commit 2085ab1ef456c2f3083bd0e336d8af8f76c0b8e5
Author: ModelHub XC <noreply@modelhub.org.cn>
Date:   Fri Oct 16 09:38:14 2026 +0800

    初始化项目，由ModelHub XC社区提供模型
    
    Model: jondurbin/spicyboros-13b-2.2-checkpoints
    Source: Original Platform

diff --git a/.gitattributes b/.gitattributes
new file mode 100644
index 0000000..53d7257
--- /dev/null
+++ b/.gitattributes
@@ -0,0 +1,47 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.bz2 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
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack 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
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* 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
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zstandard filter=lfs diff=lfs merge=lfs -text
+*.tfevents* filter=lfs diff=lfs merge=lfs -text
+*.db* filter=lfs diff=lfs merge=lfs -text
+*.ark* filter=lfs diff=lfs merge=lfs -text
+**/*ckpt*data* filter=lfs diff=lfs merge=lfs -text
+**/*ckpt*.meta filter=lfs diff=lfs merge=lfs -text
+**/*ckpt*.index filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.gguf* filter=lfs diff=lfs merge=lfs -text
+*.ggml filter=lfs diff=lfs merge=lfs -text
+*.llamafile* filter=lfs diff=lfs merge=lfs -text
+*.pt2 filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
\ No newline at end of file
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..df93db9
--- /dev/null
+++ b/README.md
@@ -0,0 +1,7 @@
+---
+license: llama2
+tags:
+- not-for-all-audiences
+---
+
+Checkpoints of https://hf.co/jondurbin/spicyboros-13b-2.2
\ No newline at end of file
diff --git a/checkpoint-350/config.json b/checkpoint-350/config.json
new file mode 100644
index 0000000..87e67dc
--- /dev/null
+++ b/checkpoint-350/config.json
@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "/workspace/llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.33.1",
+  "use_cache": false,
+  "vocab_size": 32000
+}
diff --git a/checkpoint-350/generation_config.json b/checkpoint-350/generation_config.json
new file mode 100644
index 0000000..0b71b22
--- /dev/null
+++ b/checkpoint-350/generation_config.json
@@ -0,0 +1,4 @@
+{
+  "max_new_tokens": 4096,
+  "transformers_version": "4.33.1"
+}
diff --git a/checkpoint-350/latest b/checkpoint-350/latest
new file mode 100644
index 0000000..60987f8
--- /dev/null
+++ b/checkpoint-350/latest
@@ -0,0 +1 @@
+global_step350
\ No newline at end of file
diff --git a/checkpoint-350/pytorch_model-00001-of-00003.bin b/checkpoint-350/pytorch_model-00001-of-00003.bin
new file mode 100644
index 0000000..6af8136
--- /dev/null
+++ b/checkpoint-350/pytorch_model-00001-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3beb1f7977a68d07ca6405476b8832f0fb05d41cea54af699bf387bd231057af
+size 9948719433
diff --git a/checkpoint-350/pytorch_model-00002-of-00003.bin b/checkpoint-350/pytorch_model-00002-of-00003.bin
new file mode 100644
index 0000000..3f89e85
--- /dev/null
+++ b/checkpoint-350/pytorch_model-00002-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01e5db58cb847f4df55521655e047fdd8416c7321242968b9cbc2d7dd9a35021
+size 9904155408
diff --git a/checkpoint-350/pytorch_model-00003-of-00003.bin b/checkpoint-350/pytorch_model-00003-of-00003.bin
new file mode 100644
index 0000000..1ec2e5f
--- /dev/null
+++ b/checkpoint-350/pytorch_model-00003-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b7a5debc9ba2ff2585d358bd09532bf2f63083cfb9eb04bc6773a1bd3b92117
+size 6178978207
diff --git a/checkpoint-350/pytorch_model.bin.index.json b/checkpoint-350/pytorch_model.bin.index.json
new file mode 100644
index 0000000..5ebaea3
--- /dev/null
+++ b/checkpoint-350/pytorch_model.bin.index.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a81b85fdd21a6ffc94396229e2cac8f8135c4a542730e957ede126202c6bf475
+size 29894
diff --git a/checkpoint-350/special_tokens_map.json b/checkpoint-350/special_tokens_map.json
new file mode 100644
index 0000000..14761dc
--- /dev/null
+++ b/checkpoint-350/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}
diff --git a/checkpoint-350/tokenizer.model b/checkpoint-350/tokenizer.model
new file mode 100644
index 0000000..6c00c74
--- /dev/null
+++ b/checkpoint-350/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/checkpoint-350/tokenizer_config.json b/checkpoint-350/tokenizer_config.json
new file mode 100644
index 0000000..16ada5c
--- /dev/null
+++ b/checkpoint-350/tokenizer_config.json
@@ -0,0 +1,37 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_default_system_prompt": true
+}
diff --git a/checkpoint-350/trainer_state.json b/checkpoint-350/trainer_state.json
new file mode 100644
index 0000000..fab524b
--- /dev/null
+++ b/checkpoint-350/trainer_state.json
@@ -0,0 +1,2231 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.216867469879518,
+  "eval_steps": 25,
+  "global_step": 350,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 7.333333333333333e-06,
+      "loss": 0.9353,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 1.4666666666666666e-05,
+      "loss": 0.8037,
+      "step": 2
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 2.2e-05,
+      "loss": 0.7535,
+      "step": 3
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 2.1999680209249292e-05,
+      "loss": 0.6836,
+      "step": 4
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 2.1998720855591014e-05,
+      "loss": 0.8619,
+      "step": 5
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 2.1997121994805608e-05,
+      "loss": 1.3258,
+      "step": 6
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 2.1994883719856868e-05,
+      "loss": 1.1205,
+      "step": 7
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 2.199200616088654e-05,
+      "loss": 0.7097,
+      "step": 8
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 2.1988489485206775e-05,
+      "loss": 0.6175,
+      "step": 9
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 2.198433389729035e-05,
+      "loss": 0.6068,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 2.1979539638758833e-05,
+      "loss": 0.713,
+      "step": 11
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 2.19741069883685e-05,
+      "loss": 1.0415,
+      "step": 12
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 2.1968036261994137e-05,
+      "loss": 1.2173,
+      "step": 13
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 2.1961327812610683e-05,
+      "loss": 0.6987,
+      "step": 14
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 2.19539820302727e-05,
+      "loss": 0.6693,
+      "step": 15
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 2.1945999342091685e-05,
+      "loss": 0.6082,
+      "step": 16
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 2.1937380212211243e-05,
+      "loss": 0.6565,
+      "step": 17
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 2.1928125141780108e-05,
+      "loss": 0.9413,
+      "step": 18
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 2.191823466892299e-05,
+      "loss": 1.3279,
+      "step": 19
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 2.1907709368709293e-05,
+      "loss": 0.7384,
+      "step": 20
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.189654985311969e-05,
+      "loss": 0.7139,
+      "step": 21
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.1884756771010514e-05,
+      "loss": 0.5926,
+      "step": 22
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.1872330808076058e-05,
+      "loss": 0.5966,
+      "step": 23
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.1859272686808682e-05,
+      "loss": 0.843,
+      "step": 24
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.184558316645683e-05,
+      "loss": 1.2463,
+      "step": 25
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.7440170645713806,
+      "eval_runtime": 49.2892,
+      "eval_samples_per_second": 17.631,
+      "eval_steps_per_second": 0.284,
+      "step": 25
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.1831263042980858e-05,
+      "loss": 0.8548,
+      "step": 26
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.1816313149006775e-05,
+      "loss": 0.6954,
+      "step": 27
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.180073435377782e-05,
+      "loss": 0.666,
+      "step": 28
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.1784527563103927e-05,
+      "loss": 0.6027,
+      "step": 29
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.1767693719309048e-05,
+      "loss": 0.6978,
+      "step": 30
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.175023380117637e-05,
+      "loss": 0.9855,
+      "step": 31
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.1732148823891417e-05,
+      "loss": 1.0321,
+      "step": 32
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.171343983898299e-05,
+      "loss": 0.6891,
+      "step": 33
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.1694107934262066e-05,
+      "loss": 0.6134,
+      "step": 34
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.1674154233758516e-05,
+      "loss": 0.5408,
+      "step": 35
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.165357989765577e-05,
+      "loss": 0.6853,
+      "step": 36
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1632386122223363e-05,
+      "loss": 0.9126,
+      "step": 37
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1610574139747343e-05,
+      "loss": 1.2144,
+      "step": 38
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1588145218458675e-05,
+      "loss": 0.6852,
+      "step": 39
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.156510066245946e-05,
+      "loss": 0.6938,
+      "step": 40
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.1541441811647135e-05,
+      "loss": 0.599,
+      "step": 41
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.1517170041636537e-05,
+      "loss": 0.6201,
+      "step": 42
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.1492286763679958e-05,
+      "loss": 0.8481,
+      "step": 43
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.1466793424585053e-05,
+      "loss": 1.3248,
+      "step": 44
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.1440691506630743e-05,
+      "loss": 0.6589,
+      "step": 45
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.1413982527481012e-05,
+      "loss": 0.6731,
+      "step": 46
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.1386668040096685e-05,
+      "loss": 0.6028,
+      "step": 47
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.1358749632645103e-05,
+      "loss": 0.5619,
+      "step": 48
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 2.1330228928407817e-05,
+      "loss": 0.7259,
+      "step": 49
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 2.1301107585686164e-05,
+      "loss": 1.2589,
+      "step": 50
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.7223495841026306,
+      "eval_runtime": 45.8793,
+      "eval_samples_per_second": 18.941,
+      "eval_steps_per_second": 0.305,
+      "step": 50
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 2.127138729770489e-05,
+      "loss": 0.8173,
+      "step": 51
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 2.1241069792513666e-05,
+      "loss": 0.6887,
+      "step": 52
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 2.1210156832886625e-05,
+      "loss": 0.6537,
+      "step": 53
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 2.1178650216219872e-05,
+      "loss": 0.5452,
+      "step": 54
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 2.114655177442697e-05,
+      "loss": 0.6558,
+      "step": 55
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 2.111386337383243e-05,
+      "loss": 1.0381,
+      "step": 56
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 2.1080586915063186e-05,
+      "loss": 0.99,
+      "step": 57
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.1046724332938103e-05,
+      "loss": 0.657,
+      "step": 58
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.101227759635546e-05,
+      "loss": 0.6018,
+      "step": 59
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.0977248708178485e-05,
+      "loss": 0.5436,
+      "step": 60
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.09416397051189e-05,
+      "loss": 0.6401,
+      "step": 61
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.0905452657618488e-05,
+      "loss": 0.9314,
+      "step": 62
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.0868689669728716e-05,
+      "loss": 1.2026,
+      "step": 63
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 2.0831352878988404e-05,
+      "loss": 0.6308,
+      "step": 64
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 2.0793444456299434e-05,
+      "loss": 0.6889,
+      "step": 65
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 2.0754966605800523e-05,
+      "loss": 0.5788,
+      "step": 66
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 2.071592156473907e-05,
+      "loss": 0.5568,
+      "step": 67
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 2.067631160334107e-05,
+      "loss": 0.8306,
+      "step": 68
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 2.0636139024679125e-05,
+      "loss": 1.2907,
+      "step": 69
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 2.0595406164538524e-05,
+      "loss": 0.6435,
+      "step": 70
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.0554115391281437e-05,
+      "loss": 0.6731,
+      "step": 71
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.05122691057092e-05,
+      "loss": 0.5612,
+      "step": 72
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 2.046986974092274e-05,
+      "loss": 0.548,
+      "step": 73
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 2.042691976218109e-05,
+      "loss": 0.7796,
+      "step": 74
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 2.0383421666758057e-05,
+      "loss": 1.2264,
+      "step": 75
+    },
+    {
+      "epoch": 0.9,
+      "eval_loss": 0.7080472707748413,
+      "eval_runtime": 45.869,
+      "eval_samples_per_second": 18.945,
+      "eval_steps_per_second": 0.305,
+      "step": 75
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 2.0339377983797022e-05,
+      "loss": 0.8322,
+      "step": 76
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 2.029479127416389e-05,
+      "loss": 0.7071,
+      "step": 77
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.0249664130298165e-05,
+      "loss": 0.6058,
+      "step": 78
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.020399917606226e-05,
+      "loss": 0.5243,
+      "step": 79
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.01577990665889e-05,
+      "loss": 0.6636,
+      "step": 80
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 2.0111066488126753e-05,
+      "loss": 1.0678,
+      "step": 81
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.0063804157884248e-05,
+      "loss": 0.853,
+      "step": 82
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.0016014823871578e-05,
+      "loss": 0.8473,
+      "step": 83
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.9967701264740925e-05,
+      "loss": 0.7626,
+      "step": 84
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.99188662896249e-05,
+      "loss": 0.6879,
+      "step": 85
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.9869512737973196e-05,
+      "loss": 0.5818,
+      "step": 86
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.9819643479387516e-05,
+      "loss": 0.5151,
+      "step": 87
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.9769261413454705e-05,
+      "loss": 0.6649,
+      "step": 88
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.971836946957816e-05,
+      "loss": 1.0307,
+      "step": 89
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.96669706068075e-05,
+      "loss": 1.0031,
+      "step": 90
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.961506781366653e-05,
+      "loss": 0.67,
+      "step": 91
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.9562664107979465e-05,
+      "loss": 0.6331,
+      "step": 92
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.950976253669547e-05,
+      "loss": 0.5232,
+      "step": 93
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.9456366175711482e-05,
+      "loss": 0.6291,
+      "step": 94
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.9402478129693395e-05,
+      "loss": 0.8978,
+      "step": 95
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.9348101531895503e-05,
+      "loss": 1.1345,
+      "step": 96
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.929323954397837e-05,
+      "loss": 0.6327,
+      "step": 97
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.9237895355824952e-05,
+      "loss": 0.6239,
+      "step": 98
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.9182072185355157e-05,
+      "loss": 0.5745,
+      "step": 99
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.9125773278338728e-05,
+      "loss": 0.5543,
+      "step": 100
+    },
+    {
+      "epoch": 1.2,
+      "eval_loss": 0.7009603977203369,
+      "eval_runtime": 46.0556,
+      "eval_samples_per_second": 18.868,
+      "eval_steps_per_second": 0.304,
+      "step": 100
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.9069001908206527e-05,
+      "loss": 0.8478,
+      "step": 101
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.90117613758602e-05,
+      "loss": 1.2408,
+      "step": 102
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.8954055009480263e-05,
+      "loss": 0.5991,
+      "step": 103
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.8895886164332572e-05,
+      "loss": 0.6304,
+      "step": 104
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.883725822257325e-05,
+      "loss": 0.5863,
+      "step": 105
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.877817459305202e-05,
+      "loss": 0.5341,
+      "step": 106
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.8718638711114027e-05,
+      "loss": 0.6937,
+      "step": 107
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.865865403840005e-05,
+      "loss": 1.12,
+      "step": 108
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.8598224062645293e-05,
+      "loss": 0.8003,
+      "step": 109
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.8537352297476526e-05,
+      "loss": 0.6984,
+      "step": 110
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.8476042282207843e-05,
+      "loss": 0.5954,
+      "step": 111
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.8414297581634844e-05,
+      "loss": 0.5379,
+      "step": 112
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.8352121785827374e-05,
+      "loss": 0.6839,
+      "step": 113
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.828951850992078e-05,
+      "loss": 0.9745,
+      "step": 114
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.822649139390572e-05,
+      "loss": 0.9575,
+      "step": 115
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.8163044102416508e-05,
+      "loss": 0.647,
+      "step": 116
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.809918032451805e-05,
+      "loss": 0.5792,
+      "step": 117
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.8034903773491342e-05,
+      "loss": 0.5287,
+      "step": 118
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.797021818661757e-05,
+      "loss": 0.6415,
+      "step": 119
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.790512732496079e-05,
+      "loss": 0.8963,
+      "step": 120
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.78396349731493e-05,
+      "loss": 1.1426,
+      "step": 121
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.7773744939155508e-05,
+      "loss": 0.6213,
+      "step": 122
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.77074610540746e-05,
+      "loss": 0.6494,
+      "step": 123
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.7640787171901715e-05,
+      "loss": 0.5484,
+      "step": 124
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.7573727169307932e-05,
+      "loss": 0.5539,
+      "step": 125
