Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+checkpoint-210/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-315/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-420/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-522/tokenizer.json filter=lfs diff=lfs merge=lfs -text

checkpoint-210/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-210/config.json

@@ -0,0 +1,27 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

checkpoint-210/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.1,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.51.3"
+}

checkpoint-210/latest

@@ -0,0 +1 @@
+global_step210

checkpoint-210/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a06c293277e50009d04ed093425fc41eecc8905751f0a0333c48dea1ce508ac
+size 3554214752

checkpoint-210/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad792af33c7cfa8b15298ecc9d976ebdcdeb444ca0e704c7b0657f41ee6547eb
+size 14512

checkpoint-210/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:722c924fceffd85f8ab1a5445f1ea1e6c502644b6a42e2ff6b5a9a76ea26e1fe
+size 14512

checkpoint-210/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd9619b29ba8eccd0bac55eb76eb51a451347acd8a0824d109fe4121ffbee803
+size 1064

checkpoint-210/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-210/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

checkpoint-210/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-210/trainer_state.json

@@ -0,0 +1,1504 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.40210627094303497,
+  "eval_steps": 500,
+  "global_step": 210,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0019147917663954045,
+      "grad_norm": 2.4612040519714355,
+      "learning_rate": 0.0,
+      "loss": 0.6068,
+      "step": 1
+    },
+    {
+      "epoch": 0.003829583532790809,
+      "grad_norm": 2.5184295177459717,
+      "learning_rate": 3.846153846153847e-07,
+      "loss": 0.5979,
+      "step": 2
+    },
+    {
+      "epoch": 0.0057443752991862135,
+      "grad_norm": 2.515437602996826,
+      "learning_rate": 7.692307692307694e-07,
+      "loss": 0.6487,
+      "step": 3
+    },
+    {
+      "epoch": 0.007659167065581618,
+      "grad_norm": 2.2364346981048584,
+      "learning_rate": 1.153846153846154e-06,
+      "loss": 0.6041,
+      "step": 4
+    },
+    {
+      "epoch": 0.009573958831977022,
+      "grad_norm": 2.1685123443603516,
+      "learning_rate": 1.5384615384615387e-06,
+      "loss": 0.5414,
+      "step": 5
+    },
+    {
+      "epoch": 0.011488750598372427,
+      "grad_norm": 2.6281416416168213,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.5794,
+      "step": 6
+    },
+    {
+      "epoch": 0.013403542364767831,
+      "grad_norm": 2.0852179527282715,
+      "learning_rate": 2.307692307692308e-06,
+      "loss": 0.5518,
+      "step": 7
+    },
+    {
+      "epoch": 0.015318334131163236,
+      "grad_norm": 1.9601969718933105,
+      "learning_rate": 2.6923076923076923e-06,
+      "loss": 0.557,
+      "step": 8
+    },
+    {
+      "epoch": 0.01723312589755864,
+      "grad_norm": 1.822036862373352,
+      "learning_rate": 3.0769230769230774e-06,
+      "loss": 0.5732,
+      "step": 9
+    },
+    {
+      "epoch": 0.019147917663954045,
+      "grad_norm": 1.6944836378097534,
+      "learning_rate": 3.4615384615384617e-06,
+      "loss": 0.5451,
+      "step": 10
+    },
+    {
+      "epoch": 0.02106270943034945,
+      "grad_norm": 1.521175503730774,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.5148,
+      "step": 11
+    },
+    {
+      "epoch": 0.022977501196744854,
+      "grad_norm": 1.573475956916809,
+      "learning_rate": 4.230769230769231e-06,
+      "loss": 0.512,
+      "step": 12
+    },
+    {
+      "epoch": 0.02489229296314026,
+      "grad_norm": 1.4469544887542725,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.5542,
+      "step": 13
+    },
+    {
+      "epoch": 0.026807084729535663,
+      "grad_norm": 1.3531206846237183,
+      "learning_rate": 5e-06,
+      "loss": 0.4868,
+      "step": 14
+    },
+    {
+      "epoch": 0.028721876495931067,
+      "grad_norm": 1.3297739028930664,
+      "learning_rate": 5.384615384615385e-06,
+      "loss": 0.4713,
+      "step": 15
+    },
+    {
+      "epoch": 0.030636668262326472,
+      "grad_norm": 1.430997610092163,
+      "learning_rate": 5.769230769230769e-06,
+      "loss": 0.4683,
+      "step": 16
+    },
+    {
+      "epoch": 0.032551460028721876,
+      "grad_norm": 1.2832906246185303,
+      "learning_rate": 6.153846153846155e-06,
+      "loss": 0.4275,
+      "step": 17
+    },
+    {
+      "epoch": 0.03446625179511728,
+      "grad_norm": 1.1449981927871704,
+      "learning_rate": 6.538461538461539e-06,
+      "loss": 0.4692,
+      "step": 18
+    },
+    {
+      "epoch": 0.036381043561512685,
+      "grad_norm": 1.1069403886795044,
+      "learning_rate": 6.923076923076923e-06,
+      "loss": 0.4925,
+      "step": 19
+    },
+    {
+      "epoch": 0.03829583532790809,
+      "grad_norm": 1.066596508026123,
+      "learning_rate": 7.307692307692308e-06,
+      "loss": 0.4809,
+      "step": 20
+    },
+    {
+      "epoch": 0.040210627094303494,
+      "grad_norm": 1.0530707836151123,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.4645,
+      "step": 21
+    },
+    {
+      "epoch": 0.0421254188606989,
+      "grad_norm": 1.0063157081604004,
+      "learning_rate": 8.076923076923077e-06,
+      "loss": 0.465,
+      "step": 22
+    },
+    {
+      "epoch": 0.0440402106270943,
+      "grad_norm": 1.1088693141937256,
+      "learning_rate": 8.461538461538462e-06,
+      "loss": 0.4824,
+      "step": 23
+    },
+    {
+      "epoch": 0.04595500239348971,
+      "grad_norm": 1.0253574848175049,
+      "learning_rate": 8.846153846153847e-06,
+      "loss": 0.4559,
+      "step": 24
+    },
+    {
+      "epoch": 0.04786979415988511,
+      "grad_norm": 1.0317028760910034,
+      "learning_rate": 9.230769230769232e-06,
+      "loss": 0.4687,
+      "step": 25
+    },
+    {
+      "epoch": 0.04978458592628052,
+      "grad_norm": 0.9329833388328552,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.4785,
+      "step": 26
+    },
+    {
+      "epoch": 0.05169937769267592,
+      "grad_norm": 0.9087225794792175,
+      "learning_rate": 1e-05,
+      "loss": 0.4505,
+      "step": 27
+    },
+    {
+      "epoch": 0.053614169459071326,
+      "grad_norm": 1.0503875017166138,
+      "learning_rate": 9.999899706000774e-06,
+      "loss": 0.4866,
+      "step": 28
+    },
+    {
+      "epoch": 0.05552896122546673,
+      "grad_norm": 0.9580711722373962,
+      "learning_rate": 9.999598828026644e-06,
+      "loss": 0.4759,
+      "step": 29
+    },
+    {
+      "epoch": 0.057443752991862135,
+      "grad_norm": 0.8906879425048828,
+      "learning_rate": 9.999097378148116e-06,
+      "loss": 0.4455,
+      "step": 30
+    },
+    {
+      "epoch": 0.05935854475825754,
+      "grad_norm": 0.8807913064956665,
+      "learning_rate": 9.998395376482152e-06,
+      "loss": 0.4335,
+      "step": 31
+    },
+    {
+      "epoch": 0.061273336524652944,
+      "grad_norm": 0.8745759129524231,
+      "learning_rate": 9.99749285119138e-06,
+      "loss": 0.43,
+      "step": 32
+    },
+    {
+      "epoch": 0.06318812829104835,
+      "grad_norm": 0.9661962389945984,
+      "learning_rate": 9.996389838482942e-06,
+      "loss": 0.5295,
+      "step": 33
+    },
+    {
+      "epoch": 0.06510292005744375,
+      "grad_norm": 0.9019286036491394,
+      "learning_rate": 9.995086382607064e-06,
+      "loss": 0.4774,
+      "step": 34
+    },
+    {
+      "epoch": 0.06701771182383916,
+      "grad_norm": 0.8906053900718689,
+      "learning_rate": 9.993582535855265e-06,
+      "loss": 0.4577,
+      "step": 35
+    },
+    {
+      "epoch": 0.06893250359023456,
+      "grad_norm": 0.8857560753822327,
+      "learning_rate": 9.991878358558267e-06,
+      "loss": 0.4785,
+      "step": 36
+    },
+    {
+      "epoch": 0.07084729535662997,
+      "grad_norm": 0.9806991219520569,
+      "learning_rate": 9.989973919083576e-06,
+      "loss": 0.4662,
+      "step": 37
+    },
+    {
+      "epoch": 0.07276208712302537,
+      "grad_norm": 0.9356399774551392,
+      "learning_rate": 9.987869293832727e-06,
+      "loss": 0.4669,
+      "step": 38
+    },
+    {
+      "epoch": 0.07467687888942078,
+      "grad_norm": 0.7924108505249023,
+      "learning_rate": 9.985564567238237e-06,
+      "loss": 0.4451,
+      "step": 39
+    },
+    {
+      "epoch": 0.07659167065581618,
+      "grad_norm": 0.9577414393424988,
+      "learning_rate": 9.983059831760205e-06,
+      "loss": 0.4837,
+      "step": 40
+    },
+    {
+      "epoch": 0.07850646242221158,
+      "grad_norm": 0.8300902843475342,
+      "learning_rate": 9.980355187882606e-06,
+      "loss": 0.4438,
+      "step": 41
+    },
+    {
+      "epoch": 0.08042125418860699,
+      "grad_norm": 0.8420023918151855,
+      "learning_rate": 9.977450744109258e-06,
+      "loss": 0.4223,
+      "step": 42
+    },
+    {
+      "epoch": 0.0823360459550024,
+      "grad_norm": 0.8297982811927795,
+      "learning_rate": 9.974346616959476e-06,
+      "loss": 0.4366,
+      "step": 43
+    },
+    {
+      "epoch": 0.0842508377213978,
+      "grad_norm": 0.9187960624694824,
+      "learning_rate": 9.97104293096339e-06,
+      "loss": 0.4739,
+      "step": 44
+    },
+    {
+      "epoch": 0.0861656294877932,
+      "grad_norm": 0.9509177803993225,
+      "learning_rate": 9.967539818656953e-06,
+      "loss": 0.4587,
+      "step": 45
+    },
+    {
+      "epoch": 0.0880804212541886,
+      "grad_norm": 0.9072842001914978,
+      "learning_rate": 9.96383742057662e-06,
+      "loss": 0.5177,
+      "step": 46
+    },
+    {
+      "epoch": 0.08999521302058401,
+      "grad_norm": 0.8744245767593384,
+      "learning_rate": 9.959935885253715e-06,
+      "loss": 0.4454,
+      "step": 47
+    },
+    {
+      "epoch": 0.09191000478697942,
+      "grad_norm": 0.8209521174430847,
+      "learning_rate": 9.955835369208475e-06,
+      "loss": 0.4243,
+      "step": 48
+    },
+    {
+      "epoch": 0.09382479655337482,
+      "grad_norm": 0.793670654296875,
+      "learning_rate": 9.951536036943753e-06,
+      "loss": 0.4256,
+      "step": 49
+    },
+    {
+      "epoch": 0.09573958831977022,
+      "grad_norm": 0.8487685322761536,
+      "learning_rate": 9.94703806093845e-06,
+      "loss": 0.4625,
+      "step": 50
+    },
+    {
+      "epoch": 0.09765438008616563,
+      "grad_norm": 0.8209853768348694,
+      "learning_rate": 9.942341621640558e-06,
+      "loss": 0.4385,
+      "step": 51
+    },
+    {
+      "epoch": 0.09956917185256103,
+      "grad_norm": 0.8096733689308167,
+      "learning_rate": 9.937446907459954e-06,
+      "loss": 0.4575,
+      "step": 52
+    },
+    {
+      "epoch": 0.10148396361895644,
+      "grad_norm": 0.8126389384269714,
+      "learning_rate": 9.932354114760819e-06,
+      "loss": 0.4262,
+      "step": 53
+    },
+    {
+      "epoch": 0.10339875538535184,
+      "grad_norm": 0.968154788017273,
+      "learning_rate": 9.92706344785377e-06,
+      "loss": 0.5304,
+      "step": 54
+    },
+    {
+      "epoch": 0.10531354715174725,
+      "grad_norm": 0.7662584781646729,
+      "learning_rate": 9.921575118987672e-06,
+      "loss": 0.4076,
+      "step": 55
+    },
+    {
+      "epoch": 0.10722833891814265,
+      "grad_norm": 0.8463670015335083,
+      "learning_rate": 9.915889348341098e-06,
+      "loss": 0.4434,
+      "step": 56
+    },
+    {
+      "epoch": 0.10914313068453806,
+      "grad_norm": 0.8167222142219543,
+      "learning_rate": 9.910006364013522e-06,
+      "loss": 0.4079,
+      "step": 57
+    },
+    {
+      "epoch": 0.11105792245093346,
+      "grad_norm": 0.8205484747886658,
+      "learning_rate": 9.903926402016153e-06,
+      "loss": 0.4167,
+      "step": 58
+    },
+    {
+      "epoch": 0.11297271421732887,
+      "grad_norm": 0.901168942451477,
+      "learning_rate": 9.897649706262474e-06,
+      "loss": 0.4762,
+      "step": 59
+    },
+    {
+      "epoch": 0.11488750598372427,
+      "grad_norm": 0.8437636494636536,
+      "learning_rate": 9.891176528558451e-06,
+      "loss": 0.4327,
+      "step": 60
+    },
+    {
+      "epoch": 0.11680229775011967,
+      "grad_norm": 0.8409866094589233,
+      "learning_rate": 9.884507128592435e-06,
+      "loss": 0.446,
+      "step": 61
+    },
+    {
+      "epoch": 0.11871708951651508,
+      "grad_norm": 0.8613469004631042,
+      "learning_rate": 9.877641773924748e-06,
+      "loss": 0.4217,
+      "step": 62
+    },
+    {
+      "epoch": 0.12063188128291048,
+      "grad_norm": 0.8663684129714966,
+      "learning_rate": 9.870580739976936e-06,
+      "loss": 0.4211,
+      "step": 63
+    },
+    {
+      "epoch": 0.12254667304930589,
+      "grad_norm": 0.8156937956809998,
+      "learning_rate": 9.863324310020735e-06,
+      "loss": 0.4273,
+      "step": 64
+    },
+    {
+      "epoch": 0.12446146481570129,
+      "grad_norm": 0.8753077387809753,
+      "learning_rate": 9.855872775166696e-06,
+      "loss": 0.4663,
+      "step": 65
+    },
+    {
+      "epoch": 0.1263762565820967,
+      "grad_norm": 0.81593918800354,
+      "learning_rate": 9.848226434352513e-06,
+      "loss": 0.4398,
+      "step": 66
+    },
+    {
+      "epoch": 0.12829104834849211,
+      "grad_norm": 0.8915799260139465,
+      "learning_rate": 9.840385594331022e-06,
+      "loss": 0.4758,
+      "step": 67
+    },
+    {
+      "epoch": 0.1302058401148875,
+      "grad_norm": 0.8952628374099731,
+      "learning_rate": 9.83235056965791e-06,
+      "loss": 0.4884,
+      "step": 68
+    },
+    {
+      "epoch": 0.13212063188128292,
+      "grad_norm": 0.874907910823822,
+      "learning_rate": 9.824121682679072e-06,
+      "loss": 0.4411,
+      "step": 69
+    },
+    {
+      "epoch": 0.13403542364767831,
+      "grad_norm": 0.8349279761314392,
+      "learning_rate": 9.815699263517712e-06,
+      "loss": 0.438,
+      "step": 70
+    },
+    {
+      "epoch": 0.13595021541407373,
+      "grad_norm": 0.8123260736465454,
+      "learning_rate": 9.807083650061063e-06,
+      "loss": 0.4483,
+      "step": 71
+    },
+    {
+      "epoch": 0.13786500718046912,
+      "grad_norm": 0.8606418371200562,
+      "learning_rate": 9.798275187946859e-06,
+      "loss": 0.44,
+      "step": 72
+    },
+    {
+      "epoch": 0.13977979894686454,
+      "grad_norm": 0.778163492679596,
+      "learning_rate": 9.789274230549456e-06,
+      "loss": 0.4054,
+      "step": 73
+    },
+    {
+      "epoch": 0.14169459071325993,
+      "grad_norm": 0.7578093409538269,
+      "learning_rate": 9.780081138965663e-06,
+      "loss": 0.3794,
+      "step": 74
+    },
+    {
+      "epoch": 0.14360938247965535,
+      "grad_norm": 0.904350221157074,
+      "learning_rate": 9.770696282000245e-06,
+      "loss": 0.4538,
+      "step": 75
+    },
+    {
+      "epoch": 0.14552417424605074,
+      "grad_norm": 0.8594577312469482,
+      "learning_rate": 9.761120036151138e-06,
+      "loss": 0.4211,
+      "step": 76
+    },
+    {
+      "epoch": 0.14743896601244616,
+      "grad_norm": 0.7970111966133118,
+      "learning_rate": 9.751352785594337e-06,
+      "loss": 0.401,
+      "step": 77
+    },
+    {
+      "epoch": 0.14935375777884155,
+      "grad_norm": 0.9434993267059326,
+      "learning_rate": 9.741394922168495e-06,
+      "loss": 0.4868,
+      "step": 78
+    },
+    {
+      "epoch": 0.15126854954523697,
+      "grad_norm": 0.7988068461418152,
+      "learning_rate": 9.731246845359187e-06,
+      "loss": 0.4088,
+      "step": 79
+    },
+    {
+      "epoch": 0.15318334131163236,
+      "grad_norm": 0.7606924176216125,
+      "learning_rate": 9.720908962282893e-06,
+      "loss": 0.4026,
+      "step": 80
+    },
+    {
+      "epoch": 0.15509813307802778,
+      "grad_norm": 0.8167851567268372,
+      "learning_rate": 9.710381687670675e-06,
+      "loss": 0.4347,
+      "step": 81
+    },
+    {
+      "epoch": 0.15701292484442317,
+      "grad_norm": 0.910528838634491,
+      "learning_rate": 9.699665443851518e-06,
+      "loss": 0.4445,
+      "step": 82
+    },
+    {
+      "epoch": 0.1589277166108186,
+      "grad_norm": 0.77234947681427,
+      "learning_rate": 9.688760660735403e-06,
+      "loss": 0.4022,
+      "step": 83
+    },
+    {
+      "epoch": 0.16084250837721398,
+      "grad_norm": 0.7472870945930481,
+      "learning_rate": 9.677667775796052e-06,
+      "loss": 0.3996,
+      "step": 84
+    },
+    {
+      "epoch": 0.1627573001436094,
+      "grad_norm": 0.874338686466217,
+      "learning_rate": 9.666387234053385e-06,
+      "loss": 0.45,
+      "step": 85
+    },
+    {
+      "epoch": 0.1646720919100048,
+      "grad_norm": 0.8861207962036133,
+      "learning_rate": 9.654919488055656e-06,
+      "loss": 0.4388,
+      "step": 86
+    },
+    {
+      "epoch": 0.1665868836764002,
+      "grad_norm": 0.8397772908210754,
+      "learning_rate": 9.643264997861312e-06,
+      "loss": 0.4182,
+      "step": 87
+    },
+    {
+      "epoch": 0.1685016754427956,
+      "grad_norm": 0.8450121283531189,
+      "learning_rate": 9.631424231020523e-06,
+      "loss": 0.4433,
+      "step": 88
+    },
+    {
+      "epoch": 0.170416467209191,
+      "grad_norm": 0.908657431602478,
+      "learning_rate": 9.619397662556434e-06,
+      "loss": 0.4481,
+      "step": 89
+    },
+    {
+      "epoch": 0.1723312589755864,
+      "grad_norm": 0.9613133072853088,
+      "learning_rate": 9.607185774946106e-06,
+      "loss": 0.5182,
+      "step": 90
+    },
+    {
+      "epoch": 0.17424605074198182,
+      "grad_norm": 0.9044798016548157,
+      "learning_rate": 9.594789058101154e-06,
+      "loss": 0.4445,
+      "step": 91
+    },
+    {
+      "epoch": 0.1761608425083772,
+      "grad_norm": 0.8073885440826416,
+      "learning_rate": 9.582208009348104e-06,
+      "loss": 0.4103,
+      "step": 92
+    },
+    {
+      "epoch": 0.17807563427477263,
+      "grad_norm": 0.866552472114563,
+      "learning_rate": 9.569443133408434e-06,
+      "loss": 0.4573,
+      "step": 93
+    },
+    {
+      "epoch": 0.17999042604116802,
+      "grad_norm": 0.8651822805404663,
+      "learning_rate": 9.556494942378328e-06,
+      "loss": 0.4369,
+      "step": 94
+    },
+    {
+      "epoch": 0.18190521780756344,
+      "grad_norm": 0.8955625295639038,
+      "learning_rate": 9.543363955708124e-06,
+      "loss": 0.4493,
+      "step": 95
+    },
+    {
+      "epoch": 0.18382000957395883,
+      "grad_norm": 0.7363105416297913,
+      "learning_rate": 9.530050700181499e-06,
+      "loss": 0.3666,
+      "step": 96
+    },
+    {
+      "epoch": 0.18573480134035425,
+      "grad_norm": 0.7756189107894897,
+      "learning_rate": 9.5165557098943e-06,
+      "loss": 0.4105,
+      "step": 97
+    },
+    {
+      "epoch": 0.18764959310674964,
+      "grad_norm": 0.8105975985527039,
+      "learning_rate": 9.502879526233151e-06,
+      "loss": 0.4026,
+      "step": 98
+    },
+    {
+      "epoch": 0.18956438487314506,
+      "grad_norm": 0.8235639929771423,
+      "learning_rate": 9.48902269785371e-06,
+      "loss": 0.4235,
+      "step": 99
+    },
+    {
+      "epoch": 0.19147917663954045,
+      "grad_norm": 0.9181493520736694,
+      "learning_rate": 9.47498578065867e-06,
+      "loss": 0.4123,
+      "step": 100
+    },
+    {
+      "epoch": 0.19339396840593587,
+      "grad_norm": 0.8257604241371155,
+      "learning_rate": 9.460769337775461e-06,
+      "loss": 0.4304,
+      "step": 101
+    },
+    {
+      "epoch": 0.19530876017233126,
+      "grad_norm": 0.8084312081336975,
+      "learning_rate": 9.446373939533642e-06,
+      "loss": 0.3962,
+      "step": 102
+    },
