "grad_norm": 8.260685920715332, + "learning_rate": 1e-06, + "loss": 0.2222, + "mean_token_accuracy": 0.9044591188430786, + "num_tokens": 9363104.0, + "step": 2911 + }, + { + "epoch": 1.4895140664961637, + "grad_norm": 9.374205589294434, + "learning_rate": 1e-06, + "loss": 0.2372, + "mean_token_accuracy": 0.8963411450386047, + "num_tokens": 9366704.0, + "step": 2912 + }, + { + "epoch": 1.4900255754475702, + "grad_norm": 8.502230644226074, + "learning_rate": 1e-06, + "loss": 0.2415, + "mean_token_accuracy": 0.9067124128341675, + "num_tokens": 9370427.0, + "step": 2913 + }, + { + "epoch": 1.490537084398977, + "grad_norm": 9.597463607788086, + "learning_rate": 1e-06, + "loss": 0.2963, + "mean_token_accuracy": 0.8912819027900696, + "num_tokens": 9373563.0, + "step": 2914 + }, + { + "epoch": 1.4910485933503836, + "grad_norm": 9.056981086730957, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.911340594291687, + "num_tokens": 9376201.0, + "step": 2915 + }, + { + "epoch": 1.4915601023017904, + "grad_norm": 6.942068099975586, + "learning_rate": 1e-06, + "loss": 0.1757, + "mean_token_accuracy": 0.9134489297866821, + "num_tokens": 9379810.0, + "step": 2916 + }, + { + "epoch": 1.4920716112531969, + "grad_norm": 12.243064880371094, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.8744865655899048, + "num_tokens": 9383309.0, + "step": 2917 + }, + { + "epoch": 1.4925831202046036, + "grad_norm": 8.860383987426758, + "learning_rate": 1e-06, + "loss": 0.2268, + "mean_token_accuracy": 0.9098861217498779, + "num_tokens": 9387964.0, + "step": 2918 + }, + { + "epoch": 1.49309462915601, + "grad_norm": 7.988664627075195, + "learning_rate": 1e-06, + "loss": 0.2168, + "mean_token_accuracy": 0.9136420488357544, + "num_tokens": 9390680.0, + "step": 2919 + }, + { + "epoch": 1.4936061381074168, + "grad_norm": 16.456571578979492, + "learning_rate": 1e-06, + "loss": 0.2391, + "mean_token_accuracy": 0.9140481948852539, + "num_tokens": 9393257.0, + "step": 2920 + }, + { + "epoch": 1.4941176470588236, + "grad_norm": 10.745260238647461, + "learning_rate": 1e-06, + "loss": 0.221, + "mean_token_accuracy": 0.9304088354110718, + "num_tokens": 9396249.0, + "step": 2921 + }, + { + "epoch": 1.4946291560102303, + "grad_norm": 7.983025074005127, + "learning_rate": 1e-06, + "loss": 0.2171, + "mean_token_accuracy": 0.9192070364952087, + "num_tokens": 9400104.0, + "step": 2922 + }, + { + "epoch": 1.4951406649616368, + "grad_norm": 9.807036399841309, + "learning_rate": 1e-06, + "loss": 0.2949, + "mean_token_accuracy": 0.8850447535514832, + "num_tokens": 9402670.0, + "step": 2923 + }, + { + "epoch": 1.4956521739130435, + "grad_norm": 8.641883850097656, + "learning_rate": 1e-06, + "loss": 0.2074, + "mean_token_accuracy": 0.8916740417480469, + "num_tokens": 9405755.0, + "step": 2924 + }, + { + "epoch": 1.49616368286445, + "grad_norm": 7.595441818237305, + "learning_rate": 1e-06, + "loss": 0.1943, + "mean_token_accuracy": 0.9276456832885742, + "num_tokens": 9408854.0, + "step": 2925 + }, + { + "epoch": 1.4966751918158567, + "grad_norm": 6.539834022521973, + "learning_rate": 1e-06, + "loss": 0.1411, + "mean_token_accuracy": 0.8757612705230713, + "num_tokens": 9412117.0, + "step": 2926 + }, + { + "epoch": 1.4971867007672635, + "grad_norm": 6.851874351501465, + "learning_rate": 1e-06, + "loss": 0.1859, + "mean_token_accuracy": 0.9055823683738708, + "num_tokens": 9415537.0, + "step": 2927 + }, + { + "epoch": 1.4976982097186702, + "grad_norm": 8.732760429382324, + "learning_rate": 1e-06, + "loss": 0.2125, + "mean_token_accuracy": 0.8993335962295532, + "num_tokens": 9418334.0, + "step": 2928 + }, + { + "epoch": 1.4982097186700767, + "grad_norm": 11.014547348022461, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8344665765762329, + "num_tokens": 9421206.0, + "step": 2929 + }, + { + "epoch": 1.4987212276214834, + "grad_norm": 10.617002487182617, + "learning_rate": 1e-06, + "loss": 0.2765, + "mean_token_accuracy": 0.8978285789489746, + "num_tokens": 9424305.0, + "step": 2930 + }, + { + "epoch": 1.49923273657289, + "grad_norm": 12.274026870727539, + "learning_rate": 1e-06, + "loss": 0.2956, + "mean_token_accuracy": 0.8883122801780701, + "num_tokens": 9427387.0, + "step": 2931 + }, + { + "epoch": 1.4997442455242966, + "grad_norm": 12.507925987243652, + "learning_rate": 1e-06, + "loss": 0.2909, + "mean_token_accuracy": 0.8877824544906616, + "num_tokens": 9430467.0, + "step": 2932 + }, + { + "epoch": 1.5002557544757034, + "grad_norm": 6.426373481750488, + "learning_rate": 1e-06, + "loss": 0.2034, + "mean_token_accuracy": 0.8922048807144165, + "num_tokens": 9434232.0, + "step": 2933 + }, + { + "epoch": 1.50076726342711, + "grad_norm": 6.912188529968262, + "learning_rate": 1e-06, + "loss": 0.1586, + "mean_token_accuracy": 0.8974236249923706, + "num_tokens": 9437996.0, + "step": 2934 + }, + { + "epoch": 1.5012787723785166, + "grad_norm": 11.379992485046387, + "learning_rate": 1e-06, + "loss": 0.3248, + "mean_token_accuracy": 0.8582661151885986, + "num_tokens": 9441612.0, + "step": 2935 + }, + { + "epoch": 1.5017902813299233, + "grad_norm": 9.949040412902832, + "learning_rate": 1e-06, + "loss": 0.2593, + "mean_token_accuracy": 0.8946558237075806, + "num_tokens": 9444856.0, + "step": 2936 + }, + { + "epoch": 1.5023017902813298, + "grad_norm": 7.831646919250488, + "learning_rate": 1e-06, + "loss": 0.1841, + "mean_token_accuracy": 0.9163055419921875, + "num_tokens": 9448361.0, + "step": 2937 + }, + { + "epoch": 1.5028132992327365, + "grad_norm": 8.500174522399902, + "learning_rate": 1e-06, + "loss": 0.1778, + "mean_token_accuracy": 0.9409736394882202, + "num_tokens": 9450920.0, + "step": 2938 + }, + { + "epoch": 1.5033248081841433, + "grad_norm": 9.532085418701172, + "learning_rate": 1e-06, + "loss": 0.2513, + "mean_token_accuracy": 0.9046088457107544, + "num_tokens": 9453705.0, + "step": 2939 + }, + { + "epoch": 1.50383631713555, + "grad_norm": 9.731687545776367, + "learning_rate": 1e-06, + "loss": 0.1557, + "mean_token_accuracy": 0.9260950088500977, + "num_tokens": 9457413.0, + "step": 2940 + }, + { + "epoch": 1.5043478260869565, + "grad_norm": 13.918618202209473, + "learning_rate": 1e-06, + "loss": 0.3868, + "mean_token_accuracy": 0.8749106526374817, + "num_tokens": 9460098.0, + "step": 2941 + }, + { + "epoch": 1.5048593350383632, + "grad_norm": 7.83819580078125, + "learning_rate": 1e-06, + "loss": 0.183, + "mean_token_accuracy": 0.9133318066596985, + "num_tokens": 9463765.0, + "step": 2942 + }, + { + "epoch": 1.5053708439897697, + "grad_norm": 11.990787506103516, + "learning_rate": 1e-06, + "loss": 0.3703, + "mean_token_accuracy": 0.8667997121810913, + "num_tokens": 9466897.0, + "step": 2943 + }, + { + "epoch": 1.5058823529411764, + "grad_norm": 8.56093692779541, + "learning_rate": 1e-06, + "loss": 0.2147, + "mean_token_accuracy": 0.9201338291168213, + "num_tokens": 9470241.0, + "step": 2944 + }, + { + "epoch": 1.5063938618925832, + "grad_norm": 9.805789947509766, + "learning_rate": 1e-06, + "loss": 0.201, + "mean_token_accuracy": 0.9113770127296448, + "num_tokens": 9473416.0, + "step": 2945 + }, + { + "epoch": 1.50690537084399, + "grad_norm": 9.315298080444336, + "learning_rate": 1e-06, + "loss": 0.2635, + "mean_token_accuracy": 0.8920457363128662, + "num_tokens": 9476963.0, + "step": 2946 + }, + { + "epoch": 1.5074168797953964, + "grad_norm": 10.139205932617188, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8602679967880249, + "num_tokens": 9479628.0, + "step": 2947 + }, + { + "epoch": 1.507928388746803, + "grad_norm": 10.486778259277344, + "learning_rate": 1e-06, + "loss": 0.2708, + "mean_token_accuracy": 0.902316689491272, + "num_tokens": 9482397.0, + "step": 2948 + }, + { + "epoch": 1.5084398976982096, + "grad_norm": 13.015390396118164, + "learning_rate": 1e-06, + "loss": 0.3309, + "mean_token_accuracy": 0.8748836517333984, + "num_tokens": 9485165.0, + "step": 2949 + }, + { + "epoch": 1.5089514066496164, + "grad_norm": 9.933533668518066, + "learning_rate": 1e-06, + "loss": 0.2751, + "mean_token_accuracy": 0.9041036367416382, + "num_tokens": 9488515.0, + "step": 2950 + }, + { + "epoch": 1.509462915601023, + "grad_norm": 10.344608306884766, + "learning_rate": 1e-06, + "loss": 0.2041, + "mean_token_accuracy": 0.9271876215934753, + "num_tokens": 9491520.0, + "step": 2951 + }, + { + "epoch": 1.5099744245524298, + "grad_norm": 5.872512340545654, + "learning_rate": 1e-06, + "loss": 0.1451, + "mean_token_accuracy": 0.9226297736167908, + "num_tokens": 9494825.0, + "step": 2952 + }, + { + "epoch": 1.5104859335038363, + "grad_norm": 8.233858108520508, + "learning_rate": 1e-06, + "loss": 0.211, + "mean_token_accuracy": 0.9534205198287964, + "num_tokens": 9497387.0, + "step": 2953 + }, + { + "epoch": 1.5109974424552428, + "grad_norm": 9.9365816116333, + "learning_rate": 1e-06, + "loss": 0.3098, + "mean_token_accuracy": 0.9004008769989014, + "num_tokens": 9500332.0, + "step": 2954 + }, + { + "epoch": 1.5115089514066495, + "grad_norm": 8.840529441833496, + "learning_rate": 1e-06, + "loss": 0.2387, + "mean_token_accuracy": 0.9101779460906982, + "num_tokens": 9503310.0, + "step": 2955 + }, + { + "epoch": 1.5120204603580563, + "grad_norm": 4.1635823249816895, + "learning_rate": 1e-06, + "loss": 0.1296, + "mean_token_accuracy": 0.9421769380569458, + "num_tokens": 9508613.0, + "step": 2956 + }, + { + "epoch": 1.512531969309463, + "grad_norm": 8.701842308044434, + "learning_rate": 1e-06, + "loss": 0.1885, + "mean_token_accuracy": 0.9233746528625488, + "num_tokens": 9511828.0, + "step": 2957 + }, + { + "epoch": 1.5130434782608697, + "grad_norm": 9.801673889160156, + "learning_rate": 1e-06, + "loss": 0.2581, + "mean_token_accuracy": 0.9015085697174072, + "num_tokens": 9514388.0, + "step": 2958 + }, + { + "epoch": 1.5135549872122762, + "grad_norm": 17.139806747436523, + "learning_rate": 1e-06, + "loss": 0.3418, + "mean_token_accuracy": 0.8884871602058411, + "num_tokens": 9518346.0, + "step": 2959 + }, + { + "epoch": 1.5140664961636827, + "grad_norm": 10.531225204467773, + "learning_rate": 1e-06, + "loss": 0.2129, + "mean_token_accuracy": 0.9219048619270325, + "num_tokens": 9521410.0, + "step": 2960 + }, + { + "epoch": 1.5145780051150894, + "grad_norm": 9.016901969909668, + "learning_rate": 1e-06, + "loss": 0.2112, + "mean_token_accuracy": 0.9003004431724548, + "num_tokens": 9524927.0, + "step": 2961 + }, + { + "epoch": 1.5150895140664962, + "grad_norm": 11.113777160644531, + "learning_rate": 1e-06, + "loss": 0.2652, + "mean_token_accuracy": 0.8898568153381348, + "num_tokens": 9528235.0, + "step": 2962 + }, + { + "epoch": 1.515601023017903, + "grad_norm": 7.476075172424316, + "learning_rate": 1e-06, + "loss": 0.1578, + "mean_token_accuracy": 0.9266808032989502, + "num_tokens": 9532489.0, + "step": 2963 + }, + { + "epoch": 1.5161125319693096, + "grad_norm": 9.57284164428711, + "learning_rate": 1e-06, + "loss": 0.2427, + "mean_token_accuracy": 0.9085447788238525, + "num_tokens": 9535402.0, + "step": 2964 + }, + { + "epoch": 1.5166240409207161, + "grad_norm": 9.348052024841309, + "learning_rate": 1e-06, + "loss": 0.2488, + "mean_token_accuracy": 0.8991104364395142, + "num_tokens": 9538470.0, + "step": 2965 + }, + { + "epoch": 1.5171355498721226, + "grad_norm": 9.943816184997559, + "learning_rate": 1e-06, + "loss": 0.2286, + "mean_token_accuracy": 0.8889056444168091, + "num_tokens": 9542243.0, + "step": 2966 + }, + { + "epoch": 1.5176470588235293, + "grad_norm": 10.500575065612793, + "learning_rate": 1e-06, + "loss": 0.2671, + "mean_token_accuracy": 0.8919262886047363, + "num_tokens": 9544782.0, + "step": 2967 + }, + { + "epoch": 1.518158567774936, + "grad_norm": 8.282591819763184, + "learning_rate": 1e-06, + "loss": 0.1719, + "mean_token_accuracy": 0.9335013031959534, + "num_tokens": 9547855.0, + "step": 2968 + }, + { + "epoch": 1.5186700767263428, + "grad_norm": 8.499592781066895, + "learning_rate": 1e-06, + "loss": 0.1982, + "mean_token_accuracy": 0.930389940738678, + "num_tokens": 9550919.0, + "step": 2969 + }, + { + "epoch": 1.5191815856777495, + "grad_norm": 10.109574317932129, + "learning_rate": 1e-06, + "loss": 0.2096, + "mean_token_accuracy": 0.8968086838722229, + "num_tokens": 9553979.0, + "step": 2970 + }, + { + "epoch": 1.519693094629156, + "grad_norm": 9.528036117553711, + "learning_rate": 1e-06, + "loss": 0.1967, + "mean_token_accuracy": 0.9088143110275269, + "num_tokens": 9557008.0, + "step": 2971 + }, + { + "epoch": 1.5202046035805625, + "grad_norm": 6.747566223144531, + "learning_rate": 1e-06, + "loss": 0.1765, + "mean_token_accuracy": 0.9244707226753235, + "num_tokens": 9560056.0, + "step": 2972 + }, + { + "epoch": 1.5207161125319693, + "grad_norm": 8.42009162902832, + "learning_rate": 1e-06, + "loss": 0.2082, + "mean_token_accuracy": 0.9064043760299683, + "num_tokens": 9563234.0, + "step": 2973 + }, + { + "epoch": 1.521227621483376, + "grad_norm": 9.042177200317383, + "learning_rate": 1e-06, + "loss": 0.2056, + "mean_token_accuracy": 0.9152294397354126, + "num_tokens": 9565917.0, + "step": 2974 + }, + { + "epoch": 1.5217391304347827, + "grad_norm": 10.040491104125977, + "learning_rate": 1e-06, + "loss": 0.2039, + "mean_token_accuracy": 0.9133328199386597, + "num_tokens": 9570040.0, + "step": 2975 + }, + { + "epoch": 1.5222506393861892, + "grad_norm": 9.038538932800293, + "learning_rate": 1e-06, + "loss": 0.2177, + "mean_token_accuracy": 0.908561110496521, + "num_tokens": 9574980.0, + "step": 2976 + }, + { + "epoch": 1.522762148337596, + "grad_norm": 11.133055686950684, + "learning_rate": 1e-06, + "loss": 0.2652, + "mean_token_accuracy": 0.8935723304748535, + "num_tokens": 9578559.0, + "step": 2977 + }, + { + "epoch": 1.5232736572890024, + "grad_norm": 10.553840637207031, + "learning_rate": 1e-06, + "loss": 0.164, + "mean_token_accuracy": 0.9155797958374023, + "num_tokens": 9582206.0, + "step": 2978 + }, + { + "epoch": 1.5237851662404092, + "grad_norm": 8.8803071975708, + "learning_rate": 1e-06, + "loss": 0.254, + "mean_token_accuracy": 0.8778273463249207, + "num_tokens": 9585164.0, + "step": 2979 + }, + { + "epoch": 1.5242966751918159, + "grad_norm": 13.61034870147705, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8836358189582825, + "num_tokens": 9588155.0, + "step": 2980 + }, + { + "epoch": 1.5248081841432226, + "grad_norm": 8.65895938873291, + "learning_rate": 1e-06, + "loss": 0.2274, + "mean_token_accuracy": 0.9064152240753174, + "num_tokens": 9590912.0, + "step": 2981 + }, + { + "epoch": 1.5253196930946291, + "grad_norm": 6.1537346839904785, + "learning_rate": 1e-06, + "loss": 0.1849, + "mean_token_accuracy": 0.8950072526931763, + "num_tokens": 9595111.0, + "step": 2982 + }, + { + "epoch": 1.5258312020460358, + "grad_norm": 7.775795936584473, + "learning_rate": 1e-06, + "loss": 0.2223, + "mean_token_accuracy": 0.9114144444465637, + "num_tokens": 9598991.0, + "step": 2983 + }, + { + "epoch": 1.5263427109974423, + "grad_norm": 11.52584171295166, + "learning_rate": 1e-06, + "loss": 0.222, + "mean_token_accuracy": 0.9136835336685181, + "num_tokens": 9602585.0, + "step": 2984 + }, + { + "epoch": 1.526854219948849, + "grad_norm": 9.933723449707031, + "learning_rate": 1e-06, + "loss": 0.2393, + "mean_token_accuracy": 0.911053478717804, + "num_tokens": 9605128.0, + "step": 2985 + }, + { + "epoch": 1.5273657289002558, + "grad_norm": 11.357743263244629, + "learning_rate": 1e-06, + "loss": 0.2847, + "mean_token_accuracy": 0.9018988609313965, + "num_tokens": 9607941.0, + "step": 2986 + }, + { + "epoch": 1.5278772378516625, + "grad_norm": 6.550184726715088, + "learning_rate": 1e-06, + "loss": 0.1654, + "mean_token_accuracy": 0.9205410480499268, + "num_tokens": 9612136.0, + "step": 2987 + }, + { + "epoch": 1.528388746803069, + "grad_norm": 8.889971733093262, + "learning_rate": 1e-06, + "loss": 0.2152, + "mean_token_accuracy": 0.881904125213623, + "num_tokens": 9616199.0, + "step": 2988 + }, + { + "epoch": 1.5289002557544757, + "grad_norm": 14.543840408325195, + "learning_rate": 1e-06, + "loss": 0.3585, + "mean_token_accuracy": 0.8618497848510742, + "num_tokens": 9618884.0, + "step": 2989 + }, + { + "epoch": 1.5294117647058822, + "grad_norm": 13.192959785461426, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8742536902427673, + "num_tokens": 9622395.0, + "step": 2990 + }, + { + "epoch": 1.529923273657289, + "grad_norm": 9.818873405456543, + "learning_rate": 1e-06, + "loss": 0.2774, + "mean_token_accuracy": 0.8957535028457642, + "num_tokens": 9625432.0, + "step": 2991 + }, + { + "epoch": 1.5304347826086957, + "grad_norm": 10.240565299987793, + "learning_rate": 1e-06, + "loss": 0.2344, + "mean_token_accuracy": 0.9054609537124634, + "num_tokens": 9628828.0, + "step": 2992 + }, + { + "epoch": 1.5309462915601024, + "grad_norm": 8.597281455993652, + "learning_rate": 1e-06, + "loss": 0.2074, + "mean_token_accuracy": 0.9083966016769409, + "num_tokens": 9631777.0, + "step": 2993 + }, + { + "epoch": 1.531457800511509, + "grad_norm": 10.351947784423828, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8770867586135864, + "num_tokens": 9635225.0, + "step": 2994 + }, + { + "epoch": 1.5319693094629157, + "grad_norm": 6.562121391296387, + "learning_rate": 1e-06, + "loss": 0.1396, + "mean_token_accuracy": 0.9323230385780334, + "num_tokens": 9639387.0, + "step": 2995 + }, + { + "epoch": 1.5324808184143222, + "grad_norm": 11.039229393005371, + "learning_rate": 1e-06, + "loss": 0.2751, + "mean_token_accuracy": 0.8805804252624512, + "num_tokens": 9642583.0, + "step": 2996 + }, + { + "epoch": 1.5329923273657289, + "grad_norm": 12.240857124328613, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.8823949098587036, + "num_tokens": 9645171.0, + "step": 2997 + }, + { + "epoch": 1.5335038363171356, + "grad_norm": 6.268917083740234, + "learning_rate": 1e-06, + "loss": 0.1722, + "mean_token_accuracy": 0.9113789796829224, + "num_tokens": 9649057.0, + "step": 2998 + }, + { + "epoch": 1.5340153452685423, + "grad_norm": 10.918238639831543, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8793227672576904, + "num_tokens": 9651989.0, + "step": 2999 + }, + { + "epoch": 1.5345268542199488, + "grad_norm": 7.283839702606201, + "learning_rate": 1e-06, + "loss": 0.236, + "mean_token_accuracy": 0.8857045769691467, + "num_tokens": 9655943.0, + "step": 3000 + }, + { + "epoch": 1.5350383631713553, + "grad_norm": 7.93113899230957, + "learning_rate": 1e-06, + "loss": 0.1881, + "mean_token_accuracy": 0.9389165639877319, + "num_tokens": 9659658.0, + "step": 3001 + }, + { + "epoch": 1.535549872122762, + "grad_norm": 6.998528957366943, + "learning_rate": 1e-06, + "loss": 0.205, + "mean_token_accuracy": 0.879045844078064, + "num_tokens": 9663004.0, + "step": 3002 + }, + { + "epoch": 1.5360613810741688, + "grad_norm": 9.425460815429688, + "learning_rate": 1e-06, + "loss": 0.239, + "mean_token_accuracy": 0.8982094526290894, + "num_tokens": 9666322.0, + "step": 3003 + }, + { + "epoch": 1.5365728900255755, + "grad_norm": 12.453216552734375, + "learning_rate": 1e-06, + "loss": 0.265, + "mean_token_accuracy": 0.8860942125320435, + "num_tokens": 9669339.0, + "step": 3004 + }, + { + "epoch": 1.5370843989769822, + "grad_norm": 9.336000442504883, + "learning_rate": 1e-06, + "loss": 0.2039, + "mean_token_accuracy": 0.916731595993042, + "num_tokens": 9672585.0, + "step": 3005 + }, + { + "epoch": 1.5375959079283887, + "grad_norm": 6.581609725952148, + "learning_rate": 1e-06, + "loss": 0.148, + "mean_token_accuracy": 0.9307796955108643, + "num_tokens": 9676236.0, + "step": 3006 + }, + { + "epoch": 1.5381074168797952, + "grad_norm": 12.109413146972656, + "learning_rate": 1e-06, + "loss": 0.3745, + "mean_token_accuracy": 0.8744027018547058, + "num_tokens": 9679377.0, + "step": 3007 + }, + { + "epoch": 1.538618925831202, + "grad_norm": 8.632303237915039, + "learning_rate": 1e-06, + "loss": 0.1879, + "mean_token_accuracy": 0.9362889528274536, + "num_tokens": 9682586.0, + "step": 3008 + }, + { + "epoch": 1.5391304347826087, + "grad_norm": 6.560681343078613, + "learning_rate": 1e-06, + "loss": 0.1845, + "mean_token_accuracy": 0.9059913158416748, + "num_tokens": 9686084.0, + "step": 3009 + }, + { + "epoch": 1.5396419437340154, + "grad_norm": 11.464566230773926, + "learning_rate": 1e-06, + "loss": 0.2377, + "mean_token_accuracy": 0.9129694104194641, + "num_tokens": 9688740.0, + "step": 3010 + }, + { + "epoch": 1.5401534526854221, + "grad_norm": 11.22123908996582, + "learning_rate": 1e-06, + "loss": 0.3054, + "mean_token_accuracy": 0.8776522874832153, + "num_tokens": 9691374.0, + "step": 3011 + }, + { + "epoch": 1.5406649616368286, + "grad_norm": 8.610074043273926, + "learning_rate": 1e-06, + "loss": 0.1902, + "mean_token_accuracy": 0.8914388418197632, + "num_tokens": 9695061.0, + "step": 3012 + }, + { + "epoch": 1.5411764705882351, + "grad_norm": 7.192942142486572, + "learning_rate": 1e-06, + "loss": 0.2187, + "mean_token_accuracy": 0.9018328785896301, + "num_tokens": 9698168.0, + "step": 3013 + }, + { + "epoch": 1.5416879795396419, + "grad_norm": 11.484277725219727, + "learning_rate": 1e-06, + "loss": 0.3057, + "mean_token_accuracy": 0.8679574131965637, + "num_tokens": 9701961.0, + "step": 3014 + }, + { + "epoch": 1.5421994884910486, + "grad_norm": 10.260034561157227, + "learning_rate": 1e-06, + "loss": 0.2824, + "mean_token_accuracy": 0.9001743793487549, + "num_tokens": 9705193.0, + "step": 3015 + }, + { + "epoch": 1.5427109974424553, + "grad_norm": 8.996676445007324, + "learning_rate": 1e-06, + "loss": 0.2425, + "mean_token_accuracy": 0.9071991443634033, + "num_tokens": 9708417.0, + "step": 3016 + }, + { + "epoch": 1.543222506393862, + "grad_norm": 8.25013542175293, + "learning_rate": 1e-06, + "loss": 0.1772, + "mean_token_accuracy": 0.9256689548492432, + "num_tokens": 9712692.0, + "step": 3017 + }, + { + "epoch": 1.5437340153452686, + "grad_norm": 9.815691947937012, + "learning_rate": 1e-06, + "loss": 0.174, + "mean_token_accuracy": 0.9144695401191711, + "num_tokens": 9715314.0, + "step": 3018 + }, + { + "epoch": 1.544245524296675, + "grad_norm": 15.29290771484375, + "learning_rate": 1e-06, + "loss": 0.4002, + "mean_token_accuracy": 0.8607478141784668, + "num_tokens": 9718300.0, + "step": 3019 + }, + { + "epoch": 1.5447570332480818, + "grad_norm": 8.717740058898926, + "learning_rate": 1e-06, + "loss": 0.2236, + "mean_token_accuracy": 0.9214508533477783, + "num_tokens": 9721705.0, + "step": 3020 + }, + { + "epoch": 1.5452685421994885, + "grad_norm": 8.229061126708984, + "learning_rate": 1e-06, + "loss": 0.1917, + "mean_token_accuracy": 0.9356064796447754, + "num_tokens": 9725024.0, + "step": 3021 + }, + { + "epoch": 1.5457800511508952, + "grad_norm": 11.90822696685791, + "learning_rate": 1e-06, + "loss": 0.2638, + "mean_token_accuracy": 0.8805629014968872, + "num_tokens": 9728021.0, + "step": 3022 + }, + { + "epoch": 1.546291560102302, + "grad_norm": 8.228610038757324, + "learning_rate": 1e-06, + "loss": 0.2473, + "mean_token_accuracy": 0.8863067626953125, + "num_tokens": 9731289.0, + "step": 3023 + }, + { + "epoch": 1.5468030690537085, + "grad_norm": 11.288642883300781, + "learning_rate": 1e-06, + "loss": 0.3733, + "mean_token_accuracy": 0.8527617454528809, + "num_tokens": 9734082.0, + "step": 3024 + }, + { + "epoch": 1.547314578005115, + "grad_norm": 11.340495109558105, + "learning_rate": 1e-06, + "loss": 0.3215, + "mean_token_accuracy": 0.8915387988090515, + "num_tokens": 9736700.0, + "step": 3025 + }, + { + "epoch": 1.5478260869565217, + "grad_norm": 9.05723762512207, + "learning_rate": 1e-06, + "loss": 0.2623, + "mean_token_accuracy": 0.8972040414810181, + "num_tokens": 9740031.0, + "step": 3026 + }, + { + "epoch": 1.5483375959079284, + "grad_norm": 9.477959632873535, + "learning_rate": 1e-06, + "loss": 0.236, + "mean_token_accuracy": 0.9058218598365784, + "num_tokens": 9742759.0, + "step": 3027 + }, + { + "epoch": 1.5488491048593351, + "grad_norm": 12.231276512145996, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8725732564926147, + "num_tokens": 9745306.0, + "step": 3028 + }, + { + "epoch": 1.5493606138107416, + "grad_norm": 7.642716884613037, + "learning_rate": 1e-06, + "loss": 0.22, + "mean_token_accuracy": 0.896345317363739, + "num_tokens": 9747878.0, + "step": 3029 + }, + { + "epoch": 1.5498721227621484, + "grad_norm": 10.11664867401123, + "learning_rate": 1e-06, + "loss": 0.2767, + "mean_token_accuracy": 0.8896204233169556, + "num_tokens": 9751702.0, + "step": 3030 + }, + { + "epoch": 1.5503836317135549, + "grad_norm": 10.805331230163574, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8442392945289612, + "num_tokens": 9754617.0, + "step": 3031 + }, + { + "epoch": 1.5508951406649616, + "grad_norm": 11.693099975585938, + "learning_rate": 1e-06, + "loss": 0.2987, + "mean_token_accuracy": 0.896803617477417, + "num_tokens": 9757410.0, + "step": 3032 + }, + { + "epoch": 1.5514066496163683, + "grad_norm": 9.366107940673828, + "learning_rate": 1e-06, + "loss": 0.2145, + "mean_token_accuracy": 0.9175437688827515, + "num_tokens": 9760515.0, + "step": 3033 + }, + { + "epoch": 1.551918158567775, + "grad_norm": 8.578768730163574, + "learning_rate": 1e-06, + "loss": 0.1662, + "mean_token_accuracy": 0.9103432893753052, + "num_tokens": 9764668.0, + "step": 3034 + }, + { + "epoch": 1.5524296675191815, + "grad_norm": 16.108379364013672, + "learning_rate": 1e-06, + "loss": 0.2421, + "mean_token_accuracy": 0.9146167635917664, + "num_tokens": 9767654.0, + "step": 3035 + }, + { + "epoch": 1.5529411764705883, + "grad_norm": 10.584540367126465, + "learning_rate": 1e-06, + "loss": 0.2958, + "mean_token_accuracy": 0.8953818678855896, + "num_tokens": 9771152.0, + "step": 3036 + }, + { + "epoch": 1.5534526854219948, + "grad_norm": 10.175375938415527, + "learning_rate": 1e-06, + "loss": 