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.6961802244186401,
+      "eval_runtime": 45.816,
+      "eval_samples_per_second": 18.967,
+      "eval_steps_per_second": 0.306,
+      "step": 125
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.7506284945414796e-05,
+      "loss": 0.8497,
+      "step": 126
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.743846442156766e-05,
+      "loss": 1.1775,
+      "step": 127
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.7370269541107663e-05,
+      "loss": 0.6627,
+      "step": 128
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1.7301704269142434e-05,
+      "loss": 0.6425,
+      "step": 129
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 1.7232772592315577e-05,
+      "loss": 0.5537,
+      "step": 130
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 1.716347851857485e-05,
+      "loss": 0.5311,
+      "step": 131
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 1.7093826076939137e-05,
+      "loss": 0.7583,
+      "step": 132
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 1.7023819317264186e-05,
+      "loss": 1.1448,
+      "step": 133
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1.695346231000713e-05,
+      "loss": 0.7377,
+      "step": 134
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1.6882759145989816e-05,
+      "loss": 0.6445,
+      "step": 135
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 1.681171393616095e-05,
+      "loss": 0.5787,
+      "step": 136
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.674033081135708e-05,
+      "loss": 0.5084,
+      "step": 137
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.6668613922062398e-05,
+      "loss": 0.6423,
+      "step": 138
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 1.6596567438167426e-05,
+      "loss": 0.9592,
+      "step": 139
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1.6524195548726557e-05,
+      "loss": 1.0,
+      "step": 140
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1.645150246171449e-05,
+      "loss": 0.6274,
+      "step": 141
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1.637849240378157e-05,
+      "loss": 0.597,
+      "step": 142
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1.630516962000802e-05,
+      "loss": 0.524,
+      "step": 143
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1.623153837365712e-05,
+      "loss": 0.5914,
+      "step": 144
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.6157602945927335e-05,
+      "loss": 0.8503,
+      "step": 145
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 1.6083367635703386e-05,
+      "loss": 1.058,
+      "step": 146
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1.600883675930628e-05,
+      "loss": 0.6366,
+      "step": 147
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.593401465024237e-05,
+      "loss": 0.6191,
+      "step": 148
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 1.5858905658951377e-05,
+      "loss": 0.5433,
+      "step": 149
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1.578351415255342e-05,
+      "loss": 0.5568,
+      "step": 150
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.6907634735107422,
+      "eval_runtime": 45.8388,
+      "eval_samples_per_second": 18.958,
+      "eval_steps_per_second": 0.305,
+      "step": 150
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1.570784451459512e-05,
+      "loss": 0.8225,
+      "step": 151
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 1.5631901144794727e-05,
+      "loss": 1.1446,
+      "step": 152
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 1.5555688458786283e-05,
+      "loss": 0.6378,
+      "step": 153
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 1.547921088786289e-05,
+      "loss": 0.6675,
+      "step": 154
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1.5402472878719058e-05,
+      "loss": 0.5395,
+      "step": 155
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1.532547889319216e-05,
+      "loss": 0.5378,
+      "step": 156
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1.524823340800301e-05,
+      "loss": 0.7644,
+      "step": 157
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1.5170740914495543e-05,
+      "loss": 1.1105,
+      "step": 158
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 1.5093005918375706e-05,
+      "loss": 0.7881,
+      "step": 159
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1.5015032939449451e-05,
+      "loss": 0.6366,
+      "step": 160
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 1.493682651135996e-05,
+      "loss": 0.5853,
+      "step": 161
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 1.485839118132402e-05,
+      "loss": 0.5138,
+      "step": 162
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1.4779731509867645e-05,
+      "loss": 0.6444,
+      "step": 163
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1.4700852070560903e-05,
+      "loss": 0.9914,
+      "step": 164
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.4621757449752e-05,
+      "loss": 0.8894,
+      "step": 165
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 1.4542452246300597e-05,
+      "loss": 0.8105,
+      "step": 166
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 1.4462941071310434e-05,
+      "loss": 0.7536,
+      "step": 167
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 1.438322854786121e-05,
+      "loss": 0.6122,
+      "step": 168
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 1.4303319310739777e-05,
+      "loss": 0.575,
+      "step": 169
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 1.4223218006170675e-05,
+      "loss": 0.5015,
+      "step": 170
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 1.4142929291545953e-05,
+      "loss": 0.6682,
+      "step": 171
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 1.4062457835154405e-05,
+      "loss": 1.0041,
+      "step": 172
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1.3981808315910112e-05,
+      "loss": 0.9622,
+      "step": 173
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 1.3900985423080396e-05,
+      "loss": 0.6121,
+      "step": 174
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 1.3819993856013184e-05,
+      "loss": 0.6222,
+      "step": 175
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.6892703175544739,
+      "eval_runtime": 45.9318,
+      "eval_samples_per_second": 18.919,
+      "eval_steps_per_second": 0.305,
+      "step": 175
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1.3738838323863754e-05,
+      "loss": 0.5335,
+      "step": 176
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 1.3657523545320932e-05,
+      "loss": 0.5978,
+      "step": 177
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 1.3576054248332733e-05,
+      "loss": 0.9021,
+      "step": 178
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 1.3494435169831453e-05,
+      "loss": 1.0881,
+      "step": 179
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 1.3412671055458253e-05,
+      "loss": 0.6492,
+      "step": 180
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 1.3330766659287234e-05,
+      "loss": 0.6129,
+      "step": 181
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 1.3248726743548996e-05,
+      "loss": 0.5737,
+      "step": 182
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 1.3166556078353768e-05,
+      "loss": 0.5407,
+      "step": 183
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 1.3084259441414036e-05,
+      "loss": 0.7442,
+      "step": 184
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1.3001841617766777e-05,
+      "loss": 1.0671,
+      "step": 185
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 1.2919307399495196e-05,
+      "loss": 0.6467,
+      "step": 186
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 1.2836661585450144e-05,
+      "loss": 0.6384,
+      "step": 187
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1.275390898097105e-05,
+      "loss": 0.5394,
+      "step": 188
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 1.267105439760656e-05,
+      "loss": 0.5238,
+      "step": 189
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 1.2588102652834745e-05,
+      "loss": 0.7205,
+      "step": 190
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 1.2505058569783012e-05,
+      "loss": 1.0399,
+      "step": 191
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 1.2421926976947664e-05,
+      "loss": 0.7659,
+      "step": 192
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 1.2338712707913146e-05,
+      "loss": 0.6225,
+      "step": 193
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 1.2255420601071026e-05,
+      "loss": 0.5688,
+      "step": 194
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1.2172055499338637e-05,
+      "loss": 0.5092,
+      "step": 195
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 1.2088622249877521e-05,
+      "loss": 0.623,
+      "step": 196
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 1.200512570381158e-05,
+      "loss": 0.9878,
+      "step": 197
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 1.192157071594503e-05,
+      "loss": 0.9355,
+      "step": 198
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 1.1837962144480099e-05,
+      "loss": 0.666,
+      "step": 199
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.1754304850734577e-05,
+      "loss": 0.5828,
+      "step": 200
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.6850700974464417,
+      "eval_runtime": 45.894,
+      "eval_samples_per_second": 18.935,
+      "eval_steps_per_second": 0.305,
+      "step": 200
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.167060369885915e-05,
+      "loss": 0.4919,
+      "step": 201
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1.1586863555554581e-05,
+      "loss": 0.5812,
+      "step": 202
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.1503089289788741e-05,
+      "loss": 0.8558,
+      "step": 203
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.1419285772513506e-05,
+      "loss": 1.1615,
+      "step": 204
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1.1335457876381556e-05,
+      "loss": 0.5879,
+      "step": 205
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.1251610475463037e-05,
+      "loss": 0.6408,
+      "step": 206
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.1167748444962185e-05,
+      "loss": 0.5349,
+      "step": 207
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.1083876660933852e-05,
+      "loss": 0.5493,
+      "step": 208
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.1e-05,
+      "loss": 0.8208,
+      "step": 209
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.0916123339066147e-05,
+      "loss": 1.0668,
+      "step": 210
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.0832251555037816e-05,
+      "loss": 0.6051,
+      "step": 211
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.0748389524536964e-05,
+      "loss": 0.6071,
+      "step": 212
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.0664542123618447e-05,
+      "loss": 0.5666,
+      "step": 213
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.0580714227486495e-05,
+      "loss": 0.5134,
+      "step": 214
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1.0496910710211261e-05,
+      "loss": 0.7728,
+      "step": 215
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 1.041313644444542e-05,
+      "loss": 1.0725,
+      "step": 216
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1.032939630114085e-05,
+      "loss": 0.744,
+      "step": 217
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1.0245695149265425e-05,
+      "loss": 0.646,
+      "step": 218
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 1.0162037855519904e-05,
+      "loss": 0.5654,
+      "step": 219
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 1.0078429284054971e-05,
+      "loss": 0.5043,
+      "step": 220
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 9.99487429618842e-06,
+      "loss": 0.6779,
+      "step": 221
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 9.91137775012248e-06,
+      "loss": 0.9569,
+      "step": 222
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 9.827944500661365e-06,
+      "loss": 0.966,
+      "step": 223
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 9.744579398928975e-06,
+      "loss": 0.5635,
+      "step": 224
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 9.661287292086853e-06,
+      "loss": 0.6108,
+      "step": 225
+    },
+    {
+      "epoch": 2.71,
+      "eval_loss": 0.6831873655319214,
+      "eval_runtime": 45.9501,
+      "eval_samples_per_second": 18.912,
+      "eval_steps_per_second": 0.305,
+      "step": 225
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 9.578073023052337e-06,
+      "loss": 0.515,
+      "step": 226
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 9.494941430216989e-06,
+      "loss": 0.5935,
+      "step": 227
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 9.411897347165256e-06,
+      "loss": 0.8431,
+      "step": 228
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 9.328945602393441e-06,
+      "loss": 1.1265,
+      "step": 229
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 9.246091019028949e-06,
+      "loss": 0.5872,
+      "step": 230
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 9.163338414549857e-06,
+      "loss": 0.6158,
+      "step": 231
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 9.080692600504803e-06,
+      "loss": 0.5229,
+      "step": 232
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 8.998158382233226e-06,
+      "loss": 0.5407,
+      "step": 233
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 8.915740558585963e-06,
+      "loss": 0.8395,
+      "step": 234
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 8.833443921646237e-06,
+      "loss": 1.162,
+      "step": 235
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 8.751273256451007e-06,
+      "loss": 0.6233,
+      "step": 236
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 8.669233340712766e-06,
+      "loss": 0.6508,
+      "step": 237
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 8.587328944541747e-06,
+      "loss": 0.5542,
+      "step": 238
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.50556483016855e-06,
+      "loss": 0.5076,
+      "step": 239
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.423945751667272e-06,
+      "loss": 0.7069,
+      "step": 240
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 8.342476454679073e-06,
+      "loss": 1.0609,
+      "step": 241
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 8.26116167613625e-06,
+      "loss": 0.7234,
+      "step": 242
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 8.18000614398682e-06,
+      "loss": 0.6374,
+      "step": 243
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 8.099014576919607e-06,
+      "loss": 0.5869,
+      "step": 244
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 8.018191684089893e-06,
+      "loss": 0.4982,
+      "step": 245
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 7.937542164845596e-06,
+      "loss": 0.615,
+      "step": 246
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 7.857070708454048e-06,
+      "loss": 0.9688,
+      "step": 247
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 7.776781993829328e-06,
+      "loss": 0.865,
+      "step": 248
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 7.696680689260222e-06,
+      "loss": 0.7174,
+      "step": 249
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 7.616771452138793e-06,
+      "loss": 0.6979,
+      "step": 250
+    },
+    {
+      "epoch": 3.01,
+      "eval_loss": 0.6813020706176758,
+      "eval_runtime": 46.0472,
+      "eval_samples_per_second": 18.872,
+      "eval_steps_per_second": 0.304,
+      "step": 250
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 7.537058928689568e-06,
+      "loss": 0.6459,
+      "step": 251
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 7.457547753699405e-06,
+      "loss": 0.5596,
+      "step": 252
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 7.378242550248006e-06,
+      "loss": 0.5273,
+      "step": 253
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 7.2991479294391e-06,
+      "loss": 0.6464,
+      "step": 254
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 7.220268490132359e-06,
+      "loss": 0.8977,
+      "step": 255
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 7.141608818675983e-06,
+      "loss": 0.962,
+      "step": 256
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 7.063173488640043e-06,
+      "loss": 0.5978,
+      "step": 257
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 6.98496706055055e-06,
+      "loss": 0.6181,
+      "step": 258
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 6.906994081624296e-06,
+      "loss": 0.5011,
+      "step": 259
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 6.829259085504458e-06,
+      "loss": 0.5811,
+      "step": 260
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 6.751766591996991e-06,
+      "loss": 0.7713,
+      "step": 261
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 6.674521106807838e-06,
+      "loss": 1.1051,
+      "step": 262
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 6.597527121280941e-06,
+      "loss": 0.6117,
+      "step": 263
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 6.52078911213711e-06,
+      "loss": 0.5819,
+      "step": 264
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 6.4443115412137146e-06,
+      "loss": 0.5601,
+      "step": 265
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 6.36809885520527e-06,
+      "loss": 0.547,
+      "step": 266
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 6.292155485404879e-06,
+      "loss": 0.81,
+      "step": 267
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 6.216485847446581e-06,
+      "loss": 1.1568,
+      "step": 268
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 6.14109434104862e-06,
+      "loss": 0.5418,
+      "step": 269
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 6.0659853497576245e-06,
+      "loss": 0.6514,
+      "step": 270
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 5.991163240693715e-06,
+      "loss": 0.538,
+      "step": 271
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 5.916632364296616e-06,
+      "loss": 0.4997,
+      "step": 272
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 5.842397054072663e-06,
+      "loss": 0.6937,
+      "step": 273
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 5.768461626342883e-06,
+      "loss": 1.0214,
+      "step": 274
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 5.694830379991981e-06,
+      "loss": 0.711,
+      "step": 275
+    },
+    {
+      "epoch": 3.31,
+      "eval_loss": 0.6800615787506104,
+      "eval_runtime": 46.0873,
+      "eval_samples_per_second": 18.856,
+      "eval_steps_per_second": 0.304,
+      "step": 275
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 5.621507596218431e-06,
+      "loss": 0.6605,
+      "step": 276
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 5.548497538285508e-06,
+      "loss": 0.5435,
+      "step": 277
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 5.475804451273445e-06,
+      "loss": 0.4961,
+      "step": 278
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 5.403432561832573e-06,
+      "loss": 0.61,
+      "step": 279
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 5.331386077937598e-06,
+      "loss": 1.0412,
+      "step": 280
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 5.259669188642919e-06,
+      "loss": 0.9235,
+      "step": 281
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 5.188286063839047e-06,
+      "loss": 0.6159,
+      "step": 282
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 5.117240854010184e-06,
+      "loss": 0.6207,
+      "step": 283
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 5.046537689992867e-06,
+      "loss": 0.5115,
+      "step": 284