+    {
+      "epoch": 0.19722355193872668,
+      "grad_norm": 0.8189854025840759,
+      "learning_rate": 9.431800163442043e-06,
+      "loss": 0.4111,
+      "step": 103
+    },
+    {
+      "epoch": 0.19913834370512207,
+      "grad_norm": 0.8247345089912415,
+      "learning_rate": 9.417048594165572e-06,
+      "loss": 0.4485,
+      "step": 104
+    },
+    {
+      "epoch": 0.20105313547151749,
+      "grad_norm": 0.8686959147453308,
+      "learning_rate": 9.402119823501787e-06,
+      "loss": 0.4304,
+      "step": 105
+    },
+    {
+      "epoch": 0.20296792723791288,
+      "grad_norm": 0.839127779006958,
+      "learning_rate": 9.387014450357128e-06,
+      "loss": 0.4148,
+      "step": 106
+    },
+    {
+      "epoch": 0.2048827190043083,
+      "grad_norm": 0.9121819734573364,
+      "learning_rate": 9.371733080722911e-06,
+      "loss": 0.4842,
+      "step": 107
+    },
+    {
+      "epoch": 0.20679751077070369,
+      "grad_norm": 0.8987030982971191,
+      "learning_rate": 9.356276327651006e-06,
+      "loss": 0.4387,
+      "step": 108
+    },
+    {
+      "epoch": 0.2087123025370991,
+      "grad_norm": 0.8091103434562683,
+      "learning_rate": 9.340644811229243e-06,
+      "loss": 0.4024,
+      "step": 109
+    },
+    {
+      "epoch": 0.2106270943034945,
+      "grad_norm": 0.7413907051086426,
+      "learning_rate": 9.324839158556542e-06,
+      "loss": 0.3791,
+      "step": 110
+    },
+    {
+      "epoch": 0.2125418860698899,
+      "grad_norm": 0.8281195163726807,
+      "learning_rate": 9.308860003717748e-06,
+      "loss": 0.4151,
+      "step": 111
+    },
+    {
+      "epoch": 0.2144566778362853,
+      "grad_norm": 0.8579381704330444,
+      "learning_rate": 9.292707987758202e-06,
+      "loss": 0.4446,
+      "step": 112
+    },
+    {
+      "epoch": 0.21637146960268072,
+      "grad_norm": 0.833625853061676,
+      "learning_rate": 9.27638375865801e-06,
+      "loss": 0.4311,
+      "step": 113
+    },
+    {
+      "epoch": 0.2182862613690761,
+      "grad_norm": 0.8704817891120911,
+      "learning_rate": 9.259887971306064e-06,
+      "loss": 0.4852,
+      "step": 114
+    },
+    {
+      "epoch": 0.22020105313547153,
+      "grad_norm": 0.8960813283920288,
+      "learning_rate": 9.243221287473755e-06,
+      "loss": 0.4477,
+      "step": 115
+    },
+    {
+      "epoch": 0.22211584490186692,
+      "grad_norm": 0.8218123316764832,
+      "learning_rate": 9.226384375788435e-06,
+      "loss": 0.4169,
+      "step": 116
+    },
+    {
+      "epoch": 0.22403063666826234,
+      "grad_norm": 0.8349615931510925,
+      "learning_rate": 9.209377911706585e-06,
+      "loss": 0.4045,
+      "step": 117
+    },
+    {
+      "epoch": 0.22594542843465773,
+      "grad_norm": 0.8087317943572998,
+      "learning_rate": 9.192202577486725e-06,
+      "loss": 0.3923,
+      "step": 118
+    },
+    {
+      "epoch": 0.22786022020105315,
+      "grad_norm": 0.8312190175056458,
+      "learning_rate": 9.174859062162037e-06,
+      "loss": 0.3958,
+      "step": 119
+    },
+    {
+      "epoch": 0.22977501196744854,
+      "grad_norm": 0.8147414326667786,
+      "learning_rate": 9.157348061512728e-06,
+      "loss": 0.4417,
+      "step": 120
+    },
+    {
+      "epoch": 0.23168980373384396,
+      "grad_norm": 0.860390305519104,
+      "learning_rate": 9.139670278038109e-06,
+      "loss": 0.4404,
+      "step": 121
+    },
+    {
+      "epoch": 0.23360459550023935,
+      "grad_norm": 0.7553086280822754,
+      "learning_rate": 9.121826420928421e-06,
+      "loss": 0.3684,
+      "step": 122
+    },
+    {
+      "epoch": 0.23551938726663477,
+      "grad_norm": 0.8265523314476013,
+      "learning_rate": 9.103817206036383e-06,
+      "loss": 0.4025,
+      "step": 123
+    },
+    {
+      "epoch": 0.23743417903303016,
+      "grad_norm": 0.8537978529930115,
+      "learning_rate": 9.085643355848468e-06,
+      "loss": 0.4419,
+      "step": 124
+    },
+    {
+      "epoch": 0.23934897079942558,
+      "grad_norm": 0.8403013348579407,
+      "learning_rate": 9.06730559945592e-06,
+      "loss": 0.4013,
+      "step": 125
+    },
+    {
+      "epoch": 0.24126376256582097,
+      "grad_norm": 0.831799328327179,
+      "learning_rate": 9.048804672525513e-06,
+      "loss": 0.417,
+      "step": 126
+    },
+    {
+      "epoch": 0.24317855433221638,
+      "grad_norm": 0.7995438575744629,
+      "learning_rate": 9.030141317270026e-06,
+      "loss": 0.3941,
+      "step": 127
+    },
+    {
+      "epoch": 0.24509334609861178,
+      "grad_norm": 0.8485874533653259,
+      "learning_rate": 9.011316282418474e-06,
+      "loss": 0.4128,
+      "step": 128
+    },
+    {
+      "epoch": 0.2470081378650072,
+      "grad_norm": 0.8879005908966064,
+      "learning_rate": 8.992330323186069e-06,
+      "loss": 0.4445,
+      "step": 129
+    },
+    {
+      "epoch": 0.24892292963140258,
+      "grad_norm": 0.9082502722740173,
+      "learning_rate": 8.973184201243922e-06,
+      "loss": 0.4825,
+      "step": 130
+    },
+    {
+      "epoch": 0.250837721397798,
+      "grad_norm": 0.8748418688774109,
+      "learning_rate": 8.953878684688492e-06,
+      "loss": 0.4209,
+      "step": 131
+    },
+    {
+      "epoch": 0.2527525131641934,
+      "grad_norm": 0.8313506245613098,
+      "learning_rate": 8.934414548010764e-06,
+      "loss": 0.4083,
+      "step": 132
+    },
+    {
+      "epoch": 0.2546673049305888,
+      "grad_norm": 0.8159312009811401,
+      "learning_rate": 8.914792572065178e-06,
+      "loss": 0.4166,
+      "step": 133
+    },
+    {
+      "epoch": 0.25658209669698423,
+      "grad_norm": 0.9007822275161743,
+      "learning_rate": 8.89501354403831e-06,
+      "loss": 0.4591,
+      "step": 134
+    },
+    {
+      "epoch": 0.2584968884633796,
+      "grad_norm": 0.9515888690948486,
+      "learning_rate": 8.875078257417294e-06,
+      "loss": 0.4659,
+      "step": 135
+    },
+    {
+      "epoch": 0.260411680229775,
+      "grad_norm": 0.8637048602104187,
+      "learning_rate": 8.854987511957974e-06,
+      "loss": 0.4387,
+      "step": 136
+    },
+    {
+      "epoch": 0.26232647199617043,
+      "grad_norm": 0.8459029793739319,
+      "learning_rate": 8.834742113652835e-06,
+      "loss": 0.4285,
+      "step": 137
+    },
+    {
+      "epoch": 0.26424126376256585,
+      "grad_norm": 0.7771206498146057,
+      "learning_rate": 8.81434287469866e-06,
+      "loss": 0.3803,
+      "step": 138
+    },
+    {
+      "epoch": 0.2661560555289612,
+      "grad_norm": 0.8040123581886292,
+      "learning_rate": 8.793790613463956e-06,
+      "loss": 0.411,
+      "step": 139
+    },
+    {
+      "epoch": 0.26807084729535663,
+      "grad_norm": 0.8000430464744568,
+      "learning_rate": 8.773086154456106e-06,
+      "loss": 0.4165,
+      "step": 140
+    },
+    {
+      "epoch": 0.26998563906175205,
+      "grad_norm": 0.9271606802940369,
+      "learning_rate": 8.752230328288314e-06,
+      "loss": 0.4755,
+      "step": 141
+    },
+    {
+      "epoch": 0.27190043082814747,
+      "grad_norm": 0.7970160841941833,
+      "learning_rate": 8.731223971646261e-06,
+      "loss": 0.3916,
+      "step": 142
+    },
+    {
+      "epoch": 0.27381522259454283,
+      "grad_norm": 0.7897955775260925,
+      "learning_rate": 8.710067927254555e-06,
+      "loss": 0.3846,
+      "step": 143
+    },
+    {
+      "epoch": 0.27573001436093825,
+      "grad_norm": 0.8004978895187378,
+      "learning_rate": 8.688763043842916e-06,
+      "loss": 0.4077,
+      "step": 144
+    },
+    {
+      "epoch": 0.27764480612733367,
+      "grad_norm": 0.8318747282028198,
+      "learning_rate": 8.66731017611213e-06,
+      "loss": 0.4334,
+      "step": 145
+    },
+    {
+      "epoch": 0.2795595978937291,
+      "grad_norm": 0.8479062914848328,
+      "learning_rate": 8.645710184699756e-06,
+      "loss": 0.4172,
+      "step": 146
+    },
+    {
+      "epoch": 0.28147438966012445,
+      "grad_norm": 0.8077636361122131,
+      "learning_rate": 8.6239639361456e-06,
+      "loss": 0.4104,
+      "step": 147
+    },
+    {
+      "epoch": 0.28338918142651986,
+      "grad_norm": 0.8098336458206177,
+      "learning_rate": 8.602072302856961e-06,
+      "loss": 0.406,
+      "step": 148
+    },
+    {
+      "epoch": 0.2853039731929153,
+      "grad_norm": 0.8435089588165283,
+      "learning_rate": 8.580036163073615e-06,
+      "loss": 0.4312,
+      "step": 149
+    },
+    {
+      "epoch": 0.2872187649593107,
+      "grad_norm": 0.8346257209777832,
+      "learning_rate": 8.5578564008326e-06,
+      "loss": 0.3888,
+      "step": 150
+    },
+    {
+      "epoch": 0.28913355672570606,
+      "grad_norm": 0.8100122213363647,
+      "learning_rate": 8.535533905932739e-06,
+      "loss": 0.4042,
+      "step": 151
+    },
+    {
+      "epoch": 0.2910483484921015,
+      "grad_norm": 0.8516131043434143,
+      "learning_rate": 8.513069573898944e-06,
+      "loss": 0.4161,
+      "step": 152
+    },
+    {
+      "epoch": 0.2929631402584969,
+      "grad_norm": 0.8350062966346741,
+      "learning_rate": 8.490464305946296e-06,
+      "loss": 0.4231,
+      "step": 153
+    },
+    {
+      "epoch": 0.2948779320248923,
+      "grad_norm": 0.8145307302474976,
+      "learning_rate": 8.467719008943886e-06,
+      "loss": 0.4125,
+      "step": 154
+    },
+    {
+      "epoch": 0.2967927237912877,
+      "grad_norm": 0.8515341281890869,
+      "learning_rate": 8.444834595378434e-06,
+      "loss": 0.4192,
+      "step": 155
+    },
+    {
+      "epoch": 0.2987075155576831,
+      "grad_norm": 0.8017379641532898,
+      "learning_rate": 8.421811983317682e-06,
+      "loss": 0.4007,
+      "step": 156
+    },
+    {
+      "epoch": 0.3006223073240785,
+      "grad_norm": 0.8179863691329956,
+      "learning_rate": 8.398652096373566e-06,
+      "loss": 0.4188,
+      "step": 157
+    },
+    {
+      "epoch": 0.30253709909047394,
+      "grad_norm": 0.8272606730461121,
+      "learning_rate": 8.375355863665155e-06,
+      "loss": 0.4035,
+      "step": 158
+    },
+    {
+      "epoch": 0.3044518908568693,
+      "grad_norm": 0.8340199589729309,
+      "learning_rate": 8.351924219781393e-06,
+      "loss": 0.4414,
+      "step": 159
+    },
+    {
+      "epoch": 0.3063666826232647,
+      "grad_norm": 0.82643723487854,
+      "learning_rate": 8.328358104743588e-06,
+      "loss": 0.4143,
+      "step": 160
+    },
+    {
+      "epoch": 0.30828147438966014,
+      "grad_norm": 0.7954344153404236,
+      "learning_rate": 8.304658463967705e-06,
+      "loss": 0.4128,
+      "step": 161
+    },
+    {
+      "epoch": 0.31019626615605556,
+      "grad_norm": 0.7908764481544495,
+      "learning_rate": 8.28082624822645e-06,
+      "loss": 0.3813,
+      "step": 162
+    },
+    {
+      "epoch": 0.3121110579224509,
+      "grad_norm": 0.7368812561035156,
+      "learning_rate": 8.256862413611113e-06,
+      "loss": 0.3879,
+      "step": 163
+    },
+    {
+      "epoch": 0.31402584968884634,
+      "grad_norm": 0.8204508423805237,
+      "learning_rate": 8.232767921493216e-06,
+      "loss": 0.4308,
+      "step": 164
+    },
+    {
+      "epoch": 0.31594064145524176,
+      "grad_norm": 0.8691006302833557,
+      "learning_rate": 8.20854373848595e-06,
+      "loss": 0.4518,
+      "step": 165
+    },
+    {
+      "epoch": 0.3178554332216372,
+      "grad_norm": 0.7581012845039368,
+      "learning_rate": 8.184190836405394e-06,
+      "loss": 0.3711,
+      "step": 166
+    },
+    {
+      "epoch": 0.31977022498803254,
+      "grad_norm": 0.7817586660385132,
+      "learning_rate": 8.15971019223152e-06,
+      "loss": 0.4057,
+      "step": 167
+    },
+    {
+      "epoch": 0.32168501675442795,
+      "grad_norm": 0.7559137344360352,
+      "learning_rate": 8.135102788069015e-06,
+      "loss": 0.361,
+      "step": 168
+    },
+    {
+      "epoch": 0.3235998085208234,
+      "grad_norm": 0.7424903512001038,
+      "learning_rate": 8.110369611107869e-06,
+      "loss": 0.3667,
+      "step": 169
+    },
+    {
+      "epoch": 0.3255146002872188,
+      "grad_norm": 0.7965791821479797,
+      "learning_rate": 8.085511653583772e-06,
+      "loss": 0.3814,
+      "step": 170
+    },
+    {
+      "epoch": 0.32742939205361415,
+      "grad_norm": 0.8504881262779236,
+      "learning_rate": 8.060529912738316e-06,
+      "loss": 0.4453,
+      "step": 171
+    },
+    {
+      "epoch": 0.3293441838200096,
+      "grad_norm": 0.8762128353118896,
+      "learning_rate": 8.035425390778975e-06,
+      "loss": 0.4516,
+      "step": 172
+    },
+    {
+      "epoch": 0.331258975586405,
+      "grad_norm": 0.8157339692115784,
+      "learning_rate": 8.010199094838915e-06,
+      "loss": 0.4204,
+      "step": 173
+    },
+    {
+      "epoch": 0.3331737673528004,
+      "grad_norm": 0.8493944406509399,
+      "learning_rate": 7.984852036936578e-06,
+      "loss": 0.39,
+      "step": 174
+    },
+    {
+      "epoch": 0.3350885591191958,
+      "grad_norm": 0.838901937007904,
+      "learning_rate": 7.959385233935087e-06,
+      "loss": 0.4417,
+      "step": 175
+    },
+    {
+      "epoch": 0.3370033508855912,
+      "grad_norm": 0.7574141621589661,
+      "learning_rate": 7.933799707501448e-06,
+      "loss": 0.3515,
+      "step": 176
+    },
+    {
+      "epoch": 0.3389181426519866,
+      "grad_norm": 0.8099052906036377,
+      "learning_rate": 7.908096484065569e-06,
+      "loss": 0.4091,
+      "step": 177
+    },
+    {
+      "epoch": 0.340832934418382,
+      "grad_norm": 0.8116398453712463,
+      "learning_rate": 7.88227659477908e-06,
+      "loss": 0.4137,
+      "step": 178
+    },
+    {
+      "epoch": 0.3427477261847774,
+      "grad_norm": 0.7785760164260864,
+      "learning_rate": 7.856341075473963e-06,
+      "loss": 0.3829,
+      "step": 179
+    },
+    {
+      "epoch": 0.3446625179511728,
+      "grad_norm": 0.7929257154464722,
+      "learning_rate": 7.830290966620997e-06,
+      "loss": 0.3739,
+      "step": 180
+    },
+    {
+      "epoch": 0.3465773097175682,
+      "grad_norm": 0.8796236515045166,
+      "learning_rate": 7.804127313288023e-06,
+      "loss": 0.4027,
+      "step": 181
+    },
+    {
+      "epoch": 0.34849210148396365,
+      "grad_norm": 0.8447411060333252,
+      "learning_rate": 7.777851165098012e-06,
+      "loss": 0.4202,
+      "step": 182
+    },
+    {
+      "epoch": 0.350406893250359,
+      "grad_norm": 0.7473250031471252,
+      "learning_rate": 7.751463576186957e-06,
+      "loss": 0.3777,
+      "step": 183
+    },
+    {
+      "epoch": 0.3523216850167544,
+      "grad_norm": 0.8254420757293701,
+      "learning_rate": 7.72496560516159e-06,
+      "loss": 0.4147,
+      "step": 184
+    },
+    {
+      "epoch": 0.35423647678314985,
+      "grad_norm": 0.8766903281211853,
+      "learning_rate": 7.6983583150569e-06,
+      "loss": 0.4212,
+      "step": 185
+    },
+    {
+      "epoch": 0.35615126854954526,
+      "grad_norm": 0.8106472492218018,
+      "learning_rate": 7.671642773293506e-06,
+      "loss": 0.3897,
+      "step": 186
+    },
+    {
+      "epoch": 0.3580660603159406,
+      "grad_norm": 0.8969345092773438,
+      "learning_rate": 7.644820051634813e-06,
+      "loss": 0.4166,
+      "step": 187
+    },
+    {
+      "epoch": 0.35998085208233604,
+      "grad_norm": 0.9210174679756165,
+      "learning_rate": 7.617891226144034e-06,
+      "loss": 0.4744,
+      "step": 188
+    },
+    {
+      "epoch": 0.36189564384873146,
+      "grad_norm": 0.8183197379112244,
+      "learning_rate": 7.59085737714101e-06,
+      "loss": 0.3909,
+      "step": 189
+    },
+    {
+      "epoch": 0.3638104356151269,
+      "grad_norm": 0.9258884191513062,
+      "learning_rate": 7.563719589158874e-06,
+      "loss": 0.4191,
+      "step": 190
+    },
+    {
+      "epoch": 0.36572522738152224,
+      "grad_norm": 0.8379626274108887,
+      "learning_rate": 7.536478950900537e-06,
+      "loss": 0.4081,
+      "step": 191
+    },
+    {
+      "epoch": 0.36764001914791766,
+      "grad_norm": 0.8194741010665894,
+      "learning_rate": 7.509136555195025e-06,
+      "loss": 0.3983,
+      "step": 192
+    },
+    {
+      "epoch": 0.3695548109143131,
+      "grad_norm": 0.8078919649124146,
+      "learning_rate": 7.481693498953621e-06,
+      "loss": 0.4117,
+      "step": 193
+    },
+    {
+      "epoch": 0.3714696026807085,
+      "grad_norm": 0.8059271574020386,
+      "learning_rate": 7.4541508831258695e-06,
+      "loss": 0.3919,
+      "step": 194
+    },
+    {
+      "epoch": 0.37338439444710386,
+      "grad_norm": 0.7913739681243896,
+      "learning_rate": 7.4265098126554065e-06,
+      "loss": 0.3786,
+      "step": 195
+    },
+    {
+      "epoch": 0.3752991862134993,
+      "grad_norm": 0.8568047285079956,
+      "learning_rate": 7.3987713964356335e-06,
+      "loss": 0.4505,
+      "step": 196
+    },
+    {
+      "epoch": 0.3772139779798947,
+      "grad_norm": 0.9244000315666199,
+      "learning_rate": 7.370936747265226e-06,
+      "loss": 0.4546,
+      "step": 197
+    },
+    {
+      "epoch": 0.3791287697462901,
+      "grad_norm": 0.7795801162719727,
+      "learning_rate": 7.3430069818035e-06,
+      "loss": 0.3955,
+      "step": 198
+    },
+    {
+      "epoch": 0.3810435615126855,
+      "grad_norm": 0.7940351366996765,
+      "learning_rate": 7.314983220525604e-06,
+      "loss": 0.4043,
+      "step": 199
+    },
+    {
+      "epoch": 0.3829583532790809,
+      "grad_norm": 0.7889821529388428,
+      "learning_rate": 7.286866587677576e-06,
+      "loss": 0.3883,
+      "step": 200
+    },
+    {
+      "epoch": 0.3848731450454763,
+      "grad_norm": 0.857402503490448,
+      "learning_rate": 7.2586582112312355e-06,
+      "loss": 0.4288,
+      "step": 201
+    },
+    {
+      "epoch": 0.38678793681187174,
+      "grad_norm": 0.7854819893836975,
+      "learning_rate": 7.230359222838939e-06,
+      "loss": 0.3884,
+      "step": 202
+    },
+    {
+      "epoch": 0.3887027285782671,
+      "grad_norm": 0.9061365723609924,
+      "learning_rate": 7.201970757788172e-06,
+      "loss": 0.458,
+      "step": 203
+    },
+    {
+      "epoch": 0.3906175203446625,
+      "grad_norm": 0.7978100776672363,
+      "learning_rate": 7.173493954956012e-06,
+      "loss": 0.3903,
+      "step": 204
+    },
+    {
+      "epoch": 0.39253231211105793,
+      "grad_norm": 0.8213270902633667,
+      "learning_rate": 7.144929956763438e-06,
+      "loss": 0.4035,
+      "step": 205
+    },
+    {
+      "epoch": 0.39444710387745335,
+      "grad_norm": 0.9091420769691467,
+      "learning_rate": 7.116279909129492e-06,
+      "loss": 0.4505,
+      "step": 206
+    },
+    {
+      "epoch": 0.3963618956438487,
+      "grad_norm": 0.8489376306533813,
+      "learning_rate": 7.087544961425317e-06,
+      "loss": 0.4037,
+      "step": 207
+    },
+    {
+      "epoch": 0.39827668741024413,
+      "grad_norm": 0.805548369884491,
+      "learning_rate": 7.058726266428042e-06,
+      "loss": 0.4053,
+      "step": 208
+    },
+    {
+      "epoch": 0.40019147917663955,
+      "grad_norm": 0.7785957455635071,
+      "learning_rate": 7.029824980274536e-06,
+      "loss": 0.3719,
+      "step": 209
+    },
+    {
+      "epoch": 0.40210627094303497,
+      "grad_norm": 0.8519039154052734,
+      "learning_rate": 7.0008422624150285e-06,
+      "loss": 0.4176,
+      "step": 210
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 522,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 105,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.1029906388628275e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-210/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66ffb9355835ef1e788c4b73fef523b9594f231dd9bde187500393bdf3018899
+size 10936