0.2747, + "mean_token_accuracy": 0.8753173351287842, + "num_tokens": 9774323.0, + "step": 3037 + }, + { + "epoch": 1.5539641943734015, + "grad_norm": 14.594674110412598, + "learning_rate": 1e-06, + "loss": 0.2931, + "mean_token_accuracy": 0.8987932205200195, + "num_tokens": 9777233.0, + "step": 3038 + }, + { + "epoch": 1.5544757033248082, + "grad_norm": 12.849802017211914, + "learning_rate": 1e-06, + "loss": 0.3387, + "mean_token_accuracy": 0.8673677444458008, + "num_tokens": 9780801.0, + "step": 3039 + }, + { + "epoch": 1.554987212276215, + "grad_norm": 4.3951311111450195, + "learning_rate": 1e-06, + "loss": 0.1151, + "mean_token_accuracy": 0.9261860251426697, + "num_tokens": 9785734.0, + "step": 3040 + }, + { + "epoch": 1.5554987212276215, + "grad_norm": 15.909781455993652, + "learning_rate": 1e-06, + "loss": 0.3426, + "mean_token_accuracy": 0.9069306254386902, + "num_tokens": 9789030.0, + "step": 3041 + }, + { + "epoch": 1.5560102301790282, + "grad_norm": 7.756892204284668, + "learning_rate": 1e-06, + "loss": 0.2382, + "mean_token_accuracy": 0.9157264232635498, + "num_tokens": 9794018.0, + "step": 3042 + }, + { + "epoch": 1.5565217391304347, + "grad_norm": 9.45541000366211, + "learning_rate": 1e-06, + "loss": 0.2644, + "mean_token_accuracy": 0.912011444568634, + "num_tokens": 9796404.0, + "step": 3043 + }, + { + "epoch": 1.5570332480818414, + "grad_norm": 9.818737983703613, + "learning_rate": 1e-06, + "loss": 0.1731, + "mean_token_accuracy": 0.9332444071769714, + "num_tokens": 9800228.0, + "step": 3044 + }, + { + "epoch": 1.5575447570332481, + "grad_norm": 9.685802459716797, + "learning_rate": 1e-06, + "loss": 0.2825, + "mean_token_accuracy": 0.9030320644378662, + "num_tokens": 9803352.0, + "step": 3045 + }, + { + "epoch": 1.5580562659846549, + "grad_norm": 10.511833190917969, + "learning_rate": 1e-06, + "loss": 0.2652, + "mean_token_accuracy": 0.8982608318328857, + "num_tokens": 9806289.0, + "step": 3046 + }, + { + "epoch": 1.5585677749360614, + "grad_norm": 9.765327453613281, + "learning_rate": 1e-06, + "loss": 0.2142, + "mean_token_accuracy": 0.9050107002258301, + "num_tokens": 9809583.0, + "step": 3047 + }, + { + "epoch": 1.559079283887468, + "grad_norm": 11.478885650634766, + "learning_rate": 1e-06, + "loss": 0.2784, + "mean_token_accuracy": 0.9080254435539246, + "num_tokens": 9812190.0, + "step": 3048 + }, + { + "epoch": 1.5595907928388746, + "grad_norm": 12.935345649719238, + "learning_rate": 1e-06, + "loss": 0.25, + "mean_token_accuracy": 0.9026979207992554, + "num_tokens": 9815174.0, + "step": 3049 + }, + { + "epoch": 1.5601023017902813, + "grad_norm": 8.214735984802246, + "learning_rate": 1e-06, + "loss": 0.2501, + "mean_token_accuracy": 0.8953521847724915, + "num_tokens": 9817959.0, + "step": 3050 + }, + { + "epoch": 1.560613810741688, + "grad_norm": 10.556377410888672, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.8787655830383301, + "num_tokens": 9820639.0, + "step": 3051 + }, + { + "epoch": 1.5611253196930948, + "grad_norm": 12.467019081115723, + "learning_rate": 1e-06, + "loss": 0.3323, + "mean_token_accuracy": 0.8873552083969116, + "num_tokens": 9823653.0, + "step": 3052 + }, + { + "epoch": 1.5616368286445013, + "grad_norm": 9.491520881652832, + "learning_rate": 1e-06, + "loss": 0.1917, + "mean_token_accuracy": 0.9354743957519531, + "num_tokens": 9827570.0, + "step": 3053 + }, + { + "epoch": 1.5621483375959078, + "grad_norm": 8.992819786071777, + "learning_rate": 1e-06, + "loss": 0.2559, + "mean_token_accuracy": 0.893684446811676, + "num_tokens": 9830531.0, + "step": 3054 + }, + { + "epoch": 1.5626598465473145, + "grad_norm": 9.309272766113281, + "learning_rate": 1e-06, + "loss": 0.1942, + "mean_token_accuracy": 0.8979969024658203, + "num_tokens": 9833928.0, + "step": 3055 + }, + { + "epoch": 1.5631713554987212, + "grad_norm": 9.934090614318848, + "learning_rate": 1e-06, + "loss": 0.2347, + "mean_token_accuracy": 0.9045335054397583, + "num_tokens": 9836511.0, + "step": 3056 + }, + { + "epoch": 1.563682864450128, + "grad_norm": 8.150444030761719, + "learning_rate": 1e-06, + "loss": 0.2665, + "mean_token_accuracy": 0.8736503720283508, + "num_tokens": 9839882.0, + "step": 3057 + }, + { + "epoch": 1.5641943734015347, + "grad_norm": 7.284010887145996, + "learning_rate": 1e-06, + "loss": 0.2176, + "mean_token_accuracy": 0.8921781182289124, + "num_tokens": 9842516.0, + "step": 3058 + }, + { + "epoch": 1.5647058823529412, + "grad_norm": 12.149635314941406, + "learning_rate": 1e-06, + "loss": 0.3495, + "mean_token_accuracy": 0.8518050312995911, + "num_tokens": 9845374.0, + "step": 3059 + }, + { + "epoch": 1.5652173913043477, + "grad_norm": 7.8963470458984375, + "learning_rate": 1e-06, + "loss": 0.2057, + "mean_token_accuracy": 0.9002891778945923, + "num_tokens": 9848828.0, + "step": 3060 + }, + { + "epoch": 1.5657289002557544, + "grad_norm": 7.064937591552734, + "learning_rate": 1e-06, + "loss": 0.2018, + "mean_token_accuracy": 0.917493462562561, + "num_tokens": 9852838.0, + "step": 3061 + }, + { + "epoch": 1.5662404092071611, + "grad_norm": 10.856348991394043, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.9011961221694946, + "num_tokens": 9857094.0, + "step": 3062 + }, + { + "epoch": 1.5667519181585678, + "grad_norm": 13.496193885803223, + "learning_rate": 1e-06, + "loss": 0.2884, + "mean_token_accuracy": 0.8955926895141602, + "num_tokens": 9859754.0, + "step": 3063 + }, + { + "epoch": 1.5672634271099746, + "grad_norm": 8.299346923828125, + "learning_rate": 1e-06, + "loss": 0.1434, + "mean_token_accuracy": 0.9111707806587219, + "num_tokens": 9862886.0, + "step": 3064 + }, + { + "epoch": 1.567774936061381, + "grad_norm": 11.383841514587402, + "learning_rate": 1e-06, + "loss": 0.3433, + "mean_token_accuracy": 0.8369016647338867, + "num_tokens": 9865625.0, + "step": 3065 + }, + { + "epoch": 1.5682864450127876, + "grad_norm": 11.694772720336914, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.8818703889846802, + "num_tokens": 9868412.0, + "step": 3066 + }, + { + "epoch": 1.5687979539641943, + "grad_norm": 8.898903846740723, + "learning_rate": 1e-06, + "loss": 0.2119, + "mean_token_accuracy": 0.9025357365608215, + "num_tokens": 9871603.0, + "step": 3067 + }, + { + "epoch": 1.569309462915601, + "grad_norm": 9.02415657043457, + "learning_rate": 1e-06, + "loss": 0.2311, + "mean_token_accuracy": 0.8962329626083374, + "num_tokens": 9874564.0, + "step": 3068 + }, + { + "epoch": 1.5698209718670078, + "grad_norm": 9.892559051513672, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.9127856492996216, + "num_tokens": 9878730.0, + "step": 3069 + }, + { + "epoch": 1.5703324808184145, + "grad_norm": 8.1226167678833, + "learning_rate": 1e-06, + "loss": 0.1983, + "mean_token_accuracy": 0.9019536375999451, + "num_tokens": 9881754.0, + "step": 3070 + }, + { + "epoch": 1.570843989769821, + "grad_norm": 11.533504486083984, + "learning_rate": 1e-06, + "loss": 0.2341, + "mean_token_accuracy": 0.9123857617378235, + "num_tokens": 9884512.0, + "step": 3071 + }, + { + "epoch": 1.5713554987212275, + "grad_norm": 9.929969787597656, + "learning_rate": 1e-06, + "loss": 0.1924, + "mean_token_accuracy": 0.9227875471115112, + "num_tokens": 9887760.0, + "step": 3072 + }, + { + "epoch": 1.5718670076726342, + "grad_norm": 8.801955223083496, + "learning_rate": 1e-06, + "loss": 0.2227, + "mean_token_accuracy": 0.877159833908081, + "num_tokens": 9890686.0, + "step": 3073 + }, + { + "epoch": 1.572378516624041, + "grad_norm": 6.731752395629883, + "learning_rate": 1e-06, + "loss": 0.1816, + "mean_token_accuracy": 0.9046540260314941, + "num_tokens": 9893790.0, + "step": 3074 + }, + { + "epoch": 1.5728900255754477, + "grad_norm": 10.7772798538208, + "learning_rate": 1e-06, + "loss": 0.2241, + "mean_token_accuracy": 0.9133120179176331, + "num_tokens": 9896794.0, + "step": 3075 + }, + { + "epoch": 1.5734015345268544, + "grad_norm": 10.055270195007324, + "learning_rate": 1e-06, + "loss": 0.2366, + "mean_token_accuracy": 0.9280559420585632, + "num_tokens": 9899809.0, + "step": 3076 + }, + { + "epoch": 1.5739130434782609, + "grad_norm": 8.55172348022461, + "learning_rate": 1e-06, + "loss": 0.2229, + "mean_token_accuracy": 0.9246398210525513, + "num_tokens": 9902841.0, + "step": 3077 + }, + { + "epoch": 1.5744245524296674, + "grad_norm": 9.699663162231445, + "learning_rate": 1e-06, + "loss": 0.2957, + "mean_token_accuracy": 0.8760600090026855, + "num_tokens": 9906007.0, + "step": 3078 + }, + { + "epoch": 1.5749360613810741, + "grad_norm": 9.756325721740723, + "learning_rate": 1e-06, + "loss": 0.1893, + "mean_token_accuracy": 0.919608473777771, + "num_tokens": 9909377.0, + "step": 3079 + }, + { + "epoch": 1.5754475703324808, + "grad_norm": 12.058891296386719, + "learning_rate": 1e-06, + "loss": 0.2403, + "mean_token_accuracy": 0.903043270111084, + "num_tokens": 9913004.0, + "step": 3080 + }, + { + "epoch": 1.5759590792838876, + "grad_norm": 12.249202728271484, + "learning_rate": 1e-06, + "loss": 0.221, + "mean_token_accuracy": 0.8856365084648132, + "num_tokens": 9916052.0, + "step": 3081 + }, + { + "epoch": 1.576470588235294, + "grad_norm": 9.723875999450684, + "learning_rate": 1e-06, + "loss": 0.2069, + "mean_token_accuracy": 0.9176441431045532, + "num_tokens": 9919256.0, + "step": 3082 + }, + { + "epoch": 1.5769820971867008, + "grad_norm": 8.092965126037598, + "learning_rate": 1e-06, + "loss": 0.2536, + "mean_token_accuracy": 0.9183851480484009, + "num_tokens": 9922639.0, + "step": 3083 + }, + { + "epoch": 1.5774936061381073, + "grad_norm": 8.3773193359375, + "learning_rate": 1e-06, + "loss": 0.1988, + "mean_token_accuracy": 0.9270491600036621, + "num_tokens": 9925949.0, + "step": 3084 + }, + { + "epoch": 1.578005115089514, + "grad_norm": 9.59123706817627, + "learning_rate": 1e-06, + "loss": 0.2362, + "mean_token_accuracy": 0.8855547904968262, + "num_tokens": 9928406.0, + "step": 3085 + }, + { + "epoch": 1.5785166240409207, + "grad_norm": 9.219832420349121, + "learning_rate": 1e-06, + "loss": 0.22, + "mean_token_accuracy": 0.8918853402137756, + "num_tokens": 9931221.0, + "step": 3086 + }, + { + "epoch": 1.5790281329923275, + "grad_norm": 17.426239013671875, + "learning_rate": 1e-06, + "loss": 0.2945, + "mean_token_accuracy": 0.869101881980896, + "num_tokens": 9934153.0, + "step": 3087 + }, + { + "epoch": 1.579539641943734, + "grad_norm": 6.5605034828186035, + "learning_rate": 1e-06, + "loss": 0.1899, + "mean_token_accuracy": 0.905315637588501, + "num_tokens": 9937925.0, + "step": 3088 + }, + { + "epoch": 1.5800511508951407, + "grad_norm": 6.889161586761475, + "learning_rate": 1e-06, + "loss": 0.1861, + "mean_token_accuracy": 0.9297389984130859, + "num_tokens": 9941440.0, + "step": 3089 + }, + { + "epoch": 1.5805626598465472, + "grad_norm": 11.701288223266602, + "learning_rate": 1e-06, + "loss": 0.257, + "mean_token_accuracy": 0.9142080545425415, + "num_tokens": 9943994.0, + "step": 3090 + }, + { + "epoch": 1.581074168797954, + "grad_norm": 11.913290977478027, + "learning_rate": 1e-06, + "loss": 0.2819, + "mean_token_accuracy": 0.9123924374580383, + "num_tokens": 9946464.0, + "step": 3091 + }, + { + "epoch": 1.5815856777493607, + "grad_norm": 10.472043991088867, + "learning_rate": 1e-06, + "loss": 0.2576, + "mean_token_accuracy": 0.9288227558135986, + "num_tokens": 9949645.0, + "step": 3092 + }, + { + "epoch": 1.5820971867007674, + "grad_norm": 11.927746772766113, + "learning_rate": 1e-06, + "loss": 0.2396, + "mean_token_accuracy": 0.902508020401001, + "num_tokens": 9952975.0, + "step": 3093 + }, + { + "epoch": 1.5826086956521739, + "grad_norm": 14.120912551879883, + "learning_rate": 1e-06, + "loss": 0.3666, + "mean_token_accuracy": 0.8595572710037231, + "num_tokens": 9955603.0, + "step": 3094 + }, + { + "epoch": 1.5831202046035806, + "grad_norm": 10.014267921447754, + "learning_rate": 1e-06, + "loss": 0.2348, + "mean_token_accuracy": 0.9017889499664307, + "num_tokens": 9958695.0, + "step": 3095 + }, + { + "epoch": 1.583631713554987, + "grad_norm": 8.58243179321289, + "learning_rate": 1e-06, + "loss": 0.1828, + "mean_token_accuracy": 0.932138204574585, + "num_tokens": 9962213.0, + "step": 3096 + }, + { + "epoch": 1.5841432225063938, + "grad_norm": 8.255656242370605, + "learning_rate": 1e-06, + "loss": 0.2425, + "mean_token_accuracy": 0.9021299481391907, + "num_tokens": 9965255.0, + "step": 3097 + }, + { + "epoch": 1.5846547314578006, + "grad_norm": 10.618401527404785, + "learning_rate": 1e-06, + "loss": 0.235, + "mean_token_accuracy": 0.9043095111846924, + "num_tokens": 9968525.0, + "step": 3098 + }, + { + "epoch": 1.5851662404092073, + "grad_norm": 10.548704147338867, + "learning_rate": 1e-06, + "loss": 0.2318, + "mean_token_accuracy": 0.9004417061805725, + "num_tokens": 9971949.0, + "step": 3099 + }, + { + "epoch": 1.5856777493606138, + "grad_norm": 10.970016479492188, + "learning_rate": 1e-06, + "loss": 0.3266, + "mean_token_accuracy": 0.8753887414932251, + "num_tokens": 9975769.0, + "step": 3100 + }, + { + "epoch": 1.5861892583120205, + "grad_norm": 9.856474876403809, + "learning_rate": 1e-06, + "loss": 0.336, + "mean_token_accuracy": 0.8842858672142029, + "num_tokens": 9979064.0, + "step": 3101 + }, + { + "epoch": 1.586700767263427, + "grad_norm": 9.91379165649414, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.8879377841949463, + "num_tokens": 9982427.0, + "step": 3102 + }, + { + "epoch": 1.5872122762148337, + "grad_norm": 10.50040340423584, + "learning_rate": 1e-06, + "loss": 0.2739, + "mean_token_accuracy": 0.8907811641693115, + "num_tokens": 9985506.0, + "step": 3103 + }, + { + "epoch": 1.5877237851662405, + "grad_norm": 9.274961471557617, + "learning_rate": 1e-06, + "loss": 0.2487, + "mean_token_accuracy": 0.8883101940155029, + "num_tokens": 9988687.0, + "step": 3104 + }, + { + "epoch": 1.5882352941176472, + "grad_norm": 7.931965351104736, + "learning_rate": 1e-06, + "loss": 0.243, + "mean_token_accuracy": 0.9024320840835571, + "num_tokens": 9991426.0, + "step": 3105 + }, + { + "epoch": 1.5887468030690537, + "grad_norm": 7.29335880279541, + "learning_rate": 1e-06, + "loss": 0.1456, + "mean_token_accuracy": 0.9071899056434631, + "num_tokens": 9994431.0, + "step": 3106 + }, + { + "epoch": 1.5892583120204602, + "grad_norm": 10.194864273071289, + "learning_rate": 1e-06, + "loss": 0.2265, + "mean_token_accuracy": 0.9237654209136963, + "num_tokens": 9997430.0, + "step": 3107 + }, + { + "epoch": 1.589769820971867, + "grad_norm": 6.878809928894043, + "learning_rate": 1e-06, + "loss": 0.1817, + "mean_token_accuracy": 0.9070170521736145, + "num_tokens": 10001146.0, + "step": 3108 + }, + { + "epoch": 1.5902813299232736, + "grad_norm": 8.35495376586914, + "learning_rate": 1e-06, + "loss": 0.2353, + "mean_token_accuracy": 0.8963255882263184, + "num_tokens": 10004418.0, + "step": 3109 + }, + { + "epoch": 1.5907928388746804, + "grad_norm": 10.641709327697754, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.862542450428009, + "num_tokens": 10007335.0, + "step": 3110 + }, + { + "epoch": 1.591304347826087, + "grad_norm": 12.763171195983887, + "learning_rate": 1e-06, + "loss": 0.3081, + "mean_token_accuracy": 0.8952069282531738, + "num_tokens": 10010147.0, + "step": 3111 + }, + { + "epoch": 1.5918158567774936, + "grad_norm": 7.600194454193115, + "learning_rate": 1e-06, + "loss": 0.2161, + "mean_token_accuracy": 0.905221700668335, + "num_tokens": 10013227.0, + "step": 3112 + }, + { + "epoch": 1.5923273657289, + "grad_norm": 10.490277290344238, + "learning_rate": 1e-06, + "loss": 0.269, + "mean_token_accuracy": 0.9109852313995361, + "num_tokens": 10015917.0, + "step": 3113 + }, + { + "epoch": 1.5928388746803068, + "grad_norm": 8.361187934875488, + "learning_rate": 1e-06, + "loss": 0.2401, + "mean_token_accuracy": 0.8886577486991882, + "num_tokens": 10019374.0, + "step": 3114 + }, + { + "epoch": 1.5933503836317136, + "grad_norm": 7.792890548706055, + "learning_rate": 1e-06, + "loss": 0.1821, + "mean_token_accuracy": 0.9184344410896301, + "num_tokens": 10023223.0, + "step": 3115 + }, + { + "epoch": 1.5938618925831203, + "grad_norm": 10.878156661987305, + "learning_rate": 1e-06, + "loss": 0.29, + "mean_token_accuracy": 0.8788076639175415, + "num_tokens": 10026324.0, + "step": 3116 + }, + { + "epoch": 1.594373401534527, + "grad_norm": 7.128375053405762, + "learning_rate": 1e-06, + "loss": 0.19, + "mean_token_accuracy": 0.9296363592147827, + "num_tokens": 10029989.0, + "step": 3117 + }, + { + "epoch": 1.5948849104859335, + "grad_norm": 11.690553665161133, + "learning_rate": 1e-06, + "loss": 0.2166, + "mean_token_accuracy": 0.9200379848480225, + "num_tokens": 10033377.0, + "step": 3118 + }, + { + "epoch": 1.59539641943734, + "grad_norm": 11.242384910583496, + "learning_rate": 1e-06, + "loss": 0.265, + "mean_token_accuracy": 0.9082083702087402, + "num_tokens": 10036707.0, + "step": 3119 + }, + { + "epoch": 1.5959079283887467, + "grad_norm": 7.450801849365234, + "learning_rate": 1e-06, + "loss": 0.1293, + "mean_token_accuracy": 0.9442641735076904, + "num_tokens": 10042139.0, + "step": 3120 + }, + { + "epoch": 1.5964194373401535, + "grad_norm": 9.633010864257812, + "learning_rate": 1e-06, + "loss": 0.2579, + "mean_token_accuracy": 0.8868436813354492, + "num_tokens": 10045726.0, + "step": 3121 + }, + { + "epoch": 1.5969309462915602, + "grad_norm": 6.320951461791992, + "learning_rate": 1e-06, + "loss": 0.1425, + "mean_token_accuracy": 0.9277617931365967, + "num_tokens": 10048889.0, + "step": 3122 + }, + { + "epoch": 1.597442455242967, + "grad_norm": 7.862182140350342, + "learning_rate": 1e-06, + "loss": 0.2311, + "mean_token_accuracy": 0.9023782014846802, + "num_tokens": 10052093.0, + "step": 3123 + }, + { + "epoch": 1.5979539641943734, + "grad_norm": 8.623570442199707, + "learning_rate": 1e-06, + "loss": 0.2447, + "mean_token_accuracy": 0.8775570392608643, + "num_tokens": 10055401.0, + "step": 3124 + }, + { + "epoch": 1.59846547314578, + "grad_norm": 11.349642753601074, + "learning_rate": 1e-06, + "loss": 0.2719, + "mean_token_accuracy": 0.8724378347396851, + "num_tokens": 10058447.0, + "step": 3125 + }, + { + "epoch": 1.5989769820971866, + "grad_norm": 11.253616333007812, + "learning_rate": 1e-06, + "loss": 0.2673, + "mean_token_accuracy": 0.9046159982681274, + "num_tokens": 10061234.0, + "step": 3126 + }, + { + "epoch": 1.5994884910485934, + "grad_norm": 7.190959930419922, + "learning_rate": 1e-06, + "loss": 0.2357, + "mean_token_accuracy": 0.8908618688583374, + "num_tokens": 10064243.0, + "step": 3127 + }, + { + "epoch": 1.6, + "grad_norm": 10.933338165283203, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8914289474487305, + "num_tokens": 10067001.0, + "step": 3128 + }, + { + "epoch": 1.6005115089514068, + "grad_norm": 12.404141426086426, + "learning_rate": 1e-06, + "loss": 0.2488, + "mean_token_accuracy": 0.9133620262145996, + "num_tokens": 10071492.0, + "step": 3129 + }, + { + "epoch": 1.6010230179028133, + "grad_norm": 9.235118865966797, + "learning_rate": 1e-06, + "loss": 0.2691, + "mean_token_accuracy": 0.9054509401321411, + "num_tokens": 10074175.0, + "step": 3130 + }, + { + "epoch": 1.6015345268542198, + "grad_norm": 11.210700035095215, + "learning_rate": 1e-06, + "loss": 0.2323, + "mean_token_accuracy": 0.9012575149536133, + "num_tokens": 10077786.0, + "step": 3131 + }, + { + "epoch": 1.6020460358056265, + "grad_norm": 6.5098748207092285, + "learning_rate": 1e-06, + "loss": 0.2043, + "mean_token_accuracy": 0.892155647277832, + "num_tokens": 10081484.0, + "step": 3132 + }, + { + "epoch": 1.6025575447570333, + "grad_norm": 11.71800708770752, + "learning_rate": 1e-06, + "loss": 0.2666, + "mean_token_accuracy": 0.8856448531150818, + "num_tokens": 10084390.0, + "step": 3133 + }, + { + "epoch": 1.60306905370844, + "grad_norm": 12.12025260925293, + "learning_rate": 1e-06, + "loss": 0.2458, + "mean_token_accuracy": 0.8860739469528198, + "num_tokens": 10087938.0, + "step": 3134 + }, + { + "epoch": 1.6035805626598465, + "grad_norm": 6.103675365447998, + "learning_rate": 1e-06, + "loss": 0.1895, + "mean_token_accuracy": 0.9371147155761719, + "num_tokens": 10090926.0, + "step": 3135 + }, + { + "epoch": 1.6040920716112532, + "grad_norm": 5.567747592926025, + "learning_rate": 1e-06, + "loss": 0.1595, + "mean_token_accuracy": 0.9276198148727417, + "num_tokens": 10095127.0, + "step": 3136 + }, + { + "epoch": 1.6046035805626597, + "grad_norm": 9.921294212341309, + "learning_rate": 1e-06, + "loss": 0.2812, + "mean_token_accuracy": 0.9165963530540466, + "num_tokens": 10098387.0, + "step": 3137 + }, + { + "epoch": 1.6051150895140665, + "grad_norm": 9.233597755432129, + "learning_rate": 1e-06, + "loss": 0.2374, + "mean_token_accuracy": 0.8714087605476379, + "num_tokens": 10101271.0, + "step": 3138 + }, + { + "epoch": 1.6056265984654732, + "grad_norm": 12.030258178710938, + "learning_rate": 1e-06, + "loss": 0.228, + "mean_token_accuracy": 0.9129461646080017, + "num_tokens": 10104234.0, + "step": 3139 + }, + { + "epoch": 1.60613810741688, + "grad_norm": 10.138401985168457, + "learning_rate": 1e-06, + "loss": 0.2582, + "mean_token_accuracy": 0.8889690637588501, + "num_tokens": 10107855.0, + "step": 3140 + }, + { + "epoch": 1.6066496163682864, + "grad_norm": 18.113155364990234, + "learning_rate": 1e-06, + "loss": 0.4392, + "mean_token_accuracy": 0.8509005308151245, + "num_tokens": 10110630.0, + "step": 3141 + }, + { + "epoch": 1.6071611253196931, + "grad_norm": 12.418489456176758, + "learning_rate": 1e-06, + "loss": 0.2488, + "mean_token_accuracy": 0.8915014266967773, + "num_tokens": 10114086.0, + "step": 3142 + }, + { + "epoch": 1.6076726342710996, + "grad_norm": 9.503786087036133, + "learning_rate": 1e-06, + "loss": 0.3028, + "mean_token_accuracy": 0.8717502951622009, + "num_tokens": 10116869.0, + "step": 3143 + }, + { + "epoch": 1.6081841432225064, + "grad_norm": 7.530333042144775, + "learning_rate": 1e-06, + "loss": 0.1909, + "mean_token_accuracy": 0.901962161064148, + "num_tokens": 10120510.0, + "step": 3144 + }, + { + "epoch": 1.608695652173913, + "grad_norm": 10.641215324401855, + "learning_rate": 1e-06, + "loss": 0.2889, + "mean_token_accuracy": 0.8864670991897583, + "num_tokens": 10123147.0, + "step": 3145 + }, + { + "epoch": 1.6092071611253198, + "grad_norm": 8.43759822845459, + "learning_rate": 1e-06, + "loss": 0.2147, + "mean_token_accuracy": 0.9084576368331909, + "num_tokens": 10126001.0, + "step": 3146 + }, + { + "epoch": 1.6097186700767263, + "grad_norm": 10.676647186279297, + "learning_rate": 1e-06, + "loss": 0.3335, + "mean_token_accuracy": 0.8704566955566406, + "num_tokens": 10129704.0, + "step": 3147 + }, + { + "epoch": 1.610230179028133, + "grad_norm": 11.079033851623535, + "learning_rate": 1e-06, + "loss": 0.2693, + "mean_token_accuracy": 0.8614329695701599, + "num_tokens": 10132888.0, + "step": 3148 + }, + { + "epoch": 1.6107416879795395, + "grad_norm": 11.82433032989502, + "learning_rate": 1e-06, + "loss": 0.3543, + "mean_token_accuracy": 0.8501625061035156, + "num_tokens": 10135364.0, + "step": 3149 + }, + { + "epoch": 1.6112531969309463, + "grad_norm": 6.274867057800293, + "learning_rate": 1e-06, + "loss": 0.1658, + "mean_token_accuracy": 0.9241483211517334, + "num_tokens": 10138516.0, + "step": 3150 + }, + { + "epoch": 1.611764705882353, + "grad_norm": 10.298154830932617, + "learning_rate": 1e-06, + "loss": 0.2685, + "mean_token_accuracy": 0.9087077379226685, + "num_tokens": 10141774.0, + "step": 3151 + }, + { + "epoch": 1.6122762148337597, + "grad_norm": 6.754014015197754, + "learning_rate": 1e-06, + "loss": 0.173, + "mean_token_accuracy": 0.9324764013290405, + "num_tokens": 10144882.0, + "step": 3152 + }, + { + "epoch": 1.6127877237851662, + "grad_norm": 10.195101737976074, + "learning_rate": 1e-06, + "loss": 0.232, + "mean_token_accuracy": 0.901889979839325, + "num_tokens": 10148497.0, + "step": 3153 + }, + { + "epoch": 1.613299232736573, + "grad_norm": 12.917954444885254, + "learning_rate": 1e-06, + "loss": 0.4574, + "mean_token_accuracy": 0.8194377422332764, + "num_tokens": 10151438.0, + "step": 3154 + }, + { + "epoch": 1.6138107416879794, + "grad_norm": 8.512420654296875, + "learning_rate": 1e-06, + "loss": 0.2133, + "mean_token_accuracy": 0.890508234500885, + "num_tokens": 10154857.0, + "step": 3155 + }, + { + "epoch": 1.6143222506393862, + "grad_norm": 12.688522338867188, + "learning_rate": 1e-06, + "loss": 0.292, + "mean_token_accuracy": 0.8784877061843872, + "num_tokens": 10157512.0, + "step": 3156 + }, + { + "epoch": 1.614833759590793, + "grad_norm": 14.015625, + "learning_rate": 1e-06, + "loss": 0.2595, + "mean_token_accuracy": 0.864558219909668, + "num_tokens": 10160612.0, + "step": 3157 + }, + { + "epoch": 1.6153452685421996, + "grad_norm": 13.414484977722168, + "learning_rate": 1e-06, + "loss": 0.345, + "mean_token_accuracy": 0.8780720233917236, + "num_tokens": 10163673.0, + "step": 3158 + }, + { + "epoch": 1.6158567774936061, + "grad_norm": 7.657265663146973, + "learning_rate": 1e-06, + "loss": 0.1805, + "mean_token_accuracy": 0.9136742949485779, + "num_tokens": 10166878.0, + "step": 3159 + }, + { + "epoch": 1.6163682864450126, + "grad_norm": 11.726642608642578, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.8914451599121094, + "num_tokens": 10169761.0, + "step": 3160 + }, + { + "epoch": 1.6168797953964194, + "grad_norm": 7.551482677459717, + "learning_rate": 1e-06, + "loss": 0.1902, + "mean_token_accuracy": 0.9112461805343628, + "num_tokens": 10172967.0, + "step": 3161 + }, + { + "epoch": 1.617391304347826, + "grad_norm": 10.769936561584473, + "learning_rate": 1e-06, + "loss": 0.3511, + "mean_token_accuracy": 0.8857441544532776, + "num_tokens": 10176041.0, + "step": 3162 + }, + { + "epoch": 1.6179028132992328, + "grad_norm": 9.823403358459473, + "learning_rate": 1e-06, + "loss": 