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 4.976180682735812e-06,
+      "loss": 0.5823,
+      "step": 285
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 4.906173923060861e-06,
+      "loss": 0.8708,
+      "step": 286
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 4.836521481425152e-06,
+      "loss": 1.0297,
+      "step": 287
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 4.7672274076844245e-06,
+      "loss": 0.5713,
+      "step": 288
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 4.698295730857569e-06,
+      "loss": 0.616,
+      "step": 289
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 4.629730458892339e-06,
+      "loss": 0.5142,
+      "step": 290
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 4.561535578432336e-06,
+      "loss": 0.5395,
+      "step": 291
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 4.493715054585204e-06,
+      "loss": 0.7795,
+      "step": 292
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 4.4262728306920685e-06,
+      "loss": 1.1353,
+      "step": 293
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 4.359212828098284e-06,
+      "loss": 0.6227,
+      "step": 294
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 4.292538945925402e-06,
+      "loss": 0.6017,
+      "step": 295
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 4.2262550608444916e-06,
+      "loss": 0.5493,
+      "step": 296
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 4.1603650268507e-06,
+      "loss": 0.4872,
+      "step": 297
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 4.094872675039208e-06,
+      "loss": 0.711,
+      "step": 298
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 4.029781813382433e-06,
+      "loss": 1.1265,
+      "step": 299
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 3.96509622650866e-06,
+      "loss": 0.7292,
+      "step": 300
+    },
+    {
+      "epoch": 3.61,
+      "eval_loss": 0.6790695190429688,
+      "eval_runtime": 46.0896,
+      "eval_samples_per_second": 18.855,
+      "eval_steps_per_second": 0.304,
+      "step": 300
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 3.90081967548195e-06,
+      "loss": 0.6189,
+      "step": 301
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 3.836955897583491e-06,
+      "loss": 0.5988,
+      "step": 302
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 3.7735086060942807e-06,
+      "loss": 0.504,
+      "step": 303
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 3.7104814900792186e-06,
+      "loss": 0.6354,
+      "step": 304
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 3.6478782141726272e-06,
+      "loss": 0.9963,
+      "step": 305
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 3.585702418365156e-06,
+      "loss": 0.9168,
+      "step": 306
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 3.5239577177921574e-06,
+      "loss": 0.5816,
+      "step": 307
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 3.462647702523473e-06,
+      "loss": 0.5856,
+      "step": 308
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 3.401775937354709e-06,
+      "loss": 0.5252,
+      "step": 309
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 3.3413459615999465e-06,
+      "loss": 0.6033,
+      "step": 310
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 3.2813612888859768e-06,
+      "loss": 0.7483,
+      "step": 311
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 3.221825406947978e-06,
+      "loss": 1.1016,
+      "step": 312
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 3.162741777426749e-06,
+      "loss": 0.6132,
+      "step": 313
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 3.1041138356674276e-06,
+      "loss": 0.5922,
+      "step": 314
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 3.0459449905197367e-06,
+      "loss": 0.5414,
+      "step": 315
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 2.988238624139802e-06,
+      "loss": 0.525,
+      "step": 316
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 2.930998091793475e-06,
+      "loss": 0.7916,
+      "step": 317
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 2.874226721661275e-06,
+      "loss": 1.0491,
+      "step": 318
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 2.817927814644844e-06,
+      "loss": 0.6115,
+      "step": 319
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 2.762104644175049e-06,
+      "loss": 0.6208,
+      "step": 320
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 2.7067604560216305e-06,
+      "loss": 0.5477,
+      "step": 321
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 2.651898468104497e-06,
+      "loss": 0.5065,
+      "step": 322
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 2.5975218703066078e-06,
+      "loss": 0.7619,
+      "step": 323
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 2.5436338242885146e-06,
+      "loss": 1.0427,
+      "step": 324
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 2.490237463304532e-06,
+      "loss": 0.7504,
+      "step": 325
+    },
+    {
+      "epoch": 3.92,
+      "eval_loss": 0.6780019998550415,
+      "eval_runtime": 46.0417,
+      "eval_samples_per_second": 18.874,
+      "eval_steps_per_second": 0.304,
+      "step": 325
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 2.437335892020534e-06,
+      "loss": 0.6214,
+      "step": 326
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 2.3849321863334734e-06,
+      "loss": 0.5806,
+      "step": 327
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 2.333029393192503e-06,
+      "loss": 0.5048,
+      "step": 328
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 2.281630530421846e-06,
+      "loss": 0.6738,
+      "step": 329
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 2.230738586545296e-06,
+      "loss": 0.9792,
+      "step": 330
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 2.1803565206124854e-06,
+      "loss": 0.7933,
+      "step": 331
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 2.130487262026805e-06,
+      "loss": 0.7724,
+      "step": 332
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 2.081133710375105e-06,
+      "loss": 0.7386,
+      "step": 333
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 2.032298735259076e-06,
+      "loss": 0.644,
+      "step": 334
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 1.9839851761284216e-06,
+      "loss": 0.5484,
+      "step": 335
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1.9361958421157527e-06,
+      "loss": 0.495,
+      "step": 336
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 1.8889335118732466e-06,
+      "loss": 0.6507,
+      "step": 337
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 1.8422009334111009e-06,
+      "loss": 0.9705,
+      "step": 338
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 1.796000823937737e-06,
+      "loss": 0.8893,
+      "step": 339
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 1.7503358697018337e-06,
+      "loss": 0.5689,
+      "step": 340
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 1.7052087258361115e-06,
+      "loss": 0.5853,
+      "step": 341
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1.6606220162029773e-06,
+      "loss": 0.5154,
+      "step": 342
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 1.6165783332419437e-06,
+      "loss": 0.5699,
+      "step": 343
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 1.5730802378189136e-06,
+      "loss": 0.8551,
+      "step": 344
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 1.5301302590772631e-06,
+      "loss": 1.0386,
+      "step": 345
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1.4877308942908004e-06,
+      "loss": 0.5727,
+      "step": 346
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 1.445884608718565e-06,
+      "loss": 0.6391,
+      "step": 347
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 1.4045938354614744e-06,
+      "loss": 0.5394,
+      "step": 348
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 1.3638609753208755e-06,
+      "loss": 0.5419,
+      "step": 349
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 1.3236883966589302e-06,
+      "loss": 0.8315,
+      "step": 350
+    },
+    {
+      "epoch": 4.22,
+      "eval_loss": 0.6777500510215759,
+      "eval_runtime": 45.8022,
+      "eval_samples_per_second": 18.973,
+      "eval_steps_per_second": 0.306,
+      "step": 350
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 415,
+  "num_train_epochs": 5,
+  "save_steps": 25,
+  "total_flos": 7.911012149353251e+18,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/checkpoint-350/training_args.bin b/checkpoint-350/training_args.bin
new file mode 100644
index 0000000..c951a58
--- /dev/null
+++ b/checkpoint-350/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c98be9e90d62a814f5e99d55200fc54cf1e034ee18489beefd52468c142448bf
+size 6907
diff --git a/checkpoint-350/zero_to_fp32.py b/checkpoint-350/zero_to_fp32.py
new file mode 100644
index 0000000..c98caae
--- /dev/null
+++ b/checkpoint-350/zero_to_fp32.py
@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)
diff --git a/checkpoint-375/config.json b/checkpoint-375/config.json
new file mode 100644
index 0000000..87e67dc
--- /dev/null
+++ b/checkpoint-375/config.json
@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "/workspace/llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.33.1",
+  "use_cache": false,
+  "vocab_size": 32000
+}
diff --git a/checkpoint-375/generation_config.json b/checkpoint-375/generation_config.json
new file mode 100644
index 0000000..0b71b22
--- /dev/null
+++ b/checkpoint-375/generation_config.json
@@ -0,0 +1,4 @@
+{
+  "max_new_tokens": 4096,
+  "transformers_version": "4.33.1"
+}
diff --git a/checkpoint-375/latest b/checkpoint-375/latest
new file mode 100644
index 0000000..2e9ab08
--- /dev/null
+++ b/checkpoint-375/latest
@@ -0,0 +1 @@
+global_step375
\ No newline at end of file
diff --git a/checkpoint-375/pytorch_model-00001-of-00003.bin b/checkpoint-375/pytorch_model-00001-of-00003.bin
new file mode 100644
index 0000000..d7af461
--- /dev/null
+++ b/checkpoint-375/pytorch_model-00001-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b52898b18929f9d849e8fdafbfdbcb4655c90569abf4a425221d4e0aaae2cf1
+size 9948719433
diff --git a/checkpoint-375/pytorch_model-00002-of-00003.bin b/checkpoint-375/pytorch_model-00002-of-00003.bin
new file mode 100644
index 0000000..8249d43
--- /dev/null
+++ b/checkpoint-375/pytorch_model-00002-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed5a78f77a56d67668cc7ccfaf9218b358c038c0ca096cbc073956c11bdaf5e7
+size 9904155408
diff --git a/checkpoint-375/pytorch_model-00003-of-00003.bin b/checkpoint-375/pytorch_model-00003-of-00003.bin
new file mode 100644
index 0000000..7bcc936
--- /dev/null
+++ b/checkpoint-375/pytorch_model-00003-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:447de333c219da8c5932c56f1db5431c3cf2b67eaec59cd74c1a822a641a495b
+size 6178978207
diff --git a/checkpoint-375/pytorch_model.bin.index.json b/checkpoint-375/pytorch_model.bin.index.json
new file mode 100644
index 0000000..5ebaea3
--- /dev/null
+++ b/checkpoint-375/pytorch_model.bin.index.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a81b85fdd21a6ffc94396229e2cac8f8135c4a542730e957ede126202c6bf475
+size 29894
diff --git a/checkpoint-375/special_tokens_map.json b/checkpoint-375/special_tokens_map.json
new file mode 100644
index 0000000..14761dc
--- /dev/null
+++ b/checkpoint-375/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}
diff --git a/checkpoint-375/tokenizer.model b/checkpoint-375/tokenizer.model
new file mode 100644
index 0000000..6c00c74
--- /dev/null
+++ b/checkpoint-375/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/checkpoint-375/tokenizer_config.json b/checkpoint-375/tokenizer_config.json
new file mode 100644
index 0000000..16ada5c
--- /dev/null
+++ b/checkpoint-375/tokenizer_config.json
@@ -0,0 +1,37 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_default_system_prompt": true
+}
diff --git a/checkpoint-375/trainer_state.json b/checkpoint-375/trainer_state.json
new file mode 100644
index 0000000..14172bf
--- /dev/null
+++ b/checkpoint-375/trainer_state.json
@@ -0,0 +1,2389 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.518072289156627,
+  "eval_steps": 25,
+  "global_step": 375,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 7.333333333333333e-06,
+      "loss": 0.9353,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 1.4666666666666666e-05,
+      "loss": 0.8037,
+      "step": 2
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 2.2e-05,
+      "loss": 0.7535,
+      "step": 3
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 2.1999680209249292e-05,
+      "loss": 0.6836,
+      "step": 4
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 2.1998720855591014e-05,
+      "loss": 0.8619,
+      "step": 5
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 2.1997121994805608e-05,
+      "loss": 1.3258,
+      "step": 6
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 2.1994883719856868e-05,
+      "loss": 1.1205,
+      "step": 7
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 2.199200616088654e-05,
+      "loss": 0.7097,
+      "step": 8
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 2.1988489485206775e-05,
+      "loss": 0.6175,
+      "step": 9
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 2.198433389729035e-05,
+      "loss": 0.6068,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 2.1979539638758833e-05,
+      "loss": 0.713,
+      "step": 11
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 2.19741069883685e-05,
+      "loss": 1.0415,
+      "step": 12
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 2.1968036261994137e-05,
+      "loss": 1.2173,
+      "step": 13
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 2.1961327812610683e-05,
+      "loss": 0.6987,
+      "step": 14
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 2.19539820302727e-05,
+      "loss": 0.6693,
+      "step": 15
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 2.1945999342091685e-05,
+      "loss": 0.6082,
+      "step": 16
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 2.1937380212211243e-05,
+      "loss": 0.6565,
+      "step": 17
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 2.1928125141780108e-05,
+      "loss": 0.9413,
+      "step": 18
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 2.191823466892299e-05,
+      "loss": 1.3279,
+      "step": 19
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 2.1907709368709293e-05,
+      "loss": 0.7384,
+      "step": 20
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.189654985311969e-05,
+      "loss": 0.7139,
+      "step": 21
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.1884756771010514e-05,
+      "loss": 0.5926,
+      "step": 22
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.1872330808076058e-05,
+      "loss": 0.5966,
+      "step": 23
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.1859272686808682e-05,
+      "loss": 0.843,
+      "step": 24
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.184558316645683e-05,
+      "loss": 1.2463,
+      "step": 25
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.7440170645713806,
+      "eval_runtime": 49.2892,
+      "eval_samples_per_second": 17.631,
+      "eval_steps_per_second": 0.284,
+      "step": 25
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.1831263042980858e-05,
+      "loss": 0.8548,
+      "step": 26
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.1816313149006775e-05,
+      "loss": 0.6954,
+      "step": 27
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.180073435377782e-05,
+      "loss": 0.666,
+      "step": 28
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.1784527563103927e-05,
+      "loss": 0.6027,
+      "step": 29
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.1767693719309048e-05,
+      "loss": 0.6978,
+      "step": 30
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.175023380117637e-05,
+      "loss": 0.9855,
+      "step": 31
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.1732148823891417e-05,
+      "loss": 1.0321,
+      "step": 32
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.171343983898299e-05,
+      "loss": 0.6891,
+      "step": 33
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.1694107934262066e-05,
+      "loss": 0.6134,
+      "step": 34
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.1674154233758516e-05,
+      "loss": 0.5408,
+      "step": 35
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.165357989765577e-05,
+      "loss": 0.6853,
+      "step": 36
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1632386122223363e-05,
+      "loss": 0.9126,
+      "step": 37
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1610574139747343e-05,
+      "loss": 1.2144,
+      "step": 38
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1588145218458675e-05,
+      "loss": 0.6852,
+      "step": 39
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.156510066245946e-05,
+      "loss": 0.6938,
+      "step": 40
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.1541441811647135e-05,
+      "loss": 0.599,
+      "step": 41
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.1517170041636537e-05,
+      "loss": 0.6201,
+      "step": 42
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.1492286763679958e-05,
+      "loss": 0.8481,
+      "step": 43
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.1466793424585053e-05,
+      "loss": 1.3248,
+      "step": 44
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.1440691506630743e-05,
+      "loss": 0.6589,
+      "step": 45
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.1413982527481012e-05,
+      "loss": 0.6731,
+      "step": 46
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.1386668040096685e-05,
+      "loss": 0.6028,
+      "step": 47
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.1358749632645103e-05,
+      "loss": 0.5619,
+      "step": 48
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 2.1330228928407817e-05,
+      "loss": 0.7259,
+      "step": 49
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 2.1301107585686164e-05,
+      "loss": 1.2589,
+      "step": 50
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.7223495841026306,
+      "eval_runtime": 45.8793,
+      "eval_samples_per_second": 18.941,
+      "eval_steps_per_second": 0.305,
+      "step": 50
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 2.127138729770489e-05,
+      "loss": 0.8173,
+      "step": 51
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 2.1241069792513666e-05,
+      "loss": 0.6887,
+      "step": 52
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 2.1210156832886625e-05,
+      "loss": 0.6537,
+      "step": 53
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 2.1178650216219872e-05,
+      "loss": 0.5452,
+      "step": 54
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 2.114655177442697e-05,
+      "loss": 0.6558,
+      "step": 55
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 2.111386337383243e-05,
+      "loss": 1.0381,
+      "step": 56
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 2.1080586915063186e-05,
+      "loss": 0.99,
+      "step": 57
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.1046724332938103e-05,
+      "loss": 0.657,
+      "step": 58
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.101227759635546e-05,
+      "loss": 0.6018,
+      "step": 59
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.0977248708178485e-05,
+      "loss": 0.5436,
+      "step": 60
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.09416397051189e-05,
+      "loss": 0.6401,
+      "step": 61
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.0905452657618488e-05,
+      "loss": 0.9314,
+      "step": 62
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.0868689669728716e-05,
+      "loss": 1.2026,
+      "step": 63
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 2.0831352878988404e-05,
+      "loss": 0.6308,
+      "step": 64