checkpoint-210/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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 _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+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() if _has_callable(shape, 'numel') else math.prod(shape)
+            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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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 tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

checkpoint-315/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-315/config.json

@@ -0,0 +1,27 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

checkpoint-315/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.1,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.51.3"
+}

checkpoint-315/latest

@@ -0,0 +1 @@
+global_step315

checkpoint-315/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7458f4d552c6240cc4a4ec473ccc86badb2cc7e33ec73fd4591f3deccea7dda1
+size 3554214752

checkpoint-315/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad792af33c7cfa8b15298ecc9d976ebdcdeb444ca0e704c7b0657f41ee6547eb
+size 14512

checkpoint-315/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:722c924fceffd85f8ab1a5445f1ea1e6c502644b6a42e2ff6b5a9a76ea26e1fe
+size 14512

checkpoint-315/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a4ec420f7da6d05dd8e17b1cc8fc882ab1f031dca3a4f53381815b864453c833
+size 1064

checkpoint-315/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-315/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

checkpoint-315/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-315/trainer_state.json

@@ -0,0 +1,2239 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6031594064145525,
+  "eval_steps": 500,
+  "global_step": 315,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0019147917663954045,
+      "grad_norm": 2.4612040519714355,
+      "learning_rate": 0.0,
+      "loss": 0.6068,
+      "step": 1
+    },
+    {
+      "epoch": 0.003829583532790809,
+      "grad_norm": 2.5184295177459717,
+      "learning_rate": 3.846153846153847e-07,
+      "loss": 0.5979,
+      "step": 2
+    },
+    {
+      "epoch": 0.0057443752991862135,
+      "grad_norm": 2.515437602996826,
+      "learning_rate": 7.692307692307694e-07,
+      "loss": 0.6487,
+      "step": 3
+    },
+    {
+      "epoch": 0.007659167065581618,
+      "grad_norm": 2.2364346981048584,
+      "learning_rate": 1.153846153846154e-06,
+      "loss": 0.6041,
+      "step": 4
+    },
+    {
+      "epoch": 0.009573958831977022,
+      "grad_norm": 2.1685123443603516,
+      "learning_rate": 1.5384615384615387e-06,
+      "loss": 0.5414,
+      "step": 5
+    },
+    {
+      "epoch": 0.011488750598372427,
+      "grad_norm": 2.6281416416168213,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.5794,
+      "step": 6
+    },
+    {
+      "epoch": 0.013403542364767831,
+      "grad_norm": 2.0852179527282715,
+      "learning_rate": 2.307692307692308e-06,
+      "loss": 0.5518,
+      "step": 7
+    },
+    {
+      "epoch": 0.015318334131163236,
+      "grad_norm": 1.9601969718933105,
+      "learning_rate": 2.6923076923076923e-06,
+      "loss": 0.557,
+      "step": 8
+    },
+    {
+      "epoch": 0.01723312589755864,
+      "grad_norm": 1.822036862373352,
+      "learning_rate": 3.0769230769230774e-06,
+      "loss": 0.5732,
+      "step": 9
+    },
+    {
+      "epoch": 0.019147917663954045,
+      "grad_norm": 1.6944836378097534,
+      "learning_rate": 3.4615384615384617e-06,
+      "loss": 0.5451,
+      "step": 10
+    },
+    {
+      "epoch": 0.02106270943034945,
+      "grad_norm": 1.521175503730774,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.5148,
+      "step": 11
+    },
+    {
+      "epoch": 0.022977501196744854,
+      "grad_norm": 1.573475956916809,
+      "learning_rate": 4.230769230769231e-06,
+      "loss": 0.512,
+      "step": 12
+    },
+    {
+      "epoch": 0.02489229296314026,
+      "grad_norm": 1.4469544887542725,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.5542,
+      "step": 13
+    },
+    {
+      "epoch": 0.026807084729535663,
+      "grad_norm": 1.3531206846237183,
+      "learning_rate": 5e-06,
+      "loss": 0.4868,
+      "step": 14
+    },
+    {
+      "epoch": 0.028721876495931067,
+      "grad_norm": 1.3297739028930664,
+      "learning_rate": 5.384615384615385e-06,
+      "loss": 0.4713,
+      "step": 15
+    },
+    {
+      "epoch": 0.030636668262326472,
+      "grad_norm": 1.430997610092163,
+      "learning_rate": 5.769230769230769e-06,
+      "loss": 0.4683,
+      "step": 16
+    },
+    {
+      "epoch": 0.032551460028721876,
+      "grad_norm": 1.2832906246185303,
+      "learning_rate": 6.153846153846155e-06,
+      "loss": 0.4275,
+      "step": 17
+    },
+    {
+      "epoch": 0.03446625179511728,
+      "grad_norm": 1.1449981927871704,
+      "learning_rate": 6.538461538461539e-06,
+      "loss": 0.4692,
+      "step": 18
+    },
+    {
+      "epoch": 0.036381043561512685,
+      "grad_norm": 1.1069403886795044,
+      "learning_rate": 6.923076923076923e-06,
+      "loss": 0.4925,
+      "step": 19
+    },
+    {
+      "epoch": 0.03829583532790809,
+      "grad_norm": 1.066596508026123,
+      "learning_rate": 7.307692307692308e-06,
+      "loss": 0.4809,
+      "step": 20
+    },
+    {
+      "epoch": 0.040210627094303494,
+      "grad_norm": 1.0530707836151123,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.4645,
+      "step": 21
+    },
+    {
+      "epoch": 0.0421254188606989,
+      "grad_norm": 1.0063157081604004,
+      "learning_rate": 8.076923076923077e-06,
+      "loss": 0.465,
+      "step": 22
+    },
+    {
+      "epoch": 0.0440402106270943,
+      "grad_norm": 1.1088693141937256,
+      "learning_rate": 8.461538461538462e-06,
+      "loss": 0.4824,
+      "step": 23
+    },
+    {
+      "epoch": 0.04595500239348971,
+      "grad_norm": 1.0253574848175049,
+      "learning_rate": 8.846153846153847e-06,
+      "loss": 0.4559,
+      "step": 24
+    },
+    {
+      "epoch": 0.04786979415988511,
+      "grad_norm": 1.0317028760910034,
+      "learning_rate": 9.230769230769232e-06,
+      "loss": 0.4687,
+      "step": 25
+    },
+    {
+      "epoch": 0.04978458592628052,
+      "grad_norm": 0.9329833388328552,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.4785,
+      "step": 26
+    },
+    {
+      "epoch": 0.05169937769267592,
+      "grad_norm": 0.9087225794792175,
+      "learning_rate": 1e-05,
+      "loss": 0.4505,
+      "step": 27
+    },
+    {
+      "epoch": 0.053614169459071326,
+      "grad_norm": 1.0503875017166138,
+      "learning_rate": 9.999899706000774e-06,
+      "loss": 0.4866,
+      "step": 28
+    },
+    {
+      "epoch": 0.05552896122546673,
+      "grad_norm": 0.9580711722373962,
+      "learning_rate": 9.999598828026644e-06,
+      "loss": 0.4759,
+      "step": 29
+    },
+    {
+      "epoch": 0.057443752991862135,
+      "grad_norm": 0.8906879425048828,
+      "learning_rate": 9.999097378148116e-06,
+      "loss": 0.4455,
+      "step": 30
+    },
+    {
+      "epoch": 0.05935854475825754,
+      "grad_norm": 0.8807913064956665,
+      "learning_rate": 9.998395376482152e-06,
+      "loss": 0.4335,
+      "step": 31
+    },
+    {
+      "epoch": 0.061273336524652944,
+      "grad_norm": 0.8745759129524231,
+      "learning_rate": 9.99749285119138e-06,
+      "loss": 0.43,
+      "step": 32
+    },
+    {
+      "epoch": 0.06318812829104835,
+      "grad_norm": 0.9661962389945984,
+      "learning_rate": 9.996389838482942e-06,
+      "loss": 0.5295,
+      "step": 33
+    },
+    {
+      "epoch": 0.06510292005744375,
+      "grad_norm": 0.9019286036491394,
+      "learning_rate": 9.995086382607064e-06,
+      "loss": 0.4774,
+      "step": 34
+    },
+    {
+      "epoch": 0.06701771182383916,
+      "grad_norm": 0.8906053900718689,
+      "learning_rate": 9.993582535855265e-06,
+      "loss": 0.4577,
+      "step": 35
+    },
+    {
+      "epoch": 0.06893250359023456,
+      "grad_norm": 0.8857560753822327,
+      "learning_rate": 9.991878358558267e-06,
+      "loss": 0.4785,
+      "step": 36
+    },
+    {
+      "epoch": 0.07084729535662997,
+      "grad_norm": 0.9806991219520569,
+      "learning_rate": 9.989973919083576e-06,
+      "loss": 0.4662,
+      "step": 37
+    },
+    {
+      "epoch": 0.07276208712302537,
+      "grad_norm": 0.9356399774551392,
+      "learning_rate": 9.987869293832727e-06,
+      "loss": 0.4669,
+      "step": 38
+    },
+    {
+      "epoch": 0.07467687888942078,
+      "grad_norm": 0.7924108505249023,
+      "learning_rate": 9.985564567238237e-06,
+      "loss": 0.4451,
+      "step": 39
+    },
+    {
+      "epoch": 0.07659167065581618,
+      "grad_norm": 0.9577414393424988,
+      "learning_rate": 9.983059831760205e-06,
+      "loss": 0.4837,
+      "step": 40
+    },
+    {
+      "epoch": 0.07850646242221158,
+      "grad_norm": 0.8300902843475342,
+      "learning_rate": 9.980355187882606e-06,
+      "loss": 0.4438,
+      "step": 41
+    },
+    {
+      "epoch": 0.08042125418860699,
+      "grad_norm": 0.8420023918151855,
+      "learning_rate": 9.977450744109258e-06,
+      "loss": 0.4223,
+      "step": 42
+    },
+    {
+      "epoch": 0.0823360459550024,
+      "grad_norm": 0.8297982811927795,
+      "learning_rate": 9.974346616959476e-06,
+      "loss": 0.4366,
+      "step": 43
+    },
+    {
+      "epoch": 0.0842508377213978,
+      "grad_norm": 0.9187960624694824,
+      "learning_rate": 9.97104293096339e-06,
+      "loss": 0.4739,
+      "step": 44
+    },
+    {
+      "epoch": 0.0861656294877932,
+      "grad_norm": 0.9509177803993225,
+      "learning_rate": 9.967539818656953e-06,
+      "loss": 0.4587,
+      "step": 45
+    },
+    {
+      "epoch": 0.0880804212541886,
+      "grad_norm": 0.9072842001914978,
+      "learning_rate": 9.96383742057662e-06,
+      "loss": 0.5177,
+      "step": 46
+    },
+    {
+      "epoch": 0.08999521302058401,
+      "grad_norm": 0.8744245767593384,
+      "learning_rate": 9.959935885253715e-06,
+      "loss": 0.4454,
+      "step": 47
+    },
+    {
+      "epoch": 0.09191000478697942,
+      "grad_norm": 0.8209521174430847,
+      "learning_rate": 9.955835369208475e-06,
+      "loss": 0.4243,
+      "step": 48
+    },
+    {
+      "epoch": 0.09382479655337482,
+      "grad_norm": 0.793670654296875,
+      "learning_rate": 9.951536036943753e-06,
+      "loss": 0.4256,
+      "step": 49
+    },
+    {
+      "epoch": 0.09573958831977022,
+      "grad_norm": 0.8487685322761536,
+      "learning_rate": 9.94703806093845e-06,
+      "loss": 0.4625,
+      "step": 50
+    },
+    {
+      "epoch": 0.09765438008616563,
+      "grad_norm": 0.8209853768348694,
+      "learning_rate": 9.942341621640558e-06,
+      "loss": 0.4385,
+      "step": 51
+    },
+    {
+      "epoch": 0.09956917185256103,
+      "grad_norm": 0.8096733689308167,
+      "learning_rate": 9.937446907459954e-06,
+      "loss": 0.4575,
+      "step": 52
+    },
+    {
+      "epoch": 0.10148396361895644,
+      "grad_norm": 0.8126389384269714,
+      "learning_rate": 9.932354114760819e-06,
+      "loss": 0.4262,
+      "step": 53
+    },
+    {
+      "epoch": 0.10339875538535184,
+      "grad_norm": 0.968154788017273,
+      "learning_rate": 9.92706344785377e-06,
+      "loss": 0.5304,
+      "step": 54
+    },
+    {
+      "epoch": 0.10531354715174725,
+      "grad_norm": 0.7662584781646729,
+      "learning_rate": 9.921575118987672e-06,
+      "loss": 0.4076,
+      "step": 55
+    },
+    {
+      "epoch": 0.10722833891814265,
+      "grad_norm": 0.8463670015335083,
+      "learning_rate": 9.915889348341098e-06,
+      "loss": 0.4434,
+      "step": 56
+    },
+    {
+      "epoch": 0.10914313068453806,
+      "grad_norm": 0.8167222142219543,
+      "learning_rate": 9.910006364013522e-06,
+      "loss": 0.4079,
+      "step": 57
+    },
+    {
+      "epoch": 0.11105792245093346,
+      "grad_norm": 0.8205484747886658,
+      "learning_rate": 9.903926402016153e-06,
+      "loss": 0.4167,
+      "step": 58
+    },
+    {
+      "epoch": 0.11297271421732887,
+      "grad_norm": 0.901168942451477,
+      "learning_rate": 9.897649706262474e-06,
+      "loss": 0.4762,
+      "step": 59
+    },
+    {
+      "epoch": 0.11488750598372427,
+      "grad_norm": 0.8437636494636536,
+      "learning_rate": 9.891176528558451e-06,
+      "loss": 0.4327,
+      "step": 60
+    },
+    {
+      "epoch": 0.11680229775011967,
+      "grad_norm": 0.8409866094589233,
+      "learning_rate": 9.884507128592435e-06,
+      "loss": 0.446,
+      "step": 61
+    },
+    {
+      "epoch": 0.11871708951651508,
+      "grad_norm": 0.8613469004631042,
+      "learning_rate": 9.877641773924748e-06,
+      "loss": 0.4217,
+      "step": 62
+    },
+    {
+      "epoch": 0.12063188128291048,
+      "grad_norm": 0.8663684129714966,
+      "learning_rate": 9.870580739976936e-06,
+      "loss": 0.4211,
+      "step": 63
+    },
+    {
+      "epoch": 0.12254667304930589,
+      "grad_norm": 0.8156937956809998,
+      "learning_rate": 9.863324310020735e-06,
+      "loss": 0.4273,
+      "step": 64
+    },
+    {
+      "epoch": 0.12446146481570129,
+      "grad_norm": 0.8753077387809753,
+      "learning_rate": 9.855872775166696e-06,
+      "loss": 0.4663,
+      "step": 65
+    },
+    {
+      "epoch": 0.1263762565820967,
+      "grad_norm": 0.81593918800354,
+      "learning_rate": 9.848226434352513e-06,
+      "loss": 0.4398,
+      "step": 66
+    },
+    {
+      "epoch": 0.12829104834849211,
+      "grad_norm": 0.8915799260139465,
+      "learning_rate": 9.840385594331022e-06,
+      "loss": 0.4758,
+      "step": 67
+    },
+    {
+      "epoch": 0.1302058401148875,
+      "grad_norm": 0.8952628374099731,
+      "learning_rate": 9.83235056965791e-06,
+      "loss": 0.4884,
+      "step": 68
+    },
+    {
+      "epoch": 0.13212063188128292,
+      "grad_norm": 0.874907910823822,
+      "learning_rate": 9.824121682679072e-06,
+      "loss": 0.4411,
+      "step": 69
+    },
+    {
+      "epoch": 0.13403542364767831,
+      "grad_norm": 0.8349279761314392,
+      "learning_rate": 9.815699263517712e-06,
+      "loss": 0.438,
+      "step": 70
+    },
+    {
+      "epoch": 0.13595021541407373,
+      "grad_norm": 0.8123260736465454,
+      "learning_rate": 9.807083650061063e-06,
+      "loss": 0.4483,
+      "step": 71
+    },
+    {
+      "epoch": 0.13786500718046912,
+      "grad_norm": 0.8606418371200562,
+      "learning_rate": 9.798275187946859e-06,
+      "loss": 0.44,
+      "step": 72
+    },
+    {
+      "epoch": 0.13977979894686454,
+      "grad_norm": 0.778163492679596,
+      "learning_rate": 9.789274230549456e-06,
+      "loss": 0.4054,
+      "step": 73
+    },
+    {
+      "epoch": 0.14169459071325993,
+      "grad_norm": 0.7578093409538269,
+      "learning_rate": 9.780081138965663e-06,
+      "loss": 0.3794,
+      "step": 74
+    },
+    {
+      "epoch": 0.14360938247965535,
+      "grad_norm": 0.904350221157074,
+      "learning_rate": 9.770696282000245e-06,
+      "loss": 0.4538,
+      "step": 75
+    },
+    {
+      "epoch": 0.14552417424605074,
+      "grad_norm": 0.8594577312469482,
+      "learning_rate": 9.761120036151138e-06,
+      "loss": 0.4211,
+      "step": 76
+    },
+    {
+      "epoch": 0.14743896601244616,
+      "grad_norm": 0.7970111966133118,
+      "learning_rate": 9.751352785594337e-06,
+      "loss": 0.401,
+      "step": 77
+    },
+    {
+      "epoch": 0.14935375777884155,
+      "grad_norm": 0.9434993267059326,
+      "learning_rate": 9.741394922168495e-06,
+      "loss": 0.4868,
+      "step": 78
+    },
+    {
+      "epoch": 0.15126854954523697,
+      "grad_norm": 0.7988068461418152,
+      "learning_rate": 9.731246845359187e-06,
+      "loss": 0.4088,
+      "step": 79
+    },
+    {
+      "epoch": 0.15318334131163236,
+      "grad_norm": 0.7606924176216125,
+      "learning_rate": 9.720908962282893e-06,
+      "loss": 0.4026,
+      "step": 80
+    },
+    {
+      "epoch": 0.15509813307802778,
+      "grad_norm": 0.8167851567268372,
+      "learning_rate": 9.710381687670675e-06,
+      "loss": 0.4347,
+      "step": 81
+    },
+    {
+      "epoch": 0.15701292484442317,
+      "grad_norm": 0.910528838634491,
+      "learning_rate": 9.699665443851518e-06,
+      "loss": 0.4445,
+      "step": 82
+    },
+    {
+      "epoch": 0.1589277166108186,
+      "grad_norm": 0.77234947681427,
+      "learning_rate": 9.688760660735403e-06,
+      "loss": 0.4022,
+      "step": 83
+    },
+    {
+      "epoch": 0.16084250837721398,
+      "grad_norm": 0.7472870945930481,
+      "learning_rate": 9.677667775796052e-06,
+      "loss": 0.3996,
+      "step": 84
+    },
+    {
+      "epoch": 0.1627573001436094,
+      "grad_norm": 0.874338686466217,
+      "learning_rate": 9.666387234053385e-06,
+      "loss": 0.45,
+      "step": 85
+    },
+    {
+      "epoch": 0.1646720919100048,
+      "grad_norm": 0.8861207962036133,
+      "learning_rate": 9.654919488055656e-06,
+      "loss": 0.4388,
+      "step": 86
+    },
+    {
+      "epoch": 0.1665868836764002,
+      "grad_norm": 0.8397772908210754,
+      "learning_rate": 9.643264997861312e-06,
+      "loss": 0.4182,
+      "step": 87
+    },
+    {
+      "epoch": 0.1685016754427956,
+      "grad_norm": 0.8450121283531189,
+      "learning_rate": 9.631424231020523e-06,
+      "loss": 0.4433,
+      "step": 88
+    },
+    {
+      "epoch": 0.170416467209191,
+      "grad_norm": 0.908657431602478,
+      "learning_rate": 9.619397662556434e-06,
+      "loss": 0.4481,
+      "step": 89
+    },
+    {
+      "epoch": 0.1723312589755864,
+      "grad_norm": 0.9613133072853088,
+      "learning_rate": 9.607185774946106e-06,
+      "loss": 0.5182,
+      "step": 90
+    },
+    {
+      "epoch": 0.17424605074198182,
+      "grad_norm": 0.9044798016548157,
+      "learning_rate": 9.594789058101154e-06,
+      "loss": 0.4445,
+      "step": 91
+    },
+    {
+      "epoch": 0.1761608425083772,
+      "grad_norm": 0.8073885440826416,
+      "learning_rate": 9.582208009348104e-06,
+      "loss": 0.4103,
+      "step": 92
+    },
+    {
+      "epoch": 0.17807563427477263,
+      "grad_norm": 0.866552472114563,
+      "learning_rate": 9.569443133408434e-06,
+      "loss": 0.4573,
+      "step": 93
+    },
+    {
+      "epoch": 0.17999042604116802,
+      "grad_norm": 0.8651822805404663,
+      "learning_rate": 9.556494942378328e-06,
+      "loss": 0.4369,
+      "step": 94
+    },
+    {
+      "epoch": 0.18190521780756344,
+      "grad_norm": 0.8955625295639038,
+      "learning_rate": 9.543363955708124e-06,
+      "loss": 0.4493,
+      "step": 95
+    },
+    {
+      "epoch": 0.18382000957395883,
+      "grad_norm": 0.7363105416297913,
+      "learning_rate": 9.530050700181499e-06,
+      "loss": 0.3666,
+      "step": 96
+    },
+    {
+      "epoch": 0.18573480134035425,
+      "grad_norm": 0.7756189107894897,
+      "learning_rate": 9.5165557098943e-06,
+      "loss": 0.4105,
+      "step": 97
+    },
+    {
+      "epoch": 0.18764959310674964,
+      "grad_norm": 0.8105975985527039,
+      "learning_rate": 9.502879526233151e-06,
+      "loss": 0.4026,
+      "step": 98
+    },
+    {
+      "epoch": 0.18956438487314506,
+      "grad_norm": 0.8235639929771423,
+      "learning_rate": 9.48902269785371e-06,
+      "loss": 0.4235,
+      "step": 99
+    },
+    {
+      "epoch": 0.19147917663954045,
+      "grad_norm": 0.9181493520736694,