0.2271, + "mean_token_accuracy": 0.8966392278671265, + "num_tokens": 10179616.0, + "step": 3163 + }, + { + "epoch": 1.6184143222506395, + "grad_norm": 13.235735893249512, + "learning_rate": 1e-06, + "loss": 0.2328, + "mean_token_accuracy": 0.9064418077468872, + "num_tokens": 10182024.0, + "step": 3164 + }, + { + "epoch": 1.618925831202046, + "grad_norm": 11.238419532775879, + "learning_rate": 1e-06, + "loss": 0.3004, + "mean_token_accuracy": 0.8925251364707947, + "num_tokens": 10185228.0, + "step": 3165 + }, + { + "epoch": 1.6194373401534525, + "grad_norm": 8.159570693969727, + "learning_rate": 1e-06, + "loss": 0.2075, + "mean_token_accuracy": 0.9085150957107544, + "num_tokens": 10189463.0, + "step": 3166 + }, + { + "epoch": 1.6199488491048593, + "grad_norm": 8.680652618408203, + "learning_rate": 1e-06, + "loss": 0.1783, + "mean_token_accuracy": 0.9169919490814209, + "num_tokens": 10193665.0, + "step": 3167 + }, + { + "epoch": 1.620460358056266, + "grad_norm": 14.1497802734375, + "learning_rate": 1e-06, + "loss": 0.2538, + "mean_token_accuracy": 0.8977670669555664, + "num_tokens": 10196974.0, + "step": 3168 + }, + { + "epoch": 1.6209718670076727, + "grad_norm": 6.8415937423706055, + "learning_rate": 1e-06, + "loss": 0.2408, + "mean_token_accuracy": 0.910830020904541, + "num_tokens": 10200953.0, + "step": 3169 + }, + { + "epoch": 1.6214833759590794, + "grad_norm": 6.729895114898682, + "learning_rate": 1e-06, + "loss": 0.1359, + "mean_token_accuracy": 0.9343070387840271, + "num_tokens": 10204051.0, + "step": 3170 + }, + { + "epoch": 1.621994884910486, + "grad_norm": 8.308314323425293, + "learning_rate": 1e-06, + "loss": 0.2728, + "mean_token_accuracy": 0.8842716813087463, + "num_tokens": 10207662.0, + "step": 3171 + }, + { + "epoch": 1.6225063938618924, + "grad_norm": 9.508468627929688, + "learning_rate": 1e-06, + "loss": 0.2162, + "mean_token_accuracy": 0.8975169658660889, + "num_tokens": 10210626.0, + "step": 3172 + }, + { + "epoch": 1.6230179028132992, + "grad_norm": 5.0620269775390625, + "learning_rate": 1e-06, + "loss": 0.1219, + "mean_token_accuracy": 0.926517128944397, + "num_tokens": 10214213.0, + "step": 3173 + }, + { + "epoch": 1.6235294117647059, + "grad_norm": 6.2819600105285645, + "learning_rate": 1e-06, + "loss": 0.142, + "mean_token_accuracy": 0.9134677052497864, + "num_tokens": 10218321.0, + "step": 3174 + }, + { + "epoch": 1.6240409207161126, + "grad_norm": 10.49717903137207, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8903806209564209, + "num_tokens": 10222363.0, + "step": 3175 + }, + { + "epoch": 1.6245524296675193, + "grad_norm": 8.593362808227539, + "learning_rate": 1e-06, + "loss": 0.2348, + "mean_token_accuracy": 0.8927780389785767, + "num_tokens": 10225377.0, + "step": 3176 + }, + { + "epoch": 1.6250639386189258, + "grad_norm": 9.911316871643066, + "learning_rate": 1e-06, + "loss": 0.2043, + "mean_token_accuracy": 0.9107654094696045, + "num_tokens": 10228295.0, + "step": 3177 + }, + { + "epoch": 1.6255754475703323, + "grad_norm": 12.165268898010254, + "learning_rate": 1e-06, + "loss": 0.3672, + "mean_token_accuracy": 0.8769474029541016, + "num_tokens": 10231772.0, + "step": 3178 + }, + { + "epoch": 1.626086956521739, + "grad_norm": 9.36999225616455, + "learning_rate": 1e-06, + "loss": 0.2836, + "mean_token_accuracy": 0.8813974857330322, + "num_tokens": 10234325.0, + "step": 3179 + }, + { + "epoch": 1.6265984654731458, + "grad_norm": 6.491453647613525, + "learning_rate": 1e-06, + "loss": 0.1764, + "mean_token_accuracy": 0.8816123008728027, + "num_tokens": 10238181.0, + "step": 3180 + }, + { + "epoch": 1.6271099744245525, + "grad_norm": 9.220442771911621, + "learning_rate": 1e-06, + "loss": 0.229, + "mean_token_accuracy": 0.9006528854370117, + "num_tokens": 10242782.0, + "step": 3181 + }, + { + "epoch": 1.6276214833759592, + "grad_norm": 10.52321720123291, + "learning_rate": 1e-06, + "loss": 0.3074, + "mean_token_accuracy": 0.9009988903999329, + "num_tokens": 10246252.0, + "step": 3182 + }, + { + "epoch": 1.6281329923273657, + "grad_norm": 12.518316268920898, + "learning_rate": 1e-06, + "loss": 0.2647, + "mean_token_accuracy": 0.8956186175346375, + "num_tokens": 10249700.0, + "step": 3183 + }, + { + "epoch": 1.6286445012787722, + "grad_norm": 11.28691577911377, + "learning_rate": 1e-06, + "loss": 0.2539, + "mean_token_accuracy": 0.9062544703483582, + "num_tokens": 10252515.0, + "step": 3184 + }, + { + "epoch": 1.629156010230179, + "grad_norm": 8.462470054626465, + "learning_rate": 1e-06, + "loss": 0.2574, + "mean_token_accuracy": 0.8851587176322937, + "num_tokens": 10255687.0, + "step": 3185 + }, + { + "epoch": 1.6296675191815857, + "grad_norm": 7.967087268829346, + "learning_rate": 1e-06, + "loss": 0.2039, + "mean_token_accuracy": 0.8955816030502319, + "num_tokens": 10260619.0, + "step": 3186 + }, + { + "epoch": 1.6301790281329924, + "grad_norm": 8.599660873413086, + "learning_rate": 1e-06, + "loss": 0.1971, + "mean_token_accuracy": 0.9092192053794861, + "num_tokens": 10264173.0, + "step": 3187 + }, + { + "epoch": 1.630690537084399, + "grad_norm": 21.20624542236328, + "learning_rate": 1e-06, + "loss": 0.3628, + "mean_token_accuracy": 0.8766456842422485, + "num_tokens": 10267317.0, + "step": 3188 + }, + { + "epoch": 1.6312020460358057, + "grad_norm": 8.183389663696289, + "learning_rate": 1e-06, + "loss": 0.2193, + "mean_token_accuracy": 0.9105823040008545, + "num_tokens": 10269622.0, + "step": 3189 + }, + { + "epoch": 1.6317135549872122, + "grad_norm": 9.577706336975098, + "learning_rate": 1e-06, + "loss": 0.2149, + "mean_token_accuracy": 0.9230014085769653, + "num_tokens": 10272370.0, + "step": 3190 + }, + { + "epoch": 1.6322250639386189, + "grad_norm": 7.963017463684082, + "learning_rate": 1e-06, + "loss": 0.2045, + "mean_token_accuracy": 0.906679093837738, + "num_tokens": 10275781.0, + "step": 3191 + }, + { + "epoch": 1.6327365728900256, + "grad_norm": 13.23799991607666, + "learning_rate": 1e-06, + "loss": 0.3638, + "mean_token_accuracy": 0.8887779712677002, + "num_tokens": 10279542.0, + "step": 3192 + }, + { + "epoch": 1.6332480818414323, + "grad_norm": 7.042800426483154, + "learning_rate": 1e-06, + "loss": 0.2086, + "mean_token_accuracy": 0.9137497544288635, + "num_tokens": 10283375.0, + "step": 3193 + }, + { + "epoch": 1.6337595907928388, + "grad_norm": 9.139140129089355, + "learning_rate": 1e-06, + "loss": 0.2618, + "mean_token_accuracy": 0.9024206399917603, + "num_tokens": 10286415.0, + "step": 3194 + }, + { + "epoch": 1.6342710997442456, + "grad_norm": 9.734764099121094, + "learning_rate": 1e-06, + "loss": 0.233, + "mean_token_accuracy": 0.8954989314079285, + "num_tokens": 10289886.0, + "step": 3195 + }, + { + "epoch": 1.634782608695652, + "grad_norm": 10.05602741241455, + "learning_rate": 1e-06, + "loss": 0.2481, + "mean_token_accuracy": 0.9067478179931641, + "num_tokens": 10292714.0, + "step": 3196 + }, + { + "epoch": 1.6352941176470588, + "grad_norm": 9.973955154418945, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.8836252689361572, + "num_tokens": 10296601.0, + "step": 3197 + }, + { + "epoch": 1.6358056265984655, + "grad_norm": 7.9865498542785645, + "learning_rate": 1e-06, + "loss": 0.1617, + "mean_token_accuracy": 0.9188215136528015, + "num_tokens": 10299819.0, + "step": 3198 + }, + { + "epoch": 1.6363171355498722, + "grad_norm": 8.442412376403809, + "learning_rate": 1e-06, + "loss": 0.1756, + "mean_token_accuracy": 0.9356071352958679, + "num_tokens": 10303570.0, + "step": 3199 + }, + { + "epoch": 1.6368286445012787, + "grad_norm": 8.995307922363281, + "learning_rate": 1e-06, + "loss": 0.2651, + "mean_token_accuracy": 0.9141217470169067, + "num_tokens": 10306578.0, + "step": 3200 + }, + { + "epoch": 1.6373401534526855, + "grad_norm": 8.56017017364502, + "learning_rate": 1e-06, + "loss": 0.2036, + "mean_token_accuracy": 0.9381052255630493, + "num_tokens": 10311291.0, + "step": 3201 + }, + { + "epoch": 1.637851662404092, + "grad_norm": 10.013983726501465, + "learning_rate": 1e-06, + "loss": 0.2651, + "mean_token_accuracy": 0.9029793739318848, + "num_tokens": 10314134.0, + "step": 3202 + }, + { + "epoch": 1.6383631713554987, + "grad_norm": 10.149547576904297, + "learning_rate": 1e-06, + "loss": 0.2551, + "mean_token_accuracy": 0.8690071105957031, + "num_tokens": 10316995.0, + "step": 3203 + }, + { + "epoch": 1.6388746803069054, + "grad_norm": 9.268896102905273, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.9090691804885864, + "num_tokens": 10319729.0, + "step": 3204 + }, + { + "epoch": 1.6393861892583121, + "grad_norm": 6.9746413230896, + "learning_rate": 1e-06, + "loss": 0.1748, + "mean_token_accuracy": 0.9308493137359619, + "num_tokens": 10322367.0, + "step": 3205 + }, + { + "epoch": 1.6398976982097186, + "grad_norm": 10.43873405456543, + "learning_rate": 1e-06, + "loss": 0.243, + "mean_token_accuracy": 0.895817756652832, + "num_tokens": 10324966.0, + "step": 3206 + }, + { + "epoch": 1.6404092071611254, + "grad_norm": 11.35728931427002, + "learning_rate": 1e-06, + "loss": 0.2304, + "mean_token_accuracy": 0.9025325179100037, + "num_tokens": 10327662.0, + "step": 3207 + }, + { + "epoch": 1.6409207161125319, + "grad_norm": 10.488473892211914, + "learning_rate": 1e-06, + "loss": 0.2882, + "mean_token_accuracy": 0.9109772443771362, + "num_tokens": 10331402.0, + "step": 3208 + }, + { + "epoch": 1.6414322250639386, + "grad_norm": 8.649422645568848, + "learning_rate": 1e-06, + "loss": 0.225, + "mean_token_accuracy": 0.8997557163238525, + "num_tokens": 10333589.0, + "step": 3209 + }, + { + "epoch": 1.6419437340153453, + "grad_norm": 8.0997953414917, + "learning_rate": 1e-06, + "loss": 0.1923, + "mean_token_accuracy": 0.9191041588783264, + "num_tokens": 10336751.0, + "step": 3210 + }, + { + "epoch": 1.642455242966752, + "grad_norm": 10.813031196594238, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8828383684158325, + "num_tokens": 10340166.0, + "step": 3211 + }, + { + "epoch": 1.6429667519181586, + "grad_norm": 11.09253978729248, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8754423260688782, + "num_tokens": 10343425.0, + "step": 3212 + }, + { + "epoch": 1.643478260869565, + "grad_norm": 8.182897567749023, + "learning_rate": 1e-06, + "loss": 0.1716, + "mean_token_accuracy": 0.9039928317070007, + "num_tokens": 10346420.0, + "step": 3213 + }, + { + "epoch": 1.6439897698209718, + "grad_norm": 8.532514572143555, + "learning_rate": 1e-06, + "loss": 0.1829, + "mean_token_accuracy": 0.9352399110794067, + "num_tokens": 10349386.0, + "step": 3214 + }, + { + "epoch": 1.6445012787723785, + "grad_norm": 7.055219650268555, + "learning_rate": 1e-06, + "loss": 0.1939, + "mean_token_accuracy": 0.9064986705780029, + "num_tokens": 10352867.0, + "step": 3215 + }, + { + "epoch": 1.6450127877237852, + "grad_norm": 7.145852565765381, + "learning_rate": 1e-06, + "loss": 0.1605, + "mean_token_accuracy": 0.9095931649208069, + "num_tokens": 10356276.0, + "step": 3216 + }, + { + "epoch": 1.645524296675192, + "grad_norm": 12.642600059509277, + "learning_rate": 1e-06, + "loss": 0.4274, + "mean_token_accuracy": 0.8332319855690002, + "num_tokens": 10359234.0, + "step": 3217 + }, + { + "epoch": 1.6460358056265985, + "grad_norm": 12.714743614196777, + "learning_rate": 1e-06, + "loss": 0.3459, + "mean_token_accuracy": 0.8895676136016846, + "num_tokens": 10362154.0, + "step": 3218 + }, + { + "epoch": 1.646547314578005, + "grad_norm": 13.326123237609863, + "learning_rate": 1e-06, + "loss": 0.4189, + "mean_token_accuracy": 0.8525851964950562, + "num_tokens": 10364958.0, + "step": 3219 + }, + { + "epoch": 1.6470588235294117, + "grad_norm": 16.217830657958984, + "learning_rate": 1e-06, + "loss": 0.2573, + "mean_token_accuracy": 0.908940315246582, + "num_tokens": 10368147.0, + "step": 3220 + }, + { + "epoch": 1.6475703324808184, + "grad_norm": 8.3870210647583, + "learning_rate": 1e-06, + "loss": 0.2273, + "mean_token_accuracy": 0.9174612760543823, + "num_tokens": 10370595.0, + "step": 3221 + }, + { + "epoch": 1.6480818414322251, + "grad_norm": 10.200784683227539, + "learning_rate": 1e-06, + "loss": 0.264, + "mean_token_accuracy": 0.897019624710083, + "num_tokens": 10373270.0, + "step": 3222 + }, + { + "epoch": 1.6485933503836319, + "grad_norm": 6.509662628173828, + "learning_rate": 1e-06, + "loss": 0.1566, + "mean_token_accuracy": 0.9050028324127197, + "num_tokens": 10376487.0, + "step": 3223 + }, + { + "epoch": 1.6491048593350384, + "grad_norm": 11.57075309753418, + "learning_rate": 1e-06, + "loss": 0.2313, + "mean_token_accuracy": 0.9127698540687561, + "num_tokens": 10379245.0, + "step": 3224 + }, + { + "epoch": 1.6496163682864449, + "grad_norm": 10.738129615783691, + "learning_rate": 1e-06, + "loss": 0.2495, + "mean_token_accuracy": 0.9080731272697449, + "num_tokens": 10382163.0, + "step": 3225 + }, + { + "epoch": 1.6501278772378516, + "grad_norm": 9.01251220703125, + "learning_rate": 1e-06, + "loss": 0.1925, + "mean_token_accuracy": 0.9453808665275574, + "num_tokens": 10386089.0, + "step": 3226 + }, + { + "epoch": 1.6506393861892583, + "grad_norm": 8.903592109680176, + "learning_rate": 1e-06, + "loss": 0.2495, + "mean_token_accuracy": 0.8908267021179199, + "num_tokens": 10390386.0, + "step": 3227 + }, + { + "epoch": 1.651150895140665, + "grad_norm": 16.294158935546875, + "learning_rate": 1e-06, + "loss": 0.4309, + "mean_token_accuracy": 0.8750125169754028, + "num_tokens": 10393912.0, + "step": 3228 + }, + { + "epoch": 1.6516624040920718, + "grad_norm": 10.324244499206543, + "learning_rate": 1e-06, + "loss": 0.3155, + "mean_token_accuracy": 0.8476574420928955, + "num_tokens": 10396875.0, + "step": 3229 + }, + { + "epoch": 1.6521739130434783, + "grad_norm": 10.378252983093262, + "learning_rate": 1e-06, + "loss": 0.3038, + "mean_token_accuracy": 0.8798875212669373, + "num_tokens": 10399534.0, + "step": 3230 + }, + { + "epoch": 1.6526854219948848, + "grad_norm": 7.500706672668457, + "learning_rate": 1e-06, + "loss": 0.1922, + "mean_token_accuracy": 0.9170474410057068, + "num_tokens": 10403779.0, + "step": 3231 + }, + { + "epoch": 1.6531969309462915, + "grad_norm": 8.270329475402832, + "learning_rate": 1e-06, + "loss": 0.2141, + "mean_token_accuracy": 0.9046338796615601, + "num_tokens": 10406253.0, + "step": 3232 + }, + { + "epoch": 1.6537084398976982, + "grad_norm": 6.811947822570801, + "learning_rate": 1e-06, + "loss": 0.2442, + "mean_token_accuracy": 0.9146692752838135, + "num_tokens": 10410527.0, + "step": 3233 + }, + { + "epoch": 1.654219948849105, + "grad_norm": 11.386951446533203, + "learning_rate": 1e-06, + "loss": 0.2586, + "mean_token_accuracy": 0.8968353867530823, + "num_tokens": 10413631.0, + "step": 3234 + }, + { + "epoch": 1.6547314578005117, + "grad_norm": 12.206825256347656, + "learning_rate": 1e-06, + "loss": 0.2871, + "mean_token_accuracy": 0.8862193822860718, + "num_tokens": 10416577.0, + "step": 3235 + }, + { + "epoch": 1.6552429667519182, + "grad_norm": 6.892454624176025, + "learning_rate": 1e-06, + "loss": 0.2137, + "mean_token_accuracy": 0.9002392888069153, + "num_tokens": 10420605.0, + "step": 3236 + }, + { + "epoch": 1.6557544757033247, + "grad_norm": 6.446257591247559, + "learning_rate": 1e-06, + "loss": 0.1933, + "mean_token_accuracy": 0.9245491027832031, + "num_tokens": 10423983.0, + "step": 3237 + }, + { + "epoch": 1.6562659846547314, + "grad_norm": 25.503562927246094, + "learning_rate": 1e-06, + "loss": 0.2854, + "mean_token_accuracy": 0.8945444822311401, + "num_tokens": 10426427.0, + "step": 3238 + }, + { + "epoch": 1.6567774936061381, + "grad_norm": 6.904967308044434, + "learning_rate": 1e-06, + "loss": 0.1594, + "mean_token_accuracy": 0.9140629768371582, + "num_tokens": 10429866.0, + "step": 3239 + }, + { + "epoch": 1.6572890025575449, + "grad_norm": 10.415745735168457, + "learning_rate": 1e-06, + "loss": 0.2683, + "mean_token_accuracy": 0.8927760124206543, + "num_tokens": 10433342.0, + "step": 3240 + }, + { + "epoch": 1.6578005115089514, + "grad_norm": 14.238094329833984, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8677879571914673, + "num_tokens": 10436698.0, + "step": 3241 + }, + { + "epoch": 1.658312020460358, + "grad_norm": 8.085211753845215, + "learning_rate": 1e-06, + "loss": 0.1873, + "mean_token_accuracy": 0.9255909323692322, + "num_tokens": 10439423.0, + "step": 3242 + }, + { + "epoch": 1.6588235294117646, + "grad_norm": 15.203873634338379, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.8649330139160156, + "num_tokens": 10442686.0, + "step": 3243 + }, + { + "epoch": 1.6593350383631713, + "grad_norm": 10.27798080444336, + "learning_rate": 1e-06, + "loss": 0.225, + "mean_token_accuracy": 0.9276858568191528, + "num_tokens": 10445742.0, + "step": 3244 + }, + { + "epoch": 1.659846547314578, + "grad_norm": 8.475693702697754, + "learning_rate": 1e-06, + "loss": 0.2475, + "mean_token_accuracy": 0.8895602226257324, + "num_tokens": 10449751.0, + "step": 3245 + }, + { + "epoch": 1.6603580562659848, + "grad_norm": 6.294806003570557, + "learning_rate": 1e-06, + "loss": 0.136, + "mean_token_accuracy": 0.9375141859054565, + "num_tokens": 10453372.0, + "step": 3246 + }, + { + "epoch": 1.6608695652173913, + "grad_norm": 8.333237648010254, + "learning_rate": 1e-06, + "loss": 0.2703, + "mean_token_accuracy": 0.8845622539520264, + "num_tokens": 10456358.0, + "step": 3247 + }, + { + "epoch": 1.661381074168798, + "grad_norm": 7.138317108154297, + "learning_rate": 1e-06, + "loss": 0.1545, + "mean_token_accuracy": 0.9334011077880859, + "num_tokens": 10460362.0, + "step": 3248 + }, + { + "epoch": 1.6618925831202045, + "grad_norm": 11.949570655822754, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8671740293502808, + "num_tokens": 10463862.0, + "step": 3249 + }, + { + "epoch": 1.6624040920716112, + "grad_norm": 9.940445899963379, + "learning_rate": 1e-06, + "loss": 0.2764, + "mean_token_accuracy": 0.8709439635276794, + "num_tokens": 10466489.0, + "step": 3250 + }, + { + "epoch": 1.662915601023018, + "grad_norm": 7.622842311859131, + "learning_rate": 1e-06, + "loss": 0.2444, + "mean_token_accuracy": 0.8970901966094971, + "num_tokens": 10470860.0, + "step": 3251 + }, + { + "epoch": 1.6634271099744247, + "grad_norm": 11.714641571044922, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.9102819561958313, + "num_tokens": 10474152.0, + "step": 3252 + }, + { + "epoch": 1.6639386189258312, + "grad_norm": 9.811553955078125, + "learning_rate": 1e-06, + "loss": 0.2603, + "mean_token_accuracy": 0.9033886194229126, + "num_tokens": 10477016.0, + "step": 3253 + }, + { + "epoch": 1.664450127877238, + "grad_norm": 11.798426628112793, + "learning_rate": 1e-06, + "loss": 0.3593, + "mean_token_accuracy": 0.8654283285140991, + "num_tokens": 10479867.0, + "step": 3254 + }, + { + "epoch": 1.6649616368286444, + "grad_norm": 8.025064468383789, + "learning_rate": 1e-06, + "loss": 0.165, + "mean_token_accuracy": 0.9226697683334351, + "num_tokens": 10482616.0, + "step": 3255 + }, + { + "epoch": 1.6654731457800511, + "grad_norm": 7.225887775421143, + "learning_rate": 1e-06, + "loss": 0.234, + "mean_token_accuracy": 0.9163985252380371, + "num_tokens": 10485598.0, + "step": 3256 + }, + { + "epoch": 1.6659846547314578, + "grad_norm": 9.038694381713867, + "learning_rate": 1e-06, + "loss": 0.238, + "mean_token_accuracy": 0.9012943506240845, + "num_tokens": 10488291.0, + "step": 3257 + }, + { + "epoch": 1.6664961636828646, + "grad_norm": 4.9458513259887695, + "learning_rate": 1e-06, + "loss": 0.1597, + "mean_token_accuracy": 0.9134212136268616, + "num_tokens": 10492614.0, + "step": 3258 + }, + { + "epoch": 1.667007672634271, + "grad_norm": 10.541753768920898, + "learning_rate": 1e-06, + "loss": 0.3159, + "mean_token_accuracy": 0.8843485116958618, + "num_tokens": 10495639.0, + "step": 3259 + }, + { + "epoch": 1.6675191815856778, + "grad_norm": 15.274483680725098, + "learning_rate": 1e-06, + "loss": 0.3229, + "mean_token_accuracy": 0.8782124519348145, + "num_tokens": 10499075.0, + "step": 3260 + }, + { + "epoch": 1.6680306905370843, + "grad_norm": 8.167170524597168, + "learning_rate": 1e-06, + "loss": 0.2084, + "mean_token_accuracy": 0.917121171951294, + "num_tokens": 10501911.0, + "step": 3261 + }, + { + "epoch": 1.668542199488491, + "grad_norm": 6.893503665924072, + "learning_rate": 1e-06, + "loss": 0.1556, + "mean_token_accuracy": 0.928573727607727, + "num_tokens": 10506482.0, + "step": 3262 + }, + { + "epoch": 1.6690537084398978, + "grad_norm": 8.064567565917969, + "learning_rate": 1e-06, + "loss": 0.2684, + "mean_token_accuracy": 0.8827871680259705, + "num_tokens": 10508861.0, + "step": 3263 + }, + { + "epoch": 1.6695652173913045, + "grad_norm": 10.111151695251465, + "learning_rate": 1e-06, + "loss": 0.2335, + "mean_token_accuracy": 0.9092579483985901, + "num_tokens": 10511399.0, + "step": 3264 + }, + { + "epoch": 1.670076726342711, + "grad_norm": 7.523343563079834, + "learning_rate": 1e-06, + "loss": 0.2654, + "mean_token_accuracy": 0.8906967043876648, + "num_tokens": 10514845.0, + "step": 3265 + }, + { + "epoch": 1.6705882352941175, + "grad_norm": 6.622818470001221, + "learning_rate": 1e-06, + "loss": 0.1321, + "mean_token_accuracy": 0.9267733097076416, + "num_tokens": 10518958.0, + "step": 3266 + }, + { + "epoch": 1.6710997442455242, + "grad_norm": 7.017532825469971, + "learning_rate": 1e-06, + "loss": 0.2163, + "mean_token_accuracy": 0.9294455051422119, + "num_tokens": 10522304.0, + "step": 3267 + }, + { + "epoch": 1.671611253196931, + "grad_norm": 8.14418888092041, + "learning_rate": 1e-06, + "loss": 0.1835, + "mean_token_accuracy": 0.9236142039299011, + "num_tokens": 10525460.0, + "step": 3268 + }, + { + "epoch": 1.6721227621483377, + "grad_norm": 8.611435890197754, + "learning_rate": 1e-06, + "loss": 0.2225, + "mean_token_accuracy": 0.9059970378875732, + "num_tokens": 10528757.0, + "step": 3269 + }, + { + "epoch": 1.6726342710997444, + "grad_norm": 8.359086990356445, + "learning_rate": 1e-06, + "loss": 0.2104, + "mean_token_accuracy": 0.9042062759399414, + "num_tokens": 10531376.0, + "step": 3270 + }, + { + "epoch": 1.6731457800511509, + "grad_norm": 10.471177101135254, + "learning_rate": 1e-06, + "loss": 0.2286, + "mean_token_accuracy": 0.9119307994842529, + "num_tokens": 10534241.0, + "step": 3271 + }, + { + "epoch": 1.6736572890025574, + "grad_norm": 8.97203254699707, + "learning_rate": 1e-06, + "loss": 0.1658, + "mean_token_accuracy": 0.9049051403999329, + "num_tokens": 10537616.0, + "step": 3272 + }, + { + "epoch": 1.6741687979539641, + "grad_norm": 14.838815689086914, + "learning_rate": 1e-06, + "loss": 0.282, + "mean_token_accuracy": 0.8972179889678955, + "num_tokens": 10540594.0, + "step": 3273 + }, + { + "epoch": 1.6746803069053708, + "grad_norm": 9.114075660705566, + "learning_rate": 1e-06, + "loss": 0.147, + "mean_token_accuracy": 0.9419647455215454, + "num_tokens": 10543378.0, + "step": 3274 + }, + { + "epoch": 1.6751918158567776, + "grad_norm": 10.025110244750977, + "learning_rate": 1e-06, + "loss": 0.2115, + "mean_token_accuracy": 0.9231897592544556, + "num_tokens": 10547084.0, + "step": 3275 + }, + { + "epoch": 1.6757033248081843, + "grad_norm": 14.650654792785645, + "learning_rate": 1e-06, + "loss": 0.2892, + "mean_token_accuracy": 0.8882565498352051, + "num_tokens": 10549830.0, + "step": 3276 + }, + { + "epoch": 1.6762148337595908, + "grad_norm": 7.356710433959961, + "learning_rate": 1e-06, + "loss": 0.2157, + "mean_token_accuracy": 0.894465446472168, + "num_tokens": 10553531.0, + "step": 3277 + }, + { + "epoch": 1.6767263427109973, + "grad_norm": 9.10318660736084, + "learning_rate": 1e-06, + "loss": 0.2137, + "mean_token_accuracy": 0.9179826974868774, + "num_tokens": 10556588.0, + "step": 3278 + }, + { + "epoch": 1.677237851662404, + "grad_norm": 11.388260841369629, + "learning_rate": 1e-06, + "loss": 0.3619, + "mean_token_accuracy": 0.8695428967475891, + "num_tokens": 10560064.0, + "step": 3279 + }, + { + "epoch": 1.6777493606138107, + "grad_norm": 8.873132705688477, + "learning_rate": 1e-06, + "loss": 0.1789, + "mean_token_accuracy": 0.8909472227096558, + "num_tokens": 10562881.0, + "step": 3280 + }, + { + "epoch": 1.6782608695652175, + "grad_norm": 8.268045425415039, + "learning_rate": 1e-06, + "loss": 0.2406, + "mean_token_accuracy": 0.8894047737121582, + "num_tokens": 10565644.0, + "step": 3281 + }, + { + "epoch": 1.6787723785166242, + "grad_norm": 8.669053077697754, + "learning_rate": 1e-06, + "loss": 0.2219, + "mean_token_accuracy": 0.9081403017044067, + "num_tokens": 10568426.0, + "step": 3282 + }, + { + "epoch": 1.6792838874680307, + "grad_norm": 10.432759284973145, + "learning_rate": 1e-06, + "loss": 0.2465, + "mean_token_accuracy": 0.910929799079895, + "num_tokens": 10571634.0, + "step": 3283 + }, + { + "epoch": 1.6797953964194372, + "grad_norm": 9.37349796295166, + "learning_rate": 1e-06, + "loss": 0.2118, + "mean_token_accuracy": 0.9177170991897583, + "num_tokens": 10574750.0, + "step": 3284 + }, + { + "epoch": 1.680306905370844, + "grad_norm": 11.415549278259277, + "learning_rate": 1e-06, + "loss": 0.286, + "mean_token_accuracy": 0.8771886825561523, + "num_tokens": 10577571.0, + "step": 3285 + }, + { + "epoch": 1.6808184143222507, + "grad_norm": 7.816883087158203, + "learning_rate": 1e-06, + "loss": 0.1358, + "mean_token_accuracy": 0.9181559085845947, + "num_tokens": 10580865.0, + "step": 3286 + }, + { + "epoch": 1.6813299232736574, + "grad_norm": 15.816133499145508, + "learning_rate": 1e-06, + "loss": 0.2732, + "mean_token_accuracy": 0.9023298025131226, + "num_tokens": 10583813.0, + "step": 3287 + }, + { + "epoch": 1.681841432225064, + "grad_norm": 11.118762016296387, + "learning_rate": 1e-06, + "loss": 0.2562, + "mean_token_accuracy": 0.9058537483215332, + "num_tokens": 10587113.0, + "step": 3288 + }, + { + "epoch": 