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 2.0793444456299434e-05,
+      "loss": 0.6889,
+      "step": 65
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 2.0754966605800523e-05,
+      "loss": 0.5788,
+      "step": 66
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 2.071592156473907e-05,
+      "loss": 0.5568,
+      "step": 67
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 2.067631160334107e-05,
+      "loss": 0.8306,
+      "step": 68
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 2.0636139024679125e-05,
+      "loss": 1.2907,
+      "step": 69
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 2.0595406164538524e-05,
+      "loss": 0.6435,
+      "step": 70
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.0554115391281437e-05,
+      "loss": 0.6731,
+      "step": 71
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.05122691057092e-05,
+      "loss": 0.5612,
+      "step": 72
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 2.046986974092274e-05,
+      "loss": 0.548,
+      "step": 73
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 2.042691976218109e-05,
+      "loss": 0.7796,
+      "step": 74
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 2.0383421666758057e-05,
+      "loss": 1.2264,
+      "step": 75
+    },
+    {
+      "epoch": 0.9,
+      "eval_loss": 0.7080472707748413,
+      "eval_runtime": 45.869,
+      "eval_samples_per_second": 18.945,
+      "eval_steps_per_second": 0.305,
+      "step": 75
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 2.0339377983797022e-05,
+      "loss": 0.8322,
+      "step": 76
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 2.029479127416389e-05,
+      "loss": 0.7071,
+      "step": 77
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.0249664130298165e-05,
+      "loss": 0.6058,
+      "step": 78
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.020399917606226e-05,
+      "loss": 0.5243,
+      "step": 79
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.01577990665889e-05,
+      "loss": 0.6636,
+      "step": 80
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 2.0111066488126753e-05,
+      "loss": 1.0678,
+      "step": 81
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.0063804157884248e-05,
+      "loss": 0.853,
+      "step": 82
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.0016014823871578e-05,
+      "loss": 0.8473,
+      "step": 83
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.9967701264740925e-05,
+      "loss": 0.7626,
+      "step": 84
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.99188662896249e-05,
+      "loss": 0.6879,
+      "step": 85
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.9869512737973196e-05,
+      "loss": 0.5818,
+      "step": 86
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.9819643479387516e-05,
+      "loss": 0.5151,
+      "step": 87
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.9769261413454705e-05,
+      "loss": 0.6649,
+      "step": 88
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.971836946957816e-05,
+      "loss": 1.0307,
+      "step": 89
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.96669706068075e-05,
+      "loss": 1.0031,
+      "step": 90
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.961506781366653e-05,
+      "loss": 0.67,
+      "step": 91
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.9562664107979465e-05,
+      "loss": 0.6331,
+      "step": 92
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.950976253669547e-05,
+      "loss": 0.5232,
+      "step": 93
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.9456366175711482e-05,
+      "loss": 0.6291,
+      "step": 94
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.9402478129693395e-05,
+      "loss": 0.8978,
+      "step": 95
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.9348101531895503e-05,
+      "loss": 1.1345,
+      "step": 96
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.929323954397837e-05,
+      "loss": 0.6327,
+      "step": 97
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.9237895355824952e-05,
+      "loss": 0.6239,
+      "step": 98
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.9182072185355157e-05,
+      "loss": 0.5745,
+      "step": 99
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.9125773278338728e-05,
+      "loss": 0.5543,
+      "step": 100
+    },
+    {
+      "epoch": 1.2,
+      "eval_loss": 0.7009603977203369,
+      "eval_runtime": 46.0556,
+      "eval_samples_per_second": 18.868,
+      "eval_steps_per_second": 0.304,
+      "step": 100
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.9069001908206527e-05,
+      "loss": 0.8478,
+      "step": 101
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.90117613758602e-05,
+      "loss": 1.2408,
+      "step": 102
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.8954055009480263e-05,
+      "loss": 0.5991,
+      "step": 103
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.8895886164332572e-05,
+      "loss": 0.6304,
+      "step": 104
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.883725822257325e-05,
+      "loss": 0.5863,
+      "step": 105
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.877817459305202e-05,
+      "loss": 0.5341,
+      "step": 106
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.8718638711114027e-05,
+      "loss": 0.6937,
+      "step": 107
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.865865403840005e-05,
+      "loss": 1.12,
+      "step": 108
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.8598224062645293e-05,
+      "loss": 0.8003,
+      "step": 109
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.8537352297476526e-05,
+      "loss": 0.6984,
+      "step": 110
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.8476042282207843e-05,
+      "loss": 0.5954,
+      "step": 111
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.8414297581634844e-05,
+      "loss": 0.5379,
+      "step": 112
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.8352121785827374e-05,
+      "loss": 0.6839,
+      "step": 113
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.828951850992078e-05,
+      "loss": 0.9745,
+      "step": 114
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.822649139390572e-05,
+      "loss": 0.9575,
+      "step": 115
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.8163044102416508e-05,
+      "loss": 0.647,
+      "step": 116
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.809918032451805e-05,
+      "loss": 0.5792,
+      "step": 117
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.8034903773491342e-05,
+      "loss": 0.5287,
+      "step": 118
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.797021818661757e-05,
+      "loss": 0.6415,
+      "step": 119
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.790512732496079e-05,
+      "loss": 0.8963,
+      "step": 120
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.78396349731493e-05,
+      "loss": 1.1426,
+      "step": 121
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.7773744939155508e-05,
+      "loss": 0.6213,
+      "step": 122
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.77074610540746e-05,
+      "loss": 0.6494,
+      "step": 123
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.7640787171901715e-05,
+      "loss": 0.5484,
+      "step": 124
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.7573727169307932e-05,
+      "loss": 0.5539,
+      "step": 125
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.6961802244186401,
+      "eval_runtime": 45.816,
+      "eval_samples_per_second": 18.967,
+      "eval_steps_per_second": 0.306,
+      "step": 125
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.7506284945414796e-05,
+      "loss": 0.8497,
+      "step": 126
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.743846442156766e-05,
+      "loss": 1.1775,
+      "step": 127
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.7370269541107663e-05,
+      "loss": 0.6627,
+      "step": 128
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1.7301704269142434e-05,
+      "loss": 0.6425,
+      "step": 129
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 1.7232772592315577e-05,
+      "loss": 0.5537,
+      "step": 130
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 1.716347851857485e-05,
+      "loss": 0.5311,
+      "step": 131
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 1.7093826076939137e-05,
+      "loss": 0.7583,
+      "step": 132
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 1.7023819317264186e-05,
+      "loss": 1.1448,
+      "step": 133
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1.695346231000713e-05,
+      "loss": 0.7377,
+      "step": 134
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1.6882759145989816e-05,
+      "loss": 0.6445,
+      "step": 135
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 1.681171393616095e-05,
+      "loss": 0.5787,
+      "step": 136
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.674033081135708e-05,
+      "loss": 0.5084,
+      "step": 137
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.6668613922062398e-05,
+      "loss": 0.6423,
+      "step": 138
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 1.6596567438167426e-05,
+      "loss": 0.9592,
+      "step": 139
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1.6524195548726557e-05,
+      "loss": 1.0,
+      "step": 140
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1.645150246171449e-05,
+      "loss": 0.6274,
+      "step": 141
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1.637849240378157e-05,
+      "loss": 0.597,
+      "step": 142
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1.630516962000802e-05,
+      "loss": 0.524,
+      "step": 143
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1.623153837365712e-05,
+      "loss": 0.5914,
+      "step": 144
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.6157602945927335e-05,
+      "loss": 0.8503,
+      "step": 145
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 1.6083367635703386e-05,
+      "loss": 1.058,
+      "step": 146
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1.600883675930628e-05,
+      "loss": 0.6366,
+      "step": 147
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.593401465024237e-05,
+      "loss": 0.6191,
+      "step": 148
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 1.5858905658951377e-05,
+      "loss": 0.5433,
+      "step": 149
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1.578351415255342e-05,
+      "loss": 0.5568,
+      "step": 150
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.6907634735107422,
+      "eval_runtime": 45.8388,
+      "eval_samples_per_second": 18.958,
+      "eval_steps_per_second": 0.305,
+      "step": 150
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1.570784451459512e-05,
+      "loss": 0.8225,
+      "step": 151
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 1.5631901144794727e-05,
+      "loss": 1.1446,
+      "step": 152
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 1.5555688458786283e-05,
+      "loss": 0.6378,
+      "step": 153
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 1.547921088786289e-05,
+      "loss": 0.6675,
+      "step": 154
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1.5402472878719058e-05,
+      "loss": 0.5395,
+      "step": 155
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1.532547889319216e-05,
+      "loss": 0.5378,
+      "step": 156
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1.524823340800301e-05,
+      "loss": 0.7644,
+      "step": 157
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1.5170740914495543e-05,
+      "loss": 1.1105,
+      "step": 158
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 1.5093005918375706e-05,
+      "loss": 0.7881,
+      "step": 159
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1.5015032939449451e-05,
+      "loss": 0.6366,
+      "step": 160
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 1.493682651135996e-05,
+      "loss": 0.5853,
+      "step": 161
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 1.485839118132402e-05,
+      "loss": 0.5138,
+      "step": 162
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1.4779731509867645e-05,
+      "loss": 0.6444,
+      "step": 163
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1.4700852070560903e-05,
+      "loss": 0.9914,
+      "step": 164
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.4621757449752e-05,
+      "loss": 0.8894,
+      "step": 165
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 1.4542452246300597e-05,
+      "loss": 0.8105,
+      "step": 166
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 1.4462941071310434e-05,
+      "loss": 0.7536,
+      "step": 167
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 1.438322854786121e-05,
+      "loss": 0.6122,
+      "step": 168
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 1.4303319310739777e-05,
+      "loss": 0.575,
+      "step": 169
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 1.4223218006170675e-05,
+      "loss": 0.5015,
+      "step": 170
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 1.4142929291545953e-05,
+      "loss": 0.6682,
+      "step": 171
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 1.4062457835154405e-05,
+      "loss": 1.0041,
+      "step": 172
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1.3981808315910112e-05,
+      "loss": 0.9622,
+      "step": 173
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 1.3900985423080396e-05,
+      "loss": 0.6121,
+      "step": 174
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 1.3819993856013184e-05,
+      "loss": 0.6222,
+      "step": 175
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.6892703175544739,
+      "eval_runtime": 45.9318,
+      "eval_samples_per_second": 18.919,
+      "eval_steps_per_second": 0.305,
+      "step": 175
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1.3738838323863754e-05,
+      "loss": 0.5335,
+      "step": 176
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 1.3657523545320932e-05,
+      "loss": 0.5978,
+      "step": 177
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 1.3576054248332733e-05,
+      "loss": 0.9021,
+      "step": 178
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 1.3494435169831453e-05,
+      "loss": 1.0881,
+      "step": 179
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 1.3412671055458253e-05,
+      "loss": 0.6492,
+      "step": 180
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 1.3330766659287234e-05,
+      "loss": 0.6129,
+      "step": 181
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 1.3248726743548996e-05,
+      "loss": 0.5737,
+      "step": 182
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 1.3166556078353768e-05,
+      "loss": 0.5407,
+      "step": 183
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 1.3084259441414036e-05,
+      "loss": 0.7442,
+      "step": 184
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1.3001841617766777e-05,
+      "loss": 1.0671,
+      "step": 185
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 1.2919307399495196e-05,
+      "loss": 0.6467,
+      "step": 186
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 1.2836661585450144e-05,
+      "loss": 0.6384,
+      "step": 187
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1.275390898097105e-05,
+      "loss": 0.5394,
+      "step": 188
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 1.267105439760656e-05,
+      "loss": 0.5238,
+      "step": 189
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 1.2588102652834745e-05,
+      "loss": 0.7205,
+      "step": 190
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 1.2505058569783012e-05,
+      "loss": 1.0399,
+      "step": 191
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 1.2421926976947664e-05,
+      "loss": 0.7659,
+      "step": 192
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 1.2338712707913146e-05,
+      "loss": 0.6225,
+      "step": 193
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 1.2255420601071026e-05,
+      "loss": 0.5688,
+      "step": 194
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1.2172055499338637e-05,
+      "loss": 0.5092,
+      "step": 195
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 1.2088622249877521e-05,
+      "loss": 0.623,
+      "step": 196
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 1.200512570381158e-05,
+      "loss": 0.9878,
+      "step": 197
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 1.192157071594503e-05,
+      "loss": 0.9355,
+      "step": 198
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 1.1837962144480099e-05,
+      "loss": 0.666,
+      "step": 199
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.1754304850734577e-05,
+      "loss": 0.5828,
+      "step": 200
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.6850700974464417,
+      "eval_runtime": 45.894,
+      "eval_samples_per_second": 18.935,
+      "eval_steps_per_second": 0.305,
+      "step": 200
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.167060369885915e-05,
+      "loss": 0.4919,
+      "step": 201
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1.1586863555554581e-05,
+      "loss": 0.5812,
+      "step": 202
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.1503089289788741e-05,
+      "loss": 0.8558,
+      "step": 203
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.1419285772513506e-05,
+      "loss": 1.1615,
+      "step": 204
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1.1335457876381556e-05,
+      "loss": 0.5879,
+      "step": 205
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.1251610475463037e-05,
+      "loss": 0.6408,
+      "step": 206
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.1167748444962185e-05,
+      "loss": 0.5349,
+      "step": 207
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.1083876660933852e-05,
+      "loss": 0.5493,
+      "step": 208
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.1e-05,
+      "loss": 0.8208,
+      "step": 209
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.0916123339066147e-05,
+      "loss": 1.0668,
+      "step": 210
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.0832251555037816e-05,
+      "loss": 0.6051,
+      "step": 211
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.0748389524536964e-05,
+      "loss": 0.6071,
+      "step": 212
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.0664542123618447e-05,
+      "loss": 0.5666,
+      "step": 213
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.0580714227486495e-05,
+      "loss": 0.5134,
+      "step": 214
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1.0496910710211261e-05,
+      "loss": 0.7728,
+      "step": 215
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 1.041313644444542e-05,
+      "loss": 1.0725,
+      "step": 216
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1.032939630114085e-05,
+      "loss": 0.744,
+      "step": 217
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1.0245695149265425e-05,
+      "loss": 0.646,
+      "step": 218
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 1.0162037855519904e-05,
+      "loss": 0.5654,
+      "step": 219
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 1.0078429284054971e-05,
+      "loss": 0.5043,
+      "step": 220
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 9.99487429618842e-06,
+      "loss": 0.6779,
+      "step": 221
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 9.91137775012248e-06,
+      "loss": 0.9569,
+      "step": 222
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 9.827944500661365e-06,
+      "loss": 0.966,
+      "step": 223
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 9.744579398928975e-06,
+      "loss": 0.5635,
+      "step": 224
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 9.661287292086853e-06,
+      "loss": 0.6108,
+      "step": 225
+    },
+    {
+      "epoch": 2.71,