+      "learning_rate": 9.47498578065867e-06,
+      "loss": 0.4123,
+      "step": 100
+    },
+    {
+      "epoch": 0.19339396840593587,
+      "grad_norm": 0.8257604241371155,
+      "learning_rate": 9.460769337775461e-06,
+      "loss": 0.4304,
+      "step": 101
+    },
+    {
+      "epoch": 0.19530876017233126,
+      "grad_norm": 0.8084312081336975,
+      "learning_rate": 9.446373939533642e-06,
+      "loss": 0.3962,
+      "step": 102
+    },
+    {
+      "epoch": 0.19722355193872668,
+      "grad_norm": 0.8189854025840759,
+      "learning_rate": 9.431800163442043e-06,
+      "loss": 0.4111,
+      "step": 103
+    },
+    {
+      "epoch": 0.19913834370512207,
+      "grad_norm": 0.8247345089912415,
+      "learning_rate": 9.417048594165572e-06,
+      "loss": 0.4485,
+      "step": 104
+    },
+    {
+      "epoch": 0.20105313547151749,
+      "grad_norm": 0.8686959147453308,
+      "learning_rate": 9.402119823501787e-06,
+      "loss": 0.4304,
+      "step": 105
+    },
+    {
+      "epoch": 0.20296792723791288,
+      "grad_norm": 0.839127779006958,
+      "learning_rate": 9.387014450357128e-06,
+      "loss": 0.4148,
+      "step": 106
+    },
+    {
+      "epoch": 0.2048827190043083,
+      "grad_norm": 0.9121819734573364,
+      "learning_rate": 9.371733080722911e-06,
+      "loss": 0.4842,
+      "step": 107
+    },
+    {
+      "epoch": 0.20679751077070369,
+      "grad_norm": 0.8987030982971191,
+      "learning_rate": 9.356276327651006e-06,
+      "loss": 0.4387,
+      "step": 108
+    },
+    {
+      "epoch": 0.2087123025370991,
+      "grad_norm": 0.8091103434562683,
+      "learning_rate": 9.340644811229243e-06,
+      "loss": 0.4024,
+      "step": 109
+    },
+    {
+      "epoch": 0.2106270943034945,
+      "grad_norm": 0.7413907051086426,
+      "learning_rate": 9.324839158556542e-06,
+      "loss": 0.3791,
+      "step": 110
+    },
+    {
+      "epoch": 0.2125418860698899,
+      "grad_norm": 0.8281195163726807,
+      "learning_rate": 9.308860003717748e-06,
+      "loss": 0.4151,
+      "step": 111
+    },
+    {
+      "epoch": 0.2144566778362853,
+      "grad_norm": 0.8579381704330444,
+      "learning_rate": 9.292707987758202e-06,
+      "loss": 0.4446,
+      "step": 112
+    },
+    {
+      "epoch": 0.21637146960268072,
+      "grad_norm": 0.833625853061676,
+      "learning_rate": 9.27638375865801e-06,
+      "loss": 0.4311,
+      "step": 113
+    },
+    {
+      "epoch": 0.2182862613690761,
+      "grad_norm": 0.8704817891120911,
+      "learning_rate": 9.259887971306064e-06,
+      "loss": 0.4852,
+      "step": 114
+    },
+    {
+      "epoch": 0.22020105313547153,
+      "grad_norm": 0.8960813283920288,
+      "learning_rate": 9.243221287473755e-06,
+      "loss": 0.4477,
+      "step": 115
+    },
+    {
+      "epoch": 0.22211584490186692,
+      "grad_norm": 0.8218123316764832,
+      "learning_rate": 9.226384375788435e-06,
+      "loss": 0.4169,
+      "step": 116
+    },
+    {
+      "epoch": 0.22403063666826234,
+      "grad_norm": 0.8349615931510925,
+      "learning_rate": 9.209377911706585e-06,
+      "loss": 0.4045,
+      "step": 117
+    },
+    {
+      "epoch": 0.22594542843465773,
+      "grad_norm": 0.8087317943572998,
+      "learning_rate": 9.192202577486725e-06,
+      "loss": 0.3923,
+      "step": 118
+    },
+    {
+      "epoch": 0.22786022020105315,
+      "grad_norm": 0.8312190175056458,
+      "learning_rate": 9.174859062162037e-06,
+      "loss": 0.3958,
+      "step": 119
+    },
+    {
+      "epoch": 0.22977501196744854,
+      "grad_norm": 0.8147414326667786,
+      "learning_rate": 9.157348061512728e-06,
+      "loss": 0.4417,
+      "step": 120
+    },
+    {
+      "epoch": 0.23168980373384396,
+      "grad_norm": 0.860390305519104,
+      "learning_rate": 9.139670278038109e-06,
+      "loss": 0.4404,
+      "step": 121
+    },
+    {
+      "epoch": 0.23360459550023935,
+      "grad_norm": 0.7553086280822754,
+      "learning_rate": 9.121826420928421e-06,
+      "loss": 0.3684,
+      "step": 122
+    },
+    {
+      "epoch": 0.23551938726663477,
+      "grad_norm": 0.8265523314476013,
+      "learning_rate": 9.103817206036383e-06,
+      "loss": 0.4025,
+      "step": 123
+    },
+    {
+      "epoch": 0.23743417903303016,
+      "grad_norm": 0.8537978529930115,
+      "learning_rate": 9.085643355848468e-06,
+      "loss": 0.4419,
+      "step": 124
+    },
+    {
+      "epoch": 0.23934897079942558,
+      "grad_norm": 0.8403013348579407,
+      "learning_rate": 9.06730559945592e-06,
+      "loss": 0.4013,
+      "step": 125
+    },
+    {
+      "epoch": 0.24126376256582097,
+      "grad_norm": 0.831799328327179,
+      "learning_rate": 9.048804672525513e-06,
+      "loss": 0.417,
+      "step": 126
+    },
+    {
+      "epoch": 0.24317855433221638,
+      "grad_norm": 0.7995438575744629,
+      "learning_rate": 9.030141317270026e-06,
+      "loss": 0.3941,
+      "step": 127
+    },
+    {
+      "epoch": 0.24509334609861178,
+      "grad_norm": 0.8485874533653259,
+      "learning_rate": 9.011316282418474e-06,
+      "loss": 0.4128,
+      "step": 128
+    },
+    {
+      "epoch": 0.2470081378650072,
+      "grad_norm": 0.8879005908966064,
+      "learning_rate": 8.992330323186069e-06,
+      "loss": 0.4445,
+      "step": 129
+    },
+    {
+      "epoch": 0.24892292963140258,
+      "grad_norm": 0.9082502722740173,
+      "learning_rate": 8.973184201243922e-06,
+      "loss": 0.4825,
+      "step": 130
+    },
+    {
+      "epoch": 0.250837721397798,
+      "grad_norm": 0.8748418688774109,
+      "learning_rate": 8.953878684688492e-06,
+      "loss": 0.4209,
+      "step": 131
+    },
+    {
+      "epoch": 0.2527525131641934,
+      "grad_norm": 0.8313506245613098,
+      "learning_rate": 8.934414548010764e-06,
+      "loss": 0.4083,
+      "step": 132
+    },
+    {
+      "epoch": 0.2546673049305888,
+      "grad_norm": 0.8159312009811401,
+      "learning_rate": 8.914792572065178e-06,
+      "loss": 0.4166,
+      "step": 133
+    },
+    {
+      "epoch": 0.25658209669698423,
+      "grad_norm": 0.9007822275161743,
+      "learning_rate": 8.89501354403831e-06,
+      "loss": 0.4591,
+      "step": 134
+    },
+    {
+      "epoch": 0.2584968884633796,
+      "grad_norm": 0.9515888690948486,
+      "learning_rate": 8.875078257417294e-06,
+      "loss": 0.4659,
+      "step": 135
+    },
+    {
+      "epoch": 0.260411680229775,
+      "grad_norm": 0.8637048602104187,
+      "learning_rate": 8.854987511957974e-06,
+      "loss": 0.4387,
+      "step": 136
+    },
+    {
+      "epoch": 0.26232647199617043,
+      "grad_norm": 0.8459029793739319,
+      "learning_rate": 8.834742113652835e-06,
+      "loss": 0.4285,
+      "step": 137
+    },
+    {
+      "epoch": 0.26424126376256585,
+      "grad_norm": 0.7771206498146057,
+      "learning_rate": 8.81434287469866e-06,
+      "loss": 0.3803,
+      "step": 138
+    },
+    {
+      "epoch": 0.2661560555289612,
+      "grad_norm": 0.8040123581886292,
+      "learning_rate": 8.793790613463956e-06,
+      "loss": 0.411,
+      "step": 139
+    },
+    {
+      "epoch": 0.26807084729535663,
+      "grad_norm": 0.8000430464744568,
+      "learning_rate": 8.773086154456106e-06,
+      "loss": 0.4165,
+      "step": 140
+    },
+    {
+      "epoch": 0.26998563906175205,
+      "grad_norm": 0.9271606802940369,
+      "learning_rate": 8.752230328288314e-06,
+      "loss": 0.4755,
+      "step": 141
+    },
+    {
+      "epoch": 0.27190043082814747,
+      "grad_norm": 0.7970160841941833,
+      "learning_rate": 8.731223971646261e-06,
+      "loss": 0.3916,
+      "step": 142
+    },
+    {
+      "epoch": 0.27381522259454283,
+      "grad_norm": 0.7897955775260925,
+      "learning_rate": 8.710067927254555e-06,
+      "loss": 0.3846,
+      "step": 143
+    },
+    {
+      "epoch": 0.27573001436093825,
+      "grad_norm": 0.8004978895187378,
+      "learning_rate": 8.688763043842916e-06,
+      "loss": 0.4077,
+      "step": 144
+    },
+    {
+      "epoch": 0.27764480612733367,
+      "grad_norm": 0.8318747282028198,
+      "learning_rate": 8.66731017611213e-06,
+      "loss": 0.4334,
+      "step": 145
+    },
+    {
+      "epoch": 0.2795595978937291,
+      "grad_norm": 0.8479062914848328,
+      "learning_rate": 8.645710184699756e-06,
+      "loss": 0.4172,
+      "step": 146
+    },
+    {
+      "epoch": 0.28147438966012445,
+      "grad_norm": 0.8077636361122131,
+      "learning_rate": 8.6239639361456e-06,
+      "loss": 0.4104,
+      "step": 147
+    },
+    {
+      "epoch": 0.28338918142651986,
+      "grad_norm": 0.8098336458206177,
+      "learning_rate": 8.602072302856961e-06,
+      "loss": 0.406,
+      "step": 148
+    },
+    {
+      "epoch": 0.2853039731929153,
+      "grad_norm": 0.8435089588165283,
+      "learning_rate": 8.580036163073615e-06,
+      "loss": 0.4312,
+      "step": 149
+    },
+    {
+      "epoch": 0.2872187649593107,
+      "grad_norm": 0.8346257209777832,
+      "learning_rate": 8.5578564008326e-06,
+      "loss": 0.3888,
+      "step": 150
+    },
+    {
+      "epoch": 0.28913355672570606,
+      "grad_norm": 0.8100122213363647,
+      "learning_rate": 8.535533905932739e-06,
+      "loss": 0.4042,
+      "step": 151
+    },
+    {
+      "epoch": 0.2910483484921015,
+      "grad_norm": 0.8516131043434143,
+      "learning_rate": 8.513069573898944e-06,
+      "loss": 0.4161,
+      "step": 152
+    },
+    {
+      "epoch": 0.2929631402584969,
+      "grad_norm": 0.8350062966346741,
+      "learning_rate": 8.490464305946296e-06,
+      "loss": 0.4231,
+      "step": 153
+    },
+    {
+      "epoch": 0.2948779320248923,
+      "grad_norm": 0.8145307302474976,
+      "learning_rate": 8.467719008943886e-06,
+      "loss": 0.4125,
+      "step": 154
+    },
+    {
+      "epoch": 0.2967927237912877,
+      "grad_norm": 0.8515341281890869,
+      "learning_rate": 8.444834595378434e-06,
+      "loss": 0.4192,
+      "step": 155
+    },
+    {
+      "epoch": 0.2987075155576831,
+      "grad_norm": 0.8017379641532898,
+      "learning_rate": 8.421811983317682e-06,
+      "loss": 0.4007,
+      "step": 156
+    },
+    {
+      "epoch": 0.3006223073240785,
+      "grad_norm": 0.8179863691329956,
+      "learning_rate": 8.398652096373566e-06,
+      "loss": 0.4188,
+      "step": 157
+    },
+    {
+      "epoch": 0.30253709909047394,
+      "grad_norm": 0.8272606730461121,
+      "learning_rate": 8.375355863665155e-06,
+      "loss": 0.4035,
+      "step": 158
+    },
+    {
+      "epoch": 0.3044518908568693,
+      "grad_norm": 0.8340199589729309,
+      "learning_rate": 8.351924219781393e-06,
+      "loss": 0.4414,
+      "step": 159
+    },
+    {
+      "epoch": 0.3063666826232647,
+      "grad_norm": 0.82643723487854,
+      "learning_rate": 8.328358104743588e-06,
+      "loss": 0.4143,
+      "step": 160
+    },
+    {
+      "epoch": 0.30828147438966014,
+      "grad_norm": 0.7954344153404236,
+      "learning_rate": 8.304658463967705e-06,
+      "loss": 0.4128,
+      "step": 161
+    },
+    {
+      "epoch": 0.31019626615605556,
+      "grad_norm": 0.7908764481544495,
+      "learning_rate": 8.28082624822645e-06,
+      "loss": 0.3813,
+      "step": 162
+    },
+    {
+      "epoch": 0.3121110579224509,
+      "grad_norm": 0.7368812561035156,
+      "learning_rate": 8.256862413611113e-06,
+      "loss": 0.3879,
+      "step": 163
+    },
+    {
+      "epoch": 0.31402584968884634,
+      "grad_norm": 0.8204508423805237,
+      "learning_rate": 8.232767921493216e-06,
+      "loss": 0.4308,
+      "step": 164
+    },
+    {
+      "epoch": 0.31594064145524176,
+      "grad_norm": 0.8691006302833557,
+      "learning_rate": 8.20854373848595e-06,
+      "loss": 0.4518,
+      "step": 165
+    },
+    {
+      "epoch": 0.3178554332216372,
+      "grad_norm": 0.7581012845039368,
+      "learning_rate": 8.184190836405394e-06,
+      "loss": 0.3711,
+      "step": 166
+    },
+    {
+      "epoch": 0.31977022498803254,
+      "grad_norm": 0.7817586660385132,
+      "learning_rate": 8.15971019223152e-06,
+      "loss": 0.4057,
+      "step": 167
+    },
+    {
+      "epoch": 0.32168501675442795,
+      "grad_norm": 0.7559137344360352,
+      "learning_rate": 8.135102788069015e-06,
+      "loss": 0.361,
+      "step": 168
+    },
+    {
+      "epoch": 0.3235998085208234,
+      "grad_norm": 0.7424903512001038,
+      "learning_rate": 8.110369611107869e-06,
+      "loss": 0.3667,
+      "step": 169
+    },
+    {
+      "epoch": 0.3255146002872188,
+      "grad_norm": 0.7965791821479797,
+      "learning_rate": 8.085511653583772e-06,
+      "loss": 0.3814,
+      "step": 170
+    },
+    {
+      "epoch": 0.32742939205361415,
+      "grad_norm": 0.8504881262779236,
+      "learning_rate": 8.060529912738316e-06,
+      "loss": 0.4453,
+      "step": 171
+    },
+    {
+      "epoch": 0.3293441838200096,
+      "grad_norm": 0.8762128353118896,
+      "learning_rate": 8.035425390778975e-06,
+      "loss": 0.4516,
+      "step": 172
+    },
+    {
+      "epoch": 0.331258975586405,
+      "grad_norm": 0.8157339692115784,
+      "learning_rate": 8.010199094838915e-06,
+      "loss": 0.4204,
+      "step": 173
+    },
+    {
+      "epoch": 0.3331737673528004,
+      "grad_norm": 0.8493944406509399,
+      "learning_rate": 7.984852036936578e-06,
+      "loss": 0.39,
+      "step": 174
+    },
+    {
+      "epoch": 0.3350885591191958,
+      "grad_norm": 0.838901937007904,
+      "learning_rate": 7.959385233935087e-06,
+      "loss": 0.4417,
+      "step": 175
+    },
+    {
+      "epoch": 0.3370033508855912,
+      "grad_norm": 0.7574141621589661,
+      "learning_rate": 7.933799707501448e-06,
+      "loss": 0.3515,
+      "step": 176
+    },
+    {
+      "epoch": 0.3389181426519866,
+      "grad_norm": 0.8099052906036377,
+      "learning_rate": 7.908096484065569e-06,
+      "loss": 0.4091,
+      "step": 177
+    },
+    {
+      "epoch": 0.340832934418382,
+      "grad_norm": 0.8116398453712463,
+      "learning_rate": 7.88227659477908e-06,
+      "loss": 0.4137,
+      "step": 178
+    },
+    {
+      "epoch": 0.3427477261847774,
+      "grad_norm": 0.7785760164260864,
+      "learning_rate": 7.856341075473963e-06,
+      "loss": 0.3829,
+      "step": 179
+    },
+    {
+      "epoch": 0.3446625179511728,
+      "grad_norm": 0.7929257154464722,
+      "learning_rate": 7.830290966620997e-06,
+      "loss": 0.3739,
+      "step": 180
+    },
+    {
+      "epoch": 0.3465773097175682,
+      "grad_norm": 0.8796236515045166,
+      "learning_rate": 7.804127313288023e-06,
+      "loss": 0.4027,
+      "step": 181
+    },
+    {
+      "epoch": 0.34849210148396365,
+      "grad_norm": 0.8447411060333252,
+      "learning_rate": 7.777851165098012e-06,
+      "loss": 0.4202,
+      "step": 182
+    },
+    {
+      "epoch": 0.350406893250359,
+      "grad_norm": 0.7473250031471252,
+      "learning_rate": 7.751463576186957e-06,
+      "loss": 0.3777,
+      "step": 183
+    },
+    {
+      "epoch": 0.3523216850167544,
+      "grad_norm": 0.8254420757293701,
+      "learning_rate": 7.72496560516159e-06,
+      "loss": 0.4147,
+      "step": 184
+    },
+    {
+      "epoch": 0.35423647678314985,
+      "grad_norm": 0.8766903281211853,
+      "learning_rate": 7.6983583150569e-06,
+      "loss": 0.4212,
+      "step": 185
+    },
+    {
+      "epoch": 0.35615126854954526,
+      "grad_norm": 0.8106472492218018,
+      "learning_rate": 7.671642773293506e-06,
+      "loss": 0.3897,
+      "step": 186
+    },
+    {
+      "epoch": 0.3580660603159406,
+      "grad_norm": 0.8969345092773438,
+      "learning_rate": 7.644820051634813e-06,
+      "loss": 0.4166,
+      "step": 187
+    },
+    {
+      "epoch": 0.35998085208233604,
+      "grad_norm": 0.9210174679756165,
+      "learning_rate": 7.617891226144034e-06,
+      "loss": 0.4744,
+      "step": 188
+    },
+    {
+      "epoch": 0.36189564384873146,
+      "grad_norm": 0.8183197379112244,
+      "learning_rate": 7.59085737714101e-06,
+      "loss": 0.3909,
+      "step": 189
+    },
+    {
+      "epoch": 0.3638104356151269,
+      "grad_norm": 0.9258884191513062,
+      "learning_rate": 7.563719589158874e-06,
+      "loss": 0.4191,
+      "step": 190
+    },
+    {
+      "epoch": 0.36572522738152224,
+      "grad_norm": 0.8379626274108887,
+      "learning_rate": 7.536478950900537e-06,
+      "loss": 0.4081,
+      "step": 191
+    },
+    {
+      "epoch": 0.36764001914791766,
+      "grad_norm": 0.8194741010665894,
+      "learning_rate": 7.509136555195025e-06,
+      "loss": 0.3983,
+      "step": 192
+    },
+    {
+      "epoch": 0.3695548109143131,
+      "grad_norm": 0.8078919649124146,
+      "learning_rate": 7.481693498953621e-06,
+      "loss": 0.4117,
+      "step": 193
+    },
+    {
+      "epoch": 0.3714696026807085,
+      "grad_norm": 0.8059271574020386,
+      "learning_rate": 7.4541508831258695e-06,
+      "loss": 0.3919,
+      "step": 194
+    },
+    {
+      "epoch": 0.37338439444710386,
+      "grad_norm": 0.7913739681243896,
+      "learning_rate": 7.4265098126554065e-06,
+      "loss": 0.3786,
+      "step": 195
+    },
+    {
+      "epoch": 0.3752991862134993,
+      "grad_norm": 0.8568047285079956,
+      "learning_rate": 7.3987713964356335e-06,
+      "loss": 0.4505,
+      "step": 196
+    },
+    {
+      "epoch": 0.3772139779798947,
+      "grad_norm": 0.9244000315666199,
+      "learning_rate": 7.370936747265226e-06,
+      "loss": 0.4546,
+      "step": 197
+    },
+    {
+      "epoch": 0.3791287697462901,
+      "grad_norm": 0.7795801162719727,
+      "learning_rate": 7.3430069818035e-06,
+      "loss": 0.3955,
+      "step": 198
+    },
+    {
+      "epoch": 0.3810435615126855,
+      "grad_norm": 0.7940351366996765,
+      "learning_rate": 7.314983220525604e-06,
+      "loss": 0.4043,
+      "step": 199
+    },
+    {
+      "epoch": 0.3829583532790809,
+      "grad_norm": 0.7889821529388428,
+      "learning_rate": 7.286866587677576e-06,
+      "loss": 0.3883,
+      "step": 200
+    },
+    {
+      "epoch": 0.3848731450454763,
+      "grad_norm": 0.857402503490448,
+      "learning_rate": 7.2586582112312355e-06,
+      "loss": 0.4288,
+      "step": 201
+    },
+    {
+      "epoch": 0.38678793681187174,
+      "grad_norm": 0.7854819893836975,
+      "learning_rate": 7.230359222838939e-06,
+      "loss": 0.3884,
+      "step": 202
+    },
+    {
+      "epoch": 0.3887027285782671,
+      "grad_norm": 0.9061365723609924,
+      "learning_rate": 7.201970757788172e-06,
+      "loss": 0.458,
+      "step": 203
+    },
+    {
+      "epoch": 0.3906175203446625,
+      "grad_norm": 0.7978100776672363,
+      "learning_rate": 7.173493954956012e-06,
+      "loss": 0.3903,
+      "step": 204
+    },
+    {
+      "epoch": 0.39253231211105793,
+      "grad_norm": 0.8213270902633667,
+      "learning_rate": 7.144929956763438e-06,
+      "loss": 0.4035,
+      "step": 205
+    },
+    {
+      "epoch": 0.39444710387745335,
+      "grad_norm": 0.9091420769691467,
+      "learning_rate": 7.116279909129492e-06,
+      "loss": 0.4505,
+      "step": 206
+    },
+    {
+      "epoch": 0.3963618956438487,
+      "grad_norm": 0.8489376306533813,
+      "learning_rate": 7.087544961425317e-06,
+      "loss": 0.4037,
+      "step": 207
+    },
+    {
+      "epoch": 0.39827668741024413,
+      "grad_norm": 0.805548369884491,
+      "learning_rate": 7.058726266428042e-06,
+      "loss": 0.4053,
+      "step": 208
+    },
+    {