1.6823529411764706, + "grad_norm": 9.310400009155273, + "learning_rate": 1e-06, + "loss": 0.2491, + "mean_token_accuracy": 0.8800464272499084, + "num_tokens": 10590241.0, + "step": 3289 + }, + { + "epoch": 1.682864450127877, + "grad_norm": 9.05167293548584, + "learning_rate": 1e-06, + "loss": 0.2653, + "mean_token_accuracy": 0.907317042350769, + "num_tokens": 10593845.0, + "step": 3290 + }, + { + "epoch": 1.6833759590792838, + "grad_norm": 12.338994026184082, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.8632688522338867, + "num_tokens": 10597235.0, + "step": 3291 + }, + { + "epoch": 1.6838874680306906, + "grad_norm": 9.818622589111328, + "learning_rate": 1e-06, + "loss": 0.2448, + "mean_token_accuracy": 0.8972561359405518, + "num_tokens": 10600084.0, + "step": 3292 + }, + { + "epoch": 1.6843989769820973, + "grad_norm": 13.336523056030273, + "learning_rate": 1e-06, + "loss": 0.3867, + "mean_token_accuracy": 0.8618988990783691, + "num_tokens": 10604107.0, + "step": 3293 + }, + { + "epoch": 1.6849104859335038, + "grad_norm": 9.134251594543457, + "learning_rate": 1e-06, + "loss": 0.2239, + "mean_token_accuracy": 0.8688053488731384, + "num_tokens": 10608825.0, + "step": 3294 + }, + { + "epoch": 1.6854219948849105, + "grad_norm": 13.0199556350708, + "learning_rate": 1e-06, + "loss": 0.2597, + "mean_token_accuracy": 0.8978614211082458, + "num_tokens": 10612571.0, + "step": 3295 + }, + { + "epoch": 1.685933503836317, + "grad_norm": 10.98143482208252, + "learning_rate": 1e-06, + "loss": 0.3351, + "mean_token_accuracy": 0.8903392553329468, + "num_tokens": 10615967.0, + "step": 3296 + }, + { + "epoch": 1.6864450127877237, + "grad_norm": 9.044495582580566, + "learning_rate": 1e-06, + "loss": 0.2564, + "mean_token_accuracy": 0.8998270630836487, + "num_tokens": 10619307.0, + "step": 3297 + }, + { + "epoch": 1.6869565217391305, + "grad_norm": 9.788028717041016, + "learning_rate": 1e-06, + "loss": 0.3386, + "mean_token_accuracy": 0.873449444770813, + "num_tokens": 10622819.0, + "step": 3298 + }, + { + "epoch": 1.6874680306905372, + "grad_norm": 9.60610580444336, + "learning_rate": 1e-06, + "loss": 0.2793, + "mean_token_accuracy": 0.8993198871612549, + "num_tokens": 10625434.0, + "step": 3299 + }, + { + "epoch": 1.6879795396419437, + "grad_norm": 10.709124565124512, + "learning_rate": 1e-06, + "loss": 0.3445, + "mean_token_accuracy": 0.8599823713302612, + "num_tokens": 10627985.0, + "step": 3300 + }, + { + "epoch": 1.6884910485933504, + "grad_norm": 9.156899452209473, + "learning_rate": 1e-06, + "loss": 0.1747, + "mean_token_accuracy": 0.9369765520095825, + "num_tokens": 10630309.0, + "step": 3301 + }, + { + "epoch": 1.689002557544757, + "grad_norm": 6.212222099304199, + "learning_rate": 1e-06, + "loss": 0.1966, + "mean_token_accuracy": 0.9099303483963013, + "num_tokens": 10633649.0, + "step": 3302 + }, + { + "epoch": 1.6895140664961636, + "grad_norm": 8.372817039489746, + "learning_rate": 1e-06, + "loss": 0.2245, + "mean_token_accuracy": 0.9031444191932678, + "num_tokens": 10636727.0, + "step": 3303 + }, + { + "epoch": 1.6900255754475704, + "grad_norm": 10.403034210205078, + "learning_rate": 1e-06, + "loss": 0.1876, + "mean_token_accuracy": 0.9226542711257935, + "num_tokens": 10640418.0, + "step": 3304 + }, + { + "epoch": 1.690537084398977, + "grad_norm": 6.831384181976318, + "learning_rate": 1e-06, + "loss": 0.1994, + "mean_token_accuracy": 0.8933939933776855, + "num_tokens": 10643506.0, + "step": 3305 + }, + { + "epoch": 1.6910485933503836, + "grad_norm": 10.256697654724121, + "learning_rate": 1e-06, + "loss": 0.2986, + "mean_token_accuracy": 0.8936996459960938, + "num_tokens": 10646369.0, + "step": 3306 + }, + { + "epoch": 1.6915601023017903, + "grad_norm": 16.93004608154297, + "learning_rate": 1e-06, + "loss": 0.324, + "mean_token_accuracy": 0.8556476831436157, + "num_tokens": 10649176.0, + "step": 3307 + }, + { + "epoch": 1.6920716112531968, + "grad_norm": 9.791399955749512, + "learning_rate": 1e-06, + "loss": 0.2704, + "mean_token_accuracy": 0.8884109854698181, + "num_tokens": 10652169.0, + "step": 3308 + }, + { + "epoch": 1.6925831202046036, + "grad_norm": 13.015942573547363, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.9039245247840881, + "num_tokens": 10654961.0, + "step": 3309 + }, + { + "epoch": 1.6930946291560103, + "grad_norm": 7.218542575836182, + "learning_rate": 1e-06, + "loss": 0.1672, + "mean_token_accuracy": 0.9073101282119751, + "num_tokens": 10658568.0, + "step": 3310 + }, + { + "epoch": 1.693606138107417, + "grad_norm": 8.56124496459961, + "learning_rate": 1e-06, + "loss": 0.2753, + "mean_token_accuracy": 0.8824759721755981, + "num_tokens": 10661710.0, + "step": 3311 + }, + { + "epoch": 1.6941176470588235, + "grad_norm": 8.6348295211792, + "learning_rate": 1e-06, + "loss": 0.2159, + "mean_token_accuracy": 0.9281671047210693, + "num_tokens": 10664653.0, + "step": 3312 + }, + { + "epoch": 1.6946291560102302, + "grad_norm": 8.868280410766602, + "learning_rate": 1e-06, + "loss": 0.3134, + "mean_token_accuracy": 0.8705129623413086, + "num_tokens": 10668014.0, + "step": 3313 + }, + { + "epoch": 1.6951406649616367, + "grad_norm": 12.079963684082031, + "learning_rate": 1e-06, + "loss": 0.2953, + "mean_token_accuracy": 0.867264986038208, + "num_tokens": 10670650.0, + "step": 3314 + }, + { + "epoch": 1.6956521739130435, + "grad_norm": 15.314260482788086, + "learning_rate": 1e-06, + "loss": 0.3133, + "mean_token_accuracy": 0.8818453550338745, + "num_tokens": 10673311.0, + "step": 3315 + }, + { + "epoch": 1.6961636828644502, + "grad_norm": 10.2396821975708, + "learning_rate": 1e-06, + "loss": 0.3097, + "mean_token_accuracy": 0.8771383166313171, + "num_tokens": 10676790.0, + "step": 3316 + }, + { + "epoch": 1.696675191815857, + "grad_norm": 6.494268894195557, + "learning_rate": 1e-06, + "loss": 0.1798, + "mean_token_accuracy": 0.923265278339386, + "num_tokens": 10680972.0, + "step": 3317 + }, + { + "epoch": 1.6971867007672634, + "grad_norm": 8.285237312316895, + "learning_rate": 1e-06, + "loss": 0.2691, + "mean_token_accuracy": 0.8956252932548523, + "num_tokens": 10684002.0, + "step": 3318 + }, + { + "epoch": 1.69769820971867, + "grad_norm": 11.488415718078613, + "learning_rate": 1e-06, + "loss": 0.2939, + "mean_token_accuracy": 0.8880442976951599, + "num_tokens": 10686982.0, + "step": 3319 + }, + { + "epoch": 1.6982097186700766, + "grad_norm": 10.669903755187988, + "learning_rate": 1e-06, + "loss": 0.2594, + "mean_token_accuracy": 0.9152135848999023, + "num_tokens": 10689804.0, + "step": 3320 + }, + { + "epoch": 1.6987212276214834, + "grad_norm": 9.509424209594727, + "learning_rate": 1e-06, + "loss": 0.2643, + "mean_token_accuracy": 0.8890185952186584, + "num_tokens": 10692381.0, + "step": 3321 + }, + { + "epoch": 1.69923273657289, + "grad_norm": 9.86532974243164, + "learning_rate": 1e-06, + "loss": 0.3499, + "mean_token_accuracy": 0.863631546497345, + "num_tokens": 10694928.0, + "step": 3322 + }, + { + "epoch": 1.6997442455242968, + "grad_norm": 8.642650604248047, + "learning_rate": 1e-06, + "loss": 0.201, + "mean_token_accuracy": 0.8904297351837158, + "num_tokens": 10698495.0, + "step": 3323 + }, + { + "epoch": 1.7002557544757033, + "grad_norm": 8.414300918579102, + "learning_rate": 1e-06, + "loss": 0.2414, + "mean_token_accuracy": 0.8751279711723328, + "num_tokens": 10702433.0, + "step": 3324 + }, + { + "epoch": 1.7007672634271098, + "grad_norm": 10.832061767578125, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8959453701972961, + "num_tokens": 10705492.0, + "step": 3325 + }, + { + "epoch": 1.7012787723785165, + "grad_norm": 8.612594604492188, + "learning_rate": 1e-06, + "loss": 0.2318, + "mean_token_accuracy": 0.8992751836776733, + "num_tokens": 10708981.0, + "step": 3326 + }, + { + "epoch": 1.7017902813299233, + "grad_norm": 13.66185474395752, + "learning_rate": 1e-06, + "loss": 0.3201, + "mean_token_accuracy": 0.8743881583213806, + "num_tokens": 10711931.0, + "step": 3327 + }, + { + "epoch": 1.70230179028133, + "grad_norm": 7.314074516296387, + "learning_rate": 1e-06, + "loss": 0.2312, + "mean_token_accuracy": 0.8997883796691895, + "num_tokens": 10715191.0, + "step": 3328 + }, + { + "epoch": 1.7028132992327367, + "grad_norm": 10.338339805603027, + "learning_rate": 1e-06, + "loss": 0.3222, + "mean_token_accuracy": 0.8698407411575317, + "num_tokens": 10718754.0, + "step": 3329 + }, + { + "epoch": 1.7033248081841432, + "grad_norm": 10.577557563781738, + "learning_rate": 1e-06, + "loss": 0.2543, + "mean_token_accuracy": 0.8890658020973206, + "num_tokens": 10721411.0, + "step": 3330 + }, + { + "epoch": 1.7038363171355497, + "grad_norm": 8.078808784484863, + "learning_rate": 1e-06, + "loss": 0.2218, + "mean_token_accuracy": 0.9181358814239502, + "num_tokens": 10724772.0, + "step": 3331 + }, + { + "epoch": 1.7043478260869565, + "grad_norm": 6.476022243499756, + "learning_rate": 1e-06, + "loss": 0.1587, + "mean_token_accuracy": 0.9282214641571045, + "num_tokens": 10729876.0, + "step": 3332 + }, + { + "epoch": 1.7048593350383632, + "grad_norm": 8.143040657043457, + "learning_rate": 1e-06, + "loss": 0.2174, + "mean_token_accuracy": 0.9168786406517029, + "num_tokens": 10732506.0, + "step": 3333 + }, + { + "epoch": 1.70537084398977, + "grad_norm": 6.501911640167236, + "learning_rate": 1e-06, + "loss": 0.1979, + "mean_token_accuracy": 0.9103397727012634, + "num_tokens": 10736088.0, + "step": 3334 + }, + { + "epoch": 1.7058823529411766, + "grad_norm": 11.377431869506836, + "learning_rate": 1e-06, + "loss": 0.3109, + "mean_token_accuracy": 0.8688251972198486, + "num_tokens": 10738974.0, + "step": 3335 + }, + { + "epoch": 1.7063938618925831, + "grad_norm": 10.126946449279785, + "learning_rate": 1e-06, + "loss": 0.1912, + "mean_token_accuracy": 0.9248446822166443, + "num_tokens": 10742562.0, + "step": 3336 + }, + { + "epoch": 1.7069053708439896, + "grad_norm": 12.449790954589844, + "learning_rate": 1e-06, + "loss": 0.2285, + "mean_token_accuracy": 0.9137628078460693, + "num_tokens": 10745499.0, + "step": 3337 + }, + { + "epoch": 1.7074168797953964, + "grad_norm": 17.922771453857422, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.9018689393997192, + "num_tokens": 10748334.0, + "step": 3338 + }, + { + "epoch": 1.707928388746803, + "grad_norm": 10.998291969299316, + "learning_rate": 1e-06, + "loss": 0.257, + "mean_token_accuracy": 0.9013649225234985, + "num_tokens": 10751313.0, + "step": 3339 + }, + { + "epoch": 1.7084398976982098, + "grad_norm": 10.383049011230469, + "learning_rate": 1e-06, + "loss": 0.312, + "mean_token_accuracy": 0.8781235218048096, + "num_tokens": 10754682.0, + "step": 3340 + }, + { + "epoch": 1.7089514066496165, + "grad_norm": 11.376863479614258, + "learning_rate": 1e-06, + "loss": 0.3027, + "mean_token_accuracy": 0.8858060240745544, + "num_tokens": 10757900.0, + "step": 3341 + }, + { + "epoch": 1.709462915601023, + "grad_norm": 12.885891914367676, + "learning_rate": 1e-06, + "loss": 0.2448, + "mean_token_accuracy": 0.8986444473266602, + "num_tokens": 10760658.0, + "step": 3342 + }, + { + "epoch": 1.7099744245524295, + "grad_norm": 9.570771217346191, + "learning_rate": 1e-06, + "loss": 0.3207, + "mean_token_accuracy": 0.8820250630378723, + "num_tokens": 10763651.0, + "step": 3343 + }, + { + "epoch": 1.7104859335038363, + "grad_norm": 4.618025779724121, + "learning_rate": 1e-06, + "loss": 0.1098, + "mean_token_accuracy": 0.9026347398757935, + "num_tokens": 10768419.0, + "step": 3344 + }, + { + "epoch": 1.710997442455243, + "grad_norm": 11.242112159729004, + "learning_rate": 1e-06, + "loss": 0.2775, + "mean_token_accuracy": 0.8808941841125488, + "num_tokens": 10771342.0, + "step": 3345 + }, + { + "epoch": 1.7115089514066497, + "grad_norm": 13.751763343811035, + "learning_rate": 1e-06, + "loss": 0.3388, + "mean_token_accuracy": 0.8738539814949036, + "num_tokens": 10773632.0, + "step": 3346 + }, + { + "epoch": 1.7120204603580562, + "grad_norm": 6.12908935546875, + "learning_rate": 1e-06, + "loss": 0.1475, + "mean_token_accuracy": 0.9237092733383179, + "num_tokens": 10777207.0, + "step": 3347 + }, + { + "epoch": 1.712531969309463, + "grad_norm": 9.9677734375, + "learning_rate": 1e-06, + "loss": 0.2653, + "mean_token_accuracy": 0.8734435439109802, + "num_tokens": 10780681.0, + "step": 3348 + }, + { + "epoch": 1.7130434782608694, + "grad_norm": 9.448634147644043, + "learning_rate": 1e-06, + "loss": 0.2357, + "mean_token_accuracy": 0.9162354469299316, + "num_tokens": 10784131.0, + "step": 3349 + }, + { + "epoch": 1.7135549872122762, + "grad_norm": 9.456660270690918, + "learning_rate": 1e-06, + "loss": 0.223, + "mean_token_accuracy": 0.9106640815734863, + "num_tokens": 10787111.0, + "step": 3350 + }, + { + "epoch": 1.714066496163683, + "grad_norm": 14.421730995178223, + "learning_rate": 1e-06, + "loss": 0.3263, + "mean_token_accuracy": 0.8888721466064453, + "num_tokens": 10790140.0, + "step": 3351 + }, + { + "epoch": 1.7145780051150896, + "grad_norm": 11.965254783630371, + "learning_rate": 1e-06, + "loss": 0.2114, + "mean_token_accuracy": 0.8897820711135864, + "num_tokens": 10793333.0, + "step": 3352 + }, + { + "epoch": 1.7150895140664961, + "grad_norm": 11.19881534576416, + "learning_rate": 1e-06, + "loss": 0.2485, + "mean_token_accuracy": 0.9002727270126343, + "num_tokens": 10796382.0, + "step": 3353 + }, + { + "epoch": 1.7156010230179028, + "grad_norm": 10.813691139221191, + "learning_rate": 1e-06, + "loss": 0.2451, + "mean_token_accuracy": 0.9085078239440918, + "num_tokens": 10800158.0, + "step": 3354 + }, + { + "epoch": 1.7161125319693094, + "grad_norm": 11.073135375976562, + "learning_rate": 1e-06, + "loss": 0.2876, + "mean_token_accuracy": 0.897696852684021, + "num_tokens": 10802599.0, + "step": 3355 + }, + { + "epoch": 1.716624040920716, + "grad_norm": 22.51540756225586, + "learning_rate": 1e-06, + "loss": 0.4017, + "mean_token_accuracy": 0.8896665573120117, + "num_tokens": 10805250.0, + "step": 3356 + }, + { + "epoch": 1.7171355498721228, + "grad_norm": 10.635984420776367, + "learning_rate": 1e-06, + "loss": 0.2703, + "mean_token_accuracy": 0.905192494392395, + "num_tokens": 10808765.0, + "step": 3357 + }, + { + "epoch": 1.7176470588235295, + "grad_norm": 12.326204299926758, + "learning_rate": 1e-06, + "loss": 0.2498, + "mean_token_accuracy": 0.9167906641960144, + "num_tokens": 10811669.0, + "step": 3358 + }, + { + "epoch": 1.718158567774936, + "grad_norm": 9.57500171661377, + "learning_rate": 1e-06, + "loss": 0.253, + "mean_token_accuracy": 0.8873525857925415, + "num_tokens": 10815249.0, + "step": 3359 + }, + { + "epoch": 1.7186700767263428, + "grad_norm": 10.311376571655273, + "learning_rate": 1e-06, + "loss": 0.3015, + "mean_token_accuracy": 0.8705320358276367, + "num_tokens": 10818246.0, + "step": 3360 + }, + { + "epoch": 1.7191815856777493, + "grad_norm": 10.506475448608398, + "learning_rate": 1e-06, + "loss": 0.3279, + "mean_token_accuracy": 0.8928564786911011, + "num_tokens": 10821730.0, + "step": 3361 + }, + { + "epoch": 1.719693094629156, + "grad_norm": 11.265276908874512, + "learning_rate": 1e-06, + "loss": 0.3325, + "mean_token_accuracy": 0.8835636973381042, + "num_tokens": 10824448.0, + "step": 3362 + }, + { + "epoch": 1.7202046035805627, + "grad_norm": 8.787610054016113, + "learning_rate": 1e-06, + "loss": 0.2536, + "mean_token_accuracy": 0.8871047496795654, + "num_tokens": 10827670.0, + "step": 3363 + }, + { + "epoch": 1.7207161125319694, + "grad_norm": 8.732686996459961, + "learning_rate": 1e-06, + "loss": 0.2653, + "mean_token_accuracy": 0.8992849588394165, + "num_tokens": 10831123.0, + "step": 3364 + }, + { + "epoch": 1.721227621483376, + "grad_norm": 9.273665428161621, + "learning_rate": 1e-06, + "loss": 0.2342, + "mean_token_accuracy": 0.9098505973815918, + "num_tokens": 10834368.0, + "step": 3365 + }, + { + "epoch": 1.7217391304347827, + "grad_norm": 10.398266792297363, + "learning_rate": 1e-06, + "loss": 0.314, + "mean_token_accuracy": 0.8826853036880493, + "num_tokens": 10837247.0, + "step": 3366 + }, + { + "epoch": 1.7222506393861892, + "grad_norm": 7.501679420471191, + "learning_rate": 1e-06, + "loss": 0.2092, + "mean_token_accuracy": 0.8987691402435303, + "num_tokens": 10840307.0, + "step": 3367 + }, + { + "epoch": 1.7227621483375959, + "grad_norm": 13.519234657287598, + "learning_rate": 1e-06, + "loss": 0.3302, + "mean_token_accuracy": 0.8816460371017456, + "num_tokens": 10843284.0, + "step": 3368 + }, + { + "epoch": 1.7232736572890026, + "grad_norm": 9.698348045349121, + "learning_rate": 1e-06, + "loss": 0.2924, + "mean_token_accuracy": 0.9060351848602295, + "num_tokens": 10845887.0, + "step": 3369 + }, + { + "epoch": 1.7237851662404093, + "grad_norm": 8.214090347290039, + "learning_rate": 1e-06, + "loss": 0.2471, + "mean_token_accuracy": 0.8997019529342651, + "num_tokens": 10848977.0, + "step": 3370 + }, + { + "epoch": 1.7242966751918158, + "grad_norm": 8.353257179260254, + "learning_rate": 1e-06, + "loss": 0.2335, + "mean_token_accuracy": 0.8964954018592834, + "num_tokens": 10852361.0, + "step": 3371 + }, + { + "epoch": 1.7248081841432223, + "grad_norm": 10.006193161010742, + "learning_rate": 1e-06, + "loss": 0.2271, + "mean_token_accuracy": 0.9071008563041687, + "num_tokens": 10855525.0, + "step": 3372 + }, + { + "epoch": 1.725319693094629, + "grad_norm": 6.329987049102783, + "learning_rate": 1e-06, + "loss": 0.2345, + "mean_token_accuracy": 0.8985295295715332, + "num_tokens": 10859515.0, + "step": 3373 + }, + { + "epoch": 1.7258312020460358, + "grad_norm": 7.375669002532959, + "learning_rate": 1e-06, + "loss": 0.1791, + "mean_token_accuracy": 0.9202018976211548, + "num_tokens": 10862983.0, + "step": 3374 + }, + { + "epoch": 1.7263427109974425, + "grad_norm": 6.881039619445801, + "learning_rate": 1e-06, + "loss": 0.1938, + "mean_token_accuracy": 0.9181354641914368, + "num_tokens": 10865915.0, + "step": 3375 + }, + { + "epoch": 1.7268542199488492, + "grad_norm": 8.667181968688965, + "learning_rate": 1e-06, + "loss": 0.1747, + "mean_token_accuracy": 0.8875527381896973, + "num_tokens": 10868873.0, + "step": 3376 + }, + { + "epoch": 1.7273657289002557, + "grad_norm": 7.843281269073486, + "learning_rate": 1e-06, + "loss": 0.2272, + "mean_token_accuracy": 0.9080426692962646, + "num_tokens": 10871555.0, + "step": 3377 + }, + { + "epoch": 1.7278772378516623, + "grad_norm": 11.473377227783203, + "learning_rate": 1e-06, + "loss": 0.2657, + "mean_token_accuracy": 0.8943837881088257, + "num_tokens": 10874289.0, + "step": 3378 + }, + { + "epoch": 1.728388746803069, + "grad_norm": 13.996098518371582, + "learning_rate": 1e-06, + "loss": 0.3656, + "mean_token_accuracy": 0.860471248626709, + "num_tokens": 10877072.0, + "step": 3379 + }, + { + "epoch": 1.7289002557544757, + "grad_norm": 10.827471733093262, + "learning_rate": 1e-06, + "loss": 0.2233, + "mean_token_accuracy": 0.8951712846755981, + "num_tokens": 10880303.0, + "step": 3380 + }, + { + "epoch": 1.7294117647058824, + "grad_norm": 10.000904083251953, + "learning_rate": 1e-06, + "loss": 0.2934, + "mean_token_accuracy": 0.8810721635818481, + "num_tokens": 10883737.0, + "step": 3381 + }, + { + "epoch": 1.7299232736572892, + "grad_norm": 12.391359329223633, + "learning_rate": 1e-06, + "loss": 0.3112, + "mean_token_accuracy": 0.8939703106880188, + "num_tokens": 10887637.0, + "step": 3382 + }, + { + "epoch": 1.7304347826086957, + "grad_norm": 10.03061294555664, + "learning_rate": 1e-06, + "loss": 0.2293, + "mean_token_accuracy": 0.903667688369751, + "num_tokens": 10890921.0, + "step": 3383 + }, + { + "epoch": 1.7309462915601022, + "grad_norm": 9.643465042114258, + "learning_rate": 1e-06, + "loss": 0.2492, + "mean_token_accuracy": 0.9175122976303101, + "num_tokens": 10893625.0, + "step": 3384 + }, + { + "epoch": 1.7314578005115089, + "grad_norm": 11.157769203186035, + "learning_rate": 1e-06, + "loss": 0.3297, + "mean_token_accuracy": 0.8892337083816528, + "num_tokens": 10896354.0, + "step": 3385 + }, + { + "epoch": 1.7319693094629156, + "grad_norm": 10.424140930175781, + "learning_rate": 1e-06, + "loss": 0.2456, + "mean_token_accuracy": 0.9149927496910095, + "num_tokens": 10899925.0, + "step": 3386 + }, + { + "epoch": 1.7324808184143223, + "grad_norm": 6.60920524597168, + "learning_rate": 1e-06, + "loss": 0.1553, + "mean_token_accuracy": 0.9048413038253784, + "num_tokens": 10903940.0, + "step": 3387 + }, + { + "epoch": 1.732992327365729, + "grad_norm": 12.249272346496582, + "learning_rate": 1e-06, + "loss": 0.2383, + "mean_token_accuracy": 0.895309567451477, + "num_tokens": 10907483.0, + "step": 3388 + }, + { + "epoch": 1.7335038363171356, + "grad_norm": 10.23928165435791, + "learning_rate": 1e-06, + "loss": 0.1931, + "mean_token_accuracy": 0.9289050102233887, + "num_tokens": 10911032.0, + "step": 3389 + }, + { + "epoch": 1.734015345268542, + "grad_norm": 8.602579116821289, + "learning_rate": 1e-06, + "loss": 0.2826, + "mean_token_accuracy": 0.9175224304199219, + "num_tokens": 10914534.0, + "step": 3390 + }, + { + "epoch": 1.7345268542199488, + "grad_norm": 10.340063095092773, + "learning_rate": 1e-06, + "loss": 0.2726, + "mean_token_accuracy": 0.8947062492370605, + "num_tokens": 10918067.0, + "step": 3391 + }, + { + "epoch": 1.7350383631713555, + "grad_norm": 9.793487548828125, + "learning_rate": 1e-06, + "loss": 0.2807, + "mean_token_accuracy": 0.9084477424621582, + "num_tokens": 10921013.0, + "step": 3392 + }, + { + "epoch": 1.7355498721227622, + "grad_norm": 10.33290958404541, + "learning_rate": 1e-06, + "loss": 0.2351, + "mean_token_accuracy": 0.9067962169647217, + "num_tokens": 10924043.0, + "step": 3393 + }, + { + "epoch": 1.736061381074169, + "grad_norm": 10.661746978759766, + "learning_rate": 1e-06, + "loss": 0.2792, + "mean_token_accuracy": 0.8903595209121704, + "num_tokens": 10927379.0, + "step": 3394 + }, + { + "epoch": 1.7365728900255755, + "grad_norm": 6.460738658905029, + "learning_rate": 1e-06, + "loss": 0.1699, + "mean_token_accuracy": 0.9341155886650085, + "num_tokens": 10930966.0, + "step": 3395 + }, + { + "epoch": 1.737084398976982, + "grad_norm": 8.245430946350098, + "learning_rate": 1e-06, + "loss": 0.2109, + "mean_token_accuracy": 0.9134977459907532, + "num_tokens": 10934121.0, + "step": 3396 + }, + { + "epoch": 1.7375959079283887, + "grad_norm": 11.583303451538086, + "learning_rate": 1e-06, + "loss": 0.2558, + "mean_token_accuracy": 0.9091895818710327, + "num_tokens": 10936749.0, + "step": 3397 + }, + { + "epoch": 1.7381074168797954, + "grad_norm": 4.923713207244873, + "learning_rate": 1e-06, + "loss": 0.1389, + "mean_token_accuracy": 0.9292795658111572, + "num_tokens": 10941055.0, + "step": 3398 + }, + { + "epoch": 1.7386189258312021, + "grad_norm": 8.129268646240234, + "learning_rate": 1e-06, + "loss": 0.2111, + "mean_token_accuracy": 0.9035004377365112, + "num_tokens": 10944810.0, + "step": 3399 + }, + { + "epoch": 1.7391304347826086, + "grad_norm": 12.984702110290527, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8832036256790161, + "num_tokens": 10947950.0, + "step": 3400 + }, + { + "epoch": 1.7396419437340154, + "grad_norm": 10.681506156921387, + "learning_rate": 1e-06, + "loss": 0.2801, + "mean_token_accuracy": 0.8904967904090881, + "num_tokens": 10950809.0, + "step": 3401 + }, + { + "epoch": 1.7401534526854219, + "grad_norm": 8.497859954833984, + "learning_rate": 1e-06, + "loss": 0.2189, + "mean_token_accuracy": 0.8892068266868591, + "num_tokens": 10953803.0, + "step": 3402 + }, + { + "epoch": 1.7406649616368286, + "grad_norm": 6.80541467666626, + "learning_rate": 1e-06, + "loss": 0.2057, + "mean_token_accuracy": 0.8904709219932556, + "num_tokens": 10957166.0, + "step": 3403 + }, + { + "epoch": 1.7411764705882353, + "grad_norm": 11.639589309692383, + "learning_rate": 1e-06, + "loss": 0.2542, + "mean_token_accuracy": 0.8809888362884521, + "num_tokens": 10959621.0, + "step": 3404 + }, + { + "epoch": 1.741687979539642, + "grad_norm": 9.86166763305664, + "learning_rate": 1e-06, + "loss": 0.2548, + "mean_token_accuracy": 0.906664252281189, + "num_tokens": 10962702.0, + "step": 3405 + }, + { + "epoch": 1.7421994884910486, + "grad_norm": 8.945453643798828, + "learning_rate": 1e-06, + "loss": 0.2805, + "mean_token_accuracy": 0.8776764273643494, + "num_tokens": 10965905.0, + "step": 3406 + }, + { + "epoch": 1.7427109974424553, + "grad_norm": 7.172375679016113, + "learning_rate": 1e-06, + "loss": 0.1984, + "mean_token_accuracy": 0.9230648279190063, + "num_tokens": 10968262.0, + "step": 3407 + }, + { + "epoch": 1.7432225063938618, + "grad_norm": 10.07280445098877, + "learning_rate": 1e-06, + "loss": 0.3163, + "mean_token_accuracy": 0.8733195662498474, + "num_tokens": 10971151.0, + "step": 3408 + }, + { + "epoch": 1.7437340153452685, + "grad_norm": 7.198575019836426, + "learning_rate": 1e-06, + "loss": 0.2033, + "mean_token_accuracy": 0.9225656986236572, + "num_tokens": 10974371.0, + "step": 3409 + }, + { + "epoch": 1.7442455242966752, + "grad_norm": 9.475950241088867, + "learning_rate": 1e-06, + "loss": 0.278, + "mean_token_accuracy": 0.8945642113685608, + "num_tokens": 10977231.0, + "step": 3410 + }, + { + "epoch": 1.744757033248082, + "grad_norm": 10.008713722229004, + "learning_rate": 1e-06, + "loss": 0.2399, + "mean_token_accuracy": 0.9018012285232544, + "num_tokens": 10980180.0, + "step": 3411 + }, + { + "epoch": 1.7452685421994885, + "grad_norm": 13.16604995727539, + "learning_rate": 1e-06, + "loss": 0.3327, + "mean_token_accuracy": 0.900016188621521, + "num_tokens": 10982963.0, + "step": 3412 + }, + { + "epoch": 1.7457800511508952, + "grad_norm": 7.84549617767334, + "learning_rate": 1e-06, + "loss": 0.1919, + "mean_token_accuracy": 0.9242169857025146, + "num_tokens": 10985743.0, + "step": 3413 + }, + { + "epoch": 1.7462915601023017, + "grad_norm": 