+      "eval_loss": 0.6831873655319214,
+      "eval_runtime": 45.9501,
+      "eval_samples_per_second": 18.912,
+      "eval_steps_per_second": 0.305,
+      "step": 225
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 9.578073023052337e-06,
+      "loss": 0.515,
+      "step": 226
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 9.494941430216989e-06,
+      "loss": 0.5935,
+      "step": 227
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 9.411897347165256e-06,
+      "loss": 0.8431,
+      "step": 228
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 9.328945602393441e-06,
+      "loss": 1.1265,
+      "step": 229
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 9.246091019028949e-06,
+      "loss": 0.5872,
+      "step": 230
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 9.163338414549857e-06,
+      "loss": 0.6158,
+      "step": 231
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 9.080692600504803e-06,
+      "loss": 0.5229,
+      "step": 232
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 8.998158382233226e-06,
+      "loss": 0.5407,
+      "step": 233
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 8.915740558585963e-06,
+      "loss": 0.8395,
+      "step": 234
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 8.833443921646237e-06,
+      "loss": 1.162,
+      "step": 235
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 8.751273256451007e-06,
+      "loss": 0.6233,
+      "step": 236
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 8.669233340712766e-06,
+      "loss": 0.6508,
+      "step": 237
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 8.587328944541747e-06,
+      "loss": 0.5542,
+      "step": 238
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.50556483016855e-06,
+      "loss": 0.5076,
+      "step": 239
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.423945751667272e-06,
+      "loss": 0.7069,
+      "step": 240
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 8.342476454679073e-06,
+      "loss": 1.0609,
+      "step": 241
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 8.26116167613625e-06,
+      "loss": 0.7234,
+      "step": 242
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 8.18000614398682e-06,
+      "loss": 0.6374,
+      "step": 243
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 8.099014576919607e-06,
+      "loss": 0.5869,
+      "step": 244
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 8.018191684089893e-06,
+      "loss": 0.4982,
+      "step": 245
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 7.937542164845596e-06,
+      "loss": 0.615,
+      "step": 246
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 7.857070708454048e-06,
+      "loss": 0.9688,
+      "step": 247
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 7.776781993829328e-06,
+      "loss": 0.865,
+      "step": 248
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 7.696680689260222e-06,
+      "loss": 0.7174,
+      "step": 249
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 7.616771452138793e-06,
+      "loss": 0.6979,
+      "step": 250
+    },
+    {
+      "epoch": 3.01,
+      "eval_loss": 0.6813020706176758,
+      "eval_runtime": 46.0472,
+      "eval_samples_per_second": 18.872,
+      "eval_steps_per_second": 0.304,
+      "step": 250
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 7.537058928689568e-06,
+      "loss": 0.6459,
+      "step": 251
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 7.457547753699405e-06,
+      "loss": 0.5596,
+      "step": 252
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 7.378242550248006e-06,
+      "loss": 0.5273,
+      "step": 253
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 7.2991479294391e-06,
+      "loss": 0.6464,
+      "step": 254
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 7.220268490132359e-06,
+      "loss": 0.8977,
+      "step": 255
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 7.141608818675983e-06,
+      "loss": 0.962,
+      "step": 256
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 7.063173488640043e-06,
+      "loss": 0.5978,
+      "step": 257
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 6.98496706055055e-06,
+      "loss": 0.6181,
+      "step": 258
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 6.906994081624296e-06,
+      "loss": 0.5011,
+      "step": 259
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 6.829259085504458e-06,
+      "loss": 0.5811,
+      "step": 260
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 6.751766591996991e-06,
+      "loss": 0.7713,
+      "step": 261
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 6.674521106807838e-06,
+      "loss": 1.1051,
+      "step": 262
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 6.597527121280941e-06,
+      "loss": 0.6117,
+      "step": 263
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 6.52078911213711e-06,
+      "loss": 0.5819,
+      "step": 264
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 6.4443115412137146e-06,
+      "loss": 0.5601,
+      "step": 265
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 6.36809885520527e-06,
+      "loss": 0.547,
+      "step": 266
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 6.292155485404879e-06,
+      "loss": 0.81,
+      "step": 267
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 6.216485847446581e-06,
+      "loss": 1.1568,
+      "step": 268
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 6.14109434104862e-06,
+      "loss": 0.5418,
+      "step": 269
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 6.0659853497576245e-06,
+      "loss": 0.6514,
+      "step": 270
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 5.991163240693715e-06,
+      "loss": 0.538,
+      "step": 271
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 5.916632364296616e-06,
+      "loss": 0.4997,
+      "step": 272
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 5.842397054072663e-06,
+      "loss": 0.6937,
+      "step": 273
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 5.768461626342883e-06,
+      "loss": 1.0214,
+      "step": 274
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 5.694830379991981e-06,
+      "loss": 0.711,
+      "step": 275
+    },
+    {
+      "epoch": 3.31,
+      "eval_loss": 0.6800615787506104,
+      "eval_runtime": 46.0873,
+      "eval_samples_per_second": 18.856,
+      "eval_steps_per_second": 0.304,
+      "step": 275
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 5.621507596218431e-06,
+      "loss": 0.6605,
+      "step": 276
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 5.548497538285508e-06,
+      "loss": 0.5435,
+      "step": 277
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 5.475804451273445e-06,
+      "loss": 0.4961,
+      "step": 278
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 5.403432561832573e-06,
+      "loss": 0.61,
+      "step": 279
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 5.331386077937598e-06,
+      "loss": 1.0412,
+      "step": 280
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 5.259669188642919e-06,
+      "loss": 0.9235,
+      "step": 281
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 5.188286063839047e-06,
+      "loss": 0.6159,
+      "step": 282
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 5.117240854010184e-06,
+      "loss": 0.6207,
+      "step": 283
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 5.046537689992867e-06,
+      "loss": 0.5115,
+      "step": 284
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 4.976180682735812e-06,
+      "loss": 0.5823,
+      "step": 285
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 4.906173923060861e-06,
+      "loss": 0.8708,
+      "step": 286
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 4.836521481425152e-06,
+      "loss": 1.0297,
+      "step": 287
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 4.7672274076844245e-06,
+      "loss": 0.5713,
+      "step": 288
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 4.698295730857569e-06,
+      "loss": 0.616,
+      "step": 289
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 4.629730458892339e-06,
+      "loss": 0.5142,
+      "step": 290
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 4.561535578432336e-06,
+      "loss": 0.5395,
+      "step": 291
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 4.493715054585204e-06,
+      "loss": 0.7795,
+      "step": 292
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 4.4262728306920685e-06,
+      "loss": 1.1353,
+      "step": 293
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 4.359212828098284e-06,
+      "loss": 0.6227,
+      "step": 294
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 4.292538945925402e-06,
+      "loss": 0.6017,
+      "step": 295
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 4.2262550608444916e-06,
+      "loss": 0.5493,
+      "step": 296
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 4.1603650268507e-06,
+      "loss": 0.4872,
+      "step": 297
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 4.094872675039208e-06,
+      "loss": 0.711,
+      "step": 298
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 4.029781813382433e-06,
+      "loss": 1.1265,
+      "step": 299
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 3.96509622650866e-06,
+      "loss": 0.7292,
+      "step": 300
+    },
+    {
+      "epoch": 3.61,
+      "eval_loss": 0.6790695190429688,
+      "eval_runtime": 46.0896,
+      "eval_samples_per_second": 18.855,
+      "eval_steps_per_second": 0.304,
+      "step": 300
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 3.90081967548195e-06,
+      "loss": 0.6189,
+      "step": 301
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 3.836955897583491e-06,
+      "loss": 0.5988,
+      "step": 302
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 3.7735086060942807e-06,
+      "loss": 0.504,
+      "step": 303
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 3.7104814900792186e-06,
+      "loss": 0.6354,
+      "step": 304
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 3.6478782141726272e-06,
+      "loss": 0.9963,
+      "step": 305
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 3.585702418365156e-06,
+      "loss": 0.9168,
+      "step": 306
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 3.5239577177921574e-06,
+      "loss": 0.5816,
+      "step": 307
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 3.462647702523473e-06,
+      "loss": 0.5856,
+      "step": 308
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 3.401775937354709e-06,
+      "loss": 0.5252,
+      "step": 309
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 3.3413459615999465e-06,
+      "loss": 0.6033,
+      "step": 310
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 3.2813612888859768e-06,
+      "loss": 0.7483,
+      "step": 311
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 3.221825406947978e-06,
+      "loss": 1.1016,
+      "step": 312
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 3.162741777426749e-06,
+      "loss": 0.6132,
+      "step": 313
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 3.1041138356674276e-06,
+      "loss": 0.5922,
+      "step": 314
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 3.0459449905197367e-06,
+      "loss": 0.5414,
+      "step": 315
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 2.988238624139802e-06,
+      "loss": 0.525,
+      "step": 316
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 2.930998091793475e-06,
+      "loss": 0.7916,
+      "step": 317
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 2.874226721661275e-06,
+      "loss": 1.0491,
+      "step": 318
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 2.817927814644844e-06,
+      "loss": 0.6115,
+      "step": 319
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 2.762104644175049e-06,
+      "loss": 0.6208,
+      "step": 320
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 2.7067604560216305e-06,
+      "loss": 0.5477,
+      "step": 321
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 2.651898468104497e-06,
+      "loss": 0.5065,
+      "step": 322
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 2.5975218703066078e-06,
+      "loss": 0.7619,
+      "step": 323
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 2.5436338242885146e-06,
+      "loss": 1.0427,
+      "step": 324
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 2.490237463304532e-06,
+      "loss": 0.7504,
+      "step": 325
+    },
+    {
+      "epoch": 3.92,
+      "eval_loss": 0.6780019998550415,
+      "eval_runtime": 46.0417,
+      "eval_samples_per_second": 18.874,
+      "eval_steps_per_second": 0.304,
+      "step": 325
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 2.437335892020534e-06,
+      "loss": 0.6214,
+      "step": 326
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 2.3849321863334734e-06,
+      "loss": 0.5806,
+      "step": 327
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 2.333029393192503e-06,
+      "loss": 0.5048,
+      "step": 328
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 2.281630530421846e-06,
+      "loss": 0.6738,
+      "step": 329
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 2.230738586545296e-06,
+      "loss": 0.9792,
+      "step": 330
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 2.1803565206124854e-06,
+      "loss": 0.7933,
+      "step": 331
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 2.130487262026805e-06,
+      "loss": 0.7724,
+      "step": 332
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 2.081133710375105e-06,
+      "loss": 0.7386,
+      "step": 333
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 2.032298735259076e-06,
+      "loss": 0.644,
+      "step": 334
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 1.9839851761284216e-06,
+      "loss": 0.5484,
+      "step": 335
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1.9361958421157527e-06,
+      "loss": 0.495,
+      "step": 336
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 1.8889335118732466e-06,
+      "loss": 0.6507,
+      "step": 337
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 1.8422009334111009e-06,
+      "loss": 0.9705,
+      "step": 338
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 1.796000823937737e-06,
+      "loss": 0.8893,
+      "step": 339
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 1.7503358697018337e-06,
+      "loss": 0.5689,
+      "step": 340
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 1.7052087258361115e-06,
+      "loss": 0.5853,
+      "step": 341
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1.6606220162029773e-06,
+      "loss": 0.5154,
+      "step": 342
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 1.6165783332419437e-06,
+      "loss": 0.5699,
+      "step": 343
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 1.5730802378189136e-06,
+      "loss": 0.8551,
+      "step": 344
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 1.5301302590772631e-06,
+      "loss": 1.0386,
+      "step": 345
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1.4877308942908004e-06,
+      "loss": 0.5727,
+      "step": 346
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 1.445884608718565e-06,
+      "loss": 0.6391,
+      "step": 347
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 1.4045938354614744e-06,
+      "loss": 0.5394,
+      "step": 348
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 1.3638609753208755e-06,
+      "loss": 0.5419,
+      "step": 349
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 1.3236883966589302e-06,
+      "loss": 0.8315,
+      "step": 350
+    },
+    {
+      "epoch": 4.22,
+      "eval_loss": 0.6777500510215759,
+      "eval_runtime": 45.8022,
+      "eval_samples_per_second": 18.973,
+      "eval_steps_per_second": 0.306,
+      "step": 350
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 1.2840784352609335e-06,
+      "loss": 1.1398,
+      "step": 351
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 1.2450333941994781e-06,
+      "loss": 0.6095,
+      "step": 352
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 1.2065555437005672e-06,
+      "loss": 0.6423,
+      "step": 353
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 1.1686471210115957e-06,
+      "loss": 0.5307,
+      "step": 354
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 1.1313103302712859e-06,
+      "loss": 0.5394,
+      "step": 355
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 1.0945473423815139e-06,
+      "loss": 0.7663,
+      "step": 356
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 1.0583602948811003e-06,
+      "loss": 1.0817,
+      "step": 357
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 1.0227512918215148e-06,
+      "loss": 0.7287,
+      "step": 358
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 9.877224036445412e-07,
+      "loss": 0.6111,
+      "step": 359
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 9.532756670618988e-07,
+      "loss": 0.5945,
+      "step": 360
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 9.194130849368133e-07,
+      "loss": 0.498,
+      "step": 361
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 8.861366261675716e-07,
+      "loss": 0.6097,
+      "step": 362
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 8.534482255730291e-07,
+      "loss": 0.9819,
+      "step": 363
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 8.213497837801267e-07,
+      "loss": 0.9455,
+      "step": 364
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 7.898431671133734e-07,
+      "loss": 0.593,
+      "step": 365
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 7.589302074863344e-07,
+      "loss": 0.5929,
+      "step": 366
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 7.2861270229511e-07,
+      "loss": 0.5,
+      "step": 367
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 6.988924143138361e-07,
+      "loss": 0.6316,
+      "step": 368
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 6.697710715921853e-07,
+      "loss": 0.862,
+      "step": 369
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 6.412503673548962e-07,
+      "loss": 1.1132,
+      "step": 370
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 6.133319599033151e-07,
+      "loss": 0.5833,
+      "step": 371
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 5.860174725189841e-07,
+      "loss": 0.591,
+      "step": 372
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 5.593084933692568e-07,
+      "loss": 0.5223,
+      "step": 373
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 5.332065754149451e-07,
+      "loss": 0.5205,
+      "step": 374
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 5.077132363200403e-07,
+      "loss": 0.7827,
+      "step": 375
+    },
+    {
+      "epoch": 4.52,
+      "eval_loss": 0.6775889992713928,
+      "eval_runtime": 45.8266,
+      "eval_samples_per_second": 18.963,
+      "eval_steps_per_second": 0.305,
+      "step": 375
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 415,
+  "num_train_epochs": 5,
+  "save_steps": 25,
+  "total_flos": 8.478306776532386e+18,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/checkpoint-375/training_args.bin b/checkpoint-375/training_args.bin
new file mode 100644
index 0000000..c951a58
--- /dev/null
+++ b/checkpoint-375/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c98be9e90d62a814f5e99d55200fc54cf1e034ee18489beefd52468c142448bf
+size 6907
diff --git a/checkpoint-375/zero_to_fp32.py b/checkpoint-375/zero_to_fp32.py
new file mode 100644
index 0000000..c98caae
--- /dev/null
+++ b/checkpoint-375/zero_to_fp32.py
@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)
diff --git a/checkpoint-400/config.json b/checkpoint-400/config.json
new file mode 100644
index 0000000..87e67dc
--- /dev/null
+++ b/checkpoint-400/config.json
@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "/workspace/llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.33.1",
+  "use_cache": false,