+      "epoch": 0.40019147917663955,
+      "grad_norm": 0.7785957455635071,
+      "learning_rate": 7.029824980274536e-06,
+      "loss": 0.3719,
+      "step": 209
+    },
+    {
+      "epoch": 0.40210627094303497,
+      "grad_norm": 0.8519039154052734,
+      "learning_rate": 7.0008422624150285e-06,
+      "loss": 0.4176,
+      "step": 210
+    },
+    {
+      "epoch": 0.40402106270943033,
+      "grad_norm": 0.8210786581039429,
+      "learning_rate": 6.971779275566593e-06,
+      "loss": 0.4162,
+      "step": 211
+    },
+    {
+      "epoch": 0.40593585447582575,
+      "grad_norm": 0.8121908903121948,
+      "learning_rate": 6.9426371856665005e-06,
+      "loss": 0.4204,
+      "step": 212
+    },
+    {
+      "epoch": 0.40785064624222117,
+      "grad_norm": 0.8359207510948181,
+      "learning_rate": 6.913417161825449e-06,
+      "loss": 0.4251,
+      "step": 213
+    },
+    {
+      "epoch": 0.4097654380086166,
+      "grad_norm": 0.8265076279640198,
+      "learning_rate": 6.884120376280658e-06,
+      "loss": 0.3983,
+      "step": 214
+    },
+    {
+      "epoch": 0.41168022977501195,
+      "grad_norm": 0.8334207534790039,
+      "learning_rate": 6.85474800434884e-06,
+      "loss": 0.4287,
+      "step": 215
+    },
+    {
+      "epoch": 0.41359502154140737,
+      "grad_norm": 0.7907735705375671,
+      "learning_rate": 6.8253012243790565e-06,
+      "loss": 0.4068,
+      "step": 216
+    },
+    {
+      "epoch": 0.4155098133078028,
+      "grad_norm": 0.8516376614570618,
+      "learning_rate": 6.795781217705436e-06,
+      "loss": 0.4531,
+      "step": 217
+    },
+    {
+      "epoch": 0.4174246050741982,
+      "grad_norm": 0.7719292640686035,
+      "learning_rate": 6.76618916859979e-06,
+      "loss": 0.3641,
+      "step": 218
+    },
+    {
+      "epoch": 0.41933939684059357,
+      "grad_norm": 0.769619882106781,
+      "learning_rate": 6.736526264224101e-06,
+      "loss": 0.389,
+      "step": 219
+    },
+    {
+      "epoch": 0.421254188606989,
+      "grad_norm": 0.8193688988685608,
+      "learning_rate": 6.706793694582892e-06,
+      "loss": 0.381,
+      "step": 220
+    },
+    {
+      "epoch": 0.4231689803733844,
+      "grad_norm": 0.8246089220046997,
+      "learning_rate": 6.676992652475487e-06,
+      "loss": 0.4105,
+      "step": 221
+    },
+    {
+      "epoch": 0.4250837721397798,
+      "grad_norm": 0.7855746746063232,
+      "learning_rate": 6.647124333448165e-06,
+      "loss": 0.3711,
+      "step": 222
+    },
+    {
+      "epoch": 0.4269985639061752,
+      "grad_norm": 0.8676920533180237,
+      "learning_rate": 6.617189935746191e-06,
+      "loss": 0.4071,
+      "step": 223
+    },
+    {
+      "epoch": 0.4289133556725706,
+      "grad_norm": 0.8358131647109985,
+      "learning_rate": 6.587190660265752e-06,
+      "loss": 0.425,
+      "step": 224
+    },
+    {
+      "epoch": 0.430828147438966,
+      "grad_norm": 0.8328396677970886,
+      "learning_rate": 6.55712771050577e-06,
+      "loss": 0.4156,
+      "step": 225
+    },
+    {
+      "epoch": 0.43274293920536144,
+      "grad_norm": 0.7680657505989075,
+      "learning_rate": 6.52700229251963e-06,
+      "loss": 0.393,
+      "step": 226
+    },
+    {
+      "epoch": 0.4346577309717568,
+      "grad_norm": 0.8950127959251404,
+      "learning_rate": 6.496815614866792e-06,
+      "loss": 0.4035,
+      "step": 227
+    },
+    {
+      "epoch": 0.4365725227381522,
+      "grad_norm": 0.8672182559967041,
+      "learning_rate": 6.466568888564303e-06,
+      "loss": 0.4581,
+      "step": 228
+    },
+    {
+      "epoch": 0.43848731450454764,
+      "grad_norm": 0.8580407500267029,
+      "learning_rate": 6.436263327038225e-06,
+      "loss": 0.4039,
+      "step": 229
+    },
+    {
+      "epoch": 0.44040210627094306,
+      "grad_norm": 0.8145782351493835,
+      "learning_rate": 6.405900146074941e-06,
+      "loss": 0.3873,
+      "step": 230
+    },
+    {
+      "epoch": 0.4423168980373384,
+      "grad_norm": 0.8403171896934509,
+      "learning_rate": 6.375480563772391e-06,
+      "loss": 0.4373,
+      "step": 231
+    },
+    {
+      "epoch": 0.44423168980373384,
+      "grad_norm": 0.7500009536743164,
+      "learning_rate": 6.3450058004912004e-06,
+      "loss": 0.3661,
+      "step": 232
+    },
+    {
+      "epoch": 0.44614648157012926,
+      "grad_norm": 0.8253698945045471,
+      "learning_rate": 6.314477078805724e-06,
+      "loss": 0.4005,
+      "step": 233
+    },
+    {
+      "epoch": 0.4480612733365247,
+      "grad_norm": 0.7586554884910583,
+      "learning_rate": 6.283895623454997e-06,
+      "loss": 0.3858,
+      "step": 234
+    },
+    {
+      "epoch": 0.44997606510292004,
+      "grad_norm": 0.8090248107910156,
+      "learning_rate": 6.2532626612936035e-06,
+      "loss": 0.4085,
+      "step": 235
+    },
+    {
+      "epoch": 0.45189085686931546,
+      "grad_norm": 0.8589561581611633,
+      "learning_rate": 6.2225794212424565e-06,
+      "loss": 0.4402,
+      "step": 236
+    },
+    {
+      "epoch": 0.4538056486357109,
+      "grad_norm": 0.8317797780036926,
+      "learning_rate": 6.191847134239496e-06,
+      "loss": 0.4,
+      "step": 237
+    },
+    {
+      "epoch": 0.4557204404021063,
+      "grad_norm": 0.8422788381576538,
+      "learning_rate": 6.161067033190311e-06,
+      "loss": 0.4026,
+      "step": 238
+    },
+    {
+      "epoch": 0.45763523216850166,
+      "grad_norm": 0.763214647769928,
+      "learning_rate": 6.130240352918675e-06,
+      "loss": 0.3976,
+      "step": 239
+    },
+    {
+      "epoch": 0.4595500239348971,
+      "grad_norm": 0.8719058036804199,
+      "learning_rate": 6.0993683301170046e-06,
+      "loss": 0.4356,
+      "step": 240
+    },
+    {
+      "epoch": 0.4614648157012925,
+      "grad_norm": 0.7786772847175598,
+      "learning_rate": 6.068452203296754e-06,
+      "loss": 0.3858,
+      "step": 241
+    },
+    {
+      "epoch": 0.4633796074676879,
+      "grad_norm": 0.7639849185943604,
+      "learning_rate": 6.0374932127387234e-06,
+      "loss": 0.3936,
+      "step": 242
+    },
+    {
+      "epoch": 0.4652943992340833,
+      "grad_norm": 0.8484832644462585,
+      "learning_rate": 6.006492600443301e-06,
+      "loss": 0.4006,
+      "step": 243
+    },
+    {
+      "epoch": 0.4672091910004787,
+      "grad_norm": 0.835634708404541,
+      "learning_rate": 5.975451610080643e-06,
+      "loss": 0.3826,
+      "step": 244
+    },
+    {
+      "epoch": 0.4691239827668741,
+      "grad_norm": 0.8132709860801697,
+      "learning_rate": 5.944371486940772e-06,
+      "loss": 0.3934,
+      "step": 245
+    },
+    {
+      "epoch": 0.47103877453326953,
+      "grad_norm": 0.93300861120224,
+      "learning_rate": 5.913253477883629e-06,
+      "loss": 0.4376,
+      "step": 246
+    },
+    {
+      "epoch": 0.4729535662996649,
+      "grad_norm": 0.8036344647407532,
+      "learning_rate": 5.882098831289044e-06,
+      "loss": 0.3892,
+      "step": 247
+    },
+    {
+      "epoch": 0.4748683580660603,
+      "grad_norm": 0.7971835732460022,
+      "learning_rate": 5.850908797006656e-06,
+      "loss": 0.399,
+      "step": 248
+    },
+    {
+      "epoch": 0.47678314983245573,
+      "grad_norm": 0.8496800065040588,
+      "learning_rate": 5.819684626305776e-06,
+      "loss": 0.4385,
+      "step": 249
+    },
+    {
+      "epoch": 0.47869794159885115,
+      "grad_norm": 0.8178644776344299,
+      "learning_rate": 5.788427571825186e-06,
+      "loss": 0.3946,
+      "step": 250
+    },
+    {
+      "epoch": 0.4806127333652465,
+      "grad_norm": 0.8840231895446777,
+      "learning_rate": 5.757138887522884e-06,
+      "loss": 0.4125,
+      "step": 251
+    },
+    {
+      "epoch": 0.48252752513164193,
+      "grad_norm": 0.831671416759491,
+      "learning_rate": 5.725819828625782e-06,
+      "loss": 0.413,
+      "step": 252
+    },
+    {
+      "epoch": 0.48444231689803735,
+      "grad_norm": 0.7894381880760193,
+      "learning_rate": 5.694471651579346e-06,
+      "loss": 0.399,
+      "step": 253
+    },
+    {
+      "epoch": 0.48635710866443277,
+      "grad_norm": 0.8008851408958435,
+      "learning_rate": 5.663095613997196e-06,
+      "loss": 0.3875,
+      "step": 254
+    },
+    {
+      "epoch": 0.48827190043082813,
+      "grad_norm": 0.7766909003257751,
+      "learning_rate": 5.631692974610647e-06,
+      "loss": 0.3755,
+      "step": 255
+    },
+    {
+      "epoch": 0.49018669219722355,
+      "grad_norm": 0.8729162812232971,
+      "learning_rate": 5.600264993218215e-06,
+      "loss": 0.4103,
+      "step": 256
+    },
+    {
+      "epoch": 0.49210148396361897,
+      "grad_norm": 0.8618749380111694,
+      "learning_rate": 5.568812930635076e-06,
+      "loss": 0.3963,
+      "step": 257
+    },
+    {
+      "epoch": 0.4940162757300144,
+      "grad_norm": 0.8580813407897949,
+      "learning_rate": 5.537338048642487e-06,
+      "loss": 0.4353,
+      "step": 258
+    },
+    {
+      "epoch": 0.49593106749640975,
+      "grad_norm": 0.8418040871620178,
+      "learning_rate": 5.505841609937162e-06,
+      "loss": 0.3801,
+      "step": 259
+    },
+    {
+      "epoch": 0.49784585926280517,
+      "grad_norm": 0.7449955940246582,
+      "learning_rate": 5.474324878080623e-06,
+      "loss": 0.3355,
+      "step": 260
+    },
+    {
+      "epoch": 0.4997606510292006,
+      "grad_norm": 0.8238184452056885,
+      "learning_rate": 5.4427891174485014e-06,
+      "loss": 0.3868,
+      "step": 261
+    },
+    {
+      "epoch": 0.501675442795596,
+      "grad_norm": 0.8558638691902161,
+      "learning_rate": 5.41123559317982e-06,
+      "loss": 0.4149,
+      "step": 262
+    },
+    {
+      "epoch": 0.5035902345619914,
+      "grad_norm": 0.836510956287384,
+      "learning_rate": 5.379665571126232e-06,
+      "loss": 0.378,
+      "step": 263
+    },
+    {
+      "epoch": 0.5055050263283868,
+      "grad_norm": 0.7436261177062988,
+      "learning_rate": 5.348080317801244e-06,
+      "loss": 0.3672,
+      "step": 264
+    },
+    {
+      "epoch": 0.5074198180947822,
+      "grad_norm": 0.897144079208374,
+      "learning_rate": 5.316481100329408e-06,
+      "loss": 0.4309,
+      "step": 265
+    },
+    {
+      "epoch": 0.5093346098611776,
+      "grad_norm": 0.8236995935440063,
+      "learning_rate": 5.284869186395478e-06,
+      "loss": 0.4164,
+      "step": 266
+    },
+    {
+      "epoch": 0.511249401627573,
+      "grad_norm": 0.8249524831771851,
+      "learning_rate": 5.253245844193564e-06,
+      "loss": 0.408,
+      "step": 267
+    },
+    {
+      "epoch": 0.5131641933939685,
+      "grad_norm": 0.8149250745773315,
+      "learning_rate": 5.22161234237625e-06,
+      "loss": 0.4014,
+      "step": 268
+    },
+    {
+      "epoch": 0.5150789851603638,
+      "grad_norm": 0.8144116401672363,
+      "learning_rate": 5.189969950003697e-06,
+      "loss": 0.4013,
+      "step": 269
+    },
+    {
+      "epoch": 0.5169937769267592,
+      "grad_norm": 0.8883258700370789,
+      "learning_rate": 5.158319936492736e-06,
+      "loss": 0.4262,
+      "step": 270
+    },
+    {
+      "epoch": 0.5189085686931546,
+      "grad_norm": 0.7828565239906311,
+      "learning_rate": 5.12666357156594e-06,
+      "loss": 0.3877,
+      "step": 271
+    },
+    {
+      "epoch": 0.52082336045955,
+      "grad_norm": 0.8680728673934937,
+      "learning_rate": 5.0950021252006845e-06,
+      "loss": 0.4288,
+      "step": 272
+    },
+    {
+      "epoch": 0.5227381522259454,
+      "grad_norm": 0.8556041121482849,
+      "learning_rate": 5.063336867578201e-06,
+      "loss": 0.4397,
+      "step": 273
+    },
+    {
+      "epoch": 0.5246529439923409,
+      "grad_norm": 0.83294278383255,
+      "learning_rate": 5.0316690690326175e-06,
+      "loss": 0.3855,
+      "step": 274
+    },
+    {
+      "epoch": 0.5265677357587363,
+      "grad_norm": 0.8231511116027832,
+      "learning_rate": 5e-06,
+      "loss": 0.3712,
+      "step": 275
+    },
+    {
+      "epoch": 0.5284825275251317,
+      "grad_norm": 0.8193358182907104,
+      "learning_rate": 4.9683309309673825e-06,
+      "loss": 0.3833,
+      "step": 276
+    },
+    {
+      "epoch": 0.530397319291527,
+      "grad_norm": 0.7521802186965942,
+      "learning_rate": 4.936663132421801e-06,
+      "loss": 0.3668,
+      "step": 277
+    },
+    {
+      "epoch": 0.5323121110579224,
+      "grad_norm": 0.7623764872550964,
+      "learning_rate": 4.904997874799316e-06,
+      "loss": 0.3825,
+      "step": 278
+    },
+    {
+      "epoch": 0.5342269028243178,
+      "grad_norm": 0.8101343512535095,
+      "learning_rate": 4.873336428434062e-06,
+      "loss": 0.3864,
+      "step": 279
+    },
+    {
+      "epoch": 0.5361416945907133,
+      "grad_norm": 0.8006393909454346,
+      "learning_rate": 4.841680063507265e-06,
+      "loss": 0.4227,
+      "step": 280
+    },
+    {
+      "epoch": 0.5380564863571087,
+      "grad_norm": 0.8386540412902832,
+      "learning_rate": 4.8100300499963045e-06,
+      "loss": 0.4131,
+      "step": 281
+    },
+    {
+      "epoch": 0.5399712781235041,
+      "grad_norm": 0.7718150019645691,
+      "learning_rate": 4.778387657623751e-06,
+      "loss": 0.3768,
+      "step": 282
+    },
+    {
+      "epoch": 0.5418860698898995,
+      "grad_norm": 0.8466550707817078,
+      "learning_rate": 4.746754155806437e-06,
+      "loss": 0.4512,
+      "step": 283
+    },
+    {
+      "epoch": 0.5438008616562949,
+      "grad_norm": 0.8129767775535583,
+      "learning_rate": 4.715130813604522e-06,
+      "loss": 0.3961,
+      "step": 284
+    },
+    {
+      "epoch": 0.5457156534226902,
+      "grad_norm": 0.8341888189315796,
+      "learning_rate": 4.683518899670594e-06,
+      "loss": 0.3919,
+      "step": 285
+    },
+    {
+      "epoch": 0.5476304451890857,
+      "grad_norm": 0.8497266173362732,
+      "learning_rate": 4.651919682198756e-06,
+      "loss": 0.394,
+      "step": 286
+    },
+    {
+      "epoch": 0.5495452369554811,
+      "grad_norm": 0.8146570920944214,
+      "learning_rate": 4.62033442887377e-06,
+      "loss": 0.3989,
+      "step": 287
+    },
+    {
+      "epoch": 0.5514600287218765,
+      "grad_norm": 0.806999921798706,
+      "learning_rate": 4.588764406820181e-06,
+      "loss": 0.4288,
+      "step": 288
+    },
+    {
+      "epoch": 0.5533748204882719,
+      "grad_norm": 0.7479027509689331,
+      "learning_rate": 4.5572108825515e-06,
+      "loss": 0.3616,
+      "step": 289
+    },
+    {
+      "epoch": 0.5552896122546673,
+      "grad_norm": 0.7727148532867432,
+      "learning_rate": 4.5256751219193784e-06,
+      "loss": 0.3902,
+      "step": 290
+    },
+    {
+      "epoch": 0.5572044040210627,
+      "grad_norm": 0.8226932883262634,
+      "learning_rate": 4.49415839006284e-06,
+      "loss": 0.4038,
+      "step": 291
+    },
+    {
+      "epoch": 0.5591191957874582,
+      "grad_norm": 0.8253747224807739,
+      "learning_rate": 4.462661951357515e-06,
+      "loss": 0.4052,
+      "step": 292
+    },
+    {
+      "epoch": 0.5610339875538535,
+      "grad_norm": 0.8564761877059937,
+      "learning_rate": 4.431187069364927e-06,
+      "loss": 0.4099,
+      "step": 293
+    },
+    {
+      "epoch": 0.5629487793202489,
+      "grad_norm": 0.812239408493042,
+      "learning_rate": 4.3997350067817866e-06,
+      "loss": 0.3939,
+      "step": 294
+    },
+    {
+      "epoch": 0.5648635710866443,
+      "grad_norm": 0.8617464303970337,
+      "learning_rate": 4.368307025389355e-06,
+      "loss": 0.4189,
+      "step": 295
+    },
+    {
+      "epoch": 0.5667783628530397,
+      "grad_norm": 0.8013660907745361,
+      "learning_rate": 4.336904386002805e-06,
+      "loss": 0.3853,
+      "step": 296
+    },
+    {
+      "epoch": 0.5686931546194351,
+      "grad_norm": 0.8232107162475586,
+      "learning_rate": 4.3055283484206565e-06,
+      "loss": 0.4228,
+      "step": 297
+    },
+    {
+      "epoch": 0.5706079463858306,
+      "grad_norm": 0.7978037595748901,
+      "learning_rate": 4.27418017137422e-06,
+      "loss": 0.3745,
+      "step": 298
+    },
+    {
+      "epoch": 0.572522738152226,
+      "grad_norm": 0.786128044128418,
+      "learning_rate": 4.2428611124771184e-06,
+      "loss": 0.3708,
+      "step": 299
+    },
+    {
+      "epoch": 0.5744375299186214,
+      "grad_norm": 0.7764582633972168,
+      "learning_rate": 4.211572428174816e-06,
+      "loss": 0.3617,
+      "step": 300
+    },
+    {
+      "epoch": 0.5763523216850167,
+      "grad_norm": 0.8071704506874084,
+      "learning_rate": 4.180315373694225e-06,
+      "loss": 0.4013,
+      "step": 301
+    },
+    {
+      "epoch": 0.5782671134514121,
+      "grad_norm": 0.8064809441566467,
+      "learning_rate": 4.149091202993345e-06,
+      "loss": 0.3593,
+      "step": 302
+    },
+    {
+      "epoch": 0.5801819052178075,
+      "grad_norm": 0.8199291229248047,
+      "learning_rate": 4.11790116871096e-06,
+      "loss": 0.4174,
+      "step": 303
+    },
+    {
+      "epoch": 0.582096696984203,
+      "grad_norm": 0.9001505374908447,
+      "learning_rate": 4.086746522116372e-06,
+      "loss": 0.4535,
+      "step": 304
+    },
+    {
+      "epoch": 0.5840114887505984,
+      "grad_norm": 0.7408649325370789,
+      "learning_rate": 4.055628513059231e-06,
+      "loss": 0.3874,
+      "step": 305
+    },
+    {
+      "epoch": 0.5859262805169938,
+      "grad_norm": 0.8281508684158325,
+      "learning_rate": 4.02454838991936e-06,
+      "loss": 0.3783,
+      "step": 306
+    },
+    {
+      "epoch": 0.5878410722833892,
+      "grad_norm": 0.8515765070915222,
+      "learning_rate": 3.993507399556699e-06,
+      "loss": 0.4306,
+      "step": 307
+    },
+    {
+      "epoch": 0.5897558640497846,
+      "grad_norm": 0.8016983270645142,
+      "learning_rate": 3.962506787261278e-06,
+      "loss": 0.3988,
+      "step": 308
+    },
+    {
+      "epoch": 0.59167065581618,
+      "grad_norm": 0.7859688997268677,
+      "learning_rate": 3.931547796703245e-06,
+      "loss": 0.3801,
+      "step": 309
+    },
+    {
+      "epoch": 0.5935854475825754,
+      "grad_norm": 0.8068103194236755,
+      "learning_rate": 3.900631669882996e-06,
+      "loss": 0.3925,
+      "step": 310
+    },
+    {
+      "epoch": 0.5955002393489708,
+      "grad_norm": 0.811817467212677,
+      "learning_rate": 3.869759647081326e-06,
+      "loss": 0.3693,
+      "step": 311
+    },
+    {
+      "epoch": 0.5974150311153662,
+      "grad_norm": 0.7820096611976624,
+      "learning_rate": 3.83893296680969e-06,
+      "loss": 0.3837,
+      "step": 312
+    },
+    {
+      "epoch": 0.5993298228817616,
+      "grad_norm": 0.7839851379394531,
+      "learning_rate": 3.8081528657605045e-06,
+      "loss": 0.3762,
+      "step": 313
+    },
+    {
+      "epoch": 0.601244614648157,
+      "grad_norm": 0.7672075033187866,
+      "learning_rate": 3.7774205787575455e-06,
+      "loss": 0.3872,
+      "step": 314
+    },
+    {
+      "epoch": 0.6031594064145525,
+      "grad_norm": 0.8048450946807861,
+      "learning_rate": 3.7467373387063973e-06,
+      "loss": 0.4239,
+      "step": 315
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 522,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 105,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.6526141583628698e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-315/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66ffb9355835ef1e788c4b73fef523b9594f231dd9bde187500393bdf3018899
+size 10936