8.099442481994629, + "learning_rate": 1e-06, + "loss": 0.2151, + "mean_token_accuracy": 0.8914018869400024, + "num_tokens": 10989555.0, + "step": 3414 + }, + { + "epoch": 1.7468030690537084, + "grad_norm": 7.344818592071533, + "learning_rate": 1e-06, + "loss": 0.1798, + "mean_token_accuracy": 0.9090161323547363, + "num_tokens": 10993113.0, + "step": 3415 + }, + { + "epoch": 1.7473145780051151, + "grad_norm": 6.516770839691162, + "learning_rate": 1e-06, + "loss": 0.2025, + "mean_token_accuracy": 0.9164000749588013, + "num_tokens": 10996915.0, + "step": 3416 + }, + { + "epoch": 1.7478260869565219, + "grad_norm": 7.529324531555176, + "learning_rate": 1e-06, + "loss": 0.1912, + "mean_token_accuracy": 0.9136654734611511, + "num_tokens": 11000022.0, + "step": 3417 + }, + { + "epoch": 1.7483375959079284, + "grad_norm": 8.023058891296387, + "learning_rate": 1e-06, + "loss": 0.1442, + "mean_token_accuracy": 0.9232785701751709, + "num_tokens": 11002773.0, + "step": 3418 + }, + { + "epoch": 1.748849104859335, + "grad_norm": 6.342637062072754, + "learning_rate": 1e-06, + "loss": 0.1868, + "mean_token_accuracy": 0.9060109853744507, + "num_tokens": 11006424.0, + "step": 3419 + }, + { + "epoch": 1.7493606138107416, + "grad_norm": 11.973871231079102, + "learning_rate": 1e-06, + "loss": 0.2894, + "mean_token_accuracy": 0.8941202163696289, + "num_tokens": 11009392.0, + "step": 3420 + }, + { + "epoch": 1.7498721227621483, + "grad_norm": 11.089494705200195, + "learning_rate": 1e-06, + "loss": 0.2954, + "mean_token_accuracy": 0.9026827216148376, + "num_tokens": 11012482.0, + "step": 3421 + }, + { + "epoch": 1.750383631713555, + "grad_norm": 4.882879257202148, + "learning_rate": 1e-06, + "loss": 0.1272, + "mean_token_accuracy": 0.9184259176254272, + "num_tokens": 11017062.0, + "step": 3422 + }, + { + "epoch": 1.7508951406649618, + "grad_norm": 9.825043678283691, + "learning_rate": 1e-06, + "loss": 0.2476, + "mean_token_accuracy": 0.9002138376235962, + "num_tokens": 11020426.0, + "step": 3423 + }, + { + "epoch": 1.7514066496163683, + "grad_norm": 11.920654296875, + "learning_rate": 1e-06, + "loss": 0.2751, + "mean_token_accuracy": 0.899090588092804, + "num_tokens": 11022708.0, + "step": 3424 + }, + { + "epoch": 1.7519181585677748, + "grad_norm": 11.336398124694824, + "learning_rate": 1e-06, + "loss": 0.305, + "mean_token_accuracy": 0.8908859491348267, + "num_tokens": 11025645.0, + "step": 3425 + }, + { + "epoch": 1.7524296675191815, + "grad_norm": 6.0217132568359375, + "learning_rate": 1e-06, + "loss": 0.1674, + "mean_token_accuracy": 0.893552839756012, + "num_tokens": 11029635.0, + "step": 3426 + }, + { + "epoch": 1.7529411764705882, + "grad_norm": 7.503603458404541, + "learning_rate": 1e-06, + "loss": 0.2352, + "mean_token_accuracy": 0.8846651315689087, + "num_tokens": 11033192.0, + "step": 3427 + }, + { + "epoch": 1.753452685421995, + "grad_norm": 8.000825881958008, + "learning_rate": 1e-06, + "loss": 0.2548, + "mean_token_accuracy": 0.8985539674758911, + "num_tokens": 11036962.0, + "step": 3428 + }, + { + "epoch": 1.7539641943734017, + "grad_norm": 8.527788162231445, + "learning_rate": 1e-06, + "loss": 0.2628, + "mean_token_accuracy": 0.8875317573547363, + "num_tokens": 11040279.0, + "step": 3429 + }, + { + "epoch": 1.7544757033248082, + "grad_norm": 12.18313217163086, + "learning_rate": 1e-06, + "loss": 0.2935, + "mean_token_accuracy": 0.891338050365448, + "num_tokens": 11044319.0, + "step": 3430 + }, + { + "epoch": 1.7549872122762147, + "grad_norm": 11.427984237670898, + "learning_rate": 1e-06, + "loss": 0.3259, + "mean_token_accuracy": 0.8782624006271362, + "num_tokens": 11047124.0, + "step": 3431 + }, + { + "epoch": 1.7554987212276214, + "grad_norm": 7.644133567810059, + "learning_rate": 1e-06, + "loss": 0.1902, + "mean_token_accuracy": 0.9288667440414429, + "num_tokens": 11050804.0, + "step": 3432 + }, + { + "epoch": 1.7560102301790281, + "grad_norm": 9.164416313171387, + "learning_rate": 1e-06, + "loss": 0.2248, + "mean_token_accuracy": 0.895901620388031, + "num_tokens": 11053835.0, + "step": 3433 + }, + { + "epoch": 1.7565217391304349, + "grad_norm": 5.342543601989746, + "learning_rate": 1e-06, + "loss": 0.1617, + "mean_token_accuracy": 0.9238934516906738, + "num_tokens": 11057858.0, + "step": 3434 + }, + { + "epoch": 1.7570332480818416, + "grad_norm": 10.27299690246582, + "learning_rate": 1e-06, + "loss": 0.3071, + "mean_token_accuracy": 0.8912451267242432, + "num_tokens": 11061100.0, + "step": 3435 + }, + { + "epoch": 1.757544757033248, + "grad_norm": 7.139278888702393, + "learning_rate": 1e-06, + "loss": 0.1664, + "mean_token_accuracy": 0.9280058145523071, + "num_tokens": 11064444.0, + "step": 3436 + }, + { + "epoch": 1.7580562659846546, + "grad_norm": 8.834243774414062, + "learning_rate": 1e-06, + "loss": 0.2383, + "mean_token_accuracy": 0.9110682010650635, + "num_tokens": 11067593.0, + "step": 3437 + }, + { + "epoch": 1.7585677749360613, + "grad_norm": 8.59643268585205, + "learning_rate": 1e-06, + "loss": 0.2711, + "mean_token_accuracy": 0.8952500820159912, + "num_tokens": 11070524.0, + "step": 3438 + }, + { + "epoch": 1.759079283887468, + "grad_norm": 8.45664119720459, + "learning_rate": 1e-06, + "loss": 0.2556, + "mean_token_accuracy": 0.9029752016067505, + "num_tokens": 11073826.0, + "step": 3439 + }, + { + "epoch": 1.7595907928388748, + "grad_norm": 8.013971328735352, + "learning_rate": 1e-06, + "loss": 0.2431, + "mean_token_accuracy": 0.8949053883552551, + "num_tokens": 11077010.0, + "step": 3440 + }, + { + "epoch": 1.7601023017902815, + "grad_norm": 10.295646667480469, + "learning_rate": 1e-06, + "loss": 0.2054, + "mean_token_accuracy": 0.9019030332565308, + "num_tokens": 11079882.0, + "step": 3441 + }, + { + "epoch": 1.760613810741688, + "grad_norm": 10.364480018615723, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8915743231773376, + "num_tokens": 11083588.0, + "step": 3442 + }, + { + "epoch": 1.7611253196930945, + "grad_norm": 11.977986335754395, + "learning_rate": 1e-06, + "loss": 0.3442, + "mean_token_accuracy": 0.8791281580924988, + "num_tokens": 11086618.0, + "step": 3443 + }, + { + "epoch": 1.7616368286445012, + "grad_norm": 6.988438129425049, + "learning_rate": 1e-06, + "loss": 0.1746, + "mean_token_accuracy": 0.9240871667861938, + "num_tokens": 11089868.0, + "step": 3444 + }, + { + "epoch": 1.762148337595908, + "grad_norm": 6.11106014251709, + "learning_rate": 1e-06, + "loss": 0.2219, + "mean_token_accuracy": 0.8989766836166382, + "num_tokens": 11093635.0, + "step": 3445 + }, + { + "epoch": 1.7626598465473147, + "grad_norm": 7.480775356292725, + "learning_rate": 1e-06, + "loss": 0.2604, + "mean_token_accuracy": 0.9267736673355103, + "num_tokens": 11096892.0, + "step": 3446 + }, + { + "epoch": 1.7631713554987214, + "grad_norm": 9.78740406036377, + "learning_rate": 1e-06, + "loss": 0.2841, + "mean_token_accuracy": 0.8838596940040588, + "num_tokens": 11100279.0, + "step": 3447 + }, + { + "epoch": 1.763682864450128, + "grad_norm": 6.510028839111328, + "learning_rate": 1e-06, + "loss": 0.2291, + "mean_token_accuracy": 0.9213683605194092, + "num_tokens": 11104763.0, + "step": 3448 + }, + { + "epoch": 1.7641943734015344, + "grad_norm": 8.480972290039062, + "learning_rate": 1e-06, + "loss": 0.26, + "mean_token_accuracy": 0.8813074827194214, + "num_tokens": 11107787.0, + "step": 3449 + }, + { + "epoch": 1.7647058823529411, + "grad_norm": 6.397139549255371, + "learning_rate": 1e-06, + "loss": 0.1676, + "mean_token_accuracy": 0.9269760847091675, + "num_tokens": 11110826.0, + "step": 3450 + }, + { + "epoch": 1.7652173913043478, + "grad_norm": 7.071554183959961, + "learning_rate": 1e-06, + "loss": 0.2068, + "mean_token_accuracy": 0.9084890484809875, + "num_tokens": 11114278.0, + "step": 3451 + }, + { + "epoch": 1.7657289002557546, + "grad_norm": 6.668923854827881, + "learning_rate": 1e-06, + "loss": 0.2003, + "mean_token_accuracy": 0.9020075798034668, + "num_tokens": 11117918.0, + "step": 3452 + }, + { + "epoch": 1.766240409207161, + "grad_norm": 6.323726654052734, + "learning_rate": 1e-06, + "loss": 0.1804, + "mean_token_accuracy": 0.9319698810577393, + "num_tokens": 11121926.0, + "step": 3453 + }, + { + "epoch": 1.7667519181585678, + "grad_norm": 13.24165153503418, + "learning_rate": 1e-06, + "loss": 0.303, + "mean_token_accuracy": 0.8729612827301025, + "num_tokens": 11124977.0, + "step": 3454 + }, + { + "epoch": 1.7672634271099743, + "grad_norm": 8.153816223144531, + "learning_rate": 1e-06, + "loss": 0.1755, + "mean_token_accuracy": 0.9176641702651978, + "num_tokens": 11128495.0, + "step": 3455 + }, + { + "epoch": 1.767774936061381, + "grad_norm": 9.594566345214844, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9133480787277222, + "num_tokens": 11131481.0, + "step": 3456 + }, + { + "epoch": 1.7682864450127878, + "grad_norm": 9.934927940368652, + "learning_rate": 1e-06, + "loss": 0.2444, + "mean_token_accuracy": 0.9151551127433777, + "num_tokens": 11134384.0, + "step": 3457 + }, + { + "epoch": 1.7687979539641945, + "grad_norm": 6.121517658233643, + "learning_rate": 1e-06, + "loss": 0.1406, + "mean_token_accuracy": 0.9471389651298523, + "num_tokens": 11138358.0, + "step": 3458 + }, + { + "epoch": 1.769309462915601, + "grad_norm": 10.688363075256348, + "learning_rate": 1e-06, + "loss": 0.3411, + "mean_token_accuracy": 0.8768113851547241, + "num_tokens": 11141444.0, + "step": 3459 + }, + { + "epoch": 1.7698209718670077, + "grad_norm": 8.043453216552734, + "learning_rate": 1e-06, + "loss": 0.2121, + "mean_token_accuracy": 0.9083893299102783, + "num_tokens": 11144829.0, + "step": 3460 + }, + { + "epoch": 1.7703324808184142, + "grad_norm": 10.906465530395508, + "learning_rate": 1e-06, + "loss": 0.2993, + "mean_token_accuracy": 0.8870329856872559, + "num_tokens": 11147788.0, + "step": 3461 + }, + { + "epoch": 1.770843989769821, + "grad_norm": 9.807517051696777, + "learning_rate": 1e-06, + "loss": 0.2557, + "mean_token_accuracy": 0.9112803936004639, + "num_tokens": 11150765.0, + "step": 3462 + }, + { + "epoch": 1.7713554987212277, + "grad_norm": 8.882965087890625, + "learning_rate": 1e-06, + "loss": 0.239, + "mean_token_accuracy": 0.8968214988708496, + "num_tokens": 11153429.0, + "step": 3463 + }, + { + "epoch": 1.7718670076726344, + "grad_norm": 12.586237907409668, + "learning_rate": 1e-06, + "loss": 0.2998, + "mean_token_accuracy": 0.8899363279342651, + "num_tokens": 11156065.0, + "step": 3464 + }, + { + "epoch": 1.772378516624041, + "grad_norm": 10.224907875061035, + "learning_rate": 1e-06, + "loss": 0.2738, + "mean_token_accuracy": 0.897993803024292, + "num_tokens": 11158940.0, + "step": 3465 + }, + { + "epoch": 1.7728900255754476, + "grad_norm": 9.56519603729248, + "learning_rate": 1e-06, + "loss": 0.2906, + "mean_token_accuracy": 0.8630239963531494, + "num_tokens": 11161519.0, + "step": 3466 + }, + { + "epoch": 1.7734015345268541, + "grad_norm": 5.86713171005249, + "learning_rate": 1e-06, + "loss": 0.2218, + "mean_token_accuracy": 0.8969435691833496, + "num_tokens": 11165414.0, + "step": 3467 + }, + { + "epoch": 1.7739130434782608, + "grad_norm": 12.94452953338623, + "learning_rate": 1e-06, + "loss": 0.2827, + "mean_token_accuracy": 0.8819204568862915, + "num_tokens": 11168242.0, + "step": 3468 + }, + { + "epoch": 1.7744245524296676, + "grad_norm": 8.428577423095703, + "learning_rate": 1e-06, + "loss": 0.2985, + "mean_token_accuracy": 0.8917874693870544, + "num_tokens": 11171053.0, + "step": 3469 + }, + { + "epoch": 1.7749360613810743, + "grad_norm": 8.93050765991211, + "learning_rate": 1e-06, + "loss": 0.1908, + "mean_token_accuracy": 0.9066535830497742, + "num_tokens": 11174242.0, + "step": 3470 + }, + { + "epoch": 1.7754475703324808, + "grad_norm": 7.970930576324463, + "learning_rate": 1e-06, + "loss": 0.2417, + "mean_token_accuracy": 0.9038804769515991, + "num_tokens": 11178088.0, + "step": 3471 + }, + { + "epoch": 1.7759590792838875, + "grad_norm": 9.228598594665527, + "learning_rate": 1e-06, + "loss": 0.2896, + "mean_token_accuracy": 0.885330080986023, + "num_tokens": 11180538.0, + "step": 3472 + }, + { + "epoch": 1.776470588235294, + "grad_norm": 10.313338279724121, + "learning_rate": 1e-06, + "loss": 0.3563, + "mean_token_accuracy": 0.8643245697021484, + "num_tokens": 11183501.0, + "step": 3473 + }, + { + "epoch": 1.7769820971867007, + "grad_norm": 9.881854057312012, + "learning_rate": 1e-06, + "loss": 0.227, + "mean_token_accuracy": 0.9070590734481812, + "num_tokens": 11187311.0, + "step": 3474 + }, + { + "epoch": 1.7774936061381075, + "grad_norm": 38.47542953491211, + "learning_rate": 1e-06, + "loss": 0.3093, + "mean_token_accuracy": 0.8925277590751648, + "num_tokens": 11190621.0, + "step": 3475 + }, + { + "epoch": 1.7780051150895142, + "grad_norm": 8.81738567352295, + "learning_rate": 1e-06, + "loss": 0.245, + "mean_token_accuracy": 0.928647518157959, + "num_tokens": 11193212.0, + "step": 3476 + }, + { + "epoch": 1.7785166240409207, + "grad_norm": 6.3125410079956055, + "learning_rate": 1e-06, + "loss": 0.1506, + "mean_token_accuracy": 0.8989827036857605, + "num_tokens": 11197222.0, + "step": 3477 + }, + { + "epoch": 1.7790281329923272, + "grad_norm": 11.012521743774414, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9154804944992065, + "num_tokens": 11200877.0, + "step": 3478 + }, + { + "epoch": 1.779539641943734, + "grad_norm": 9.710805892944336, + "learning_rate": 1e-06, + "loss": 0.2368, + "mean_token_accuracy": 0.9201002717018127, + "num_tokens": 11204148.0, + "step": 3479 + }, + { + "epoch": 1.7800511508951407, + "grad_norm": 8.340277671813965, + "learning_rate": 1e-06, + "loss": 0.2437, + "mean_token_accuracy": 0.9115502834320068, + "num_tokens": 11207514.0, + "step": 3480 + }, + { + "epoch": 1.7805626598465474, + "grad_norm": 10.50871467590332, + "learning_rate": 1e-06, + "loss": 0.2587, + "mean_token_accuracy": 0.8971933126449585, + "num_tokens": 11210417.0, + "step": 3481 + }, + { + "epoch": 1.781074168797954, + "grad_norm": 9.905024528503418, + "learning_rate": 1e-06, + "loss": 0.2584, + "mean_token_accuracy": 0.9011632204055786, + "num_tokens": 11213627.0, + "step": 3482 + }, + { + "epoch": 1.7815856777493606, + "grad_norm": 7.133299827575684, + "learning_rate": 1e-06, + "loss": 0.2349, + "mean_token_accuracy": 0.8962095379829407, + "num_tokens": 11216840.0, + "step": 3483 + }, + { + "epoch": 1.782097186700767, + "grad_norm": 11.558893203735352, + "learning_rate": 1e-06, + "loss": 0.3412, + "mean_token_accuracy": 0.8752543330192566, + "num_tokens": 11219691.0, + "step": 3484 + }, + { + "epoch": 1.7826086956521738, + "grad_norm": 13.246192932128906, + "learning_rate": 1e-06, + "loss": 0.2992, + "mean_token_accuracy": 0.903319776058197, + "num_tokens": 11222666.0, + "step": 3485 + }, + { + "epoch": 1.7831202046035806, + "grad_norm": 11.23230266571045, + "learning_rate": 1e-06, + "loss": 0.2881, + "mean_token_accuracy": 0.8894901871681213, + "num_tokens": 11225698.0, + "step": 3486 + }, + { + "epoch": 1.7836317135549873, + "grad_norm": 13.13048267364502, + "learning_rate": 1e-06, + "loss": 0.2671, + "mean_token_accuracy": 0.8847620487213135, + "num_tokens": 11229178.0, + "step": 3487 + }, + { + "epoch": 1.784143222506394, + "grad_norm": 8.813063621520996, + "learning_rate": 1e-06, + "loss": 0.2586, + "mean_token_accuracy": 0.8958454728126526, + "num_tokens": 11231506.0, + "step": 3488 + }, + { + "epoch": 1.7846547314578005, + "grad_norm": 16.021163940429688, + "learning_rate": 1e-06, + "loss": 0.2577, + "mean_token_accuracy": 0.9011632204055786, + "num_tokens": 11234286.0, + "step": 3489 + }, + { + "epoch": 1.785166240409207, + "grad_norm": 9.629995346069336, + "learning_rate": 1e-06, + "loss": 0.3026, + "mean_token_accuracy": 0.8748447895050049, + "num_tokens": 11237337.0, + "step": 3490 + }, + { + "epoch": 1.7856777493606137, + "grad_norm": 8.972618103027344, + "learning_rate": 1e-06, + "loss": 0.2815, + "mean_token_accuracy": 0.8824595212936401, + "num_tokens": 11240635.0, + "step": 3491 + }, + { + "epoch": 1.7861892583120205, + "grad_norm": 6.763735771179199, + "learning_rate": 1e-06, + "loss": 0.1553, + "mean_token_accuracy": 0.9331185817718506, + "num_tokens": 11243483.0, + "step": 3492 + }, + { + "epoch": 1.7867007672634272, + "grad_norm": 7.310549259185791, + "learning_rate": 1e-06, + "loss": 0.2039, + "mean_token_accuracy": 0.9041362404823303, + "num_tokens": 11247177.0, + "step": 3493 + }, + { + "epoch": 1.787212276214834, + "grad_norm": 10.814672470092773, + "learning_rate": 1e-06, + "loss": 0.3151, + "mean_token_accuracy": 0.8761450052261353, + "num_tokens": 11250307.0, + "step": 3494 + }, + { + "epoch": 1.7877237851662404, + "grad_norm": 11.170886039733887, + "learning_rate": 1e-06, + "loss": 0.2841, + "mean_token_accuracy": 0.894669771194458, + "num_tokens": 11252866.0, + "step": 3495 + }, + { + "epoch": 1.788235294117647, + "grad_norm": 8.205924034118652, + "learning_rate": 1e-06, + "loss": 0.2394, + "mean_token_accuracy": 0.8952051997184753, + "num_tokens": 11256198.0, + "step": 3496 + }, + { + "epoch": 1.7887468030690536, + "grad_norm": 10.26211929321289, + "learning_rate": 1e-06, + "loss": 0.3101, + "mean_token_accuracy": 0.8627880811691284, + "num_tokens": 11259080.0, + "step": 3497 + }, + { + "epoch": 1.7892583120204604, + "grad_norm": 6.90516471862793, + "learning_rate": 1e-06, + "loss": 0.1382, + "mean_token_accuracy": 0.9204810261726379, + "num_tokens": 11262088.0, + "step": 3498 + }, + { + "epoch": 1.789769820971867, + "grad_norm": 6.640057563781738, + "learning_rate": 1e-06, + "loss": 0.1433, + "mean_token_accuracy": 0.920448362827301, + "num_tokens": 11265604.0, + "step": 3499 + }, + { + "epoch": 1.7902813299232738, + "grad_norm": 8.145000457763672, + "learning_rate": 1e-06, + "loss": 0.2183, + "mean_token_accuracy": 0.9035812616348267, + "num_tokens": 11269788.0, + "step": 3500 + }, + { + "epoch": 1.7907928388746803, + "grad_norm": 6.949592113494873, + "learning_rate": 1e-06, + "loss": 0.1781, + "mean_token_accuracy": 0.9324869513511658, + "num_tokens": 11274061.0, + "step": 3501 + }, + { + "epoch": 1.7913043478260868, + "grad_norm": 11.213525772094727, + "learning_rate": 1e-06, + "loss": 0.3587, + "mean_token_accuracy": 0.8849936723709106, + "num_tokens": 11277596.0, + "step": 3502 + }, + { + "epoch": 1.7918158567774936, + "grad_norm": 7.862990856170654, + "learning_rate": 1e-06, + "loss": 0.2171, + "mean_token_accuracy": 0.9168753623962402, + "num_tokens": 11281005.0, + "step": 3503 + }, + { + "epoch": 1.7923273657289003, + "grad_norm": 10.411640167236328, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.8935596942901611, + "num_tokens": 11283523.0, + "step": 3504 + }, + { + "epoch": 1.792838874680307, + "grad_norm": 6.236587047576904, + "learning_rate": 1e-06, + "loss": 0.2049, + "mean_token_accuracy": 0.9140743017196655, + "num_tokens": 11288098.0, + "step": 3505 + }, + { + "epoch": 1.7933503836317135, + "grad_norm": 10.645889282226562, + "learning_rate": 1e-06, + "loss": 0.2167, + "mean_token_accuracy": 0.9099032282829285, + "num_tokens": 11291138.0, + "step": 3506 + }, + { + "epoch": 1.7938618925831202, + "grad_norm": 10.010353088378906, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8910446763038635, + "num_tokens": 11293620.0, + "step": 3507 + }, + { + "epoch": 1.7943734015345267, + "grad_norm": 10.896549224853516, + "learning_rate": 1e-06, + "loss": 0.2223, + "mean_token_accuracy": 0.9302049279212952, + "num_tokens": 11296919.0, + "step": 3508 + }, + { + "epoch": 1.7948849104859335, + "grad_norm": 6.436622619628906, + "learning_rate": 1e-06, + "loss": 0.1978, + "mean_token_accuracy": 0.9198903441429138, + "num_tokens": 11301344.0, + "step": 3509 + }, + { + "epoch": 1.7953964194373402, + "grad_norm": 9.139886856079102, + "learning_rate": 1e-06, + "loss": 0.2431, + "mean_token_accuracy": 0.8909316062927246, + "num_tokens": 11304035.0, + "step": 3510 + }, + { + "epoch": 1.795907928388747, + "grad_norm": 10.136362075805664, + "learning_rate": 1e-06, + "loss": 0.2143, + "mean_token_accuracy": 0.9185075759887695, + "num_tokens": 11307307.0, + "step": 3511 + }, + { + "epoch": 1.7964194373401534, + "grad_norm": 7.1761322021484375, + "learning_rate": 1e-06, + "loss": 0.1934, + "mean_token_accuracy": 0.9211466312408447, + "num_tokens": 11310172.0, + "step": 3512 + }, + { + "epoch": 1.7969309462915601, + "grad_norm": 12.016803741455078, + "learning_rate": 1e-06, + "loss": 0.3552, + "mean_token_accuracy": 0.8596669435501099, + "num_tokens": 11312982.0, + "step": 3513 + }, + { + "epoch": 1.7974424552429666, + "grad_norm": 8.089478492736816, + "learning_rate": 1e-06, + "loss": 0.2148, + "mean_token_accuracy": 0.9196332693099976, + "num_tokens": 11315210.0, + "step": 3514 + }, + { + "epoch": 1.7979539641943734, + "grad_norm": 10.211487770080566, + "learning_rate": 1e-06, + "loss": 0.3047, + "mean_token_accuracy": 0.8738839626312256, + "num_tokens": 11318076.0, + "step": 3515 + }, + { + "epoch": 1.79846547314578, + "grad_norm": 8.624298095703125, + "learning_rate": 1e-06, + "loss": 0.2501, + "mean_token_accuracy": 0.896929144859314, + "num_tokens": 11320409.0, + "step": 3516 + }, + { + "epoch": 1.7989769820971868, + "grad_norm": 8.085600852966309, + "learning_rate": 1e-06, + "loss": 0.1778, + "mean_token_accuracy": 0.8966341018676758, + "num_tokens": 11323444.0, + "step": 3517 + }, + { + "epoch": 1.7994884910485933, + "grad_norm": 8.818812370300293, + "learning_rate": 1e-06, + "loss": 0.2049, + "mean_token_accuracy": 0.9135416746139526, + "num_tokens": 11326606.0, + "step": 3518 + }, + { + "epoch": 1.8, + "grad_norm": 11.243340492248535, + "learning_rate": 1e-06, + "loss": 0.3724, + "mean_token_accuracy": 0.8687419891357422, + "num_tokens": 11329516.0, + "step": 3519 + }, + { + "epoch": 1.8005115089514065, + "grad_norm": 9.194453239440918, + "learning_rate": 1e-06, + "loss": 0.2242, + "mean_token_accuracy": 0.8767285346984863, + "num_tokens": 11332798.0, + "step": 3520 + }, + { + "epoch": 1.8010230179028133, + "grad_norm": 8.369664192199707, + "learning_rate": 1e-06, + "loss": 0.2717, + "mean_token_accuracy": 0.873503565788269, + "num_tokens": 11335921.0, + "step": 3521 + }, + { + "epoch": 1.80153452685422, + "grad_norm": 11.243556022644043, + "learning_rate": 1e-06, + "loss": 0.2749, + "mean_token_accuracy": 0.8786201477050781, + "num_tokens": 11338935.0, + "step": 3522 + }, + { + "epoch": 1.8020460358056267, + "grad_norm": 7.687213897705078, + "learning_rate": 1e-06, + "loss": 0.2052, + "mean_token_accuracy": 0.9234898090362549, + "num_tokens": 11341624.0, + "step": 3523 + }, + { + "epoch": 1.8025575447570332, + "grad_norm": 12.686960220336914, + "learning_rate": 1e-06, + "loss": 0.2736, + "mean_token_accuracy": 0.8933437466621399, + "num_tokens": 11344659.0, + "step": 3524 + }, + { + "epoch": 1.80306905370844, + "grad_norm": 9.422791481018066, + "learning_rate": 1e-06, + "loss": 0.2134, + "mean_token_accuracy": 0.9202489256858826, + "num_tokens": 11347753.0, + "step": 3525 + }, + { + "epoch": 1.8035805626598465, + "grad_norm": 6.056543350219727, + "learning_rate": 1e-06, + "loss": 0.1551, + "mean_token_accuracy": 0.9232412576675415, + "num_tokens": 11351397.0, + "step": 3526 + }, + { + "epoch": 1.8040920716112532, + "grad_norm": 10.168876647949219, + "learning_rate": 1e-06, + "loss": 0.2417, + "mean_token_accuracy": 0.8945753574371338, + "num_tokens": 11354161.0, + "step": 3527 + }, + { + "epoch": 1.80460358056266, + "grad_norm": 8.476922988891602, + "learning_rate": 1e-06, + "loss": 0.2103, + "mean_token_accuracy": 0.9128614664077759, + "num_tokens": 11358421.0, + "step": 3528 + }, + { + "epoch": 1.8051150895140666, + "grad_norm": 7.216268062591553, + "learning_rate": 1e-06, + "loss": 0.1872, + "mean_token_accuracy": 0.9266526699066162, + "num_tokens": 11361687.0, + "step": 3529 + }, + { + "epoch": 1.8056265984654731, + "grad_norm": 10.917476654052734, + "learning_rate": 1e-06, + "loss": 0.3191, + "mean_token_accuracy": 0.8900440335273743, + "num_tokens": 11364528.0, + "step": 3530 + }, + { + "epoch": 1.8061381074168796, + "grad_norm": 11.153371810913086, + "learning_rate": 1e-06, + "loss": 0.3655, + "mean_token_accuracy": 0.8878105878829956, + "num_tokens": 11367179.0, + "step": 3531 + }, + { + "epoch": 1.8066496163682864, + "grad_norm": 10.008833885192871, + "learning_rate": 1e-06, + "loss": 0.2456, + "mean_token_accuracy": 0.8924864530563354, + "num_tokens": 11370916.0, + "step": 3532 + }, + { + "epoch": 1.807161125319693, + "grad_norm": 6.72349739074707, + "learning_rate": 1e-06, + "loss": 0.1522, + "mean_token_accuracy": 0.9021697044372559, + "num_tokens": 11375331.0, + "step": 3533 + }, + { + "epoch": 1.8076726342710998, + "grad_norm": 8.47546100616455, + "learning_rate": 1e-06, + "loss": 0.2459, + "mean_token_accuracy": 0.9097815155982971, + "num_tokens": 11378153.0, + "step": 3534 + }, + { + "epoch": 1.8081841432225065, + "grad_norm": 9.135295867919922, + "learning_rate": 1e-06, + "loss": 0.2222, + "mean_token_accuracy": 0.921370267868042, + "num_tokens": 11381272.0, + "step": 3535 + }, + { + "epoch": 1.808695652173913, + "grad_norm": 8.363992691040039, + "learning_rate": 1e-06, + "loss": 0.2488, + "mean_token_accuracy": 0.888218879699707, + "num_tokens": 11384296.0, + "step": 3536 + }, + { + "epoch": 1.8092071611253195, + "grad_norm": 4.912289619445801, + "learning_rate": 1e-06, + "loss": 0.1454, + "mean_token_accuracy": 0.9147193431854248, + "num_tokens": 11388340.0, + "step": 3537 + }, + { + "epoch": 1.8097186700767263, + "grad_norm": 6.2488179206848145, + "learning_rate": 1e-06, + "loss": 0.1191, + "mean_token_accuracy": 0.9441433548927307, + "num_tokens": 11391798.0, + "step": 3538 + }, + { + "epoch": 1.810230179028133, + "grad_norm": 8.665810585021973, + "learning_rate": 1e-06, + "loss": 