+  "vocab_size": 32000
+}
diff --git a/checkpoint-400/generation_config.json b/checkpoint-400/generation_config.json
new file mode 100644
index 0000000..0b71b22
--- /dev/null
+++ b/checkpoint-400/generation_config.json
@@ -0,0 +1,4 @@
+{
+  "max_new_tokens": 4096,
+  "transformers_version": "4.33.1"
+}
diff --git a/checkpoint-400/latest b/checkpoint-400/latest
new file mode 100644
index 0000000..e5bdf58
--- /dev/null
+++ b/checkpoint-400/latest
@@ -0,0 +1 @@
+global_step400
\ No newline at end of file
diff --git a/checkpoint-400/pytorch_model-00001-of-00003.bin b/checkpoint-400/pytorch_model-00001-of-00003.bin
new file mode 100644
index 0000000..0d3399a
--- /dev/null
+++ b/checkpoint-400/pytorch_model-00001-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f2a220ab75e607619e1baa83a3b313cde30b3f745519c6f5e46a6c6d68dd6575
+size 9948719433
diff --git a/checkpoint-400/pytorch_model-00002-of-00003.bin b/checkpoint-400/pytorch_model-00002-of-00003.bin
new file mode 100644
index 0000000..26cdef6
--- /dev/null
+++ b/checkpoint-400/pytorch_model-00002-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8959a41cd176f5e41a40de2e650d29bfc454d5eb956eb5a811c1c588987c9c73
+size 9904155408
diff --git a/checkpoint-400/pytorch_model-00003-of-00003.bin b/checkpoint-400/pytorch_model-00003-of-00003.bin
new file mode 100644
index 0000000..d07e15a
--- /dev/null
+++ b/checkpoint-400/pytorch_model-00003-of-00003.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3e5229fa3d61d083820235e8d9d84f124537d216d179844db48528a1b58fc7a
+size 6178978207
diff --git a/checkpoint-400/pytorch_model.bin.index.json b/checkpoint-400/pytorch_model.bin.index.json
new file mode 100644
index 0000000..5ebaea3
--- /dev/null
+++ b/checkpoint-400/pytorch_model.bin.index.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a81b85fdd21a6ffc94396229e2cac8f8135c4a542730e957ede126202c6bf475
+size 29894
diff --git a/checkpoint-400/special_tokens_map.json b/checkpoint-400/special_tokens_map.json
new file mode 100644
index 0000000..14761dc
--- /dev/null
+++ b/checkpoint-400/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}
diff --git a/checkpoint-400/tokenizer.model b/checkpoint-400/tokenizer.model
new file mode 100644
index 0000000..6c00c74
--- /dev/null
+++ b/checkpoint-400/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/checkpoint-400/tokenizer_config.json b/checkpoint-400/tokenizer_config.json
new file mode 100644
index 0000000..16ada5c
--- /dev/null
+++ b/checkpoint-400/tokenizer_config.json
@@ -0,0 +1,37 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_default_system_prompt": true
+}
diff --git a/checkpoint-400/trainer_state.json b/checkpoint-400/trainer_state.json
new file mode 100644
index 0000000..efa570e
--- /dev/null
+++ b/checkpoint-400/trainer_state.json
@@ -0,0 +1,2547 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 4.8192771084337345,
+  "eval_steps": 25,
+  "global_step": 400,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 7.333333333333333e-06,
+      "loss": 0.9353,
+      "step": 1
+    },
+    {
+      "epoch": 0.02,
+      "learning_rate": 1.4666666666666666e-05,
+      "loss": 0.8037,
+      "step": 2
+    },
+    {
+      "epoch": 0.04,
+      "learning_rate": 2.2e-05,
+      "loss": 0.7535,
+      "step": 3
+    },
+    {
+      "epoch": 0.05,
+      "learning_rate": 2.1999680209249292e-05,
+      "loss": 0.6836,
+      "step": 4
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 2.1998720855591014e-05,
+      "loss": 0.8619,
+      "step": 5
+    },
+    {
+      "epoch": 0.07,
+      "learning_rate": 2.1997121994805608e-05,
+      "loss": 1.3258,
+      "step": 6
+    },
+    {
+      "epoch": 0.08,
+      "learning_rate": 2.1994883719856868e-05,
+      "loss": 1.1205,
+      "step": 7
+    },
+    {
+      "epoch": 0.1,
+      "learning_rate": 2.199200616088654e-05,
+      "loss": 0.7097,
+      "step": 8
+    },
+    {
+      "epoch": 0.11,
+      "learning_rate": 2.1988489485206775e-05,
+      "loss": 0.6175,
+      "step": 9
+    },
+    {
+      "epoch": 0.12,
+      "learning_rate": 2.198433389729035e-05,
+      "loss": 0.6068,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 2.1979539638758833e-05,
+      "loss": 0.713,
+      "step": 11
+    },
+    {
+      "epoch": 0.14,
+      "learning_rate": 2.19741069883685e-05,
+      "loss": 1.0415,
+      "step": 12
+    },
+    {
+      "epoch": 0.16,
+      "learning_rate": 2.1968036261994137e-05,
+      "loss": 1.2173,
+      "step": 13
+    },
+    {
+      "epoch": 0.17,
+      "learning_rate": 2.1961327812610683e-05,
+      "loss": 0.6987,
+      "step": 14
+    },
+    {
+      "epoch": 0.18,
+      "learning_rate": 2.19539820302727e-05,
+      "loss": 0.6693,
+      "step": 15
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 2.1945999342091685e-05,
+      "loss": 0.6082,
+      "step": 16
+    },
+    {
+      "epoch": 0.2,
+      "learning_rate": 2.1937380212211243e-05,
+      "loss": 0.6565,
+      "step": 17
+    },
+    {
+      "epoch": 0.22,
+      "learning_rate": 2.1928125141780108e-05,
+      "loss": 0.9413,
+      "step": 18
+    },
+    {
+      "epoch": 0.23,
+      "learning_rate": 2.191823466892299e-05,
+      "loss": 1.3279,
+      "step": 19
+    },
+    {
+      "epoch": 0.24,
+      "learning_rate": 2.1907709368709293e-05,
+      "loss": 0.7384,
+      "step": 20
+    },
+    {
+      "epoch": 0.25,
+      "learning_rate": 2.189654985311969e-05,
+      "loss": 0.7139,
+      "step": 21
+    },
+    {
+      "epoch": 0.27,
+      "learning_rate": 2.1884756771010514e-05,
+      "loss": 0.5926,
+      "step": 22
+    },
+    {
+      "epoch": 0.28,
+      "learning_rate": 2.1872330808076058e-05,
+      "loss": 0.5966,
+      "step": 23
+    },
+    {
+      "epoch": 0.29,
+      "learning_rate": 2.1859272686808682e-05,
+      "loss": 0.843,
+      "step": 24
+    },
+    {
+      "epoch": 0.3,
+      "learning_rate": 2.184558316645683e-05,
+      "loss": 1.2463,
+      "step": 25
+    },
+    {
+      "epoch": 0.3,
+      "eval_loss": 0.7440170645713806,
+      "eval_runtime": 49.2892,
+      "eval_samples_per_second": 17.631,
+      "eval_steps_per_second": 0.284,
+      "step": 25
+    },
+    {
+      "epoch": 0.31,
+      "learning_rate": 2.1831263042980858e-05,
+      "loss": 0.8548,
+      "step": 26
+    },
+    {
+      "epoch": 0.33,
+      "learning_rate": 2.1816313149006775e-05,
+      "loss": 0.6954,
+      "step": 27
+    },
+    {
+      "epoch": 0.34,
+      "learning_rate": 2.180073435377782e-05,
+      "loss": 0.666,
+      "step": 28
+    },
+    {
+      "epoch": 0.35,
+      "learning_rate": 2.1784527563103927e-05,
+      "loss": 0.6027,
+      "step": 29
+    },
+    {
+      "epoch": 0.36,
+      "learning_rate": 2.1767693719309048e-05,
+      "loss": 0.6978,
+      "step": 30
+    },
+    {
+      "epoch": 0.37,
+      "learning_rate": 2.175023380117637e-05,
+      "loss": 0.9855,
+      "step": 31
+    },
+    {
+      "epoch": 0.39,
+      "learning_rate": 2.1732148823891417e-05,
+      "loss": 1.0321,
+      "step": 32
+    },
+    {
+      "epoch": 0.4,
+      "learning_rate": 2.171343983898299e-05,
+      "loss": 0.6891,
+      "step": 33
+    },
+    {
+      "epoch": 0.41,
+      "learning_rate": 2.1694107934262066e-05,
+      "loss": 0.6134,
+      "step": 34
+    },
+    {
+      "epoch": 0.42,
+      "learning_rate": 2.1674154233758516e-05,
+      "loss": 0.5408,
+      "step": 35
+    },
+    {
+      "epoch": 0.43,
+      "learning_rate": 2.165357989765577e-05,
+      "loss": 0.6853,
+      "step": 36
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 2.1632386122223363e-05,
+      "loss": 0.9126,
+      "step": 37
+    },
+    {
+      "epoch": 0.46,
+      "learning_rate": 2.1610574139747343e-05,
+      "loss": 1.2144,
+      "step": 38
+    },
+    {
+      "epoch": 0.47,
+      "learning_rate": 2.1588145218458675e-05,
+      "loss": 0.6852,
+      "step": 39
+    },
+    {
+      "epoch": 0.48,
+      "learning_rate": 2.156510066245946e-05,
+      "loss": 0.6938,
+      "step": 40
+    },
+    {
+      "epoch": 0.49,
+      "learning_rate": 2.1541441811647135e-05,
+      "loss": 0.599,
+      "step": 41
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.1517170041636537e-05,
+      "loss": 0.6201,
+      "step": 42
+    },
+    {
+      "epoch": 0.52,
+      "learning_rate": 2.1492286763679958e-05,
+      "loss": 0.8481,
+      "step": 43
+    },
+    {
+      "epoch": 0.53,
+      "learning_rate": 2.1466793424585053e-05,
+      "loss": 1.3248,
+      "step": 44
+    },
+    {
+      "epoch": 0.54,
+      "learning_rate": 2.1440691506630743e-05,
+      "loss": 0.6589,
+      "step": 45
+    },
+    {
+      "epoch": 0.55,
+      "learning_rate": 2.1413982527481012e-05,
+      "loss": 0.6731,
+      "step": 46
+    },
+    {
+      "epoch": 0.57,
+      "learning_rate": 2.1386668040096685e-05,
+      "loss": 0.6028,
+      "step": 47
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.1358749632645103e-05,
+      "loss": 0.5619,
+      "step": 48
+    },
+    {
+      "epoch": 0.59,
+      "learning_rate": 2.1330228928407817e-05,
+      "loss": 0.7259,
+      "step": 49
+    },
+    {
+      "epoch": 0.6,
+      "learning_rate": 2.1301107585686164e-05,
+      "loss": 1.2589,
+      "step": 50
+    },
+    {
+      "epoch": 0.6,
+      "eval_loss": 0.7223495841026306,
+      "eval_runtime": 45.8793,
+      "eval_samples_per_second": 18.941,
+      "eval_steps_per_second": 0.305,
+      "step": 50
+    },
+    {
+      "epoch": 0.61,
+      "learning_rate": 2.127138729770489e-05,
+      "loss": 0.8173,
+      "step": 51
+    },
+    {
+      "epoch": 0.63,
+      "learning_rate": 2.1241069792513666e-05,
+      "loss": 0.6887,
+      "step": 52
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 2.1210156832886625e-05,
+      "loss": 0.6537,
+      "step": 53
+    },
+    {
+      "epoch": 0.65,
+      "learning_rate": 2.1178650216219872e-05,
+      "loss": 0.5452,
+      "step": 54
+    },
+    {
+      "epoch": 0.66,
+      "learning_rate": 2.114655177442697e-05,
+      "loss": 0.6558,
+      "step": 55
+    },
+    {
+      "epoch": 0.67,
+      "learning_rate": 2.111386337383243e-05,
+      "loss": 1.0381,
+      "step": 56
+    },
+    {
+      "epoch": 0.69,
+      "learning_rate": 2.1080586915063186e-05,
+      "loss": 0.99,
+      "step": 57
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 2.1046724332938103e-05,
+      "loss": 0.657,
+      "step": 58
+    },
+    {
+      "epoch": 0.71,
+      "learning_rate": 2.101227759635546e-05,
+      "loss": 0.6018,
+      "step": 59
+    },
+    {
+      "epoch": 0.72,
+      "learning_rate": 2.0977248708178485e-05,
+      "loss": 0.5436,
+      "step": 60
+    },
+    {
+      "epoch": 0.73,
+      "learning_rate": 2.09416397051189e-05,
+      "loss": 0.6401,
+      "step": 61
+    },
+    {
+      "epoch": 0.75,
+      "learning_rate": 2.0905452657618488e-05,
+      "loss": 0.9314,
+      "step": 62
+    },
+    {
+      "epoch": 0.76,
+      "learning_rate": 2.0868689669728716e-05,
+      "loss": 1.2026,
+      "step": 63
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 2.0831352878988404e-05,
+      "loss": 0.6308,
+      "step": 64
+    },
+    {
+      "epoch": 0.78,
+      "learning_rate": 2.0793444456299434e-05,
+      "loss": 0.6889,
+      "step": 65
+    },
+    {
+      "epoch": 0.8,
+      "learning_rate": 2.0754966605800523e-05,
+      "loss": 0.5788,
+      "step": 66
+    },
+    {
+      "epoch": 0.81,
+      "learning_rate": 2.071592156473907e-05,
+      "loss": 0.5568,
+      "step": 67
+    },
+    {
+      "epoch": 0.82,
+      "learning_rate": 2.067631160334107e-05,
+      "loss": 0.8306,
+      "step": 68
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 2.0636139024679125e-05,
+      "loss": 1.2907,
+      "step": 69
+    },
+    {
+      "epoch": 0.84,
+      "learning_rate": 2.0595406164538524e-05,
+      "loss": 0.6435,
+      "step": 70
+    },
+    {
+      "epoch": 0.86,
+      "learning_rate": 2.0554115391281437e-05,
+      "loss": 0.6731,
+      "step": 71
+    },
+    {
+      "epoch": 0.87,
+      "learning_rate": 2.05122691057092e-05,
+      "loss": 0.5612,
+      "step": 72
+    },
+    {
+      "epoch": 0.88,
+      "learning_rate": 2.046986974092274e-05,
+      "loss": 0.548,
+      "step": 73
+    },
+    {
+      "epoch": 0.89,
+      "learning_rate": 2.042691976218109e-05,
+      "loss": 0.7796,
+      "step": 74
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 2.0383421666758057e-05,
+      "loss": 1.2264,
+      "step": 75
+    },
+    {
+      "epoch": 0.9,
+      "eval_loss": 0.7080472707748413,
+      "eval_runtime": 45.869,
+      "eval_samples_per_second": 18.945,
+      "eval_steps_per_second": 0.305,
+      "step": 75
+    },
+    {
+      "epoch": 0.92,
+      "learning_rate": 2.0339377983797022e-05,
+      "loss": 0.8322,
+      "step": 76
+    },
+    {
+      "epoch": 0.93,
+      "learning_rate": 2.029479127416389e-05,
+      "loss": 0.7071,
+      "step": 77
+    },
+    {
+      "epoch": 0.94,
+      "learning_rate": 2.0249664130298165e-05,
+      "loss": 0.6058,
+      "step": 78
+    },
+    {
+      "epoch": 0.95,
+      "learning_rate": 2.020399917606226e-05,
+      "loss": 0.5243,
+      "step": 79
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.01577990665889e-05,
+      "loss": 0.6636,
+      "step": 80
+    },
+    {
+      "epoch": 0.98,
+      "learning_rate": 2.0111066488126753e-05,
+      "loss": 1.0678,
+      "step": 81
+    },
+    {
+      "epoch": 0.99,
+      "learning_rate": 2.0063804157884248e-05,
+      "loss": 0.853,
+      "step": 82
+    },
+    {
+      "epoch": 1.0,
+      "learning_rate": 2.0016014823871578e-05,
+      "loss": 0.8473,
+      "step": 83
+    },
+    {
+      "epoch": 1.01,
+      "learning_rate": 1.9967701264740925e-05,
+      "loss": 0.7626,
+      "step": 84
+    },
+    {
+      "epoch": 1.02,
+      "learning_rate": 1.99188662896249e-05,
+      "loss": 0.6879,
+      "step": 85
+    },
+    {
+      "epoch": 1.04,
+      "learning_rate": 1.9869512737973196e-05,
+      "loss": 0.5818,
+      "step": 86
+    },
+    {
+      "epoch": 1.05,
+      "learning_rate": 1.9819643479387516e-05,
+      "loss": 0.5151,
+      "step": 87
+    },
+    {
+      "epoch": 1.06,
+      "learning_rate": 1.9769261413454705e-05,
+      "loss": 0.6649,
+      "step": 88
+    },
+    {
+      "epoch": 1.07,
+      "learning_rate": 1.971836946957816e-05,
+      "loss": 1.0307,
+      "step": 89
+    },
+    {
+      "epoch": 1.08,
+      "learning_rate": 1.96669706068075e-05,
+      "loss": 1.0031,
+      "step": 90
+    },
+    {
+      "epoch": 1.1,
+      "learning_rate": 1.961506781366653e-05,
+      "loss": 0.67,
+      "step": 91
+    },
+    {
+      "epoch": 1.11,
+      "learning_rate": 1.9562664107979465e-05,
+      "loss": 0.6331,
+      "step": 92
+    },
+    {
+      "epoch": 1.12,
+      "learning_rate": 1.950976253669547e-05,
+      "loss": 0.5232,
+      "step": 93
+    },
+    {
+      "epoch": 1.13,
+      "learning_rate": 1.9456366175711482e-05,
+      "loss": 0.6291,
+      "step": 94
+    },
+    {
+      "epoch": 1.14,
+      "learning_rate": 1.9402478129693395e-05,
+      "loss": 0.8978,
+      "step": 95
+    },
+    {
+      "epoch": 1.16,
+      "learning_rate": 1.9348101531895503e-05,
+      "loss": 1.1345,
+      "step": 96
+    },
+    {
+      "epoch": 1.17,
+      "learning_rate": 1.929323954397837e-05,
+      "loss": 0.6327,
+      "step": 97
+    },
+    {
+      "epoch": 1.18,
+      "learning_rate": 1.9237895355824952e-05,
+      "loss": 0.6239,
+      "step": 98
+    },
+    {
+      "epoch": 1.19,
+      "learning_rate": 1.9182072185355157e-05,
+      "loss": 0.5745,
+      "step": 99
+    },
+    {
+      "epoch": 1.2,
+      "learning_rate": 1.9125773278338728e-05,
+      "loss": 0.5543,
+      "step": 100
+    },
+    {
+      "epoch": 1.2,
+      "eval_loss": 0.7009603977203369,
+      "eval_runtime": 46.0556,
+      "eval_samples_per_second": 18.868,
+      "eval_steps_per_second": 0.304,
+      "step": 100
+    },
+    {
+      "epoch": 1.22,
+      "learning_rate": 1.9069001908206527e-05,
+      "loss": 0.8478,
+      "step": 101
+    },
+    {
+      "epoch": 1.23,
+      "learning_rate": 1.90117613758602e-05,
+      "loss": 1.2408,
+      "step": 102
+    },
+    {
+      "epoch": 1.24,
+      "learning_rate": 1.8954055009480263e-05,
+      "loss": 0.5991,
+      "step": 103
+    },
+    {
+      "epoch": 1.25,
+      "learning_rate": 1.8895886164332572e-05,
+      "loss": 0.6304,
+      "step": 104
+    },
+    {
+      "epoch": 1.27,
+      "learning_rate": 1.883725822257325e-05,
+      "loss": 0.5863,
+      "step": 105
+    },
+    {
+      "epoch": 1.28,
+      "learning_rate": 1.877817459305202e-05,
+      "loss": 0.5341,
+      "step": 106
+    },
+    {
+      "epoch": 1.29,
+      "learning_rate": 1.8718638711114027e-05,
+      "loss": 0.6937,
+      "step": 107
+    },
+    {
+      "epoch": 1.3,
+      "learning_rate": 1.865865403840005e-05,
+      "loss": 1.12,
+      "step": 108
+    },
+    {
+      "epoch": 1.31,
+      "learning_rate": 1.8598224062645293e-05,
+      "loss": 0.8003,
+      "step": 109
+    },
+    {
+      "epoch": 1.33,
+      "learning_rate": 1.8537352297476526e-05,
+      "loss": 0.6984,
+      "step": 110
+    },
+    {
+      "epoch": 1.34,
+      "learning_rate": 1.8476042282207843e-05,
+      "loss": 0.5954,
+      "step": 111
+    },
+    {
+      "epoch": 1.35,
+      "learning_rate": 1.8414297581634844e-05,
+      "loss": 0.5379,
+      "step": 112
+    },
+    {
+      "epoch": 1.36,
+      "learning_rate": 1.8352121785827374e-05,
+      "loss": 0.6839,
+      "step": 113
+    },
+    {
+      "epoch": 1.37,
+      "learning_rate": 1.828951850992078e-05,
+      "loss": 0.9745,
+      "step": 114
+    },
+    {
+      "epoch": 1.39,
+      "learning_rate": 1.822649139390572e-05,
+      "loss": 0.9575,
+      "step": 115
+    },
+    {
+      "epoch": 1.4,
+      "learning_rate": 1.8163044102416508e-05,
+      "loss": 0.647,
+      "step": 116
+    },
+    {
+      "epoch": 1.41,
+      "learning_rate": 1.809918032451805e-05,
+      "loss": 0.5792,
+      "step": 117
+    },
+    {
+      "epoch": 1.42,
+      "learning_rate": 1.8034903773491342e-05,
+      "loss": 0.5287,
+      "step": 118
+    },
+    {
+      "epoch": 1.43,
+      "learning_rate": 1.797021818661757e-05,
+      "loss": 0.6415,
+      "step": 119
+    },
+    {
+      "epoch": 1.45,
+      "learning_rate": 1.790512732496079e-05,
+      "loss": 0.8963,
+      "step": 120
+    },
+    {
+      "epoch": 1.46,
+      "learning_rate": 1.78396349731493e-05,
+      "loss": 1.1426,
+      "step": 121
+    },
+    {
+      "epoch": 1.47,
+      "learning_rate": 1.7773744939155508e-05,
+      "loss": 0.6213,
+      "step": 122
+    },
+    {
+      "epoch": 1.48,
+      "learning_rate": 1.77074610540746e-05,
+      "loss": 0.6494,
+      "step": 123
+    },
+    {
+      "epoch": 1.49,
+      "learning_rate": 1.7640787171901715e-05,
+      "loss": 0.5484,
+      "step": 124
+    },
+    {
+      "epoch": 1.51,
+      "learning_rate": 1.7573727169307932e-05,
+      "loss": 0.5539,
+      "step": 125
+    },
+    {
+      "epoch": 1.51,
+      "eval_loss": 0.6961802244186401,
+      "eval_runtime": 45.816,
+      "eval_samples_per_second": 18.967,
+      "eval_steps_per_second": 0.306,
+      "step": 125
+    },
+    {
+      "epoch": 1.52,
+      "learning_rate": 1.7506284945414796e-05,
+      "loss": 0.8497,
+      "step": 126
+    },
+    {
+      "epoch": 1.53,
+      "learning_rate": 1.743846442156766e-05,
+      "loss": 1.1775,
+      "step": 127
+    },
+    {
+      "epoch": 1.54,
+      "learning_rate": 1.7370269541107663e-05,
+      "loss": 0.6627,
+      "step": 128
+    },
+    {
+      "epoch": 1.55,
+      "learning_rate": 1.7301704269142434e-05,
+      "loss": 0.6425,
+      "step": 129
+    },
+    {
+      "epoch": 1.57,
+      "learning_rate": 1.7232772592315577e-05,
+      "loss": 0.5537,
+      "step": 130
+    },
+    {
+      "epoch": 1.58,
+      "learning_rate": 1.716347851857485e-05,
+      "loss": 0.5311,
+      "step": 131