checkpoint-315/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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 _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+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() if _has_callable(shape, 'numel') else math.prod(shape)
+            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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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 tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

checkpoint-420/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-420/config.json

@@ -0,0 +1,27 @@
+{
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "eos_token_id": 151645,
+  "hidden_act": "silu",
+  "hidden_size": 1536,
+  "initializer_range": 0.02,
+  "intermediate_size": 8960,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 21,
+  "model_type": "qwen2",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

checkpoint-420/generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.1,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.51.3"
+}

checkpoint-420/latest

@@ -0,0 +1 @@
+global_step420

checkpoint-420/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:104b680822dddde737cee0c31b472c44542d6f82418d9197f380cd5eca875cc6
+size 3554214752

checkpoint-420/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad792af33c7cfa8b15298ecc9d976ebdcdeb444ca0e704c7b0657f41ee6547eb
+size 14512

checkpoint-420/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:722c924fceffd85f8ab1a5445f1ea1e6c502644b6a42e2ff6b5a9a76ea26e1fe
+size 14512

checkpoint-420/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8053bdf2e6680f6d7e7620f3d7ecd8cf15c34074cc261de25bfc326ba659e816
+size 1064

checkpoint-420/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-420/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c5ae00e602b8860cbd784ba82a8aa14e8feecec692e7076590d014d7b7fdafa
+size 11421896

checkpoint-420/tokenizer_config.json

@@ -0,0 +1,208 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-420/trainer_state.json