0.2168, + "mean_token_accuracy": 0.9177147746086121, + "num_tokens": 11395131.0, + "step": 3539 + }, + { + "epoch": 1.8107416879795397, + "grad_norm": 9.027036666870117, + "learning_rate": 1e-06, + "loss": 0.3076, + "mean_token_accuracy": 0.87518709897995, + "num_tokens": 11398570.0, + "step": 3540 + }, + { + "epoch": 1.8112531969309464, + "grad_norm": 7.206195831298828, + "learning_rate": 1e-06, + "loss": 0.1272, + "mean_token_accuracy": 0.9395532608032227, + "num_tokens": 11402358.0, + "step": 3541 + }, + { + "epoch": 1.811764705882353, + "grad_norm": 11.239521980285645, + "learning_rate": 1e-06, + "loss": 0.2813, + "mean_token_accuracy": 0.9045345783233643, + "num_tokens": 11405746.0, + "step": 3542 + }, + { + "epoch": 1.8122762148337594, + "grad_norm": 9.032048225402832, + "learning_rate": 1e-06, + "loss": 0.2393, + "mean_token_accuracy": 0.9090760350227356, + "num_tokens": 11408412.0, + "step": 3543 + }, + { + "epoch": 1.8127877237851662, + "grad_norm": 8.96870231628418, + "learning_rate": 1e-06, + "loss": 0.2762, + "mean_token_accuracy": 0.8920193910598755, + "num_tokens": 11410996.0, + "step": 3544 + }, + { + "epoch": 1.813299232736573, + "grad_norm": 10.512011528015137, + "learning_rate": 1e-06, + "loss": 0.2454, + "mean_token_accuracy": 0.8928407430648804, + "num_tokens": 11414790.0, + "step": 3545 + }, + { + "epoch": 1.8138107416879796, + "grad_norm": 10.543159484863281, + "learning_rate": 1e-06, + "loss": 0.2543, + "mean_token_accuracy": 0.897681713104248, + "num_tokens": 11418014.0, + "step": 3546 + }, + { + "epoch": 1.8143222506393863, + "grad_norm": 10.495709419250488, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8898314237594604, + "num_tokens": 11420508.0, + "step": 3547 + }, + { + "epoch": 1.8148337595907928, + "grad_norm": 11.528314590454102, + "learning_rate": 1e-06, + "loss": 0.2767, + "mean_token_accuracy": 0.8825538754463196, + "num_tokens": 11424367.0, + "step": 3548 + }, + { + "epoch": 1.8153452685421994, + "grad_norm": 8.932148933410645, + "learning_rate": 1e-06, + "loss": 0.2271, + "mean_token_accuracy": 0.9026827812194824, + "num_tokens": 11428131.0, + "step": 3549 + }, + { + "epoch": 1.815856777493606, + "grad_norm": 10.088651657104492, + "learning_rate": 1e-06, + "loss": 0.2529, + "mean_token_accuracy": 0.8912490606307983, + "num_tokens": 11430918.0, + "step": 3550 + }, + { + "epoch": 1.8163682864450128, + "grad_norm": 6.046844482421875, + "learning_rate": 1e-06, + "loss": 0.1701, + "mean_token_accuracy": 0.883922815322876, + "num_tokens": 11434696.0, + "step": 3551 + }, + { + "epoch": 1.8168797953964195, + "grad_norm": 12.238465309143066, + "learning_rate": 1e-06, + "loss": 0.2759, + "mean_token_accuracy": 0.8697008490562439, + "num_tokens": 11438584.0, + "step": 3552 + }, + { + "epoch": 1.8173913043478263, + "grad_norm": 11.643585205078125, + "learning_rate": 1e-06, + "loss": 0.3181, + "mean_token_accuracy": 0.870403528213501, + "num_tokens": 11441204.0, + "step": 3553 + }, + { + "epoch": 1.8179028132992328, + "grad_norm": 10.987801551818848, + "learning_rate": 1e-06, + "loss": 0.3299, + "mean_token_accuracy": 0.8919920921325684, + "num_tokens": 11444977.0, + "step": 3554 + }, + { + "epoch": 1.8184143222506393, + "grad_norm": 15.19997787475586, + "learning_rate": 1e-06, + "loss": 0.3138, + "mean_token_accuracy": 0.8992875218391418, + "num_tokens": 11448006.0, + "step": 3555 + }, + { + "epoch": 1.818925831202046, + "grad_norm": 11.780496597290039, + "learning_rate": 1e-06, + "loss": 0.2901, + "mean_token_accuracy": 0.9092211723327637, + "num_tokens": 11451069.0, + "step": 3556 + }, + { + "epoch": 1.8194373401534527, + "grad_norm": 7.236281394958496, + "learning_rate": 1e-06, + "loss": 0.1844, + "mean_token_accuracy": 0.923366904258728, + "num_tokens": 11453627.0, + "step": 3557 + }, + { + "epoch": 1.8199488491048594, + "grad_norm": 12.804283142089844, + "learning_rate": 1e-06, + "loss": 0.3932, + "mean_token_accuracy": 0.8606334328651428, + "num_tokens": 11456263.0, + "step": 3558 + }, + { + "epoch": 1.820460358056266, + "grad_norm": 8.933087348937988, + "learning_rate": 1e-06, + "loss": 0.2273, + "mean_token_accuracy": 0.922039806842804, + "num_tokens": 11458649.0, + "step": 3559 + }, + { + "epoch": 1.8209718670076727, + "grad_norm": 7.064133167266846, + "learning_rate": 1e-06, + "loss": 0.2028, + "mean_token_accuracy": 0.9103389382362366, + "num_tokens": 11461709.0, + "step": 3560 + }, + { + "epoch": 1.8214833759590792, + "grad_norm": 10.902726173400879, + "learning_rate": 1e-06, + "loss": 0.309, + "mean_token_accuracy": 0.9023942947387695, + "num_tokens": 11464519.0, + "step": 3561 + }, + { + "epoch": 1.821994884910486, + "grad_norm": 6.923656463623047, + "learning_rate": 1e-06, + "loss": 0.1625, + "mean_token_accuracy": 0.9377305507659912, + "num_tokens": 11468597.0, + "step": 3562 + }, + { + "epoch": 1.8225063938618926, + "grad_norm": 10.251823425292969, + "learning_rate": 1e-06, + "loss": 0.2653, + "mean_token_accuracy": 0.8940713405609131, + "num_tokens": 11471723.0, + "step": 3563 + }, + { + "epoch": 1.8230179028132993, + "grad_norm": 10.973511695861816, + "learning_rate": 1e-06, + "loss": 0.3344, + "mean_token_accuracy": 0.8666602373123169, + "num_tokens": 11474781.0, + "step": 3564 + }, + { + "epoch": 1.8235294117647058, + "grad_norm": 8.940652847290039, + "learning_rate": 1e-06, + "loss": 0.2463, + "mean_token_accuracy": 0.9076093435287476, + "num_tokens": 11478454.0, + "step": 3565 + }, + { + "epoch": 1.8240409207161126, + "grad_norm": 14.631099700927734, + "learning_rate": 1e-06, + "loss": 0.3555, + "mean_token_accuracy": 0.8664946556091309, + "num_tokens": 11481251.0, + "step": 3566 + }, + { + "epoch": 1.824552429667519, + "grad_norm": 9.196334838867188, + "learning_rate": 1e-06, + "loss": 0.3377, + "mean_token_accuracy": 0.8664751648902893, + "num_tokens": 11484284.0, + "step": 3567 + }, + { + "epoch": 1.8250639386189258, + "grad_norm": 11.544623374938965, + "learning_rate": 1e-06, + "loss": 0.2914, + "mean_token_accuracy": 0.8998546600341797, + "num_tokens": 11487249.0, + "step": 3568 + }, + { + "epoch": 1.8255754475703325, + "grad_norm": 10.217717170715332, + "learning_rate": 1e-06, + "loss": 0.3136, + "mean_token_accuracy": 0.8646630048751831, + "num_tokens": 11490304.0, + "step": 3569 + }, + { + "epoch": 1.8260869565217392, + "grad_norm": 9.386857032775879, + "learning_rate": 1e-06, + "loss": 0.2723, + "mean_token_accuracy": 0.8942670822143555, + "num_tokens": 11493414.0, + "step": 3570 + }, + { + "epoch": 1.8265984654731457, + "grad_norm": 6.643747806549072, + "learning_rate": 1e-06, + "loss": 0.2119, + "mean_token_accuracy": 0.8943304419517517, + "num_tokens": 11496647.0, + "step": 3571 + }, + { + "epoch": 1.8271099744245525, + "grad_norm": 9.675459861755371, + "learning_rate": 1e-06, + "loss": 0.2776, + "mean_token_accuracy": 0.8833373785018921, + "num_tokens": 11499287.0, + "step": 3572 + }, + { + "epoch": 1.827621483375959, + "grad_norm": 7.922542572021484, + "learning_rate": 1e-06, + "loss": 0.2646, + "mean_token_accuracy": 0.8940991163253784, + "num_tokens": 11502554.0, + "step": 3573 + }, + { + "epoch": 1.8281329923273657, + "grad_norm": 11.403875350952148, + "learning_rate": 1e-06, + "loss": 0.3321, + "mean_token_accuracy": 0.8816119432449341, + "num_tokens": 11505915.0, + "step": 3574 + }, + { + "epoch": 1.8286445012787724, + "grad_norm": 8.587730407714844, + "learning_rate": 1e-06, + "loss": 0.2576, + "mean_token_accuracy": 0.8825656175613403, + "num_tokens": 11509488.0, + "step": 3575 + }, + { + "epoch": 1.8291560102301792, + "grad_norm": 11.813533782958984, + "learning_rate": 1e-06, + "loss": 0.3195, + "mean_token_accuracy": 0.8654510974884033, + "num_tokens": 11512474.0, + "step": 3576 + }, + { + "epoch": 1.8296675191815857, + "grad_norm": 6.277484893798828, + "learning_rate": 1e-06, + "loss": 0.1822, + "mean_token_accuracy": 0.9323170781135559, + "num_tokens": 11517850.0, + "step": 3577 + }, + { + "epoch": 1.8301790281329924, + "grad_norm": 7.743961811065674, + "learning_rate": 1e-06, + "loss": 0.2266, + "mean_token_accuracy": 0.8891302347183228, + "num_tokens": 11520617.0, + "step": 3578 + }, + { + "epoch": 1.8306905370843989, + "grad_norm": 10.541397094726562, + "learning_rate": 1e-06, + "loss": 0.3087, + "mean_token_accuracy": 0.8768960237503052, + "num_tokens": 11522961.0, + "step": 3579 + }, + { + "epoch": 1.8312020460358056, + "grad_norm": 5.818716049194336, + "learning_rate": 1e-06, + "loss": 0.1519, + "mean_token_accuracy": 0.9468730688095093, + "num_tokens": 11526258.0, + "step": 3580 + }, + { + "epoch": 1.8317135549872123, + "grad_norm": 9.832551002502441, + "learning_rate": 1e-06, + "loss": 0.3428, + "mean_token_accuracy": 0.8776315450668335, + "num_tokens": 11529330.0, + "step": 3581 + }, + { + "epoch": 1.832225063938619, + "grad_norm": 8.980977058410645, + "learning_rate": 1e-06, + "loss": 0.2151, + "mean_token_accuracy": 0.9142007827758789, + "num_tokens": 11532752.0, + "step": 3582 + }, + { + "epoch": 1.8327365728900256, + "grad_norm": 6.65554666519165, + "learning_rate": 1e-06, + "loss": 0.1622, + "mean_token_accuracy": 0.9256651997566223, + "num_tokens": 11539310.0, + "step": 3583 + }, + { + "epoch": 1.833248081841432, + "grad_norm": 10.390227317810059, + "learning_rate": 1e-06, + "loss": 0.252, + "mean_token_accuracy": 0.9145392179489136, + "num_tokens": 11542284.0, + "step": 3584 + }, + { + "epoch": 1.8337595907928388, + "grad_norm": 10.254159927368164, + "learning_rate": 1e-06, + "loss": 0.2874, + "mean_token_accuracy": 0.8747242093086243, + "num_tokens": 11545051.0, + "step": 3585 + }, + { + "epoch": 1.8342710997442455, + "grad_norm": 14.14637279510498, + "learning_rate": 1e-06, + "loss": 0.3536, + "mean_token_accuracy": 0.8974968791007996, + "num_tokens": 11548110.0, + "step": 3586 + }, + { + "epoch": 1.8347826086956522, + "grad_norm": 9.631081581115723, + "learning_rate": 1e-06, + "loss": 0.2563, + "mean_token_accuracy": 0.9100013971328735, + "num_tokens": 11552335.0, + "step": 3587 + }, + { + "epoch": 1.835294117647059, + "grad_norm": 8.874272346496582, + "learning_rate": 1e-06, + "loss": 0.1631, + "mean_token_accuracy": 0.9033856391906738, + "num_tokens": 11555637.0, + "step": 3588 + }, + { + "epoch": 1.8358056265984655, + "grad_norm": 10.051268577575684, + "learning_rate": 1e-06, + "loss": 0.3318, + "mean_token_accuracy": 0.8745050430297852, + "num_tokens": 11559009.0, + "step": 3589 + }, + { + "epoch": 1.836317135549872, + "grad_norm": 10.099291801452637, + "learning_rate": 1e-06, + "loss": 0.3437, + "mean_token_accuracy": 0.8745059370994568, + "num_tokens": 11562268.0, + "step": 3590 + }, + { + "epoch": 1.8368286445012787, + "grad_norm": 8.414995193481445, + "learning_rate": 1e-06, + "loss": 0.2513, + "mean_token_accuracy": 0.8978697061538696, + "num_tokens": 11565103.0, + "step": 3591 + }, + { + "epoch": 1.8373401534526854, + "grad_norm": 8.707560539245605, + "learning_rate": 1e-06, + "loss": 0.2141, + "mean_token_accuracy": 0.8712046146392822, + "num_tokens": 11567758.0, + "step": 3592 + }, + { + "epoch": 1.8378516624040921, + "grad_norm": 6.930843830108643, + "learning_rate": 1e-06, + "loss": 0.1917, + "mean_token_accuracy": 0.9168648719787598, + "num_tokens": 11570160.0, + "step": 3593 + }, + { + "epoch": 1.8383631713554989, + "grad_norm": 10.405538558959961, + "learning_rate": 1e-06, + "loss": 0.2975, + "mean_token_accuracy": 0.8948876857757568, + "num_tokens": 11572969.0, + "step": 3594 + }, + { + "epoch": 1.8388746803069054, + "grad_norm": 9.875849723815918, + "learning_rate": 1e-06, + "loss": 0.3179, + "mean_token_accuracy": 0.877596914768219, + "num_tokens": 11576466.0, + "step": 3595 + }, + { + "epoch": 1.8393861892583119, + "grad_norm": 6.124322891235352, + "learning_rate": 1e-06, + "loss": 0.1522, + "mean_token_accuracy": 0.9327276945114136, + "num_tokens": 11580465.0, + "step": 3596 + }, + { + "epoch": 1.8398976982097186, + "grad_norm": 6.600658893585205, + "learning_rate": 1e-06, + "loss": 0.1651, + "mean_token_accuracy": 0.926602303981781, + "num_tokens": 11584062.0, + "step": 3597 + }, + { + "epoch": 1.8404092071611253, + "grad_norm": 8.170326232910156, + "learning_rate": 1e-06, + "loss": 0.246, + "mean_token_accuracy": 0.9102504253387451, + "num_tokens": 11586523.0, + "step": 3598 + }, + { + "epoch": 1.840920716112532, + "grad_norm": 8.353546142578125, + "learning_rate": 1e-06, + "loss": 0.2656, + "mean_token_accuracy": 0.9009185433387756, + "num_tokens": 11589159.0, + "step": 3599 + }, + { + "epoch": 1.8414322250639388, + "grad_norm": 7.367938041687012, + "learning_rate": 1e-06, + "loss": 0.2037, + "mean_token_accuracy": 0.8965569734573364, + "num_tokens": 11592814.0, + "step": 3600 + }, + { + "epoch": 1.8419437340153453, + "grad_norm": 11.182820320129395, + "learning_rate": 1e-06, + "loss": 0.3334, + "mean_token_accuracy": 0.8483030796051025, + "num_tokens": 11596540.0, + "step": 3601 + }, + { + "epoch": 1.8424552429667518, + "grad_norm": 11.456499099731445, + "learning_rate": 1e-06, + "loss": 0.3025, + "mean_token_accuracy": 0.8846168518066406, + "num_tokens": 11599149.0, + "step": 3602 + }, + { + "epoch": 1.8429667519181585, + "grad_norm": 8.676429748535156, + "learning_rate": 1e-06, + "loss": 0.1694, + "mean_token_accuracy": 0.8736646175384521, + "num_tokens": 11603052.0, + "step": 3603 + }, + { + "epoch": 1.8434782608695652, + "grad_norm": 9.876903533935547, + "learning_rate": 1e-06, + "loss": 0.1951, + "mean_token_accuracy": 0.8967808485031128, + "num_tokens": 11606413.0, + "step": 3604 + }, + { + "epoch": 1.843989769820972, + "grad_norm": 10.025948524475098, + "learning_rate": 1e-06, + "loss": 0.198, + "mean_token_accuracy": 0.9118192195892334, + "num_tokens": 11609264.0, + "step": 3605 + }, + { + "epoch": 1.8445012787723787, + "grad_norm": 10.873363494873047, + "learning_rate": 1e-06, + "loss": 0.2416, + "mean_token_accuracy": 0.8737434148788452, + "num_tokens": 11612772.0, + "step": 3606 + }, + { + "epoch": 1.8450127877237852, + "grad_norm": 9.27293586730957, + "learning_rate": 1e-06, + "loss": 0.26, + "mean_token_accuracy": 0.8970189690589905, + "num_tokens": 11616456.0, + "step": 3607 + }, + { + "epoch": 1.8455242966751917, + "grad_norm": 14.442221641540527, + "learning_rate": 1e-06, + "loss": 0.2402, + "mean_token_accuracy": 0.9023973941802979, + "num_tokens": 11619611.0, + "step": 3608 + }, + { + "epoch": 1.8460358056265984, + "grad_norm": 7.249248504638672, + "learning_rate": 1e-06, + "loss": 0.2251, + "mean_token_accuracy": 0.8967946171760559, + "num_tokens": 11622660.0, + "step": 3609 + }, + { + "epoch": 1.8465473145780051, + "grad_norm": 11.496664047241211, + "learning_rate": 1e-06, + "loss": 0.2738, + "mean_token_accuracy": 0.8847609162330627, + "num_tokens": 11625817.0, + "step": 3610 + }, + { + "epoch": 1.8470588235294119, + "grad_norm": 14.683018684387207, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8684306740760803, + "num_tokens": 11628742.0, + "step": 3611 + }, + { + "epoch": 1.8475703324808184, + "grad_norm": 13.318556785583496, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8748564124107361, + "num_tokens": 11631388.0, + "step": 3612 + }, + { + "epoch": 1.848081841432225, + "grad_norm": 10.003447532653809, + "learning_rate": 1e-06, + "loss": 0.2946, + "mean_token_accuracy": 0.8750156760215759, + "num_tokens": 11634303.0, + "step": 3613 + }, + { + "epoch": 1.8485933503836316, + "grad_norm": 8.44718074798584, + "learning_rate": 1e-06, + "loss": 0.2544, + "mean_token_accuracy": 0.9167075157165527, + "num_tokens": 11636804.0, + "step": 3614 + }, + { + "epoch": 1.8491048593350383, + "grad_norm": 9.691593170166016, + "learning_rate": 1e-06, + "loss": 0.2581, + "mean_token_accuracy": 0.8930420875549316, + "num_tokens": 11639662.0, + "step": 3615 + }, + { + "epoch": 1.849616368286445, + "grad_norm": 11.579106330871582, + "learning_rate": 1e-06, + "loss": 0.374, + "mean_token_accuracy": 0.867537260055542, + "num_tokens": 11642508.0, + "step": 3616 + }, + { + "epoch": 1.8501278772378518, + "grad_norm": 9.735845565795898, + "learning_rate": 1e-06, + "loss": 0.2514, + "mean_token_accuracy": 0.9125068187713623, + "num_tokens": 11645564.0, + "step": 3617 + }, + { + "epoch": 1.8506393861892583, + "grad_norm": 9.848936080932617, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.8779479265213013, + "num_tokens": 11648684.0, + "step": 3618 + }, + { + "epoch": 1.851150895140665, + "grad_norm": 9.031611442565918, + "learning_rate": 1e-06, + "loss": 0.269, + "mean_token_accuracy": 0.8818897008895874, + "num_tokens": 11651178.0, + "step": 3619 + }, + { + "epoch": 1.8516624040920715, + "grad_norm": 10.587146759033203, + "learning_rate": 1e-06, + "loss": 0.2417, + "mean_token_accuracy": 0.8910243511199951, + "num_tokens": 11654402.0, + "step": 3620 + }, + { + "epoch": 1.8521739130434782, + "grad_norm": 12.873664855957031, + "learning_rate": 1e-06, + "loss": 0.2687, + "mean_token_accuracy": 0.8885464668273926, + "num_tokens": 11657510.0, + "step": 3621 + }, + { + "epoch": 1.852685421994885, + "grad_norm": 9.828344345092773, + "learning_rate": 1e-06, + "loss": 0.2302, + "mean_token_accuracy": 0.9154094457626343, + "num_tokens": 11660602.0, + "step": 3622 + }, + { + "epoch": 1.8531969309462917, + "grad_norm": 10.570615768432617, + "learning_rate": 1e-06, + "loss": 0.2695, + "mean_token_accuracy": 0.8885558247566223, + "num_tokens": 11664149.0, + "step": 3623 + }, + { + "epoch": 1.8537084398976982, + "grad_norm": 7.473998069763184, + "learning_rate": 1e-06, + "loss": 0.168, + "mean_token_accuracy": 0.9112112522125244, + "num_tokens": 11667461.0, + "step": 3624 + }, + { + "epoch": 1.854219948849105, + "grad_norm": 7.369405746459961, + "learning_rate": 1e-06, + "loss": 0.2211, + "mean_token_accuracy": 0.9258270859718323, + "num_tokens": 11671904.0, + "step": 3625 + }, + { + "epoch": 1.8547314578005114, + "grad_norm": 8.821423530578613, + "learning_rate": 1e-06, + "loss": 0.1662, + "mean_token_accuracy": 0.9504786729812622, + "num_tokens": 11674864.0, + "step": 3626 + }, + { + "epoch": 1.8552429667519181, + "grad_norm": 7.827331066131592, + "learning_rate": 1e-06, + "loss": 0.2465, + "mean_token_accuracy": 0.9149433374404907, + "num_tokens": 11677954.0, + "step": 3627 + }, + { + "epoch": 1.8557544757033249, + "grad_norm": 5.858062267303467, + "learning_rate": 1e-06, + "loss": 0.1867, + "mean_token_accuracy": 0.9212920665740967, + "num_tokens": 11681080.0, + "step": 3628 + }, + { + "epoch": 1.8562659846547316, + "grad_norm": 13.85737419128418, + "learning_rate": 1e-06, + "loss": 0.3856, + "mean_token_accuracy": 0.8545916080474854, + "num_tokens": 11684133.0, + "step": 3629 + }, + { + "epoch": 1.856777493606138, + "grad_norm": 10.783029556274414, + "learning_rate": 1e-06, + "loss": 0.3052, + "mean_token_accuracy": 0.8750392198562622, + "num_tokens": 11686833.0, + "step": 3630 + }, + { + "epoch": 1.8572890025575448, + "grad_norm": 15.0823974609375, + "learning_rate": 1e-06, + "loss": 0.4135, + "mean_token_accuracy": 0.8163043856620789, + "num_tokens": 11689042.0, + "step": 3631 + }, + { + "epoch": 1.8578005115089513, + "grad_norm": 11.921615600585938, + "learning_rate": 1e-06, + "loss": 0.3094, + "mean_token_accuracy": 0.8772485852241516, + "num_tokens": 11692309.0, + "step": 3632 + }, + { + "epoch": 1.858312020460358, + "grad_norm": 6.113570213317871, + "learning_rate": 1e-06, + "loss": 0.1962, + "mean_token_accuracy": 0.9103925228118896, + "num_tokens": 11696250.0, + "step": 3633 + }, + { + "epoch": 1.8588235294117648, + "grad_norm": 9.93496322631836, + "learning_rate": 1e-06, + "loss": 0.2834, + "mean_token_accuracy": 0.9145519733428955, + "num_tokens": 11699822.0, + "step": 3634 + }, + { + "epoch": 1.8593350383631715, + "grad_norm": 8.408037185668945, + "learning_rate": 1e-06, + "loss": 0.2599, + "mean_token_accuracy": 0.9104129076004028, + "num_tokens": 11702924.0, + "step": 3635 + }, + { + "epoch": 1.859846547314578, + "grad_norm": 7.58071231842041, + "learning_rate": 1e-06, + "loss": 0.2306, + "mean_token_accuracy": 0.8962576389312744, + "num_tokens": 11706496.0, + "step": 3636 + }, + { + "epoch": 1.8603580562659845, + "grad_norm": 9.47937297821045, + "learning_rate": 1e-06, + "loss": 0.2591, + "mean_token_accuracy": 0.8916844725608826, + "num_tokens": 11709919.0, + "step": 3637 + }, + { + "epoch": 1.8608695652173912, + "grad_norm": 7.640621185302734, + "learning_rate": 1e-06, + "loss": 0.185, + "mean_token_accuracy": 0.9299065470695496, + "num_tokens": 11713650.0, + "step": 3638 + }, + { + "epoch": 1.861381074168798, + "grad_norm": 10.777931213378906, + "learning_rate": 1e-06, + "loss": 0.3055, + "mean_token_accuracy": 0.8732759356498718, + "num_tokens": 11716783.0, + "step": 3639 + }, + { + "epoch": 1.8618925831202047, + "grad_norm": 9.9373140335083, + "learning_rate": 1e-06, + "loss": 0.2968, + "mean_token_accuracy": 0.9013194441795349, + "num_tokens": 11719361.0, + "step": 3640 + }, + { + "epoch": 1.8624040920716114, + "grad_norm": 8.770453453063965, + "learning_rate": 1e-06, + "loss": 0.2652, + "mean_token_accuracy": 0.8950571417808533, + "num_tokens": 11722857.0, + "step": 3641 + }, + { + "epoch": 1.862915601023018, + "grad_norm": 9.606868743896484, + "learning_rate": 1e-06, + "loss": 0.236, + "mean_token_accuracy": 0.9035901427268982, + "num_tokens": 11726267.0, + "step": 3642 + }, + { + "epoch": 1.8634271099744244, + "grad_norm": 9.463525772094727, + "learning_rate": 1e-06, + "loss": 0.2084, + "mean_token_accuracy": 0.9140627384185791, + "num_tokens": 11728853.0, + "step": 3643 + }, + { + "epoch": 1.8639386189258311, + "grad_norm": 10.05209732055664, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.8929417133331299, + "num_tokens": 11732090.0, + "step": 3644 + }, + { + "epoch": 1.8644501278772379, + "grad_norm": 7.732579231262207, + "learning_rate": 1e-06, + "loss": 0.1704, + "mean_token_accuracy": 0.9145733714103699, + "num_tokens": 11735381.0, + "step": 3645 + }, + { + "epoch": 1.8649616368286446, + "grad_norm": 13.609593391418457, + "learning_rate": 1e-06, + "loss": 0.3048, + "mean_token_accuracy": 0.8894767761230469, + "num_tokens": 11738486.0, + "step": 3646 + }, + { + "epoch": 1.8654731457800513, + "grad_norm": 6.789805889129639, + "learning_rate": 1e-06, + "loss": 0.1601, + "mean_token_accuracy": 0.9351794123649597, + "num_tokens": 11741717.0, + "step": 3647 + }, + { + "epoch": 1.8659846547314578, + "grad_norm": 9.82552719116211, + "learning_rate": 1e-06, + "loss": 0.209, + "mean_token_accuracy": 0.9100666046142578, + "num_tokens": 11744399.0, + "step": 3648 + }, + { + "epoch": 1.8664961636828643, + "grad_norm": 9.601494789123535, + "learning_rate": 1e-06, + "loss": 0.2611, + "mean_token_accuracy": 0.8995043039321899, + "num_tokens": 11747078.0, + "step": 3649 + }, + { + "epoch": 1.867007672634271, + "grad_norm": 9.920660018920898, + "learning_rate": 1e-06, + "loss": 0.2561, + "mean_token_accuracy": 0.8977635502815247, + "num_tokens": 11750076.0, + "step": 3650 + }, + { + "epoch": 1.8675191815856778, + "grad_norm": 12.19914722442627, + "learning_rate": 1e-06, + "loss": 0.2959, + "mean_token_accuracy": 0.8849743604660034, + "num_tokens": 11753149.0, + "step": 3651 + }, + { + "epoch": 1.8680306905370845, + "grad_norm": 6.675117492675781, + "learning_rate": 1e-06, + "loss": 0.1851, + "mean_token_accuracy": 0.9121737480163574, + "num_tokens": 11756144.0, + "step": 3652 + }, + { + "epoch": 1.8685421994884912, + "grad_norm": 9.709856033325195, + "learning_rate": 1e-06, + "loss": 0.266, + "mean_token_accuracy": 0.8947376012802124, + "num_tokens": 11759757.0, + "step": 3653 + }, + { + "epoch": 1.8690537084398977, + "grad_norm": 11.683932304382324, + "learning_rate": 1e-06, + "loss": 0.2529, + "mean_token_accuracy": 0.912817656993866, + "num_tokens": 11763097.0, + "step": 3654 + }, + { + "epoch": 1.8695652173913042, + "grad_norm": 10.447263717651367, + "learning_rate": 1e-06, + "loss": 0.2547, + "mean_token_accuracy": 0.8871591687202454, + "num_tokens": 11766028.0, + "step": 3655 + }, + { + "epoch": 1.870076726342711, + "grad_norm": 6.979384899139404, + "learning_rate": 1e-06, + "loss": 0.2325, + "mean_token_accuracy": 0.8831960558891296, + "num_tokens": 11769637.0, + "step": 3656 + }, + { + "epoch": 1.8705882352941177, + "grad_norm": 9.771505355834961, + "learning_rate": 1e-06, + "loss": 0.2506, + "mean_token_accuracy": 0.9210999011993408, + "num_tokens": 11773255.0, + "step": 3657 + }, + { + "epoch": 1.8710997442455244, + "grad_norm": 13.873270988464355, + "learning_rate": 1e-06, + "loss": 0.3538, + "mean_token_accuracy": 0.8676401376724243, + "num_tokens": 11775867.0, + "step": 3658 + }, + { + "epoch": 1.8716112531969311, + "grad_norm": 9.235882759094238, + "learning_rate": 1e-06, + "loss": 0.2502, + "mean_token_accuracy": 0.8954727649688721, + "num_tokens": 11778836.0, + "step": 3659 + }, + { + "epoch": 1.8721227621483376, + "grad_norm": 11.229999542236328, + "learning_rate": 1e-06, + "loss": 0.2607, + "mean_token_accuracy": 0.8705745935440063, + "num_tokens": 11782197.0, + "step": 3660 + }, + { + "epoch": 1.8726342710997441, + "grad_norm": 9.794234275817871, + "learning_rate": 1e-06, + "loss": 0.2999, + "mean_token_accuracy": 0.8844281435012817, + "num_tokens": 11784853.0, + "step": 3661 + }, + { + "epoch": 1.8731457800511508, + "grad_norm": 10.517922401428223, + "learning_rate": 1e-06, + "loss": 0.3224, + "mean_token_accuracy": 0.8732835054397583, + "num_tokens": 11787865.0, + "step": 3662 + }, + { + "epoch": 1.8736572890025576, + "grad_norm": 11.366837501525879, + "learning_rate": 1e-06, + "loss": 0.2853, + "mean_token_accuracy": 0.9022891521453857, + "num_tokens": 11791322.0, + "step": 3663 + }, + { + "epoch": 1.8741687979539643, + "grad_norm": 9.044485092163086, + "learning_rate": 1e-06, + "loss": 0.271, + "mean_token_accuracy": 