+    },
+    {
+      "epoch": 1.59,
+      "learning_rate": 1.7093826076939137e-05,
+      "loss": 0.7583,
+      "step": 132
+    },
+    {
+      "epoch": 1.6,
+      "learning_rate": 1.7023819317264186e-05,
+      "loss": 1.1448,
+      "step": 133
+    },
+    {
+      "epoch": 1.61,
+      "learning_rate": 1.695346231000713e-05,
+      "loss": 0.7377,
+      "step": 134
+    },
+    {
+      "epoch": 1.63,
+      "learning_rate": 1.6882759145989816e-05,
+      "loss": 0.6445,
+      "step": 135
+    },
+    {
+      "epoch": 1.64,
+      "learning_rate": 1.681171393616095e-05,
+      "loss": 0.5787,
+      "step": 136
+    },
+    {
+      "epoch": 1.65,
+      "learning_rate": 1.674033081135708e-05,
+      "loss": 0.5084,
+      "step": 137
+    },
+    {
+      "epoch": 1.66,
+      "learning_rate": 1.6668613922062398e-05,
+      "loss": 0.6423,
+      "step": 138
+    },
+    {
+      "epoch": 1.67,
+      "learning_rate": 1.6596567438167426e-05,
+      "loss": 0.9592,
+      "step": 139
+    },
+    {
+      "epoch": 1.69,
+      "learning_rate": 1.6524195548726557e-05,
+      "loss": 1.0,
+      "step": 140
+    },
+    {
+      "epoch": 1.7,
+      "learning_rate": 1.645150246171449e-05,
+      "loss": 0.6274,
+      "step": 141
+    },
+    {
+      "epoch": 1.71,
+      "learning_rate": 1.637849240378157e-05,
+      "loss": 0.597,
+      "step": 142
+    },
+    {
+      "epoch": 1.72,
+      "learning_rate": 1.630516962000802e-05,
+      "loss": 0.524,
+      "step": 143
+    },
+    {
+      "epoch": 1.73,
+      "learning_rate": 1.623153837365712e-05,
+      "loss": 0.5914,
+      "step": 144
+    },
+    {
+      "epoch": 1.75,
+      "learning_rate": 1.6157602945927335e-05,
+      "loss": 0.8503,
+      "step": 145
+    },
+    {
+      "epoch": 1.76,
+      "learning_rate": 1.6083367635703386e-05,
+      "loss": 1.058,
+      "step": 146
+    },
+    {
+      "epoch": 1.77,
+      "learning_rate": 1.600883675930628e-05,
+      "loss": 0.6366,
+      "step": 147
+    },
+    {
+      "epoch": 1.78,
+      "learning_rate": 1.593401465024237e-05,
+      "loss": 0.6191,
+      "step": 148
+    },
+    {
+      "epoch": 1.8,
+      "learning_rate": 1.5858905658951377e-05,
+      "loss": 0.5433,
+      "step": 149
+    },
+    {
+      "epoch": 1.81,
+      "learning_rate": 1.578351415255342e-05,
+      "loss": 0.5568,
+      "step": 150
+    },
+    {
+      "epoch": 1.81,
+      "eval_loss": 0.6907634735107422,
+      "eval_runtime": 45.8388,
+      "eval_samples_per_second": 18.958,
+      "eval_steps_per_second": 0.305,
+      "step": 150
+    },
+    {
+      "epoch": 1.82,
+      "learning_rate": 1.570784451459512e-05,
+      "loss": 0.8225,
+      "step": 151
+    },
+    {
+      "epoch": 1.83,
+      "learning_rate": 1.5631901144794727e-05,
+      "loss": 1.1446,
+      "step": 152
+    },
+    {
+      "epoch": 1.84,
+      "learning_rate": 1.5555688458786283e-05,
+      "loss": 0.6378,
+      "step": 153
+    },
+    {
+      "epoch": 1.86,
+      "learning_rate": 1.547921088786289e-05,
+      "loss": 0.6675,
+      "step": 154
+    },
+    {
+      "epoch": 1.87,
+      "learning_rate": 1.5402472878719058e-05,
+      "loss": 0.5395,
+      "step": 155
+    },
+    {
+      "epoch": 1.88,
+      "learning_rate": 1.532547889319216e-05,
+      "loss": 0.5378,
+      "step": 156
+    },
+    {
+      "epoch": 1.89,
+      "learning_rate": 1.524823340800301e-05,
+      "loss": 0.7644,
+      "step": 157
+    },
+    {
+      "epoch": 1.9,
+      "learning_rate": 1.5170740914495543e-05,
+      "loss": 1.1105,
+      "step": 158
+    },
+    {
+      "epoch": 1.92,
+      "learning_rate": 1.5093005918375706e-05,
+      "loss": 0.7881,
+      "step": 159
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1.5015032939449451e-05,
+      "loss": 0.6366,
+      "step": 160
+    },
+    {
+      "epoch": 1.94,
+      "learning_rate": 1.493682651135996e-05,
+      "loss": 0.5853,
+      "step": 161
+    },
+    {
+      "epoch": 1.95,
+      "learning_rate": 1.485839118132402e-05,
+      "loss": 0.5138,
+      "step": 162
+    },
+    {
+      "epoch": 1.96,
+      "learning_rate": 1.4779731509867645e-05,
+      "loss": 0.6444,
+      "step": 163
+    },
+    {
+      "epoch": 1.98,
+      "learning_rate": 1.4700852070560903e-05,
+      "loss": 0.9914,
+      "step": 164
+    },
+    {
+      "epoch": 1.99,
+      "learning_rate": 1.4621757449752e-05,
+      "loss": 0.8894,
+      "step": 165
+    },
+    {
+      "epoch": 2.0,
+      "learning_rate": 1.4542452246300597e-05,
+      "loss": 0.8105,
+      "step": 166
+    },
+    {
+      "epoch": 2.01,
+      "learning_rate": 1.4462941071310434e-05,
+      "loss": 0.7536,
+      "step": 167
+    },
+    {
+      "epoch": 2.02,
+      "learning_rate": 1.438322854786121e-05,
+      "loss": 0.6122,
+      "step": 168
+    },
+    {
+      "epoch": 2.04,
+      "learning_rate": 1.4303319310739777e-05,
+      "loss": 0.575,
+      "step": 169
+    },
+    {
+      "epoch": 2.05,
+      "learning_rate": 1.4223218006170675e-05,
+      "loss": 0.5015,
+      "step": 170
+    },
+    {
+      "epoch": 2.06,
+      "learning_rate": 1.4142929291545953e-05,
+      "loss": 0.6682,
+      "step": 171
+    },
+    {
+      "epoch": 2.07,
+      "learning_rate": 1.4062457835154405e-05,
+      "loss": 1.0041,
+      "step": 172
+    },
+    {
+      "epoch": 2.08,
+      "learning_rate": 1.3981808315910112e-05,
+      "loss": 0.9622,
+      "step": 173
+    },
+    {
+      "epoch": 2.1,
+      "learning_rate": 1.3900985423080396e-05,
+      "loss": 0.6121,
+      "step": 174
+    },
+    {
+      "epoch": 2.11,
+      "learning_rate": 1.3819993856013184e-05,
+      "loss": 0.6222,
+      "step": 175
+    },
+    {
+      "epoch": 2.11,
+      "eval_loss": 0.6892703175544739,
+      "eval_runtime": 45.9318,
+      "eval_samples_per_second": 18.919,
+      "eval_steps_per_second": 0.305,
+      "step": 175
+    },
+    {
+      "epoch": 2.12,
+      "learning_rate": 1.3738838323863754e-05,
+      "loss": 0.5335,
+      "step": 176
+    },
+    {
+      "epoch": 2.13,
+      "learning_rate": 1.3657523545320932e-05,
+      "loss": 0.5978,
+      "step": 177
+    },
+    {
+      "epoch": 2.14,
+      "learning_rate": 1.3576054248332733e-05,
+      "loss": 0.9021,
+      "step": 178
+    },
+    {
+      "epoch": 2.16,
+      "learning_rate": 1.3494435169831453e-05,
+      "loss": 1.0881,
+      "step": 179
+    },
+    {
+      "epoch": 2.17,
+      "learning_rate": 1.3412671055458253e-05,
+      "loss": 0.6492,
+      "step": 180
+    },
+    {
+      "epoch": 2.18,
+      "learning_rate": 1.3330766659287234e-05,
+      "loss": 0.6129,
+      "step": 181
+    },
+    {
+      "epoch": 2.19,
+      "learning_rate": 1.3248726743548996e-05,
+      "loss": 0.5737,
+      "step": 182
+    },
+    {
+      "epoch": 2.2,
+      "learning_rate": 1.3166556078353768e-05,
+      "loss": 0.5407,
+      "step": 183
+    },
+    {
+      "epoch": 2.22,
+      "learning_rate": 1.3084259441414036e-05,
+      "loss": 0.7442,
+      "step": 184
+    },
+    {
+      "epoch": 2.23,
+      "learning_rate": 1.3001841617766777e-05,
+      "loss": 1.0671,
+      "step": 185
+    },
+    {
+      "epoch": 2.24,
+      "learning_rate": 1.2919307399495196e-05,
+      "loss": 0.6467,
+      "step": 186
+    },
+    {
+      "epoch": 2.25,
+      "learning_rate": 1.2836661585450144e-05,
+      "loss": 0.6384,
+      "step": 187
+    },
+    {
+      "epoch": 2.27,
+      "learning_rate": 1.275390898097105e-05,
+      "loss": 0.5394,
+      "step": 188
+    },
+    {
+      "epoch": 2.28,
+      "learning_rate": 1.267105439760656e-05,
+      "loss": 0.5238,
+      "step": 189
+    },
+    {
+      "epoch": 2.29,
+      "learning_rate": 1.2588102652834745e-05,
+      "loss": 0.7205,
+      "step": 190
+    },
+    {
+      "epoch": 2.3,
+      "learning_rate": 1.2505058569783012e-05,
+      "loss": 1.0399,
+      "step": 191
+    },
+    {
+      "epoch": 2.31,
+      "learning_rate": 1.2421926976947664e-05,
+      "loss": 0.7659,
+      "step": 192
+    },
+    {
+      "epoch": 2.33,
+      "learning_rate": 1.2338712707913146e-05,
+      "loss": 0.6225,
+      "step": 193
+    },
+    {
+      "epoch": 2.34,
+      "learning_rate": 1.2255420601071026e-05,
+      "loss": 0.5688,
+      "step": 194
+    },
+    {
+      "epoch": 2.35,
+      "learning_rate": 1.2172055499338637e-05,
+      "loss": 0.5092,
+      "step": 195
+    },
+    {
+      "epoch": 2.36,
+      "learning_rate": 1.2088622249877521e-05,
+      "loss": 0.623,
+      "step": 196
+    },
+    {
+      "epoch": 2.37,
+      "learning_rate": 1.200512570381158e-05,
+      "loss": 0.9878,
+      "step": 197
+    },
+    {
+      "epoch": 2.39,
+      "learning_rate": 1.192157071594503e-05,
+      "loss": 0.9355,
+      "step": 198
+    },
+    {
+      "epoch": 2.4,
+      "learning_rate": 1.1837962144480099e-05,
+      "loss": 0.666,
+      "step": 199
+    },
+    {
+      "epoch": 2.41,
+      "learning_rate": 1.1754304850734577e-05,
+      "loss": 0.5828,
+      "step": 200
+    },
+    {
+      "epoch": 2.41,
+      "eval_loss": 0.6850700974464417,
+      "eval_runtime": 45.894,
+      "eval_samples_per_second": 18.935,
+      "eval_steps_per_second": 0.305,
+      "step": 200
+    },
+    {
+      "epoch": 2.42,
+      "learning_rate": 1.167060369885915e-05,
+      "loss": 0.4919,
+      "step": 201
+    },
+    {
+      "epoch": 2.43,
+      "learning_rate": 1.1586863555554581e-05,
+      "loss": 0.5812,
+      "step": 202
+    },
+    {
+      "epoch": 2.45,
+      "learning_rate": 1.1503089289788741e-05,
+      "loss": 0.8558,
+      "step": 203
+    },
+    {
+      "epoch": 2.46,
+      "learning_rate": 1.1419285772513506e-05,
+      "loss": 1.1615,
+      "step": 204
+    },
+    {
+      "epoch": 2.47,
+      "learning_rate": 1.1335457876381556e-05,
+      "loss": 0.5879,
+      "step": 205
+    },
+    {
+      "epoch": 2.48,
+      "learning_rate": 1.1251610475463037e-05,
+      "loss": 0.6408,
+      "step": 206
+    },
+    {
+      "epoch": 2.49,
+      "learning_rate": 1.1167748444962185e-05,
+      "loss": 0.5349,
+      "step": 207
+    },
+    {
+      "epoch": 2.51,
+      "learning_rate": 1.1083876660933852e-05,
+      "loss": 0.5493,
+      "step": 208
+    },
+    {
+      "epoch": 2.52,
+      "learning_rate": 1.1e-05,
+      "loss": 0.8208,
+      "step": 209
+    },
+    {
+      "epoch": 2.53,
+      "learning_rate": 1.0916123339066147e-05,
+      "loss": 1.0668,
+      "step": 210
+    },
+    {
+      "epoch": 2.54,
+      "learning_rate": 1.0832251555037816e-05,
+      "loss": 0.6051,
+      "step": 211
+    },
+    {
+      "epoch": 2.55,
+      "learning_rate": 1.0748389524536964e-05,
+      "loss": 0.6071,
+      "step": 212
+    },
+    {
+      "epoch": 2.57,
+      "learning_rate": 1.0664542123618447e-05,
+      "loss": 0.5666,
+      "step": 213
+    },
+    {
+      "epoch": 2.58,
+      "learning_rate": 1.0580714227486495e-05,
+      "loss": 0.5134,
+      "step": 214
+    },
+    {
+      "epoch": 2.59,
+      "learning_rate": 1.0496910710211261e-05,
+      "loss": 0.7728,
+      "step": 215
+    },
+    {
+      "epoch": 2.6,
+      "learning_rate": 1.041313644444542e-05,
+      "loss": 1.0725,
+      "step": 216
+    },
+    {
+      "epoch": 2.61,
+      "learning_rate": 1.032939630114085e-05,
+      "loss": 0.744,
+      "step": 217
+    },
+    {
+      "epoch": 2.63,
+      "learning_rate": 1.0245695149265425e-05,
+      "loss": 0.646,
+      "step": 218
+    },
+    {
+      "epoch": 2.64,
+      "learning_rate": 1.0162037855519904e-05,
+      "loss": 0.5654,
+      "step": 219
+    },
+    {
+      "epoch": 2.65,
+      "learning_rate": 1.0078429284054971e-05,
+      "loss": 0.5043,
+      "step": 220
+    },
+    {
+      "epoch": 2.66,
+      "learning_rate": 9.99487429618842e-06,
+      "loss": 0.6779,
+      "step": 221
+    },
+    {
+      "epoch": 2.67,
+      "learning_rate": 9.91137775012248e-06,
+      "loss": 0.9569,
+      "step": 222
+    },
+    {
+      "epoch": 2.69,
+      "learning_rate": 9.827944500661365e-06,
+      "loss": 0.966,
+      "step": 223
+    },
+    {
+      "epoch": 2.7,
+      "learning_rate": 9.744579398928975e-06,
+      "loss": 0.5635,
+      "step": 224
+    },
+    {
+      "epoch": 2.71,
+      "learning_rate": 9.661287292086853e-06,
+      "loss": 0.6108,
+      "step": 225
+    },
+    {
+      "epoch": 2.71,
+      "eval_loss": 0.6831873655319214,
+      "eval_runtime": 45.9501,
+      "eval_samples_per_second": 18.912,
+      "eval_steps_per_second": 0.305,
+      "step": 225
+    },
+    {
+      "epoch": 2.72,
+      "learning_rate": 9.578073023052337e-06,
+      "loss": 0.515,
+      "step": 226
+    },
+    {
+      "epoch": 2.73,
+      "learning_rate": 9.494941430216989e-06,
+      "loss": 0.5935,
+      "step": 227
+    },
+    {
+      "epoch": 2.75,
+      "learning_rate": 9.411897347165256e-06,
+      "loss": 0.8431,
+      "step": 228
+    },
+    {
+      "epoch": 2.76,
+      "learning_rate": 9.328945602393441e-06,
+      "loss": 1.1265,
+      "step": 229
+    },
+    {
+      "epoch": 2.77,
+      "learning_rate": 9.246091019028949e-06,
+      "loss": 0.5872,
+      "step": 230
+    },
+    {
+      "epoch": 2.78,
+      "learning_rate": 9.163338414549857e-06,
+      "loss": 0.6158,
+      "step": 231
+    },
+    {
+      "epoch": 2.8,
+      "learning_rate": 9.080692600504803e-06,
+      "loss": 0.5229,
+      "step": 232
+    },
+    {
+      "epoch": 2.81,
+      "learning_rate": 8.998158382233226e-06,
+      "loss": 0.5407,
+      "step": 233
+    },
+    {
+      "epoch": 2.82,
+      "learning_rate": 8.915740558585963e-06,
+      "loss": 0.8395,
+      "step": 234
+    },
+    {
+      "epoch": 2.83,
+      "learning_rate": 8.833443921646237e-06,
+      "loss": 1.162,
+      "step": 235
+    },
+    {
+      "epoch": 2.84,
+      "learning_rate": 8.751273256451007e-06,
+      "loss": 0.6233,
+      "step": 236
+    },
+    {
+      "epoch": 2.86,
+      "learning_rate": 8.669233340712766e-06,
+      "loss": 0.6508,
+      "step": 237
+    },
+    {
+      "epoch": 2.87,
+      "learning_rate": 8.587328944541747e-06,
+      "loss": 0.5542,
+      "step": 238
+    },
+    {
+      "epoch": 2.88,
+      "learning_rate": 8.50556483016855e-06,
+      "loss": 0.5076,
+      "step": 239
+    },
+    {
+      "epoch": 2.89,
+      "learning_rate": 8.423945751667272e-06,
+      "loss": 0.7069,
+      "step": 240
+    },
+    {
+      "epoch": 2.9,
+      "learning_rate": 8.342476454679073e-06,
+      "loss": 1.0609,
+      "step": 241
+    },
+    {
+      "epoch": 2.92,
+      "learning_rate": 8.26116167613625e-06,
+      "loss": 0.7234,
+      "step": 242
+    },
+    {
+      "epoch": 2.93,
+      "learning_rate": 8.18000614398682e-06,
+      "loss": 0.6374,
+      "step": 243
+    },
+    {
+      "epoch": 2.94,
+      "learning_rate": 8.099014576919607e-06,
+      "loss": 0.5869,
+      "step": 244
+    },
+    {
+      "epoch": 2.95,
+      "learning_rate": 8.018191684089893e-06,
+      "loss": 0.4982,
+      "step": 245
+    },
+    {
+      "epoch": 2.96,
+      "learning_rate": 7.937542164845596e-06,
+      "loss": 0.615,
+      "step": 246
+    },
+    {
+      "epoch": 2.98,
+      "learning_rate": 7.857070708454048e-06,
+      "loss": 0.9688,
+      "step": 247
+    },
+    {
+      "epoch": 2.99,
+      "learning_rate": 7.776781993829328e-06,
+      "loss": 0.865,
+      "step": 248
+    },
+    {
+      "epoch": 3.0,
+      "learning_rate": 7.696680689260222e-06,
+      "loss": 0.7174,
+      "step": 249
+    },
+    {
+      "epoch": 3.01,
+      "learning_rate": 7.616771452138793e-06,
+      "loss": 0.6979,
+      "step": 250
+    },
+    {
+      "epoch": 3.01,
+      "eval_loss": 0.6813020706176758,
+      "eval_runtime": 46.0472,
+      "eval_samples_per_second": 18.872,
+      "eval_steps_per_second": 0.304,
+      "step": 250
+    },
+    {
+      "epoch": 3.02,
+      "learning_rate": 7.537058928689568e-06,
+      "loss": 0.6459,
+      "step": 251
+    },
+    {
+      "epoch": 3.04,
+      "learning_rate": 7.457547753699405e-06,
+      "loss": 0.5596,
+      "step": 252
+    },
+    {
+      "epoch": 3.05,
+      "learning_rate": 7.378242550248006e-06,
+      "loss": 0.5273,
+      "step": 253
+    },
+    {
+      "epoch": 3.06,
+      "learning_rate": 7.2991479294391e-06,
+      "loss": 0.6464,
+      "step": 254
+    },
+    {
+      "epoch": 3.07,
+      "learning_rate": 7.220268490132359e-06,
+      "loss": 0.8977,
+      "step": 255
+    },
+    {
+      "epoch": 3.08,
+      "learning_rate": 7.141608818675983e-06,
+      "loss": 0.962,
+      "step": 256
+    },
+    {
+      "epoch": 3.1,
+      "learning_rate": 7.063173488640043e-06,
+      "loss": 0.5978,
+      "step": 257
+    },
+    {
+      "epoch": 3.11,
+      "learning_rate": 6.98496706055055e-06,
+      "loss": 0.6181,
+      "step": 258
+    },
+    {
+      "epoch": 3.12,
+      "learning_rate": 6.906994081624296e-06,
+      "loss": 0.5011,
+      "step": 259
+    },
+    {
+      "epoch": 3.13,
+      "learning_rate": 6.829259085504458e-06,
+      "loss": 0.5811,
+      "step": 260
+    },
+    {
+      "epoch": 3.14,
+      "learning_rate": 6.751766591996991e-06,
+      "loss": 0.7713,
+      "step": 261
+    },
+    {
+      "epoch": 3.16,
+      "learning_rate": 6.674521106807838e-06,
+      "loss": 1.1051,
+      "step": 262
+    },
+    {
+      "epoch": 3.17,
+      "learning_rate": 6.597527121280941e-06,
+      "loss": 0.6117,
+      "step": 263
+    },
+    {
+      "epoch": 3.18,
+      "learning_rate": 6.52078911213711e-06,
+      "loss": 0.5819,
+      "step": 264
+    },
+    {
+      "epoch": 3.19,
+      "learning_rate": 6.4443115412137146e-06,
+      "loss": 0.5601,
+      "step": 265
+    },
+    {
+      "epoch": 3.2,
+      "learning_rate": 6.36809885520527e-06,
+      "loss": 0.547,
+      "step": 266
+    },
+    {
+      "epoch": 3.22,
+      "learning_rate": 6.292155485404879e-06,
+      "loss": 0.81,
+      "step": 267
+    },
+    {
+      "epoch": 3.23,
+      "learning_rate": 6.216485847446581e-06,
+      "loss": 1.1568,
+      "step": 268
+    },
+    {
+      "epoch": 3.24,
+      "learning_rate": 6.14109434104862e-06,
+      "loss": 0.5418,
+      "step": 269
+    },
+    {
+      "epoch": 3.25,
+      "learning_rate": 6.0659853497576245e-06,
+      "loss": 0.6514,
+      "step": 270
+    },
+    {
+      "epoch": 3.27,
+      "learning_rate": 5.991163240693715e-06,
+      "loss": 0.538,
+      "step": 271
+    },
+    {
+      "epoch": 3.28,
+      "learning_rate": 5.916632364296616e-06,
+      "loss": 0.4997,
+      "step": 272
+    },
+    {
+      "epoch": 3.29,
+      "learning_rate": 5.842397054072663e-06,
+      "loss": 0.6937,
+      "step": 273
+    },
+    {
+      "epoch": 3.3,
+      "learning_rate": 5.768461626342883e-06,
+      "loss": 1.0214,
+      "step": 274
+    },
+    {
+      "epoch": 3.31,
+      "learning_rate": 5.694830379991981e-06,
+      "loss": 0.711,
+      "step": 275
+    },
+    {
+      "epoch": 3.31,
+      "eval_loss": 0.6800615787506104,
+      "eval_runtime": 46.0873,
+      "eval_samples_per_second": 18.856,
+      "eval_steps_per_second": 0.304,
+      "step": 275
+    },
+    {
+      "epoch": 3.33,
+      "learning_rate": 5.621507596218431e-06,
+      "loss": 0.6605,
+      "step": 276
+    },
+    {
+      "epoch": 3.34,
+      "learning_rate": 5.548497538285508e-06,
+      "loss": 0.5435,
+      "step": 277
+    },
+    {
+      "epoch": 3.35,
+      "learning_rate": 5.475804451273445e-06,
+      "loss": 0.4961,
+      "step": 278
+    },
+    {
+      "epoch": 3.36,
+      "learning_rate": 5.403432561832573e-06,
+      "loss": 0.61,
+      "step": 279
+    },
+    {
+      "epoch": 3.37,
+      "learning_rate": 5.331386077937598e-06,
+      "loss": 1.0412,
+      "step": 280
+    },
+    {
+      "epoch": 3.39,
+      "learning_rate": 5.259669188642919e-06,
+      "loss": 0.9235,
+      "step": 281
+    },
+    {
+      "epoch": 3.4,
+      "learning_rate": 5.188286063839047e-06,
+      "loss": 0.6159,
+      "step": 282
+    },
+    {