@@ -0,0 +1,2974 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.8042125418860699,
+  "eval_steps": 500,
+  "global_step": 420,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0019147917663954045,
+      "grad_norm": 2.4612040519714355,
+      "learning_rate": 0.0,
+      "loss": 0.6068,
+      "step": 1
+    },
+    {
+      "epoch": 0.003829583532790809,
+      "grad_norm": 2.5184295177459717,
+      "learning_rate": 3.846153846153847e-07,
+      "loss": 0.5979,
+      "step": 2
+    },
+    {
+      "epoch": 0.0057443752991862135,
+      "grad_norm": 2.515437602996826,
+      "learning_rate": 7.692307692307694e-07,
+      "loss": 0.6487,
+      "step": 3
+    },
+    {
+      "epoch": 0.007659167065581618,
+      "grad_norm": 2.2364346981048584,
+      "learning_rate": 1.153846153846154e-06,
+      "loss": 0.6041,
+      "step": 4
+    },
+    {
+      "epoch": 0.009573958831977022,
+      "grad_norm": 2.1685123443603516,
+      "learning_rate": 1.5384615384615387e-06,
+      "loss": 0.5414,
+      "step": 5
+    },
+    {
+      "epoch": 0.011488750598372427,
+      "grad_norm": 2.6281416416168213,
+      "learning_rate": 1.9230769230769234e-06,
+      "loss": 0.5794,
+      "step": 6
+    },
+    {
+      "epoch": 0.013403542364767831,
+      "grad_norm": 2.0852179527282715,
+      "learning_rate": 2.307692307692308e-06,
+      "loss": 0.5518,
+      "step": 7
+    },
+    {
+      "epoch": 0.015318334131163236,
+      "grad_norm": 1.9601969718933105,
+      "learning_rate": 2.6923076923076923e-06,
+      "loss": 0.557,
+      "step": 8
+    },
+    {
+      "epoch": 0.01723312589755864,
+      "grad_norm": 1.822036862373352,
+      "learning_rate": 3.0769230769230774e-06,
+      "loss": 0.5732,
+      "step": 9
+    },
+    {
+      "epoch": 0.019147917663954045,
+      "grad_norm": 1.6944836378097534,
+      "learning_rate": 3.4615384615384617e-06,
+      "loss": 0.5451,
+      "step": 10
+    },
+    {
+      "epoch": 0.02106270943034945,
+      "grad_norm": 1.521175503730774,
+      "learning_rate": 3.846153846153847e-06,
+      "loss": 0.5148,
+      "step": 11
+    },
+    {
+      "epoch": 0.022977501196744854,
+      "grad_norm": 1.573475956916809,
+      "learning_rate": 4.230769230769231e-06,
+      "loss": 0.512,
+      "step": 12
+    },
+    {
+      "epoch": 0.02489229296314026,
+      "grad_norm": 1.4469544887542725,
+      "learning_rate": 4.615384615384616e-06,
+      "loss": 0.5542,
+      "step": 13
+    },
+    {
+      "epoch": 0.026807084729535663,
+      "grad_norm": 1.3531206846237183,
+      "learning_rate": 5e-06,
+      "loss": 0.4868,
+      "step": 14
+    },
+    {
+      "epoch": 0.028721876495931067,
+      "grad_norm": 1.3297739028930664,
+      "learning_rate": 5.384615384615385e-06,
+      "loss": 0.4713,
+      "step": 15
+    },
+    {
+      "epoch": 0.030636668262326472,
+      "grad_norm": 1.430997610092163,
+      "learning_rate": 5.769230769230769e-06,
+      "loss": 0.4683,
+      "step": 16
+    },
+    {
+      "epoch": 0.032551460028721876,
+      "grad_norm": 1.2832906246185303,
+      "learning_rate": 6.153846153846155e-06,
+      "loss": 0.4275,
+      "step": 17
+    },
+    {
+      "epoch": 0.03446625179511728,
+      "grad_norm": 1.1449981927871704,
+      "learning_rate": 6.538461538461539e-06,
+      "loss": 0.4692,
+      "step": 18
+    },
+    {
+      "epoch": 0.036381043561512685,
+      "grad_norm": 1.1069403886795044,
+      "learning_rate": 6.923076923076923e-06,
+      "loss": 0.4925,
+      "step": 19
+    },
+    {
+      "epoch": 0.03829583532790809,
+      "grad_norm": 1.066596508026123,
+      "learning_rate": 7.307692307692308e-06,
+      "loss": 0.4809,
+      "step": 20
+    },
+    {
+      "epoch": 0.040210627094303494,
+      "grad_norm": 1.0530707836151123,
+      "learning_rate": 7.692307692307694e-06,
+      "loss": 0.4645,
+      "step": 21
+    },
+    {
+      "epoch": 0.0421254188606989,
+      "grad_norm": 1.0063157081604004,
+      "learning_rate": 8.076923076923077e-06,
+      "loss": 0.465,
+      "step": 22
+    },
+    {
+      "epoch": 0.0440402106270943,
+      "grad_norm": 1.1088693141937256,
+      "learning_rate": 8.461538461538462e-06,
+      "loss": 0.4824,
+      "step": 23
+    },
+    {
+      "epoch": 0.04595500239348971,
+      "grad_norm": 1.0253574848175049,
+      "learning_rate": 8.846153846153847e-06,
+      "loss": 0.4559,
+      "step": 24
+    },
+    {
+      "epoch": 0.04786979415988511,
+      "grad_norm": 1.0317028760910034,
+      "learning_rate": 9.230769230769232e-06,
+      "loss": 0.4687,
+      "step": 25
+    },
+    {
+      "epoch": 0.04978458592628052,
+      "grad_norm": 0.9329833388328552,
+      "learning_rate": 9.615384615384616e-06,
+      "loss": 0.4785,
+      "step": 26
+    },
+    {
+      "epoch": 0.05169937769267592,
+      "grad_norm": 0.9087225794792175,
+      "learning_rate": 1e-05,
+      "loss": 0.4505,
+      "step": 27
+    },
+    {
+      "epoch": 0.053614169459071326,
+      "grad_norm": 1.0503875017166138,
+      "learning_rate": 9.999899706000774e-06,
+      "loss": 0.4866,
+      "step": 28
+    },
+    {
+      "epoch": 0.05552896122546673,
+      "grad_norm": 0.9580711722373962,
+      "learning_rate": 9.999598828026644e-06,
+      "loss": 0.4759,
+      "step": 29
+    },
+    {
+      "epoch": 0.057443752991862135,
+      "grad_norm": 0.8906879425048828,
+      "learning_rate": 9.999097378148116e-06,
+      "loss": 0.4455,
+      "step": 30
+    },
+    {
+      "epoch": 0.05935854475825754,
+      "grad_norm": 0.8807913064956665,
+      "learning_rate": 9.998395376482152e-06,
+      "loss": 0.4335,
+      "step": 31
+    },
+    {
+      "epoch": 0.061273336524652944,
+      "grad_norm": 0.8745759129524231,
+      "learning_rate": 9.99749285119138e-06,
+      "loss": 0.43,
+      "step": 32
+    },
+    {
+      "epoch": 0.06318812829104835,
+      "grad_norm": 0.9661962389945984,
+      "learning_rate": 9.996389838482942e-06,
+      "loss": 0.5295,
+      "step": 33
+    },
+    {
+      "epoch": 0.06510292005744375,
+      "grad_norm": 0.9019286036491394,
+      "learning_rate": 9.995086382607064e-06,
+      "loss": 0.4774,
+      "step": 34
+    },
+    {
+      "epoch": 0.06701771182383916,
+      "grad_norm": 0.8906053900718689,
+      "learning_rate": 9.993582535855265e-06,
+      "loss": 0.4577,
+      "step": 35
+    },
+    {
+      "epoch": 0.06893250359023456,
+      "grad_norm": 0.8857560753822327,
+      "learning_rate": 9.991878358558267e-06,
+      "loss": 0.4785,
+      "step": 36
+    },
+    {
+      "epoch": 0.07084729535662997,
+      "grad_norm": 0.9806991219520569,
+      "learning_rate": 9.989973919083576e-06,
+      "loss": 0.4662,
+      "step": 37
+    },
+    {
+      "epoch": 0.07276208712302537,
+      "grad_norm": 0.9356399774551392,
+      "learning_rate": 9.987869293832727e-06,
+      "loss": 0.4669,
+      "step": 38
+    },
+    {
+      "epoch": 0.07467687888942078,
+      "grad_norm": 0.7924108505249023,
+      "learning_rate": 9.985564567238237e-06,
+      "loss": 0.4451,
+      "step": 39
+    },
+    {
+      "epoch": 0.07659167065581618,
+      "grad_norm": 0.9577414393424988,
+      "learning_rate": 9.983059831760205e-06,
+      "loss": 0.4837,
+      "step": 40
+    },
+    {
+      "epoch": 0.07850646242221158,
+      "grad_norm": 0.8300902843475342,
+      "learning_rate": 9.980355187882606e-06,
+      "loss": 0.4438,
+      "step": 41
+    },
+    {
+      "epoch": 0.08042125418860699,
+      "grad_norm": 0.8420023918151855,
+      "learning_rate": 9.977450744109258e-06,
+      "loss": 0.4223,
+      "step": 42
+    },
+    {
+      "epoch": 0.0823360459550024,
+      "grad_norm": 0.8297982811927795,
+      "learning_rate": 9.974346616959476e-06,
+      "loss": 0.4366,
+      "step": 43
+    },
+    {
+      "epoch": 0.0842508377213978,
+      "grad_norm": 0.9187960624694824,
+      "learning_rate": 9.97104293096339e-06,
+      "loss": 0.4739,
+      "step": 44
+    },
+    {
+      "epoch": 0.0861656294877932,
+      "grad_norm": 0.9509177803993225,
+      "learning_rate": 9.967539818656953e-06,
+      "loss": 0.4587,
+      "step": 45
+    },
+    {
+      "epoch": 0.0880804212541886,
+      "grad_norm": 0.9072842001914978,
+      "learning_rate": 9.96383742057662e-06,
+      "loss": 0.5177,
+      "step": 46
+    },
+    {
+      "epoch": 0.08999521302058401,
+      "grad_norm": 0.8744245767593384,
+      "learning_rate": 9.959935885253715e-06,
+      "loss": 0.4454,
+      "step": 47
+    },
+    {
+      "epoch": 0.09191000478697942,
+      "grad_norm": 0.8209521174430847,
+      "learning_rate": 9.955835369208475e-06,
+      "loss": 0.4243,
+      "step": 48
+    },
+    {
+      "epoch": 0.09382479655337482,
+      "grad_norm": 0.793670654296875,
+      "learning_rate": 9.951536036943753e-06,
+      "loss": 0.4256,
+      "step": 49
+    },
+    {
+      "epoch": 0.09573958831977022,
+      "grad_norm": 0.8487685322761536,
+      "learning_rate": 9.94703806093845e-06,
+      "loss": 0.4625,
+      "step": 50
+    },
+    {
+      "epoch": 0.09765438008616563,
+      "grad_norm": 0.8209853768348694,
+      "learning_rate": 9.942341621640558e-06,
+      "loss": 0.4385,
+      "step": 51
+    },
+    {
+      "epoch": 0.09956917185256103,
+      "grad_norm": 0.8096733689308167,
+      "learning_rate": 9.937446907459954e-06,
+      "loss": 0.4575,
+      "step": 52
+    },
+    {
+      "epoch": 0.10148396361895644,
+      "grad_norm": 0.8126389384269714,
+      "learning_rate": 9.932354114760819e-06,
+      "loss": 0.4262,
+      "step": 53
+    },
+    {
+      "epoch": 0.10339875538535184,
+      "grad_norm": 0.968154788017273,
+      "learning_rate": 9.92706344785377e-06,
+      "loss": 0.5304,
+      "step": 54
+    },
+    {
+      "epoch": 0.10531354715174725,
+      "grad_norm": 0.7662584781646729,
+      "learning_rate": 9.921575118987672e-06,
+      "loss": 0.4076,
+      "step": 55
+    },
+    {
+      "epoch": 0.10722833891814265,
+      "grad_norm": 0.8463670015335083,
+      "learning_rate": 9.915889348341098e-06,
+      "loss": 0.4434,
+      "step": 56
+    },
+    {
+      "epoch": 0.10914313068453806,
+      "grad_norm": 0.8167222142219543,
+      "learning_rate": 9.910006364013522e-06,
+      "loss": 0.4079,
+      "step": 57
+    },
+    {
+      "epoch": 0.11105792245093346,
+      "grad_norm": 0.8205484747886658,
+      "learning_rate": 9.903926402016153e-06,
+      "loss": 0.4167,
+      "step": 58
+    },
+    {
+      "epoch": 0.11297271421732887,
+      "grad_norm": 0.901168942451477,
+      "learning_rate": 9.897649706262474e-06,
+      "loss": 0.4762,
+      "step": 59
+    },
+    {
+      "epoch": 0.11488750598372427,
+      "grad_norm": 0.8437636494636536,
+      "learning_rate": 9.891176528558451e-06,
+      "loss": 0.4327,
+      "step": 60
+    },
+    {
+      "epoch": 0.11680229775011967,
+      "grad_norm": 0.8409866094589233,
+      "learning_rate": 9.884507128592435e-06,
+      "loss": 0.446,
+      "step": 61
+    },
+    {
+      "epoch": 0.11871708951651508,
+      "grad_norm": 0.8613469004631042,
+      "learning_rate": 9.877641773924748e-06,
+      "loss": 0.4217,
+      "step": 62
+    },
+    {
+      "epoch": 0.12063188128291048,
+      "grad_norm": 0.8663684129714966,
+      "learning_rate": 9.870580739976936e-06,
+      "loss": 0.4211,
+      "step": 63
+    },
+    {
+      "epoch": 0.12254667304930589,
+      "grad_norm": 0.8156937956809998,
+      "learning_rate": 9.863324310020735e-06,
+      "loss": 0.4273,
+      "step": 64
+    },
+    {
+      "epoch": 0.12446146481570129,
+      "grad_norm": 0.8753077387809753,
+      "learning_rate": 9.855872775166696e-06,
+      "loss": 0.4663,
+      "step": 65
+    },
+    {
+      "epoch": 0.1263762565820967,
+      "grad_norm": 0.81593918800354,
+      "learning_rate": 9.848226434352513e-06,
+      "loss": 0.4398,
+      "step": 66
+    },
+    {
+      "epoch": 0.12829104834849211,
+      "grad_norm": 0.8915799260139465,
+      "learning_rate": 9.840385594331022e-06,
+      "loss": 0.4758,
+      "step": 67
+    },
+    {
+      "epoch": 0.1302058401148875,
+      "grad_norm": 0.8952628374099731,
+      "learning_rate": 9.83235056965791e-06,
+      "loss": 0.4884,
+      "step": 68
+    },
+    {
+      "epoch": 0.13212063188128292,
+      "grad_norm": 0.874907910823822,
+      "learning_rate": 9.824121682679072e-06,
+      "loss": 0.4411,
+      "step": 69
+    },
+    {
+      "epoch": 0.13403542364767831,
+      "grad_norm": 0.8349279761314392,
+      "learning_rate": 9.815699263517712e-06,
+      "loss": 0.438,
+      "step": 70
+    },
+    {
+      "epoch": 0.13595021541407373,
+      "grad_norm": 0.8123260736465454,
+      "learning_rate": 9.807083650061063e-06,
+      "loss": 0.4483,
+      "step": 71
+    },
+    {
+      "epoch": 0.13786500718046912,
+      "grad_norm": 0.8606418371200562,
+      "learning_rate": 9.798275187946859e-06,
+      "loss": 0.44,
+      "step": 72
+    },
+    {
+      "epoch": 0.13977979894686454,
+      "grad_norm": 0.778163492679596,
+      "learning_rate": 9.789274230549456e-06,
+      "loss": 0.4054,
+      "step": 73
+    },
+    {
+      "epoch": 0.14169459071325993,
+      "grad_norm": 0.7578093409538269,
+      "learning_rate": 9.780081138965663e-06,
+      "loss": 0.3794,
+      "step": 74
+    },
+    {
+      "epoch": 0.14360938247965535,
+      "grad_norm": 0.904350221157074,
+      "learning_rate": 9.770696282000245e-06,
+      "loss": 0.4538,
+      "step": 75
+    },
+    {
+      "epoch": 0.14552417424605074,
+      "grad_norm": 0.8594577312469482,
+      "learning_rate": 9.761120036151138e-06,
+      "loss": 0.4211,
+      "step": 76
+    },
+    {
+      "epoch": 0.14743896601244616,
+      "grad_norm": 0.7970111966133118,
+      "learning_rate": 9.751352785594337e-06,
+      "loss": 0.401,
+      "step": 77
+    },
+    {
+      "epoch": 0.14935375777884155,
+      "grad_norm": 0.9434993267059326,
+      "learning_rate": 9.741394922168495e-06,
+      "loss": 0.4868,
+      "step": 78
+    },
+    {
+      "epoch": 0.15126854954523697,
+      "grad_norm": 0.7988068461418152,
+      "learning_rate": 9.731246845359187e-06,
+      "loss": 0.4088,
+      "step": 79
+    },
+    {
+      "epoch": 0.15318334131163236,
+      "grad_norm": 0.7606924176216125,
+      "learning_rate": 9.720908962282893e-06,
+      "loss": 0.4026,
+      "step": 80
+    },
+    {
+      "epoch": 0.15509813307802778,
+      "grad_norm": 0.8167851567268372,
+      "learning_rate": 9.710381687670675e-06,
+      "loss": 0.4347,
+      "step": 81
+    },
+    {
+      "epoch": 0.15701292484442317,
+      "grad_norm": 0.910528838634491,
+      "learning_rate": 9.699665443851518e-06,
+      "loss": 0.4445,
+      "step": 82
+    },
+    {
+      "epoch": 0.1589277166108186,
+      "grad_norm": 0.77234947681427,
+      "learning_rate": 9.688760660735403e-06,
+      "loss": 0.4022,
+      "step": 83
+    },
+    {
+      "epoch": 0.16084250837721398,
+      "grad_norm": 0.7472870945930481,
+      "learning_rate": 9.677667775796052e-06,
+      "loss": 0.3996,
+      "step": 84
+    },
+    {
+      "epoch": 0.1627573001436094,
+      "grad_norm": 0.874338686466217,
+      "learning_rate": 9.666387234053385e-06,
+      "loss": 0.45,
+      "step": 85
+    },
+    {
+      "epoch": 0.1646720919100048,
+      "grad_norm": 0.8861207962036133,
+      "learning_rate": 9.654919488055656e-06,
+      "loss": 0.4388,
+      "step": 86
+    },
+    {
+      "epoch": 0.1665868836764002,
+      "grad_norm": 0.8397772908210754,
+      "learning_rate": 9.643264997861312e-06,
+      "loss": 0.4182,
+      "step": 87
+    },
+    {
+      "epoch": 0.1685016754427956,
+      "grad_norm": 0.8450121283531189,
+      "learning_rate": 9.631424231020523e-06,
+      "loss": 0.4433,
+      "step": 88
+    },
+    {
+      "epoch": 0.170416467209191,
+      "grad_norm": 0.908657431602478,
+      "learning_rate": 9.619397662556434e-06,
+      "loss": 0.4481,
+      "step": 89
+    },
+    {
+      "epoch": 0.1723312589755864,
+      "grad_norm": 0.9613133072853088,
+      "learning_rate": 9.607185774946106e-06,
+      "loss": 0.5182,
+      "step": 90
+    },
+    {
+      "epoch": 0.17424605074198182,
+      "grad_norm": 0.9044798016548157,
+      "learning_rate": 9.594789058101154e-06,
+      "loss": 0.4445,
+      "step": 91
+    },
+    {
+      "epoch": 0.1761608425083772,
+      "grad_norm": 0.8073885440826416,
+      "learning_rate": 9.582208009348104e-06,
+      "loss": 0.4103,
+      "step": 92
+    },
+    {
+      "epoch": 0.17807563427477263,
+      "grad_norm": 0.866552472114563,
+      "learning_rate": 9.569443133408434e-06,
+      "loss": 0.4573,
+      "step": 93
+    },
+    {
+      "epoch": 0.17999042604116802,
+      "grad_norm": 0.8651822805404663,
+      "learning_rate": 9.556494942378328e-06,
+      "loss": 0.4369,
+      "step": 94
+    },
+    {
+      "epoch": 0.18190521780756344,
+      "grad_norm": 0.8955625295639038,
+      "learning_rate": 9.543363955708124e-06,
+      "loss": 0.4493,
+      "step": 95
+    },
+    {
+      "epoch": 0.18382000957395883,
+      "grad_norm": 0.7363105416297913,
+      "learning_rate": 9.530050700181499e-06,
+      "loss": 0.3666,
+      "step": 96
+    },
+    {
+      "epoch": 0.18573480134035425,
+      "grad_norm": 0.7756189107894897,
+      "learning_rate": 9.5165557098943e-06,
+      "loss": 0.4105,
+      "step": 97
+    },
+    {
+      "epoch": 0.18764959310674964,
+      "grad_norm": 0.8105975985527039,
+      "learning_rate": 9.502879526233151e-06,
+      "loss": 0.4026,
+      "step": 98
+    },
+    {
+      "epoch": 0.18956438487314506,
+      "grad_norm": 0.8235639929771423,
+      "learning_rate": 9.48902269785371e-06,
+      "loss": 0.4235,
+      "step": 99
+    },
+    {
+      "epoch": 0.19147917663954045,
+      "grad_norm": 0.9181493520736694,
+      "learning_rate": 9.47498578065867e-06,
+      "loss": 0.4123,
+      "step": 100
+    },
+    {
+      "epoch": 0.19339396840593587,
+      "grad_norm": 0.8257604241371155,
+      "learning_rate": 9.460769337775461e-06,
+      "loss": 0.4304,
+      "step": 101
+    },
+    {
+      "epoch": 0.19530876017233126,
+      "grad_norm": 0.8084312081336975,
+      "learning_rate": 9.446373939533642e-06,
+      "loss": 0.3962,
+      "step": 102
+    },
+    {
+      "epoch": 0.19722355193872668,
+      "grad_norm": 0.8189854025840759,
+      "learning_rate": 9.431800163442043e-06,
+      "loss": 0.4111,
+      "step": 103
+    },
+    {
+      "epoch": 0.19913834370512207,
+      "grad_norm": 0.8247345089912415,
+      "learning_rate": 9.417048594165572e-06,
+      "loss": 0.4485,
+      "step": 104
+    },
+    {
+      "epoch": 0.20105313547151749,
+      "grad_norm": 0.8686959147453308,
+      "learning_rate": 9.402119823501787e-06,
+      "loss": 0.4304,
+      "step": 105
+    },
+    {
+      "epoch": 0.20296792723791288,
+      "grad_norm": 0.839127779006958,
+      "learning_rate": 9.387014450357128e-06,
+      "loss": 0.4148,
+      "step": 106
+    },
+    {
+      "epoch": 0.2048827190043083,
+      "grad_norm": 0.9121819734573364,
+      "learning_rate": 9.371733080722911e-06,
+      "loss": 0.4842,
+      "step": 107