0.8856786489486694, + "num_tokens": 11794123.0, + "step": 3664 + }, + { + "epoch": 1.8746803069053708, + "grad_norm": 10.56338882446289, + "learning_rate": 1e-06, + "loss": 0.2453, + "mean_token_accuracy": 0.864343523979187, + "num_tokens": 11797629.0, + "step": 3665 + }, + { + "epoch": 1.8751918158567775, + "grad_norm": 10.524483680725098, + "learning_rate": 1e-06, + "loss": 0.2339, + "mean_token_accuracy": 0.8850342035293579, + "num_tokens": 11800682.0, + "step": 3666 + }, + { + "epoch": 1.875703324808184, + "grad_norm": 7.704439640045166, + "learning_rate": 1e-06, + "loss": 0.1705, + "mean_token_accuracy": 0.9308056831359863, + "num_tokens": 11803902.0, + "step": 3667 + }, + { + "epoch": 1.8762148337595907, + "grad_norm": 7.2276716232299805, + "learning_rate": 1e-06, + "loss": 0.1974, + "mean_token_accuracy": 0.8839926719665527, + "num_tokens": 11807321.0, + "step": 3668 + }, + { + "epoch": 1.8767263427109975, + "grad_norm": 10.028891563415527, + "learning_rate": 1e-06, + "loss": 0.2499, + "mean_token_accuracy": 0.8773775100708008, + "num_tokens": 11810361.0, + "step": 3669 + }, + { + "epoch": 1.8772378516624042, + "grad_norm": 7.606228828430176, + "learning_rate": 1e-06, + "loss": 0.2009, + "mean_token_accuracy": 0.8734731674194336, + "num_tokens": 11814045.0, + "step": 3670 + }, + { + "epoch": 1.8777493606138107, + "grad_norm": 9.274942398071289, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.908047616481781, + "num_tokens": 11816848.0, + "step": 3671 + }, + { + "epoch": 1.8782608695652174, + "grad_norm": 14.258790016174316, + "learning_rate": 1e-06, + "loss": 0.3651, + "mean_token_accuracy": 0.8671764135360718, + "num_tokens": 11819627.0, + "step": 3672 + }, + { + "epoch": 1.878772378516624, + "grad_norm": 8.718681335449219, + "learning_rate": 1e-06, + "loss": 0.2542, + "mean_token_accuracy": 0.8821719884872437, + "num_tokens": 11823065.0, + "step": 3673 + }, + { + "epoch": 1.8792838874680307, + "grad_norm": 8.623948097229004, + "learning_rate": 1e-06, + "loss": 0.2638, + "mean_token_accuracy": 0.8971561193466187, + "num_tokens": 11825714.0, + "step": 3674 + }, + { + "epoch": 1.8797953964194374, + "grad_norm": 19.4616756439209, + "learning_rate": 1e-06, + "loss": 0.2842, + "mean_token_accuracy": 0.893516480922699, + "num_tokens": 11828534.0, + "step": 3675 + }, + { + "epoch": 1.880306905370844, + "grad_norm": 8.691061019897461, + "learning_rate": 1e-06, + "loss": 0.2512, + "mean_token_accuracy": 0.888485848903656, + "num_tokens": 11831283.0, + "step": 3676 + }, + { + "epoch": 1.8808184143222506, + "grad_norm": 11.824408531188965, + "learning_rate": 1e-06, + "loss": 0.3448, + "mean_token_accuracy": 0.8948728442192078, + "num_tokens": 11834118.0, + "step": 3677 + }, + { + "epoch": 1.8813299232736573, + "grad_norm": 7.766383171081543, + "learning_rate": 1e-06, + "loss": 0.2524, + "mean_token_accuracy": 0.8833290934562683, + "num_tokens": 11837447.0, + "step": 3678 + }, + { + "epoch": 1.8818414322250638, + "grad_norm": 7.5706939697265625, + "learning_rate": 1e-06, + "loss": 0.2211, + "mean_token_accuracy": 0.9146907329559326, + "num_tokens": 11840639.0, + "step": 3679 + }, + { + "epoch": 1.8823529411764706, + "grad_norm": 8.627143859863281, + "learning_rate": 1e-06, + "loss": 0.1775, + "mean_token_accuracy": 0.8945601582527161, + "num_tokens": 11845086.0, + "step": 3680 + }, + { + "epoch": 1.8828644501278773, + "grad_norm": 8.199018478393555, + "learning_rate": 1e-06, + "loss": 0.2438, + "mean_token_accuracy": 0.9040246605873108, + "num_tokens": 11847869.0, + "step": 3681 + }, + { + "epoch": 1.883375959079284, + "grad_norm": 7.22396183013916, + "learning_rate": 1e-06, + "loss": 0.2247, + "mean_token_accuracy": 0.8980785608291626, + "num_tokens": 11850278.0, + "step": 3682 + }, + { + "epoch": 1.8838874680306905, + "grad_norm": 10.294392585754395, + "learning_rate": 1e-06, + "loss": 0.2816, + "mean_token_accuracy": 0.8647148609161377, + "num_tokens": 11853151.0, + "step": 3683 + }, + { + "epoch": 1.8843989769820972, + "grad_norm": 5.518775463104248, + "learning_rate": 1e-06, + "loss": 0.0922, + "mean_token_accuracy": 0.9213762879371643, + "num_tokens": 11858315.0, + "step": 3684 + }, + { + "epoch": 1.8849104859335037, + "grad_norm": 14.862563133239746, + "learning_rate": 1e-06, + "loss": 0.2969, + "mean_token_accuracy": 0.8888163566589355, + "num_tokens": 11861376.0, + "step": 3685 + }, + { + "epoch": 1.8854219948849105, + "grad_norm": 5.470010280609131, + "learning_rate": 1e-06, + "loss": 0.0997, + "mean_token_accuracy": 0.9495484828948975, + "num_tokens": 11865558.0, + "step": 3686 + }, + { + "epoch": 1.8859335038363172, + "grad_norm": 7.632906913757324, + "learning_rate": 1e-06, + "loss": 0.2562, + "mean_token_accuracy": 0.9198317527770996, + "num_tokens": 11869082.0, + "step": 3687 + }, + { + "epoch": 1.886445012787724, + "grad_norm": 11.10053539276123, + "learning_rate": 1e-06, + "loss": 0.2428, + "mean_token_accuracy": 0.9027749300003052, + "num_tokens": 11871899.0, + "step": 3688 + }, + { + "epoch": 1.8869565217391304, + "grad_norm": 4.816237926483154, + "learning_rate": 1e-06, + "loss": 0.1366, + "mean_token_accuracy": 0.930332601070404, + "num_tokens": 11876491.0, + "step": 3689 + }, + { + "epoch": 1.887468030690537, + "grad_norm": 9.098045349121094, + "learning_rate": 1e-06, + "loss": 0.2712, + "mean_token_accuracy": 0.882155179977417, + "num_tokens": 11879284.0, + "step": 3690 + }, + { + "epoch": 1.8879795396419436, + "grad_norm": 9.079418182373047, + "learning_rate": 1e-06, + "loss": 0.3322, + "mean_token_accuracy": 0.8517184853553772, + "num_tokens": 11883115.0, + "step": 3691 + }, + { + "epoch": 1.8884910485933504, + "grad_norm": 13.650148391723633, + "learning_rate": 1e-06, + "loss": 0.3973, + "mean_token_accuracy": 0.870989978313446, + "num_tokens": 11886384.0, + "step": 3692 + }, + { + "epoch": 1.889002557544757, + "grad_norm": 6.812595367431641, + "learning_rate": 1e-06, + "loss": 0.2242, + "mean_token_accuracy": 0.8975998163223267, + "num_tokens": 11890189.0, + "step": 3693 + }, + { + "epoch": 1.8895140664961638, + "grad_norm": 12.050154685974121, + "learning_rate": 1e-06, + "loss": 0.2357, + "mean_token_accuracy": 0.9191465377807617, + "num_tokens": 11893204.0, + "step": 3694 + }, + { + "epoch": 1.8900255754475703, + "grad_norm": 8.894072532653809, + "learning_rate": 1e-06, + "loss": 0.3063, + "mean_token_accuracy": 0.8627766370773315, + "num_tokens": 11896087.0, + "step": 3695 + }, + { + "epoch": 1.8905370843989768, + "grad_norm": 8.41044807434082, + "learning_rate": 1e-06, + "loss": 0.2599, + "mean_token_accuracy": 0.9170889258384705, + "num_tokens": 11898914.0, + "step": 3696 + }, + { + "epoch": 1.8910485933503836, + "grad_norm": 11.686128616333008, + "learning_rate": 1e-06, + "loss": 0.3298, + "mean_token_accuracy": 0.8757829666137695, + "num_tokens": 11901473.0, + "step": 3697 + }, + { + "epoch": 1.8915601023017903, + "grad_norm": 8.118382453918457, + "learning_rate": 1e-06, + "loss": 0.1994, + "mean_token_accuracy": 0.9200929403305054, + "num_tokens": 11905459.0, + "step": 3698 + }, + { + "epoch": 1.892071611253197, + "grad_norm": 10.071949005126953, + "learning_rate": 1e-06, + "loss": 0.285, + "mean_token_accuracy": 0.9063825607299805, + "num_tokens": 11908745.0, + "step": 3699 + }, + { + "epoch": 1.8925831202046037, + "grad_norm": 7.976752758026123, + "learning_rate": 1e-06, + "loss": 0.2192, + "mean_token_accuracy": 0.9047727584838867, + "num_tokens": 11912038.0, + "step": 3700 + }, + { + "epoch": 1.8930946291560102, + "grad_norm": 11.453052520751953, + "learning_rate": 1e-06, + "loss": 0.3216, + "mean_token_accuracy": 0.8720021843910217, + "num_tokens": 11914633.0, + "step": 3701 + }, + { + "epoch": 1.8936061381074167, + "grad_norm": 8.70484733581543, + "learning_rate": 1e-06, + "loss": 0.257, + "mean_token_accuracy": 0.8946979641914368, + "num_tokens": 11917614.0, + "step": 3702 + }, + { + "epoch": 1.8941176470588235, + "grad_norm": 12.32597827911377, + "learning_rate": 1e-06, + "loss": 0.2883, + "mean_token_accuracy": 0.8887953758239746, + "num_tokens": 11920300.0, + "step": 3703 + }, + { + "epoch": 1.8946291560102302, + "grad_norm": 9.87732982635498, + "learning_rate": 1e-06, + "loss": 0.2908, + "mean_token_accuracy": 0.9023362398147583, + "num_tokens": 11923766.0, + "step": 3704 + }, + { + "epoch": 1.895140664961637, + "grad_norm": 9.30306339263916, + "learning_rate": 1e-06, + "loss": 0.2936, + "mean_token_accuracy": 0.8994463682174683, + "num_tokens": 11926277.0, + "step": 3705 + }, + { + "epoch": 1.8956521739130436, + "grad_norm": 10.88073444366455, + "learning_rate": 1e-06, + "loss": 0.2859, + "mean_token_accuracy": 0.8990234136581421, + "num_tokens": 11928670.0, + "step": 3706 + }, + { + "epoch": 1.8961636828644501, + "grad_norm": 7.266307830810547, + "learning_rate": 1e-06, + "loss": 0.2142, + "mean_token_accuracy": 0.9072935581207275, + "num_tokens": 11931714.0, + "step": 3707 + }, + { + "epoch": 1.8966751918158566, + "grad_norm": 8.727803230285645, + "learning_rate": 1e-06, + "loss": 0.2541, + "mean_token_accuracy": 0.8999931216239929, + "num_tokens": 11934268.0, + "step": 3708 + }, + { + "epoch": 1.8971867007672634, + "grad_norm": 11.816347122192383, + "learning_rate": 1e-06, + "loss": 0.271, + "mean_token_accuracy": 0.8849388360977173, + "num_tokens": 11937169.0, + "step": 3709 + }, + { + "epoch": 1.89769820971867, + "grad_norm": 8.942785263061523, + "learning_rate": 1e-06, + "loss": 0.2398, + "mean_token_accuracy": 0.9022901058197021, + "num_tokens": 11940299.0, + "step": 3710 + }, + { + "epoch": 1.8982097186700768, + "grad_norm": 13.347095489501953, + "learning_rate": 1e-06, + "loss": 0.2941, + "mean_token_accuracy": 0.8944833278656006, + "num_tokens": 11942720.0, + "step": 3711 + }, + { + "epoch": 1.8987212276214833, + "grad_norm": 9.776008605957031, + "learning_rate": 1e-06, + "loss": 0.2522, + "mean_token_accuracy": 0.8969160318374634, + "num_tokens": 11945819.0, + "step": 3712 + }, + { + "epoch": 1.89923273657289, + "grad_norm": 9.306194305419922, + "learning_rate": 1e-06, + "loss": 0.3122, + "mean_token_accuracy": 0.88261878490448, + "num_tokens": 11949208.0, + "step": 3713 + }, + { + "epoch": 1.8997442455242965, + "grad_norm": 9.948315620422363, + "learning_rate": 1e-06, + "loss": 0.2558, + "mean_token_accuracy": 0.8940179347991943, + "num_tokens": 11952521.0, + "step": 3714 + }, + { + "epoch": 1.9002557544757033, + "grad_norm": 9.787467956542969, + "learning_rate": 1e-06, + "loss": 0.2576, + "mean_token_accuracy": 0.9008193016052246, + "num_tokens": 11955812.0, + "step": 3715 + }, + { + "epoch": 1.90076726342711, + "grad_norm": 9.201615333557129, + "learning_rate": 1e-06, + "loss": 0.2646, + "mean_token_accuracy": 0.8920482397079468, + "num_tokens": 11958236.0, + "step": 3716 + }, + { + "epoch": 1.9012787723785167, + "grad_norm": 9.325794219970703, + "learning_rate": 1e-06, + "loss": 0.2631, + "mean_token_accuracy": 0.9081123471260071, + "num_tokens": 11960982.0, + "step": 3717 + }, + { + "epoch": 1.9017902813299232, + "grad_norm": 7.284931659698486, + "learning_rate": 1e-06, + "loss": 0.2312, + "mean_token_accuracy": 0.9144513607025146, + "num_tokens": 11964232.0, + "step": 3718 + }, + { + "epoch": 1.90230179028133, + "grad_norm": 9.522337913513184, + "learning_rate": 1e-06, + "loss": 0.2732, + "mean_token_accuracy": 0.8852258920669556, + "num_tokens": 11967138.0, + "step": 3719 + }, + { + "epoch": 1.9028132992327365, + "grad_norm": 7.919545650482178, + "learning_rate": 1e-06, + "loss": 0.2479, + "mean_token_accuracy": 0.876085638999939, + "num_tokens": 11970377.0, + "step": 3720 + }, + { + "epoch": 1.9033248081841432, + "grad_norm": 7.708963394165039, + "learning_rate": 1e-06, + "loss": 0.1836, + "mean_token_accuracy": 0.9118179678916931, + "num_tokens": 11974115.0, + "step": 3721 + }, + { + "epoch": 1.90383631713555, + "grad_norm": 9.34872817993164, + "learning_rate": 1e-06, + "loss": 0.2528, + "mean_token_accuracy": 0.8933957815170288, + "num_tokens": 11977857.0, + "step": 3722 + }, + { + "epoch": 1.9043478260869566, + "grad_norm": 12.036394119262695, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8658952713012695, + "num_tokens": 11980333.0, + "step": 3723 + }, + { + "epoch": 1.9048593350383631, + "grad_norm": 9.706267356872559, + "learning_rate": 1e-06, + "loss": 0.3295, + "mean_token_accuracy": 0.8769296407699585, + "num_tokens": 11983009.0, + "step": 3724 + }, + { + "epoch": 1.9053708439897699, + "grad_norm": 8.124612808227539, + "learning_rate": 1e-06, + "loss": 0.2504, + "mean_token_accuracy": 0.8844562768936157, + "num_tokens": 11986537.0, + "step": 3725 + }, + { + "epoch": 1.9058823529411764, + "grad_norm": 9.346231460571289, + "learning_rate": 1e-06, + "loss": 0.3039, + "mean_token_accuracy": 0.8880915641784668, + "num_tokens": 11989698.0, + "step": 3726 + }, + { + "epoch": 1.906393861892583, + "grad_norm": 7.9554338455200195, + "learning_rate": 1e-06, + "loss": 0.1851, + "mean_token_accuracy": 0.9005017280578613, + "num_tokens": 11993744.0, + "step": 3727 + }, + { + "epoch": 1.9069053708439898, + "grad_norm": 9.099701881408691, + "learning_rate": 1e-06, + "loss": 0.2538, + "mean_token_accuracy": 0.9114139080047607, + "num_tokens": 11997255.0, + "step": 3728 + }, + { + "epoch": 1.9074168797953965, + "grad_norm": 9.478096961975098, + "learning_rate": 1e-06, + "loss": 0.2339, + "mean_token_accuracy": 0.9147727489471436, + "num_tokens": 12000772.0, + "step": 3729 + }, + { + "epoch": 1.907928388746803, + "grad_norm": 6.404082298278809, + "learning_rate": 1e-06, + "loss": 0.1738, + "mean_token_accuracy": 0.913680911064148, + "num_tokens": 12004064.0, + "step": 3730 + }, + { + "epoch": 1.9084398976982098, + "grad_norm": 7.995066165924072, + "learning_rate": 1e-06, + "loss": 0.2526, + "mean_token_accuracy": 0.8923328518867493, + "num_tokens": 12006883.0, + "step": 3731 + }, + { + "epoch": 1.9089514066496163, + "grad_norm": 9.527521133422852, + "learning_rate": 1e-06, + "loss": 0.2611, + "mean_token_accuracy": 0.9032761454582214, + "num_tokens": 12010814.0, + "step": 3732 + }, + { + "epoch": 1.909462915601023, + "grad_norm": 8.30145263671875, + "learning_rate": 1e-06, + "loss": 0.1696, + "mean_token_accuracy": 0.9023333787918091, + "num_tokens": 12014552.0, + "step": 3733 + }, + { + "epoch": 1.9099744245524297, + "grad_norm": 6.116941928863525, + "learning_rate": 1e-06, + "loss": 0.1731, + "mean_token_accuracy": 0.9173466563224792, + "num_tokens": 12019233.0, + "step": 3734 + }, + { + "epoch": 1.9104859335038364, + "grad_norm": 9.712944984436035, + "learning_rate": 1e-06, + "loss": 0.3262, + "mean_token_accuracy": 0.858416736125946, + "num_tokens": 12022491.0, + "step": 3735 + }, + { + "epoch": 1.910997442455243, + "grad_norm": 11.1172456741333, + "learning_rate": 1e-06, + "loss": 0.3505, + "mean_token_accuracy": 0.8718527555465698, + "num_tokens": 12025950.0, + "step": 3736 + }, + { + "epoch": 1.9115089514066497, + "grad_norm": 9.764303207397461, + "learning_rate": 1e-06, + "loss": 0.3257, + "mean_token_accuracy": 0.8569986820220947, + "num_tokens": 12028829.0, + "step": 3737 + }, + { + "epoch": 1.9120204603580562, + "grad_norm": 13.38602352142334, + "learning_rate": 1e-06, + "loss": 0.3709, + "mean_token_accuracy": 0.8668622374534607, + "num_tokens": 12031874.0, + "step": 3738 + }, + { + "epoch": 1.912531969309463, + "grad_norm": 8.726906776428223, + "learning_rate": 1e-06, + "loss": 0.196, + "mean_token_accuracy": 0.9193180799484253, + "num_tokens": 12034540.0, + "step": 3739 + }, + { + "epoch": 1.9130434782608696, + "grad_norm": 8.796089172363281, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.9058945775032043, + "num_tokens": 12038002.0, + "step": 3740 + }, + { + "epoch": 1.9135549872122763, + "grad_norm": 10.836335182189941, + "learning_rate": 1e-06, + "loss": 0.3379, + "mean_token_accuracy": 0.8770406246185303, + "num_tokens": 12041126.0, + "step": 3741 + }, + { + "epoch": 1.9140664961636829, + "grad_norm": 12.665724754333496, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8784002065658569, + "num_tokens": 12044134.0, + "step": 3742 + }, + { + "epoch": 1.9145780051150894, + "grad_norm": 12.31908893585205, + "learning_rate": 1e-06, + "loss": 0.3595, + "mean_token_accuracy": 0.8649741411209106, + "num_tokens": 12047017.0, + "step": 3743 + }, + { + "epoch": 1.915089514066496, + "grad_norm": 6.122111797332764, + "learning_rate": 1e-06, + "loss": 0.1971, + "mean_token_accuracy": 0.9063775539398193, + "num_tokens": 12050416.0, + "step": 3744 + }, + { + "epoch": 1.9156010230179028, + "grad_norm": 12.725532531738281, + "learning_rate": 1e-06, + "loss": 0.3399, + "mean_token_accuracy": 0.8810039758682251, + "num_tokens": 12053329.0, + "step": 3745 + }, + { + "epoch": 1.9161125319693095, + "grad_norm": 16.18485450744629, + "learning_rate": 1e-06, + "loss": 0.3598, + "mean_token_accuracy": 0.8766969442367554, + "num_tokens": 12055841.0, + "step": 3746 + }, + { + "epoch": 1.9166240409207163, + "grad_norm": 11.817486763000488, + "learning_rate": 1e-06, + "loss": 0.3872, + "mean_token_accuracy": 0.8612566590309143, + "num_tokens": 12058719.0, + "step": 3747 + }, + { + "epoch": 1.9171355498721228, + "grad_norm": 9.491918563842773, + "learning_rate": 1e-06, + "loss": 0.3106, + "mean_token_accuracy": 0.8788498044013977, + "num_tokens": 12061548.0, + "step": 3748 + }, + { + "epoch": 1.9176470588235293, + "grad_norm": 10.14240550994873, + "learning_rate": 1e-06, + "loss": 0.34, + "mean_token_accuracy": 0.8552049398422241, + "num_tokens": 12064312.0, + "step": 3749 + }, + { + "epoch": 1.918158567774936, + "grad_norm": 8.879141807556152, + "learning_rate": 1e-06, + "loss": 0.2153, + "mean_token_accuracy": 0.9164814949035645, + "num_tokens": 12068045.0, + "step": 3750 + }, + { + "epoch": 1.9186700767263427, + "grad_norm": 8.152388572692871, + "learning_rate": 1e-06, + "loss": 0.2602, + "mean_token_accuracy": 0.9101957082748413, + "num_tokens": 12071269.0, + "step": 3751 + }, + { + "epoch": 1.9191815856777494, + "grad_norm": 11.840229988098145, + "learning_rate": 1e-06, + "loss": 0.3596, + "mean_token_accuracy": 0.8398903608322144, + "num_tokens": 12073955.0, + "step": 3752 + }, + { + "epoch": 1.9196930946291562, + "grad_norm": 10.752592086791992, + "learning_rate": 1e-06, + "loss": 0.2804, + "mean_token_accuracy": 0.8838266134262085, + "num_tokens": 12076481.0, + "step": 3753 + }, + { + "epoch": 1.9202046035805627, + "grad_norm": 7.500467300415039, + "learning_rate": 1e-06, + "loss": 0.1928, + "mean_token_accuracy": 0.9244814515113831, + "num_tokens": 12079787.0, + "step": 3754 + }, + { + "epoch": 1.9207161125319692, + "grad_norm": 6.733560085296631, + "learning_rate": 1e-06, + "loss": 0.1908, + "mean_token_accuracy": 0.9155727624893188, + "num_tokens": 12083650.0, + "step": 3755 + }, + { + "epoch": 1.921227621483376, + "grad_norm": 8.47630500793457, + "learning_rate": 1e-06, + "loss": 0.2669, + "mean_token_accuracy": 0.9165948629379272, + "num_tokens": 12086431.0, + "step": 3756 + }, + { + "epoch": 1.9217391304347826, + "grad_norm": 10.41593074798584, + "learning_rate": 1e-06, + "loss": 0.3127, + "mean_token_accuracy": 0.8971494436264038, + "num_tokens": 12089585.0, + "step": 3757 + }, + { + "epoch": 1.9222506393861893, + "grad_norm": 6.248229503631592, + "learning_rate": 1e-06, + "loss": 0.2078, + "mean_token_accuracy": 0.9127901792526245, + "num_tokens": 12093021.0, + "step": 3758 + }, + { + "epoch": 1.922762148337596, + "grad_norm": 11.447075843811035, + "learning_rate": 1e-06, + "loss": 0.3067, + "mean_token_accuracy": 0.8664413690567017, + "num_tokens": 12095979.0, + "step": 3759 + }, + { + "epoch": 1.9232736572890026, + "grad_norm": 8.026132583618164, + "learning_rate": 1e-06, + "loss": 0.2134, + "mean_token_accuracy": 0.9207901954650879, + "num_tokens": 12099144.0, + "step": 3760 + }, + { + "epoch": 1.923785166240409, + "grad_norm": 10.85414981842041, + "learning_rate": 1e-06, + "loss": 0.31, + "mean_token_accuracy": 0.8828498125076294, + "num_tokens": 12102061.0, + "step": 3761 + }, + { + "epoch": 1.9242966751918158, + "grad_norm": 8.995587348937988, + "learning_rate": 1e-06, + "loss": 0.2087, + "mean_token_accuracy": 0.915449857711792, + "num_tokens": 12104834.0, + "step": 3762 + }, + { + "epoch": 1.9248081841432225, + "grad_norm": 9.22178840637207, + "learning_rate": 1e-06, + "loss": 0.1729, + "mean_token_accuracy": 0.9377037882804871, + "num_tokens": 12108085.0, + "step": 3763 + }, + { + "epoch": 1.9253196930946292, + "grad_norm": 8.742631912231445, + "learning_rate": 1e-06, + "loss": 0.2211, + "mean_token_accuracy": 0.91883784532547, + "num_tokens": 12111515.0, + "step": 3764 + }, + { + "epoch": 1.9258312020460358, + "grad_norm": 9.922121047973633, + "learning_rate": 1e-06, + "loss": 0.2315, + "mean_token_accuracy": 0.9016212821006775, + "num_tokens": 12115167.0, + "step": 3765 + }, + { + "epoch": 1.9263427109974425, + "grad_norm": 8.839345932006836, + "learning_rate": 1e-06, + "loss": 0.2595, + "mean_token_accuracy": 0.8813503980636597, + "num_tokens": 12118112.0, + "step": 3766 + }, + { + "epoch": 1.926854219948849, + "grad_norm": 8.442924499511719, + "learning_rate": 1e-06, + "loss": 0.2839, + "mean_token_accuracy": 0.8839718103408813, + "num_tokens": 12120549.0, + "step": 3767 + }, + { + "epoch": 1.9273657289002557, + "grad_norm": 9.37731647491455, + "learning_rate": 1e-06, + "loss": 0.2824, + "mean_token_accuracy": 0.9049773216247559, + "num_tokens": 12122931.0, + "step": 3768 + }, + { + "epoch": 1.9278772378516624, + "grad_norm": 9.590326309204102, + "learning_rate": 1e-06, + "loss": 0.2354, + "mean_token_accuracy": 0.890670657157898, + "num_tokens": 12125837.0, + "step": 3769 + }, + { + "epoch": 1.9283887468030692, + "grad_norm": 8.272212982177734, + "learning_rate": 1e-06, + "loss": 0.2408, + "mean_token_accuracy": 0.8837962746620178, + "num_tokens": 12129384.0, + "step": 3770 + }, + { + "epoch": 1.9289002557544757, + "grad_norm": 6.780163764953613, + "learning_rate": 1e-06, + "loss": 0.2046, + "mean_token_accuracy": 0.9081178903579712, + "num_tokens": 12132460.0, + "step": 3771 + }, + { + "epoch": 1.9294117647058824, + "grad_norm": 8.323593139648438, + "learning_rate": 1e-06, + "loss": 0.2051, + "mean_token_accuracy": 0.9148198366165161, + "num_tokens": 12136011.0, + "step": 3772 + }, + { + "epoch": 1.9299232736572889, + "grad_norm": 7.095215320587158, + "learning_rate": 1e-06, + "loss": 0.1956, + "mean_token_accuracy": 0.9087508916854858, + "num_tokens": 12139710.0, + "step": 3773 + }, + { + "epoch": 1.9304347826086956, + "grad_norm": 8.666645050048828, + "learning_rate": 1e-06, + "loss": 0.2739, + "mean_token_accuracy": 0.9044626951217651, + "num_tokens": 12142951.0, + "step": 3774 + }, + { + "epoch": 1.9309462915601023, + "grad_norm": 6.326287269592285, + "learning_rate": 1e-06, + "loss": 0.1649, + "mean_token_accuracy": 0.9241711497306824, + "num_tokens": 12147287.0, + "step": 3775 + }, + { + "epoch": 1.931457800511509, + "grad_norm": 5.685615062713623, + "learning_rate": 1e-06, + "loss": 0.1506, + "mean_token_accuracy": 0.929517388343811, + "num_tokens": 12151528.0, + "step": 3776 + }, + { + "epoch": 1.9319693094629156, + "grad_norm": 7.893721103668213, + "learning_rate": 1e-06, + "loss": 0.1942, + "mean_token_accuracy": 0.937346339225769, + "num_tokens": 12154810.0, + "step": 3777 + }, + { + "epoch": 1.9324808184143223, + "grad_norm": 13.745471954345703, + "learning_rate": 1e-06, + "loss": 0.3036, + "mean_token_accuracy": 0.903088390827179, + "num_tokens": 12157678.0, + "step": 3778 + }, + { + "epoch": 1.9329923273657288, + "grad_norm": 8.624114990234375, + "learning_rate": 1e-06, + "loss": 0.2234, + "mean_token_accuracy": 0.9179387092590332, + "num_tokens": 12161253.0, + "step": 3779 + }, + { + "epoch": 1.9335038363171355, + "grad_norm": 8.99963092803955, + "learning_rate": 1e-06, + "loss": 0.281, + "mean_token_accuracy": 0.8878580331802368, + "num_tokens": 12164351.0, + "step": 3780 + }, + { + "epoch": 1.9340153452685422, + "grad_norm": 7.388190269470215, + "learning_rate": 1e-06, + "loss": 0.1903, + "mean_token_accuracy": 0.9080736637115479, + "num_tokens": 12168737.0, + "step": 3781 + }, + { + "epoch": 1.934526854219949, + "grad_norm": 6.212324142456055, + "learning_rate": 1e-06, + "loss": 0.2052, + "mean_token_accuracy": 0.9070480465888977, + "num_tokens": 12171810.0, + "step": 3782 + }, + { + "epoch": 1.9350383631713555, + "grad_norm": 5.063010215759277, + "learning_rate": 1e-06, + "loss": 0.1614, + "mean_token_accuracy": 0.9371755123138428, + "num_tokens": 12174964.0, + "step": 3783 + }, + { + "epoch": 1.9355498721227622, + "grad_norm": 9.219164848327637, + "learning_rate": 1e-06, + "loss": 0.2649, + "mean_token_accuracy": 0.8914515376091003, + "num_tokens": 12178623.0, + "step": 3784 + }, + { + "epoch": 1.9360613810741687, + "grad_norm": 9.227806091308594, + "learning_rate": 1e-06, + "loss": 0.274, + "mean_token_accuracy": 0.8993327021598816, + "num_tokens": 12181178.0, + "step": 3785 + }, + { + "epoch": 1.9365728900255754, + "grad_norm": 17.756471633911133, + "learning_rate": 1e-06, + "loss": 0.2792, + "mean_token_accuracy": 0.8906459808349609, + "num_tokens": 12183931.0, + "step": 3786 + }, + { + "epoch": 1.9370843989769821, + "grad_norm": 9.29240608215332, + "learning_rate": 1e-06, + "loss": 0.2296, + "mean_token_accuracy": 0.9070375561714172, + "num_tokens": 12188093.0, + "step": 3787 + }, + { + "epoch": 1.9375959079283889, + "grad_norm": 7.252779006958008, + "learning_rate": 1e-06, + "loss": 0.2007, + "mean_token_accuracy": 0.9026080369949341, + "num_tokens": 12190901.0, + "step": 3788 + }, + { + "epoch": 1.9381074168797954, + "grad_norm": 16.983949661254883, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8873017430305481, + "num_tokens": 12194629.0, + "step": 3789 + }, + { + "epoch": 