+      "epoch": 3.41,
+      "learning_rate": 5.117240854010184e-06,
+      "loss": 0.6207,
+      "step": 283
+    },
+    {
+      "epoch": 3.42,
+      "learning_rate": 5.046537689992867e-06,
+      "loss": 0.5115,
+      "step": 284
+    },
+    {
+      "epoch": 3.43,
+      "learning_rate": 4.976180682735812e-06,
+      "loss": 0.5823,
+      "step": 285
+    },
+    {
+      "epoch": 3.45,
+      "learning_rate": 4.906173923060861e-06,
+      "loss": 0.8708,
+      "step": 286
+    },
+    {
+      "epoch": 3.46,
+      "learning_rate": 4.836521481425152e-06,
+      "loss": 1.0297,
+      "step": 287
+    },
+    {
+      "epoch": 3.47,
+      "learning_rate": 4.7672274076844245e-06,
+      "loss": 0.5713,
+      "step": 288
+    },
+    {
+      "epoch": 3.48,
+      "learning_rate": 4.698295730857569e-06,
+      "loss": 0.616,
+      "step": 289
+    },
+    {
+      "epoch": 3.49,
+      "learning_rate": 4.629730458892339e-06,
+      "loss": 0.5142,
+      "step": 290
+    },
+    {
+      "epoch": 3.51,
+      "learning_rate": 4.561535578432336e-06,
+      "loss": 0.5395,
+      "step": 291
+    },
+    {
+      "epoch": 3.52,
+      "learning_rate": 4.493715054585204e-06,
+      "loss": 0.7795,
+      "step": 292
+    },
+    {
+      "epoch": 3.53,
+      "learning_rate": 4.4262728306920685e-06,
+      "loss": 1.1353,
+      "step": 293
+    },
+    {
+      "epoch": 3.54,
+      "learning_rate": 4.359212828098284e-06,
+      "loss": 0.6227,
+      "step": 294
+    },
+    {
+      "epoch": 3.55,
+      "learning_rate": 4.292538945925402e-06,
+      "loss": 0.6017,
+      "step": 295
+    },
+    {
+      "epoch": 3.57,
+      "learning_rate": 4.2262550608444916e-06,
+      "loss": 0.5493,
+      "step": 296
+    },
+    {
+      "epoch": 3.58,
+      "learning_rate": 4.1603650268507e-06,
+      "loss": 0.4872,
+      "step": 297
+    },
+    {
+      "epoch": 3.59,
+      "learning_rate": 4.094872675039208e-06,
+      "loss": 0.711,
+      "step": 298
+    },
+    {
+      "epoch": 3.6,
+      "learning_rate": 4.029781813382433e-06,
+      "loss": 1.1265,
+      "step": 299
+    },
+    {
+      "epoch": 3.61,
+      "learning_rate": 3.96509622650866e-06,
+      "loss": 0.7292,
+      "step": 300
+    },
+    {
+      "epoch": 3.61,
+      "eval_loss": 0.6790695190429688,
+      "eval_runtime": 46.0896,
+      "eval_samples_per_second": 18.855,
+      "eval_steps_per_second": 0.304,
+      "step": 300
+    },
+    {
+      "epoch": 3.63,
+      "learning_rate": 3.90081967548195e-06,
+      "loss": 0.6189,
+      "step": 301
+    },
+    {
+      "epoch": 3.64,
+      "learning_rate": 3.836955897583491e-06,
+      "loss": 0.5988,
+      "step": 302
+    },
+    {
+      "epoch": 3.65,
+      "learning_rate": 3.7735086060942807e-06,
+      "loss": 0.504,
+      "step": 303
+    },
+    {
+      "epoch": 3.66,
+      "learning_rate": 3.7104814900792186e-06,
+      "loss": 0.6354,
+      "step": 304
+    },
+    {
+      "epoch": 3.67,
+      "learning_rate": 3.6478782141726272e-06,
+      "loss": 0.9963,
+      "step": 305
+    },
+    {
+      "epoch": 3.69,
+      "learning_rate": 3.585702418365156e-06,
+      "loss": 0.9168,
+      "step": 306
+    },
+    {
+      "epoch": 3.7,
+      "learning_rate": 3.5239577177921574e-06,
+      "loss": 0.5816,
+      "step": 307
+    },
+    {
+      "epoch": 3.71,
+      "learning_rate": 3.462647702523473e-06,
+      "loss": 0.5856,
+      "step": 308
+    },
+    {
+      "epoch": 3.72,
+      "learning_rate": 3.401775937354709e-06,
+      "loss": 0.5252,
+      "step": 309
+    },
+    {
+      "epoch": 3.73,
+      "learning_rate": 3.3413459615999465e-06,
+      "loss": 0.6033,
+      "step": 310
+    },
+    {
+      "epoch": 3.75,
+      "learning_rate": 3.2813612888859768e-06,
+      "loss": 0.7483,
+      "step": 311
+    },
+    {
+      "epoch": 3.76,
+      "learning_rate": 3.221825406947978e-06,
+      "loss": 1.1016,
+      "step": 312
+    },
+    {
+      "epoch": 3.77,
+      "learning_rate": 3.162741777426749e-06,
+      "loss": 0.6132,
+      "step": 313
+    },
+    {
+      "epoch": 3.78,
+      "learning_rate": 3.1041138356674276e-06,
+      "loss": 0.5922,
+      "step": 314
+    },
+    {
+      "epoch": 3.8,
+      "learning_rate": 3.0459449905197367e-06,
+      "loss": 0.5414,
+      "step": 315
+    },
+    {
+      "epoch": 3.81,
+      "learning_rate": 2.988238624139802e-06,
+      "loss": 0.525,
+      "step": 316
+    },
+    {
+      "epoch": 3.82,
+      "learning_rate": 2.930998091793475e-06,
+      "loss": 0.7916,
+      "step": 317
+    },
+    {
+      "epoch": 3.83,
+      "learning_rate": 2.874226721661275e-06,
+      "loss": 1.0491,
+      "step": 318
+    },
+    {
+      "epoch": 3.84,
+      "learning_rate": 2.817927814644844e-06,
+      "loss": 0.6115,
+      "step": 319
+    },
+    {
+      "epoch": 3.86,
+      "learning_rate": 2.762104644175049e-06,
+      "loss": 0.6208,
+      "step": 320
+    },
+    {
+      "epoch": 3.87,
+      "learning_rate": 2.7067604560216305e-06,
+      "loss": 0.5477,
+      "step": 321
+    },
+    {
+      "epoch": 3.88,
+      "learning_rate": 2.651898468104497e-06,
+      "loss": 0.5065,
+      "step": 322
+    },
+    {
+      "epoch": 3.89,
+      "learning_rate": 2.5975218703066078e-06,
+      "loss": 0.7619,
+      "step": 323
+    },
+    {
+      "epoch": 3.9,
+      "learning_rate": 2.5436338242885146e-06,
+      "loss": 1.0427,
+      "step": 324
+    },
+    {
+      "epoch": 3.92,
+      "learning_rate": 2.490237463304532e-06,
+      "loss": 0.7504,
+      "step": 325
+    },
+    {
+      "epoch": 3.92,
+      "eval_loss": 0.6780019998550415,
+      "eval_runtime": 46.0417,
+      "eval_samples_per_second": 18.874,
+      "eval_steps_per_second": 0.304,
+      "step": 325
+    },
+    {
+      "epoch": 3.93,
+      "learning_rate": 2.437335892020534e-06,
+      "loss": 0.6214,
+      "step": 326
+    },
+    {
+      "epoch": 3.94,
+      "learning_rate": 2.3849321863334734e-06,
+      "loss": 0.5806,
+      "step": 327
+    },
+    {
+      "epoch": 3.95,
+      "learning_rate": 2.333029393192503e-06,
+      "loss": 0.5048,
+      "step": 328
+    },
+    {
+      "epoch": 3.96,
+      "learning_rate": 2.281630530421846e-06,
+      "loss": 0.6738,
+      "step": 329
+    },
+    {
+      "epoch": 3.98,
+      "learning_rate": 2.230738586545296e-06,
+      "loss": 0.9792,
+      "step": 330
+    },
+    {
+      "epoch": 3.99,
+      "learning_rate": 2.1803565206124854e-06,
+      "loss": 0.7933,
+      "step": 331
+    },
+    {
+      "epoch": 4.0,
+      "learning_rate": 2.130487262026805e-06,
+      "loss": 0.7724,
+      "step": 332
+    },
+    {
+      "epoch": 4.01,
+      "learning_rate": 2.081133710375105e-06,
+      "loss": 0.7386,
+      "step": 333
+    },
+    {
+      "epoch": 4.02,
+      "learning_rate": 2.032298735259076e-06,
+      "loss": 0.644,
+      "step": 334
+    },
+    {
+      "epoch": 4.04,
+      "learning_rate": 1.9839851761284216e-06,
+      "loss": 0.5484,
+      "step": 335
+    },
+    {
+      "epoch": 4.05,
+      "learning_rate": 1.9361958421157527e-06,
+      "loss": 0.495,
+      "step": 336
+    },
+    {
+      "epoch": 4.06,
+      "learning_rate": 1.8889335118732466e-06,
+      "loss": 0.6507,
+      "step": 337
+    },
+    {
+      "epoch": 4.07,
+      "learning_rate": 1.8422009334111009e-06,
+      "loss": 0.9705,
+      "step": 338
+    },
+    {
+      "epoch": 4.08,
+      "learning_rate": 1.796000823937737e-06,
+      "loss": 0.8893,
+      "step": 339
+    },
+    {
+      "epoch": 4.1,
+      "learning_rate": 1.7503358697018337e-06,
+      "loss": 0.5689,
+      "step": 340
+    },
+    {
+      "epoch": 4.11,
+      "learning_rate": 1.7052087258361115e-06,
+      "loss": 0.5853,
+      "step": 341
+    },
+    {
+      "epoch": 4.12,
+      "learning_rate": 1.6606220162029773e-06,
+      "loss": 0.5154,
+      "step": 342
+    },
+    {
+      "epoch": 4.13,
+      "learning_rate": 1.6165783332419437e-06,
+      "loss": 0.5699,
+      "step": 343
+    },
+    {
+      "epoch": 4.14,
+      "learning_rate": 1.5730802378189136e-06,
+      "loss": 0.8551,
+      "step": 344
+    },
+    {
+      "epoch": 4.16,
+      "learning_rate": 1.5301302590772631e-06,
+      "loss": 1.0386,
+      "step": 345
+    },
+    {
+      "epoch": 4.17,
+      "learning_rate": 1.4877308942908004e-06,
+      "loss": 0.5727,
+      "step": 346
+    },
+    {
+      "epoch": 4.18,
+      "learning_rate": 1.445884608718565e-06,
+      "loss": 0.6391,
+      "step": 347
+    },
+    {
+      "epoch": 4.19,
+      "learning_rate": 1.4045938354614744e-06,
+      "loss": 0.5394,
+      "step": 348
+    },
+    {
+      "epoch": 4.2,
+      "learning_rate": 1.3638609753208755e-06,
+      "loss": 0.5419,
+      "step": 349
+    },
+    {
+      "epoch": 4.22,
+      "learning_rate": 1.3236883966589302e-06,
+      "loss": 0.8315,
+      "step": 350
+    },
+    {
+      "epoch": 4.22,
+      "eval_loss": 0.6777500510215759,
+      "eval_runtime": 45.8022,
+      "eval_samples_per_second": 18.973,
+      "eval_steps_per_second": 0.306,
+      "step": 350
+    },
+    {
+      "epoch": 4.23,
+      "learning_rate": 1.2840784352609335e-06,
+      "loss": 1.1398,
+      "step": 351
+    },
+    {
+      "epoch": 4.24,
+      "learning_rate": 1.2450333941994781e-06,
+      "loss": 0.6095,
+      "step": 352
+    },
+    {
+      "epoch": 4.25,
+      "learning_rate": 1.2065555437005672e-06,
+      "loss": 0.6423,
+      "step": 353
+    },
+    {
+      "epoch": 4.27,
+      "learning_rate": 1.1686471210115957e-06,
+      "loss": 0.5307,
+      "step": 354
+    },
+    {
+      "epoch": 4.28,
+      "learning_rate": 1.1313103302712859e-06,
+      "loss": 0.5394,
+      "step": 355
+    },
+    {
+      "epoch": 4.29,
+      "learning_rate": 1.0945473423815139e-06,
+      "loss": 0.7663,
+      "step": 356
+    },
+    {
+      "epoch": 4.3,
+      "learning_rate": 1.0583602948811003e-06,
+      "loss": 1.0817,
+      "step": 357
+    },
+    {
+      "epoch": 4.31,
+      "learning_rate": 1.0227512918215148e-06,
+      "loss": 0.7287,
+      "step": 358
+    },
+    {
+      "epoch": 4.33,
+      "learning_rate": 9.877224036445412e-07,
+      "loss": 0.6111,
+      "step": 359
+    },
+    {
+      "epoch": 4.34,
+      "learning_rate": 9.532756670618988e-07,
+      "loss": 0.5945,
+      "step": 360
+    },
+    {
+      "epoch": 4.35,
+      "learning_rate": 9.194130849368133e-07,
+      "loss": 0.498,
+      "step": 361
+    },
+    {
+      "epoch": 4.36,
+      "learning_rate": 8.861366261675716e-07,
+      "loss": 0.6097,
+      "step": 362
+    },
+    {
+      "epoch": 4.37,
+      "learning_rate": 8.534482255730291e-07,
+      "loss": 0.9819,
+      "step": 363
+    },
+    {
+      "epoch": 4.39,
+      "learning_rate": 8.213497837801267e-07,
+      "loss": 0.9455,
+      "step": 364
+    },
+    {
+      "epoch": 4.4,
+      "learning_rate": 7.898431671133734e-07,
+      "loss": 0.593,
+      "step": 365
+    },
+    {
+      "epoch": 4.41,
+      "learning_rate": 7.589302074863344e-07,
+      "loss": 0.5929,
+      "step": 366
+    },
+    {
+      "epoch": 4.42,
+      "learning_rate": 7.2861270229511e-07,
+      "loss": 0.5,
+      "step": 367
+    },
+    {
+      "epoch": 4.43,
+      "learning_rate": 6.988924143138361e-07,
+      "loss": 0.6316,
+      "step": 368
+    },
+    {
+      "epoch": 4.45,
+      "learning_rate": 6.697710715921853e-07,
+      "loss": 0.862,
+      "step": 369
+    },
+    {
+      "epoch": 4.46,
+      "learning_rate": 6.412503673548962e-07,
+      "loss": 1.1132,
+      "step": 370
+    },
+    {
+      "epoch": 4.47,
+      "learning_rate": 6.133319599033151e-07,
+      "loss": 0.5833,
+      "step": 371
+    },
+    {
+      "epoch": 4.48,
+      "learning_rate": 5.860174725189841e-07,
+      "loss": 0.591,
+      "step": 372
+    },
+    {
+      "epoch": 4.49,
+      "learning_rate": 5.593084933692568e-07,
+      "loss": 0.5223,
+      "step": 373
+    },
+    {
+      "epoch": 4.51,
+      "learning_rate": 5.332065754149451e-07,
+      "loss": 0.5205,
+      "step": 374
+    },
+    {
+      "epoch": 4.52,
+      "learning_rate": 5.077132363200403e-07,
+      "loss": 0.7827,
+      "step": 375
+    },
+    {
+      "epoch": 4.52,
+      "eval_loss": 0.6775889992713928,
+      "eval_runtime": 45.8266,
+      "eval_samples_per_second": 18.963,
+      "eval_steps_per_second": 0.305,
+      "step": 375
+    },
+    {
+      "epoch": 4.53,
+      "learning_rate": 4.828299583634595e-07,
+      "loss": 1.0985,
+      "step": 376
+    },
+    {
+      "epoch": 4.54,
+      "learning_rate": 4.585581883528651e-07,
+      "loss": 0.5764,
+      "step": 377
+    },
+    {
+      "epoch": 4.55,
+      "learning_rate": 4.3489933754053634e-07,
+      "loss": 0.6336,
+      "step": 378
+    },
+    {
+      "epoch": 4.57,
+      "learning_rate": 4.1185478154132526e-07,
+      "loss": 0.5346,
+      "step": 379
+    },
+    {
+      "epoch": 4.58,
+      "learning_rate": 3.894258602526577e-07,
+      "loss": 0.5052,
+      "step": 380
+    },
+    {
+      "epoch": 4.59,
+      "learning_rate": 3.676138777766409e-07,
+      "loss": 0.7056,
+      "step": 381
+    },
+    {
+      "epoch": 4.6,
+      "learning_rate": 3.4642010234422683e-07,
+      "loss": 1.0772,
+      "step": 382
+    },
+    {
+      "epoch": 4.61,
+      "learning_rate": 3.2584576624148175e-07,
+      "loss": 0.7051,
+      "step": 383
+    },
+    {
+      "epoch": 4.63,
+      "learning_rate": 3.058920657379335e-07,
+      "loss": 0.6147,
+      "step": 384
+    },
+    {
+      "epoch": 4.64,
+      "learning_rate": 2.8656016101700667e-07,
+      "loss": 0.5343,
+      "step": 385
+    },
+    {
+      "epoch": 4.65,
+      "learning_rate": 2.67851176108582e-07,
+      "loss": 0.513,
+      "step": 386
+    },
+    {
+      "epoch": 4.66,
+      "learning_rate": 2.497661988236255e-07,
+      "loss": 0.6182,
+      "step": 387
+    },
+    {
+      "epoch": 4.67,
+      "learning_rate": 2.3230628069095228e-07,
+      "loss": 0.9727,
+      "step": 388
+    },
+    {
+      "epoch": 4.69,
+      "learning_rate": 2.1547243689607153e-07,
+      "loss": 0.9147,
+      "step": 389
+    },
+    {
+      "epoch": 4.7,
+      "learning_rate": 1.9926564622217724e-07,
+      "loss": 0.6112,
+      "step": 390
+    },
+    {
+      "epoch": 4.71,
+      "learning_rate": 1.8368685099322335e-07,
+      "loss": 0.5866,
+      "step": 391
+    },
+    {
+      "epoch": 4.72,
+      "learning_rate": 1.6873695701914126e-07,
+      "loss": 0.5001,
+      "step": 392
+    },
+    {
+      "epoch": 4.73,
+      "learning_rate": 1.5441683354316994e-07,
+      "loss": 0.572,
+      "step": 393
+    },
+    {
+      "epoch": 4.75,
+      "learning_rate": 1.4072731319131582e-07,
+      "loss": 0.8128,
+      "step": 394
+    },
+    {
+      "epoch": 4.76,
+      "learning_rate": 1.2766919192394278e-07,
+      "loss": 1.043,
+      "step": 395
+    },
+    {
+      "epoch": 4.77,
+      "learning_rate": 1.1524322898948557e-07,
+      "loss": 0.5517,
+      "step": 396
+    },
+    {
+      "epoch": 4.78,
+      "learning_rate": 1.0345014688031172e-07,
+      "loss": 0.6207,
+      "step": 397
+    },
+    {
+      "epoch": 4.8,
+      "learning_rate": 9.229063129070691e-08,
+      "loss": 0.5444,
+      "step": 398
+    },
+    {
+      "epoch": 4.81,
+      "learning_rate": 8.176533107701356e-08,
+      "loss": 0.5625,
+      "step": 399
+    },
+    {
+      "epoch": 4.82,
+      "learning_rate": 7.187485821989436e-08,
+      "loss": 0.7863,
+      "step": 400
+    },
+    {
+      "epoch": 4.82,
+      "eval_loss": 0.6775937676429749,
+      "eval_runtime": 45.7701,
+      "eval_samples_per_second": 18.986,
+      "eval_steps_per_second": 0.306,
+      "step": 400
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 415,
+  "num_train_epochs": 5,
+  "save_steps": 25,
+  "total_flos": 9.03857834298232e+18,
+  "trial_name": null,
+  "trial_params": null
+}
diff --git a/checkpoint-400/training_args.bin b/checkpoint-400/training_args.bin
new file mode 100644
index 0000000..c951a58
--- /dev/null
+++ b/checkpoint-400/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c98be9e90d62a814f5e99d55200fc54cf1e034ee18489beefd52468c142448bf
+size 6907
diff --git a/checkpoint-400/zero_to_fp32.py b/checkpoint-400/zero_to_fp32.py
new file mode 100644
index 0000000..c98caae
--- /dev/null
+++ b/checkpoint-400/zero_to_fp32.py
@@ -0,0 +1,587 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel()
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)
diff --git a/config.json b/config.json
new file mode 100644
index 0000000..c7505a7
--- /dev/null
+++ b/config.json
@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "spicyboros-13b-2.2",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.33.1",
+  "use_cache": false,
+  "vocab_size": 32000
+}
diff --git a/configuration.json b/configuration.json
new file mode 100644
index 0000000..bbeeda1
--- /dev/null
+++ b/configuration.json
@@ -0,0 +1 @@
+{"framework": "pytorch", "task": "text-generation", "allow_remote": true}
\ No newline at end of file
diff --git a/generation_config.json b/generation_config.json
new file mode 100644
index 0000000..0b71b22
--- /dev/null
+++ b/generation_config.json
@@ -0,0 +1,4 @@
+{
+  "max_new_tokens": 4096,
+  "transformers_version": "4.33.1"
+}
diff --git a/pytorch_model-00001-of-00006.bin b/pytorch_model-00001-of-00006.bin
new file mode 100644
index 0000000..cbdca4b
--- /dev/null
+++ b/pytorch_model-00001-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef98518b29cf0eac3a134361bcbba6b49faf8210acf248a6b8a8b9b2145148cd
+size 4978280815
diff --git a/pytorch_model-00002-of-00006.bin b/pytorch_model-00002-of-00006.bin
new file mode 100644
index 0000000..e49a0c2
--- /dev/null
+++ b/pytorch_model-00002-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80e35fe9b213c42b02cfad2dd7ee19c0ee24dca3804f74cd37f73f4f5e7cb5ed
+size 4970437691
diff --git a/pytorch_model-00003-of-00006.bin b/pytorch_model-00003-of-00006.bin
new file mode 100644
index 0000000..347497b
--- /dev/null
+++ b/pytorch_model-00003-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d6c69c169086cd4d7309597997ea72447b80054fcbed09f17cdad3bcae33546b
+size 4970437691
diff --git a/pytorch_model-00004-of-00006.bin b/pytorch_model-00004-of-00006.bin
new file mode 100644
index 0000000..2591b1f
--- /dev/null
+++ b/pytorch_model-00004-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a6724b43ab26b9298c4cd84ad18b4cbb16e8bf97096e3c6f086800a718e1d53
+size 4933716737
diff --git a/pytorch_model-00005-of-00006.bin b/pytorch_model-00005-of-00006.bin
new file mode 100644
index 0000000..acb3a17
--- /dev/null
+++ b/pytorch_model-00005-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb6169eafe210d43c6cb60134081d3e1a85063fbdb480f9da38280d47e69ad80
+size 4933737821
diff --git a/pytorch_model-00006-of-00006.bin b/pytorch_model-00006-of-00006.bin
new file mode 100644
index 0000000..76f217c
--- /dev/null
+++ b/pytorch_model-00006-of-00006.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc416bbe18ed5913e11ed7c6f60b823738efd4ba6e6305df7f5f089d6699aa3e
+size 1245240477
diff --git a/pytorch_model.bin.index.json b/pytorch_model.bin.index.json
new file mode 100644
index 0000000..a40d7a8
--- /dev/null
+++ b/pytorch_model.bin.index.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa22d62a88e24625b50c46ebd9f7242c8424d5762be4d68cee2a92934dcf88ff
+size 29894
diff --git a/special_tokens_map.json b/special_tokens_map.json
new file mode 100644
index 0000000..14761dc
--- /dev/null
+++ b/special_tokens_map.json
@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}
diff --git a/tokenizer.model b/tokenizer.model
new file mode 100644
index 0000000..6c00c74
--- /dev/null
+++ b/tokenizer.model
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723
diff --git a/tokenizer_config.json b/tokenizer_config.json
new file mode 100644
index 0000000..16ada5c
--- /dev/null
+++ b/tokenizer_config.json
@@ -0,0 +1,37 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "legacy": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_default_system_prompt": true
+}