1.938618925831202, + "grad_norm": 7.837409019470215, + "learning_rate": 1e-06, + "loss": 0.1836, + "mean_token_accuracy": 0.9135836958885193, + "num_tokens": 12198175.0, + "step": 3790 + }, + { + "epoch": 1.9391304347826086, + "grad_norm": 9.965485572814941, + "learning_rate": 1e-06, + "loss": 0.2646, + "mean_token_accuracy": 0.8955537676811218, + "num_tokens": 12201789.0, + "step": 3791 + }, + { + "epoch": 1.9396419437340153, + "grad_norm": 9.60508918762207, + "learning_rate": 1e-06, + "loss": 0.2584, + "mean_token_accuracy": 0.8856832385063171, + "num_tokens": 12204632.0, + "step": 3792 + }, + { + "epoch": 1.940153452685422, + "grad_norm": 9.899654388427734, + "learning_rate": 1e-06, + "loss": 0.251, + "mean_token_accuracy": 0.892526388168335, + "num_tokens": 12208055.0, + "step": 3793 + }, + { + "epoch": 1.9406649616368288, + "grad_norm": 8.699240684509277, + "learning_rate": 1e-06, + "loss": 0.237, + "mean_token_accuracy": 0.9062892198562622, + "num_tokens": 12211663.0, + "step": 3794 + }, + { + "epoch": 1.9411764705882353, + "grad_norm": 12.18742847442627, + "learning_rate": 1e-06, + "loss": 0.3033, + "mean_token_accuracy": 0.8743887543678284, + "num_tokens": 12214362.0, + "step": 3795 + }, + { + "epoch": 1.9416879795396418, + "grad_norm": 6.804687976837158, + "learning_rate": 1e-06, + "loss": 0.2378, + "mean_token_accuracy": 0.9103504419326782, + "num_tokens": 12217263.0, + "step": 3796 + }, + { + "epoch": 1.9421994884910485, + "grad_norm": 8.451045989990234, + "learning_rate": 1e-06, + "loss": 0.2542, + "mean_token_accuracy": 0.9170119762420654, + "num_tokens": 12220808.0, + "step": 3797 + }, + { + "epoch": 1.9427109974424552, + "grad_norm": 10.709380149841309, + "learning_rate": 1e-06, + "loss": 0.271, + "mean_token_accuracy": 0.8962794542312622, + "num_tokens": 12223813.0, + "step": 3798 + }, + { + "epoch": 1.943222506393862, + "grad_norm": 8.983138084411621, + "learning_rate": 1e-06, + "loss": 0.2981, + "mean_token_accuracy": 0.8625292778015137, + "num_tokens": 12227081.0, + "step": 3799 + }, + { + "epoch": 1.9437340153452687, + "grad_norm": 9.263007164001465, + "learning_rate": 1e-06, + "loss": 0.2545, + "mean_token_accuracy": 0.892897367477417, + "num_tokens": 12230363.0, + "step": 3800 + }, + { + "epoch": 1.9442455242966752, + "grad_norm": 7.619927406311035, + "learning_rate": 1e-06, + "loss": 0.2488, + "mean_token_accuracy": 0.8862326145172119, + "num_tokens": 12232869.0, + "step": 3801 + }, + { + "epoch": 1.9447570332480817, + "grad_norm": 18.093679428100586, + "learning_rate": 1e-06, + "loss": 0.2388, + "mean_token_accuracy": 0.8880776166915894, + "num_tokens": 12235655.0, + "step": 3802 + }, + { + "epoch": 1.9452685421994884, + "grad_norm": 12.461442947387695, + "learning_rate": 1e-06, + "loss": 0.3706, + "mean_token_accuracy": 0.8617850542068481, + "num_tokens": 12238552.0, + "step": 3803 + }, + { + "epoch": 1.9457800511508951, + "grad_norm": 7.741090297698975, + "learning_rate": 1e-06, + "loss": 0.2423, + "mean_token_accuracy": 0.8962054252624512, + "num_tokens": 12241548.0, + "step": 3804 + }, + { + "epoch": 1.9462915601023019, + "grad_norm": 9.1693754196167, + "learning_rate": 1e-06, + "loss": 0.3584, + "mean_token_accuracy": 0.8747824430465698, + "num_tokens": 12244863.0, + "step": 3805 + }, + { + "epoch": 1.9468030690537086, + "grad_norm": 10.073578834533691, + "learning_rate": 1e-06, + "loss": 0.2788, + "mean_token_accuracy": 0.8900066018104553, + "num_tokens": 12248090.0, + "step": 3806 + }, + { + "epoch": 1.947314578005115, + "grad_norm": 9.196263313293457, + "learning_rate": 1e-06, + "loss": 0.2894, + "mean_token_accuracy": 0.8955129981040955, + "num_tokens": 12250719.0, + "step": 3807 + }, + { + "epoch": 1.9478260869565216, + "grad_norm": 12.042654991149902, + "learning_rate": 1e-06, + "loss": 0.344, + "mean_token_accuracy": 0.8805822134017944, + "num_tokens": 12253954.0, + "step": 3808 + }, + { + "epoch": 1.9483375959079283, + "grad_norm": 13.082877159118652, + "learning_rate": 1e-06, + "loss": 0.2531, + "mean_token_accuracy": 0.9013957381248474, + "num_tokens": 12257149.0, + "step": 3809 + }, + { + "epoch": 1.948849104859335, + "grad_norm": 9.611446380615234, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8809338808059692, + "num_tokens": 12260083.0, + "step": 3810 + }, + { + "epoch": 1.9493606138107418, + "grad_norm": 9.042582511901855, + "learning_rate": 1e-06, + "loss": 0.2263, + "mean_token_accuracy": 0.9154270887374878, + "num_tokens": 12263567.0, + "step": 3811 + }, + { + "epoch": 1.9498721227621485, + "grad_norm": 32.689064025878906, + "learning_rate": 1e-06, + "loss": 0.296, + "mean_token_accuracy": 0.8779421448707581, + "num_tokens": 12266140.0, + "step": 3812 + }, + { + "epoch": 1.950383631713555, + "grad_norm": 13.595015525817871, + "learning_rate": 1e-06, + "loss": 0.2306, + "mean_token_accuracy": 0.8831639289855957, + "num_tokens": 12268963.0, + "step": 3813 + }, + { + "epoch": 1.9508951406649615, + "grad_norm": 9.153035163879395, + "learning_rate": 1e-06, + "loss": 0.2526, + "mean_token_accuracy": 0.8850160241127014, + "num_tokens": 12272277.0, + "step": 3814 + }, + { + "epoch": 1.9514066496163682, + "grad_norm": 8.463016510009766, + "learning_rate": 1e-06, + "loss": 0.2646, + "mean_token_accuracy": 0.8925367593765259, + "num_tokens": 12275983.0, + "step": 3815 + }, + { + "epoch": 1.951918158567775, + "grad_norm": 11.143169403076172, + "learning_rate": 1e-06, + "loss": 0.2729, + "mean_token_accuracy": 0.9097453355789185, + "num_tokens": 12279195.0, + "step": 3816 + }, + { + "epoch": 1.9524296675191817, + "grad_norm": 13.13372802734375, + "learning_rate": 1e-06, + "loss": 0.263, + "mean_token_accuracy": 0.897754967212677, + "num_tokens": 12281946.0, + "step": 3817 + }, + { + "epoch": 1.9529411764705882, + "grad_norm": 6.537653923034668, + "learning_rate": 1e-06, + "loss": 0.1447, + "mean_token_accuracy": 0.9435068964958191, + "num_tokens": 12284666.0, + "step": 3818 + }, + { + "epoch": 1.953452685421995, + "grad_norm": 8.466865539550781, + "learning_rate": 1e-06, + "loss": 0.2397, + "mean_token_accuracy": 0.8861206769943237, + "num_tokens": 12287570.0, + "step": 3819 + }, + { + "epoch": 1.9539641943734014, + "grad_norm": 6.398421287536621, + "learning_rate": 1e-06, + "loss": 0.1949, + "mean_token_accuracy": 0.9079906344413757, + "num_tokens": 12291128.0, + "step": 3820 + }, + { + "epoch": 1.9544757033248081, + "grad_norm": 9.11660385131836, + "learning_rate": 1e-06, + "loss": 0.2636, + "mean_token_accuracy": 0.8940964937210083, + "num_tokens": 12293439.0, + "step": 3821 + }, + { + "epoch": 1.9549872122762149, + "grad_norm": 11.323416709899902, + "learning_rate": 1e-06, + "loss": 0.3658, + "mean_token_accuracy": 0.858650803565979, + "num_tokens": 12296186.0, + "step": 3822 + }, + { + "epoch": 1.9554987212276216, + "grad_norm": 8.663960456848145, + "learning_rate": 1e-06, + "loss": 0.1925, + "mean_token_accuracy": 0.8926619291305542, + "num_tokens": 12299744.0, + "step": 3823 + }, + { + "epoch": 1.956010230179028, + "grad_norm": 11.955145835876465, + "learning_rate": 1e-06, + "loss": 0.3312, + "mean_token_accuracy": 0.8807655572891235, + "num_tokens": 12302473.0, + "step": 3824 + }, + { + "epoch": 1.9565217391304348, + "grad_norm": 14.572118759155273, + "learning_rate": 1e-06, + "loss": 0.3084, + "mean_token_accuracy": 0.9011697769165039, + "num_tokens": 12305543.0, + "step": 3825 + }, + { + "epoch": 1.9570332480818413, + "grad_norm": 10.516602516174316, + "learning_rate": 1e-06, + "loss": 0.2749, + "mean_token_accuracy": 0.879536509513855, + "num_tokens": 12308685.0, + "step": 3826 + }, + { + "epoch": 1.957544757033248, + "grad_norm": 6.296133518218994, + "learning_rate": 1e-06, + "loss": 0.1833, + "mean_token_accuracy": 0.9128022789955139, + "num_tokens": 12312282.0, + "step": 3827 + }, + { + "epoch": 1.9580562659846548, + "grad_norm": 9.763606071472168, + "learning_rate": 1e-06, + "loss": 0.3254, + "mean_token_accuracy": 0.8675463199615479, + "num_tokens": 12315344.0, + "step": 3828 + }, + { + "epoch": 1.9585677749360615, + "grad_norm": 9.815189361572266, + "learning_rate": 1e-06, + "loss": 0.2757, + "mean_token_accuracy": 0.8966290950775146, + "num_tokens": 12318278.0, + "step": 3829 + }, + { + "epoch": 1.959079283887468, + "grad_norm": 10.051891326904297, + "learning_rate": 1e-06, + "loss": 0.1843, + "mean_token_accuracy": 0.9080369472503662, + "num_tokens": 12322239.0, + "step": 3830 + }, + { + "epoch": 1.9595907928388747, + "grad_norm": 7.15139102935791, + "learning_rate": 1e-06, + "loss": 0.2018, + "mean_token_accuracy": 0.9237651824951172, + "num_tokens": 12325305.0, + "step": 3831 + }, + { + "epoch": 1.9601023017902812, + "grad_norm": 7.866209983825684, + "learning_rate": 1e-06, + "loss": 0.2176, + "mean_token_accuracy": 0.9052408337593079, + "num_tokens": 12327989.0, + "step": 3832 + }, + { + "epoch": 1.960613810741688, + "grad_norm": 12.204204559326172, + "learning_rate": 1e-06, + "loss": 0.2675, + "mean_token_accuracy": 0.8838554620742798, + "num_tokens": 12331303.0, + "step": 3833 + }, + { + "epoch": 1.9611253196930947, + "grad_norm": 7.487768650054932, + "learning_rate": 1e-06, + "loss": 0.196, + "mean_token_accuracy": 0.8900117874145508, + "num_tokens": 12334620.0, + "step": 3834 + }, + { + "epoch": 1.9616368286445014, + "grad_norm": 8.225921630859375, + "learning_rate": 1e-06, + "loss": 0.2184, + "mean_token_accuracy": 0.9165387153625488, + "num_tokens": 12338333.0, + "step": 3835 + }, + { + "epoch": 1.962148337595908, + "grad_norm": 8.69554615020752, + "learning_rate": 1e-06, + "loss": 0.231, + "mean_token_accuracy": 0.9033557176589966, + "num_tokens": 12341585.0, + "step": 3836 + }, + { + "epoch": 1.9626598465473146, + "grad_norm": 7.608117580413818, + "learning_rate": 1e-06, + "loss": 0.1739, + "mean_token_accuracy": 0.9259748458862305, + "num_tokens": 12345448.0, + "step": 3837 + }, + { + "epoch": 1.9631713554987211, + "grad_norm": 9.176101684570312, + "learning_rate": 1e-06, + "loss": 0.2737, + "mean_token_accuracy": 0.8937789797782898, + "num_tokens": 12349699.0, + "step": 3838 + }, + { + "epoch": 1.9636828644501279, + "grad_norm": 8.940016746520996, + "learning_rate": 1e-06, + "loss": 0.222, + "mean_token_accuracy": 0.9073907136917114, + "num_tokens": 12353427.0, + "step": 3839 + }, + { + "epoch": 1.9641943734015346, + "grad_norm": 8.370539665222168, + "learning_rate": 1e-06, + "loss": 0.2228, + "mean_token_accuracy": 0.91877281665802, + "num_tokens": 12356493.0, + "step": 3840 + }, + { + "epoch": 1.9647058823529413, + "grad_norm": 8.771291732788086, + "learning_rate": 1e-06, + "loss": 0.2203, + "mean_token_accuracy": 0.8952369689941406, + "num_tokens": 12359813.0, + "step": 3841 + }, + { + "epoch": 1.9652173913043478, + "grad_norm": 17.787418365478516, + "learning_rate": 1e-06, + "loss": 0.2522, + "mean_token_accuracy": 0.8936169147491455, + "num_tokens": 12362444.0, + "step": 3842 + }, + { + "epoch": 1.9657289002557545, + "grad_norm": 9.675430297851562, + "learning_rate": 1e-06, + "loss": 0.2476, + "mean_token_accuracy": 0.9124279022216797, + "num_tokens": 12365150.0, + "step": 3843 + }, + { + "epoch": 1.966240409207161, + "grad_norm": 11.198933601379395, + "learning_rate": 1e-06, + "loss": 0.2492, + "mean_token_accuracy": 0.9019718170166016, + "num_tokens": 12368303.0, + "step": 3844 + }, + { + "epoch": 1.9667519181585678, + "grad_norm": 6.718945026397705, + "learning_rate": 1e-06, + "loss": 0.1668, + "mean_token_accuracy": 0.8930510878562927, + "num_tokens": 12371143.0, + "step": 3845 + }, + { + "epoch": 1.9672634271099745, + "grad_norm": 14.400267601013184, + "learning_rate": 1e-06, + "loss": 0.414, + "mean_token_accuracy": 0.8575010895729065, + "num_tokens": 12374460.0, + "step": 3846 + }, + { + "epoch": 1.9677749360613812, + "grad_norm": 7.3311967849731445, + "learning_rate": 1e-06, + "loss": 0.2288, + "mean_token_accuracy": 0.9104683995246887, + "num_tokens": 12377765.0, + "step": 3847 + }, + { + "epoch": 1.9682864450127877, + "grad_norm": 14.15982723236084, + "learning_rate": 1e-06, + "loss": 0.3691, + "mean_token_accuracy": 0.8665562272071838, + "num_tokens": 12380533.0, + "step": 3848 + }, + { + "epoch": 1.9687979539641942, + "grad_norm": 7.253175735473633, + "learning_rate": 1e-06, + "loss": 0.2024, + "mean_token_accuracy": 0.9077223539352417, + "num_tokens": 12384535.0, + "step": 3849 + }, + { + "epoch": 1.969309462915601, + "grad_norm": 9.456335067749023, + "learning_rate": 1e-06, + "loss": 0.3239, + "mean_token_accuracy": 0.8554844856262207, + "num_tokens": 12387358.0, + "step": 3850 + }, + { + "epoch": 1.9698209718670077, + "grad_norm": 8.29610538482666, + "learning_rate": 1e-06, + "loss": 0.2187, + "mean_token_accuracy": 0.9255608916282654, + "num_tokens": 12391815.0, + "step": 3851 + }, + { + "epoch": 1.9703324808184144, + "grad_norm": 10.00345230102539, + "learning_rate": 1e-06, + "loss": 0.3137, + "mean_token_accuracy": 0.886292576789856, + "num_tokens": 12394679.0, + "step": 3852 + }, + { + "epoch": 1.9708439897698211, + "grad_norm": 8.282341957092285, + "learning_rate": 1e-06, + "loss": 0.2014, + "mean_token_accuracy": 0.9121664762496948, + "num_tokens": 12397641.0, + "step": 3853 + }, + { + "epoch": 1.9713554987212276, + "grad_norm": 7.3266682624816895, + "learning_rate": 1e-06, + "loss": 0.2135, + "mean_token_accuracy": 0.920190691947937, + "num_tokens": 12400660.0, + "step": 3854 + }, + { + "epoch": 1.9718670076726341, + "grad_norm": 9.704001426696777, + "learning_rate": 1e-06, + "loss": 0.2633, + "mean_token_accuracy": 0.8859962224960327, + "num_tokens": 12403749.0, + "step": 3855 + }, + { + "epoch": 1.9723785166240408, + "grad_norm": 9.308663368225098, + "learning_rate": 1e-06, + "loss": 0.2313, + "mean_token_accuracy": 0.9056787490844727, + "num_tokens": 12406811.0, + "step": 3856 + }, + { + "epoch": 1.9728900255754476, + "grad_norm": 10.372913360595703, + "learning_rate": 1e-06, + "loss": 0.293, + "mean_token_accuracy": 0.890442967414856, + "num_tokens": 12409497.0, + "step": 3857 + }, + { + "epoch": 1.9734015345268543, + "grad_norm": 10.43718433380127, + "learning_rate": 1e-06, + "loss": 0.2501, + "mean_token_accuracy": 0.9070913791656494, + "num_tokens": 12412752.0, + "step": 3858 + }, + { + "epoch": 1.973913043478261, + "grad_norm": 10.157232284545898, + "learning_rate": 1e-06, + "loss": 0.2748, + "mean_token_accuracy": 0.8961309194564819, + "num_tokens": 12416400.0, + "step": 3859 + }, + { + "epoch": 1.9744245524296675, + "grad_norm": 9.97862720489502, + "learning_rate": 1e-06, + "loss": 0.3582, + "mean_token_accuracy": 0.8806290626525879, + "num_tokens": 12419141.0, + "step": 3860 + }, + { + "epoch": 1.974936061381074, + "grad_norm": 8.383177757263184, + "learning_rate": 1e-06, + "loss": 0.2753, + "mean_token_accuracy": 0.8923853635787964, + "num_tokens": 12422390.0, + "step": 3861 + }, + { + "epoch": 1.9754475703324808, + "grad_norm": 7.52064323425293, + "learning_rate": 1e-06, + "loss": 0.2395, + "mean_token_accuracy": 0.89072585105896, + "num_tokens": 12425472.0, + "step": 3862 + }, + { + "epoch": 1.9759590792838875, + "grad_norm": 13.770757675170898, + "learning_rate": 1e-06, + "loss": 0.3752, + "mean_token_accuracy": 0.870135486125946, + "num_tokens": 12428805.0, + "step": 3863 + }, + { + "epoch": 1.9764705882352942, + "grad_norm": 10.673250198364258, + "learning_rate": 1e-06, + "loss": 0.2347, + "mean_token_accuracy": 0.8974156379699707, + "num_tokens": 12432409.0, + "step": 3864 + }, + { + "epoch": 1.976982097186701, + "grad_norm": 6.912378311157227, + "learning_rate": 1e-06, + "loss": 0.2134, + "mean_token_accuracy": 0.8961132168769836, + "num_tokens": 12436673.0, + "step": 3865 + }, + { + "epoch": 1.9774936061381074, + "grad_norm": 9.845743179321289, + "learning_rate": 1e-06, + "loss": 0.2718, + "mean_token_accuracy": 0.8925209641456604, + "num_tokens": 12440111.0, + "step": 3866 + }, + { + "epoch": 1.978005115089514, + "grad_norm": 11.18609619140625, + "learning_rate": 1e-06, + "loss": 0.2961, + "mean_token_accuracy": 0.8890672326087952, + "num_tokens": 12443496.0, + "step": 3867 + }, + { + "epoch": 1.9785166240409207, + "grad_norm": 10.847535133361816, + "learning_rate": 1e-06, + "loss": 0.2161, + "mean_token_accuracy": 0.915229320526123, + "num_tokens": 12447241.0, + "step": 3868 + }, + { + "epoch": 1.9790281329923274, + "grad_norm": 9.667433738708496, + "learning_rate": 1e-06, + "loss": 0.2685, + "mean_token_accuracy": 0.9096920490264893, + "num_tokens": 12450146.0, + "step": 3869 + }, + { + "epoch": 1.979539641943734, + "grad_norm": 11.727575302124023, + "learning_rate": 1e-06, + "loss": 0.3659, + "mean_token_accuracy": 0.8579015731811523, + "num_tokens": 12452970.0, + "step": 3870 + }, + { + "epoch": 1.9800511508951406, + "grad_norm": 9.546890258789062, + "learning_rate": 1e-06, + "loss": 0.2925, + "mean_token_accuracy": 0.8746527433395386, + "num_tokens": 12455978.0, + "step": 3871 + }, + { + "epoch": 1.9805626598465473, + "grad_norm": 9.228582382202148, + "learning_rate": 1e-06, + "loss": 0.2308, + "mean_token_accuracy": 0.8990703821182251, + "num_tokens": 12458846.0, + "step": 3872 + }, + { + "epoch": 1.9810741687979538, + "grad_norm": 6.353917121887207, + "learning_rate": 1e-06, + "loss": 0.2035, + "mean_token_accuracy": 0.9115949869155884, + "num_tokens": 12462892.0, + "step": 3873 + }, + { + "epoch": 1.9815856777493606, + "grad_norm": 6.697052955627441, + "learning_rate": 1e-06, + "loss": 0.202, + "mean_token_accuracy": 0.9152620434761047, + "num_tokens": 12467507.0, + "step": 3874 + }, + { + "epoch": 1.9820971867007673, + "grad_norm": 8.984301567077637, + "learning_rate": 1e-06, + "loss": 0.2378, + "mean_token_accuracy": 0.9196711182594299, + "num_tokens": 12470964.0, + "step": 3875 + }, + { + "epoch": 1.982608695652174, + "grad_norm": 7.565044403076172, + "learning_rate": 1e-06, + "loss": 0.1708, + "mean_token_accuracy": 0.9077292084693909, + "num_tokens": 12473815.0, + "step": 3876 + }, + { + "epoch": 1.9831202046035805, + "grad_norm": 11.36252212524414, + "learning_rate": 1e-06, + "loss": 0.3679, + "mean_token_accuracy": 0.8743026852607727, + "num_tokens": 12477118.0, + "step": 3877 + }, + { + "epoch": 1.9836317135549872, + "grad_norm": 9.081700325012207, + "learning_rate": 1e-06, + "loss": 0.2158, + "mean_token_accuracy": 0.9000771045684814, + "num_tokens": 12480082.0, + "step": 3878 + }, + { + "epoch": 1.9841432225063937, + "grad_norm": 9.853203773498535, + "learning_rate": 1e-06, + "loss": 0.3095, + "mean_token_accuracy": 0.8938548564910889, + "num_tokens": 12482918.0, + "step": 3879 + }, + { + "epoch": 1.9846547314578005, + "grad_norm": 8.905927658081055, + "learning_rate": 1e-06, + "loss": 0.2278, + "mean_token_accuracy": 0.9126735925674438, + "num_tokens": 12485752.0, + "step": 3880 + }, + { + "epoch": 1.9851662404092072, + "grad_norm": 9.736785888671875, + "learning_rate": 1e-06, + "loss": 0.2651, + "mean_token_accuracy": 0.8981884121894836, + "num_tokens": 12488560.0, + "step": 3881 + }, + { + "epoch": 1.985677749360614, + "grad_norm": 11.386073112487793, + "learning_rate": 1e-06, + "loss": 0.3072, + "mean_token_accuracy": 0.8974258899688721, + "num_tokens": 12491747.0, + "step": 3882 + }, + { + "epoch": 1.9861892583120204, + "grad_norm": 11.593241691589355, + "learning_rate": 1e-06, + "loss": 0.3354, + "mean_token_accuracy": 0.8726369142532349, + "num_tokens": 12494295.0, + "step": 3883 + }, + { + "epoch": 1.9867007672634271, + "grad_norm": 7.752676486968994, + "learning_rate": 1e-06, + "loss": 0.2248, + "mean_token_accuracy": 0.8726927638053894, + "num_tokens": 12497369.0, + "step": 3884 + }, + { + "epoch": 1.9872122762148337, + "grad_norm": 11.56777572631836, + "learning_rate": 1e-06, + "loss": 0.2555, + "mean_token_accuracy": 0.8901197910308838, + "num_tokens": 12501060.0, + "step": 3885 + }, + { + "epoch": 1.9877237851662404, + "grad_norm": 7.754931449890137, + "learning_rate": 1e-06, + "loss": 0.2332, + "mean_token_accuracy": 0.9099773168563843, + "num_tokens": 12504167.0, + "step": 3886 + }, + { + "epoch": 1.988235294117647, + "grad_norm": 16.402389526367188, + "learning_rate": 1e-06, + "loss": 0.4285, + "mean_token_accuracy": 0.8554580211639404, + "num_tokens": 12507890.0, + "step": 3887 + }, + { + "epoch": 1.9887468030690538, + "grad_norm": 14.535041809082031, + "learning_rate": 1e-06, + "loss": 0.3217, + "mean_token_accuracy": 0.8845096826553345, + "num_tokens": 12510971.0, + "step": 3888 + }, + { + "epoch": 1.9892583120204603, + "grad_norm": 13.066733360290527, + "learning_rate": 1e-06, + "loss": 0.2964, + "mean_token_accuracy": 0.865615963935852, + "num_tokens": 12513907.0, + "step": 3889 + }, + { + "epoch": 1.989769820971867, + "grad_norm": 9.662347793579102, + "learning_rate": 1e-06, + "loss": 0.2359, + "mean_token_accuracy": 0.8858930468559265, + "num_tokens": 12516773.0, + "step": 3890 + }, + { + "epoch": 1.9902813299232736, + "grad_norm": 11.377290725708008, + "learning_rate": 1e-06, + "loss": 0.2222, + "mean_token_accuracy": 0.8943707942962646, + "num_tokens": 12520429.0, + "step": 3891 + }, + { + "epoch": 1.9907928388746803, + "grad_norm": 7.607141971588135, + "learning_rate": 1e-06, + "loss": 0.2192, + "mean_token_accuracy": 0.9240541458129883, + "num_tokens": 12524313.0, + "step": 3892 + }, + { + "epoch": 1.991304347826087, + "grad_norm": 10.51944637298584, + "learning_rate": 1e-06, + "loss": 0.2926, + "mean_token_accuracy": 0.8813197612762451, + "num_tokens": 12526702.0, + "step": 3893 + }, + { + "epoch": 1.9918158567774937, + "grad_norm": 9.305794715881348, + "learning_rate": 1e-06, + "loss": 0.2725, + "mean_token_accuracy": 0.8967581987380981, + "num_tokens": 12530135.0, + "step": 3894 + }, + { + "epoch": 1.9923273657289002, + "grad_norm": 9.378854751586914, + "learning_rate": 1e-06, + "loss": 0.2773, + "mean_token_accuracy": 0.8732949495315552, + "num_tokens": 12532992.0, + "step": 3895 + }, + { + "epoch": 1.992838874680307, + "grad_norm": 9.111294746398926, + "learning_rate": 1e-06, + "loss": 0.3306, + "mean_token_accuracy": 0.8751720190048218, + "num_tokens": 12535343.0, + "step": 3896 + }, + { + "epoch": 1.9933503836317135, + "grad_norm": 9.388124465942383, + "learning_rate": 1e-06, + "loss": 0.245, + "mean_token_accuracy": 0.8554875254631042, + "num_tokens": 12538073.0, + "step": 3897 + }, + { + "epoch": 1.9938618925831202, + "grad_norm": 8.899223327636719, + "learning_rate": 1e-06, + "loss": 0.2342, + "mean_token_accuracy": 0.9018076658248901, + "num_tokens": 12540773.0, + "step": 3898 + }, + { + "epoch": 1.994373401534527, + "grad_norm": 10.65693473815918, + "learning_rate": 1e-06, + "loss": 0.2029, + "mean_token_accuracy": 0.9051783084869385, + "num_tokens": 12544512.0, + "step": 3899 + }, + { + "epoch": 1.9948849104859336, + "grad_norm": 7.191842079162598, + "learning_rate": 1e-06, + "loss": 0.192, + "mean_token_accuracy": 0.9080522060394287, + "num_tokens": 12546649.0, + "step": 3900 + }, + { + "epoch": 1.9953964194373401, + "grad_norm": 10.208560943603516, + "learning_rate": 1e-06, + "loss": 0.3035, + "mean_token_accuracy": 0.8908058404922485, + "num_tokens": 12549588.0, + "step": 3901 + }, + { + "epoch": 1.9959079283887466, + "grad_norm": 6.371354103088379, + "learning_rate": 1e-06, + "loss": 0.1895, + "mean_token_accuracy": 0.9135583639144897, + "num_tokens": 12553185.0, + "step": 3902 + }, + { + "epoch": 1.9964194373401534, + "grad_norm": 11.731912612915039, + "learning_rate": 1e-06, + "loss": 0.2937, + "mean_token_accuracy": 0.8901053667068481, + "num_tokens": 12555839.0, + "step": 3903 + }, + { + "epoch": 1.99693094629156, + "grad_norm": 14.1300630569458, + "learning_rate": 1e-06, + "loss": 0.3016, + "mean_token_accuracy": 0.8852113485336304, + "num_tokens": 12558711.0, + "step": 3904 + }, + { + "epoch": 1.9974424552429668, + "grad_norm": 9.613521575927734, + "learning_rate": 1e-06, + "loss": 0.2886, + "mean_token_accuracy": 0.8924521207809448, + "num_tokens": 12561492.0, + "step": 3905 + }, + { + "epoch": 1.9979539641943735, + "grad_norm": 8.717424392700195, + "learning_rate": 1e-06, + "loss": 0.1966, + "mean_token_accuracy": 0.9139085412025452, + "num_tokens": 12564681.0, + "step": 3906 + }, + { + "epoch": 1.99846547314578, + "grad_norm": 11.830551147460938, + "learning_rate": 1e-06, + "loss": 0.1952, + "mean_token_accuracy": 0.8842551708221436, + "num_tokens": 12567871.0, + "step": 3907 + }, + { + "epoch": 1.9989769820971865, + "grad_norm": 9.991074562072754, + "learning_rate": 1e-06, + "loss": 0.2927, + "mean_token_accuracy": 0.8622512817382812, + "num_tokens": 12570048.0, + "step": 3908 + }, + { + "epoch": 1.9994884910485933, + "grad_norm": 10.109895706176758, + "learning_rate": 1e-06, + "loss": 0.228, + "mean_token_accuracy": 0.9031952023506165, + "num_tokens": 12573671.0, + "step": 3909 + }, + { + "epoch": 2.0, + "grad_norm": 9.073880195617676, + "learning_rate": 1e-06, + "loss": 0.2126, + "mean_token_accuracy": 0.906349241733551, + "num_tokens": 12575570.0, + "step": 3910 + }, + { + "epoch": 2.0, + "step": 3910, + "total_flos": 5.664122751544197e+17, + "train_loss": 0.30678034723186126, + "train_runtime": 2812.6177, + "train_samples_per_second": 22.233, + "train_steps_per_second": 1.39 + } + ], + "logging_steps": 1, + "max_steps": 3910, + "num_input_tokens_seen": 0, + "num_train_epochs": 2, + "save_steps": 1955, + "stateful_callbacks": { + "TrainerControl": { + "args": { + "should_epoch_stop": false, + "should_evaluate": false, + "should_log": false, + "should_save": true, + "should_training_stop": true + }, + "attributes": {} + } + }, + "total_flos": 5.664122751544197e+17, + "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..15b6a9b --- /dev/null +++ b/training_args.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42b22052ffa471199cf04baa540a05c2cd5a2cfcc5d66cf9d2d73ba7a9e58bab +size 13329