+    },
+    {
+      "epoch": 0.20679751077070369,
+      "grad_norm": 0.8987030982971191,
+      "learning_rate": 9.356276327651006e-06,
+      "loss": 0.4387,
+      "step": 108
+    },
+    {
+      "epoch": 0.2087123025370991,
+      "grad_norm": 0.8091103434562683,
+      "learning_rate": 9.340644811229243e-06,
+      "loss": 0.4024,
+      "step": 109
+    },
+    {
+      "epoch": 0.2106270943034945,
+      "grad_norm": 0.7413907051086426,
+      "learning_rate": 9.324839158556542e-06,
+      "loss": 0.3791,
+      "step": 110
+    },
+    {
+      "epoch": 0.2125418860698899,
+      "grad_norm": 0.8281195163726807,
+      "learning_rate": 9.308860003717748e-06,
+      "loss": 0.4151,
+      "step": 111
+    },
+    {
+      "epoch": 0.2144566778362853,
+      "grad_norm": 0.8579381704330444,
+      "learning_rate": 9.292707987758202e-06,
+      "loss": 0.4446,
+      "step": 112
+    },
+    {
+      "epoch": 0.21637146960268072,
+      "grad_norm": 0.833625853061676,
+      "learning_rate": 9.27638375865801e-06,
+      "loss": 0.4311,
+      "step": 113
+    },
+    {
+      "epoch": 0.2182862613690761,
+      "grad_norm": 0.8704817891120911,
+      "learning_rate": 9.259887971306064e-06,
+      "loss": 0.4852,
+      "step": 114
+    },
+    {
+      "epoch": 0.22020105313547153,
+      "grad_norm": 0.8960813283920288,
+      "learning_rate": 9.243221287473755e-06,
+      "loss": 0.4477,
+      "step": 115
+    },
+    {
+      "epoch": 0.22211584490186692,
+      "grad_norm": 0.8218123316764832,
+      "learning_rate": 9.226384375788435e-06,
+      "loss": 0.4169,
+      "step": 116
+    },
+    {
+      "epoch": 0.22403063666826234,
+      "grad_norm": 0.8349615931510925,
+      "learning_rate": 9.209377911706585e-06,
+      "loss": 0.4045,
+      "step": 117
+    },
+    {
+      "epoch": 0.22594542843465773,
+      "grad_norm": 0.8087317943572998,
+      "learning_rate": 9.192202577486725e-06,
+      "loss": 0.3923,
+      "step": 118
+    },
+    {
+      "epoch": 0.22786022020105315,
+      "grad_norm": 0.8312190175056458,
+      "learning_rate": 9.174859062162037e-06,
+      "loss": 0.3958,
+      "step": 119
+    },
+    {
+      "epoch": 0.22977501196744854,
+      "grad_norm": 0.8147414326667786,
+      "learning_rate": 9.157348061512728e-06,
+      "loss": 0.4417,
+      "step": 120
+    },
+    {
+      "epoch": 0.23168980373384396,
+      "grad_norm": 0.860390305519104,
+      "learning_rate": 9.139670278038109e-06,
+      "loss": 0.4404,
+      "step": 121
+    },
+    {
+      "epoch": 0.23360459550023935,
+      "grad_norm": 0.7553086280822754,
+      "learning_rate": 9.121826420928421e-06,
+      "loss": 0.3684,
+      "step": 122
+    },
+    {
+      "epoch": 0.23551938726663477,
+      "grad_norm": 0.8265523314476013,
+      "learning_rate": 9.103817206036383e-06,
+      "loss": 0.4025,
+      "step": 123
+    },
+    {
+      "epoch": 0.23743417903303016,
+      "grad_norm": 0.8537978529930115,
+      "learning_rate": 9.085643355848468e-06,
+      "loss": 0.4419,
+      "step": 124
+    },
+    {
+      "epoch": 0.23934897079942558,
+      "grad_norm": 0.8403013348579407,
+      "learning_rate": 9.06730559945592e-06,
+      "loss": 0.4013,
+      "step": 125
+    },
+    {
+      "epoch": 0.24126376256582097,
+      "grad_norm": 0.831799328327179,
+      "learning_rate": 9.048804672525513e-06,
+      "loss": 0.417,
+      "step": 126
+    },
+    {
+      "epoch": 0.24317855433221638,
+      "grad_norm": 0.7995438575744629,
+      "learning_rate": 9.030141317270026e-06,
+      "loss": 0.3941,
+      "step": 127
+    },
+    {
+      "epoch": 0.24509334609861178,
+      "grad_norm": 0.8485874533653259,
+      "learning_rate": 9.011316282418474e-06,
+      "loss": 0.4128,
+      "step": 128
+    },
+    {
+      "epoch": 0.2470081378650072,
+      "grad_norm": 0.8879005908966064,
+      "learning_rate": 8.992330323186069e-06,
+      "loss": 0.4445,
+      "step": 129
+    },
+    {
+      "epoch": 0.24892292963140258,
+      "grad_norm": 0.9082502722740173,
+      "learning_rate": 8.973184201243922e-06,
+      "loss": 0.4825,
+      "step": 130
+    },
+    {
+      "epoch": 0.250837721397798,
+      "grad_norm": 0.8748418688774109,
+      "learning_rate": 8.953878684688492e-06,
+      "loss": 0.4209,
+      "step": 131
+    },
+    {
+      "epoch": 0.2527525131641934,
+      "grad_norm": 0.8313506245613098,
+      "learning_rate": 8.934414548010764e-06,
+      "loss": 0.4083,
+      "step": 132
+    },
+    {
+      "epoch": 0.2546673049305888,
+      "grad_norm": 0.8159312009811401,
+      "learning_rate": 8.914792572065178e-06,
+      "loss": 0.4166,
+      "step": 133
+    },
+    {
+      "epoch": 0.25658209669698423,
+      "grad_norm": 0.9007822275161743,
+      "learning_rate": 8.89501354403831e-06,
+      "loss": 0.4591,
+      "step": 134
+    },
+    {
+      "epoch": 0.2584968884633796,
+      "grad_norm": 0.9515888690948486,
+      "learning_rate": 8.875078257417294e-06,
+      "loss": 0.4659,
+      "step": 135
+    },
+    {
+      "epoch": 0.260411680229775,
+      "grad_norm": 0.8637048602104187,
+      "learning_rate": 8.854987511957974e-06,
+      "loss": 0.4387,
+      "step": 136
+    },
+    {
+      "epoch": 0.26232647199617043,
+      "grad_norm": 0.8459029793739319,
+      "learning_rate": 8.834742113652835e-06,
+      "loss": 0.4285,
+      "step": 137
+    },
+    {
+      "epoch": 0.26424126376256585,
+      "grad_norm": 0.7771206498146057,
+      "learning_rate": 8.81434287469866e-06,
+      "loss": 0.3803,
+      "step": 138
+    },
+    {
+      "epoch": 0.2661560555289612,
+      "grad_norm": 0.8040123581886292,
+      "learning_rate": 8.793790613463956e-06,
+      "loss": 0.411,
+      "step": 139
+    },
+    {
+      "epoch": 0.26807084729535663,
+      "grad_norm": 0.8000430464744568,
+      "learning_rate": 8.773086154456106e-06,
+      "loss": 0.4165,
+      "step": 140
+    },
+    {
+      "epoch": 0.26998563906175205,
+      "grad_norm": 0.9271606802940369,
+      "learning_rate": 8.752230328288314e-06,
+      "loss": 0.4755,
+      "step": 141
+    },
+    {
+      "epoch": 0.27190043082814747,
+      "grad_norm": 0.7970160841941833,
+      "learning_rate": 8.731223971646261e-06,
+      "loss": 0.3916,
+      "step": 142
+    },
+    {
+      "epoch": 0.27381522259454283,
+      "grad_norm": 0.7897955775260925,
+      "learning_rate": 8.710067927254555e-06,
+      "loss": 0.3846,
+      "step": 143
+    },
+    {
+      "epoch": 0.27573001436093825,
+      "grad_norm": 0.8004978895187378,
+      "learning_rate": 8.688763043842916e-06,
+      "loss": 0.4077,
+      "step": 144
+    },
+    {
+      "epoch": 0.27764480612733367,
+      "grad_norm": 0.8318747282028198,
+      "learning_rate": 8.66731017611213e-06,
+      "loss": 0.4334,
+      "step": 145
+    },
+    {
+      "epoch": 0.2795595978937291,
+      "grad_norm": 0.8479062914848328,
+      "learning_rate": 8.645710184699756e-06,
+      "loss": 0.4172,
+      "step": 146
+    },
+    {
+      "epoch": 0.28147438966012445,
+      "grad_norm": 0.8077636361122131,
+      "learning_rate": 8.6239639361456e-06,
+      "loss": 0.4104,
+      "step": 147
+    },
+    {
+      "epoch": 0.28338918142651986,
+      "grad_norm": 0.8098336458206177,
+      "learning_rate": 8.602072302856961e-06,
+      "loss": 0.406,
+      "step": 148
+    },
+    {
+      "epoch": 0.2853039731929153,
+      "grad_norm": 0.8435089588165283,
+      "learning_rate": 8.580036163073615e-06,
+      "loss": 0.4312,
+      "step": 149
+    },
+    {
+      "epoch": 0.2872187649593107,
+      "grad_norm": 0.8346257209777832,
+      "learning_rate": 8.5578564008326e-06,
+      "loss": 0.3888,
+      "step": 150
+    },
+    {
+      "epoch": 0.28913355672570606,
+      "grad_norm": 0.8100122213363647,
+      "learning_rate": 8.535533905932739e-06,
+      "loss": 0.4042,
+      "step": 151
+    },
+    {
+      "epoch": 0.2910483484921015,
+      "grad_norm": 0.8516131043434143,
+      "learning_rate": 8.513069573898944e-06,
+      "loss": 0.4161,
+      "step": 152
+    },
+    {
+      "epoch": 0.2929631402584969,
+      "grad_norm": 0.8350062966346741,
+      "learning_rate": 8.490464305946296e-06,
+      "loss": 0.4231,
+      "step": 153
+    },
+    {
+      "epoch": 0.2948779320248923,
+      "grad_norm": 0.8145307302474976,
+      "learning_rate": 8.467719008943886e-06,
+      "loss": 0.4125,
+      "step": 154
+    },
+    {
+      "epoch": 0.2967927237912877,
+      "grad_norm": 0.8515341281890869,
+      "learning_rate": 8.444834595378434e-06,
+      "loss": 0.4192,
+      "step": 155
+    },
+    {
+      "epoch": 0.2987075155576831,
+      "grad_norm": 0.8017379641532898,
+      "learning_rate": 8.421811983317682e-06,
+      "loss": 0.4007,
+      "step": 156
+    },
+    {
+      "epoch": 0.3006223073240785,
+      "grad_norm": 0.8179863691329956,
+      "learning_rate": 8.398652096373566e-06,
+      "loss": 0.4188,
+      "step": 157
+    },
+    {
+      "epoch": 0.30253709909047394,
+      "grad_norm": 0.8272606730461121,
+      "learning_rate": 8.375355863665155e-06,
+      "loss": 0.4035,
+      "step": 158
+    },
+    {
+      "epoch": 0.3044518908568693,
+      "grad_norm": 0.8340199589729309,
+      "learning_rate": 8.351924219781393e-06,
+      "loss": 0.4414,
+      "step": 159
+    },
+    {
+      "epoch": 0.3063666826232647,
+      "grad_norm": 0.82643723487854,
+      "learning_rate": 8.328358104743588e-06,
+      "loss": 0.4143,
+      "step": 160
+    },
+    {
+      "epoch": 0.30828147438966014,
+      "grad_norm": 0.7954344153404236,
+      "learning_rate": 8.304658463967705e-06,
+      "loss": 0.4128,
+      "step": 161
+    },
+    {
+      "epoch": 0.31019626615605556,
+      "grad_norm": 0.7908764481544495,
+      "learning_rate": 8.28082624822645e-06,
+      "loss": 0.3813,
+      "step": 162
+    },
+    {
+      "epoch": 0.3121110579224509,
+      "grad_norm": 0.7368812561035156,
+      "learning_rate": 8.256862413611113e-06,
+      "loss": 0.3879,
+      "step": 163
+    },
+    {
+      "epoch": 0.31402584968884634,
+      "grad_norm": 0.8204508423805237,
+      "learning_rate": 8.232767921493216e-06,
+      "loss": 0.4308,
+      "step": 164
+    },
+    {
+      "epoch": 0.31594064145524176,
+      "grad_norm": 0.8691006302833557,
+      "learning_rate": 8.20854373848595e-06,
+      "loss": 0.4518,
+      "step": 165
+    },
+    {
+      "epoch": 0.3178554332216372,
+      "grad_norm": 0.7581012845039368,
+      "learning_rate": 8.184190836405394e-06,
+      "loss": 0.3711,
+      "step": 166
+    },
+    {
+      "epoch": 0.31977022498803254,
+      "grad_norm": 0.7817586660385132,
+      "learning_rate": 8.15971019223152e-06,
+      "loss": 0.4057,
+      "step": 167
+    },
+    {
+      "epoch": 0.32168501675442795,
+      "grad_norm": 0.7559137344360352,
+      "learning_rate": 8.135102788069015e-06,
+      "loss": 0.361,
+      "step": 168
+    },
+    {
+      "epoch": 0.3235998085208234,
+      "grad_norm": 0.7424903512001038,
+      "learning_rate": 8.110369611107869e-06,
+      "loss": 0.3667,
+      "step": 169
+    },
+    {
+      "epoch": 0.3255146002872188,
+      "grad_norm": 0.7965791821479797,
+      "learning_rate": 8.085511653583772e-06,
+      "loss": 0.3814,
+      "step": 170
+    },
+    {
+      "epoch": 0.32742939205361415,
+      "grad_norm": 0.8504881262779236,
+      "learning_rate": 8.060529912738316e-06,
+      "loss": 0.4453,
+      "step": 171
+    },
+    {
+      "epoch": 0.3293441838200096,
+      "grad_norm": 0.8762128353118896,
+      "learning_rate": 8.035425390778975e-06,
+      "loss": 0.4516,
+      "step": 172
+    },
+    {
+      "epoch": 0.331258975586405,
+      "grad_norm": 0.8157339692115784,
+      "learning_rate": 8.010199094838915e-06,
+      "loss": 0.4204,
+      "step": 173
+    },
+    {
+      "epoch": 0.3331737673528004,
+      "grad_norm": 0.8493944406509399,
+      "learning_rate": 7.984852036936578e-06,
+      "loss": 0.39,
+      "step": 174
+    },
+    {
+      "epoch": 0.3350885591191958,
+      "grad_norm": 0.838901937007904,
+      "learning_rate": 7.959385233935087e-06,
+      "loss": 0.4417,
+      "step": 175
+    },
+    {
+      "epoch": 0.3370033508855912,
+      "grad_norm": 0.7574141621589661,
+      "learning_rate": 7.933799707501448e-06,
+      "loss": 0.3515,
+      "step": 176
+    },
+    {
+      "epoch": 0.3389181426519866,
+      "grad_norm": 0.8099052906036377,
+      "learning_rate": 7.908096484065569e-06,
+      "loss": 0.4091,
+      "step": 177
+    },
+    {
+      "epoch": 0.340832934418382,
+      "grad_norm": 0.8116398453712463,
+      "learning_rate": 7.88227659477908e-06,
+      "loss": 0.4137,
+      "step": 178
+    },
+    {
+      "epoch": 0.3427477261847774,
+      "grad_norm": 0.7785760164260864,
+      "learning_rate": 7.856341075473963e-06,
+      "loss": 0.3829,
+      "step": 179
+    },
+    {
+      "epoch": 0.3446625179511728,
+      "grad_norm": 0.7929257154464722,
+      "learning_rate": 7.830290966620997e-06,
+      "loss": 0.3739,
+      "step": 180
+    },
+    {
+      "epoch": 0.3465773097175682,
+      "grad_norm": 0.8796236515045166,
+      "learning_rate": 7.804127313288023e-06,
+      "loss": 0.4027,
+      "step": 181
+    },
+    {
+      "epoch": 0.34849210148396365,
+      "grad_norm": 0.8447411060333252,
+      "learning_rate": 7.777851165098012e-06,
+      "loss": 0.4202,
+      "step": 182
+    },
+    {
+      "epoch": 0.350406893250359,
+      "grad_norm": 0.7473250031471252,
+      "learning_rate": 7.751463576186957e-06,
+      "loss": 0.3777,
+      "step": 183
+    },
+    {
+      "epoch": 0.3523216850167544,
+      "grad_norm": 0.8254420757293701,
+      "learning_rate": 7.72496560516159e-06,
+      "loss": 0.4147,
+      "step": 184
+    },
+    {
+      "epoch": 0.35423647678314985,
+      "grad_norm": 0.8766903281211853,
+      "learning_rate": 7.6983583150569e-06,
+      "loss": 0.4212,
+      "step": 185
+    },
+    {
+      "epoch": 0.35615126854954526,
+      "grad_norm": 0.8106472492218018,
+      "learning_rate": 7.671642773293506e-06,
+      "loss": 0.3897,
+      "step": 186
+    },
+    {
+      "epoch": 0.3580660603159406,
+      "grad_norm": 0.8969345092773438,
+      "learning_rate": 7.644820051634813e-06,
+      "loss": 0.4166,
+      "step": 187
+    },
+    {
+      "epoch": 0.35998085208233604,
+      "grad_norm": 0.9210174679756165,
+      "learning_rate": 7.617891226144034e-06,
+      "loss": 0.4744,
+      "step": 188
+    },
+    {
+      "epoch": 0.36189564384873146,
+      "grad_norm": 0.8183197379112244,
+      "learning_rate": 7.59085737714101e-06,
+      "loss": 0.3909,
+      "step": 189
+    },
+    {
+      "epoch": 0.3638104356151269,
+      "grad_norm": 0.9258884191513062,
+      "learning_rate": 7.563719589158874e-06,
+      "loss": 0.4191,
+      "step": 190
+    },
+    {
+      "epoch": 0.36572522738152224,
+      "grad_norm": 0.8379626274108887,
+      "learning_rate": 7.536478950900537e-06,
+      "loss": 0.4081,
+      "step": 191
+    },
+    {
+      "epoch": 0.36764001914791766,
+      "grad_norm": 0.8194741010665894,
+      "learning_rate": 7.509136555195025e-06,
+      "loss": 0.3983,
+      "step": 192
+    },
+    {
+      "epoch": 0.3695548109143131,
+      "grad_norm": 0.8078919649124146,
+      "learning_rate": 7.481693498953621e-06,
+      "loss": 0.4117,
+      "step": 193
+    },
+    {
+      "epoch": 0.3714696026807085,
+      "grad_norm": 0.8059271574020386,
+      "learning_rate": 7.4541508831258695e-06,
+      "loss": 0.3919,
+      "step": 194
+    },
+    {
+      "epoch": 0.37338439444710386,
+      "grad_norm": 0.7913739681243896,
+      "learning_rate": 7.4265098126554065e-06,
+      "loss": 0.3786,
+      "step": 195
+    },
+    {
+      "epoch": 0.3752991862134993,
+      "grad_norm": 0.8568047285079956,
+      "learning_rate": 7.3987713964356335e-06,
+      "loss": 0.4505,
+      "step": 196
+    },
+    {
+      "epoch": 0.3772139779798947,
+      "grad_norm": 0.9244000315666199,
+      "learning_rate": 7.370936747265226e-06,
+      "loss": 0.4546,
+      "step": 197
+    },
+    {
+      "epoch": 0.3791287697462901,
+      "grad_norm": 0.7795801162719727,
+      "learning_rate": 7.3430069818035e-06,
+      "loss": 0.3955,
+      "step": 198
+    },
+    {
+      "epoch": 0.3810435615126855,
+      "grad_norm": 0.7940351366996765,
+      "learning_rate": 7.314983220525604e-06,
+      "loss": 0.4043,
+      "step": 199
+    },
+    {
+      "epoch": 0.3829583532790809,
+      "grad_norm": 0.7889821529388428,
+      "learning_rate": 7.286866587677576e-06,
+      "loss": 0.3883,
+      "step": 200
+    },
+    {
+      "epoch": 0.3848731450454763,
+      "grad_norm": 0.857402503490448,
+      "learning_rate": 7.2586582112312355e-06,
+      "loss": 0.4288,
+      "step": 201
+    },
+    {
+      "epoch": 0.38678793681187174,
+      "grad_norm": 0.7854819893836975,
+      "learning_rate": 7.230359222838939e-06,
+      "loss": 0.3884,
+      "step": 202
+    },
+    {
+      "epoch": 0.3887027285782671,
+      "grad_norm": 0.9061365723609924,
+      "learning_rate": 7.201970757788172e-06,
+      "loss": 0.458,
+      "step": 203
+    },
+    {
+      "epoch": 0.3906175203446625,
+      "grad_norm": 0.7978100776672363,
+      "learning_rate": 7.173493954956012e-06,
+      "loss": 0.3903,
+      "step": 204
+    },
+    {
+      "epoch": 0.39253231211105793,
+      "grad_norm": 0.8213270902633667,
+      "learning_rate": 7.144929956763438e-06,
+      "loss": 0.4035,
+      "step": 205
+    },
+    {
+      "epoch": 0.39444710387745335,
+      "grad_norm": 0.9091420769691467,
+      "learning_rate": 7.116279909129492e-06,
+      "loss": 0.4505,
+      "step": 206
+    },
+    {
+      "epoch": 0.3963618956438487,
+      "grad_norm": 0.8489376306533813,
+      "learning_rate": 7.087544961425317e-06,
+      "loss": 0.4037,
+      "step": 207
+    },
+    {
+      "epoch": 0.39827668741024413,
+      "grad_norm": 0.805548369884491,
+      "learning_rate": 7.058726266428042e-06,
+      "loss": 0.4053,
+      "step": 208
+    },
+    {
+      "epoch": 0.40019147917663955,
+      "grad_norm": 0.7785957455635071,
+      "learning_rate": 7.029824980274536e-06,
+      "loss": 0.3719,
+      "step": 209
+    },
+    {
+      "epoch": 0.40210627094303497,
+      "grad_norm": 0.8519039154052734,
+      "learning_rate": 7.0008422624150285e-06,
+      "loss": 0.4176,
+      "step": 210
+    },
+    {
+      "epoch": 0.40402106270943033,
+      "grad_norm": 0.8210786581039429,
+      "learning_rate": 6.971779275566593e-06,
+      "loss": 0.4162,
+      "step": 211
+    },
+    {
+      "epoch": 0.40593585447582575,
+      "grad_norm": 0.8121908903121948,
+      "learning_rate": 6.9426371856665005e-06,
+      "loss": 0.4204,
+      "step": 212
+    },
+    {
+      "epoch": 0.40785064624222117,
+      "grad_norm": 0.8359207510948181,
+      "learning_rate": 6.913417161825449e-06,
+      "loss": 0.4251,
+      "step": 213
+    },
+    {
+      "epoch": 0.4097654380086166,
+      "grad_norm": 0.8265076279640198,
+      "learning_rate": 6.884120376280658e-06,
+      "loss": 0.3983,
+      "step": 214
+    },
+    {
+      "epoch": 0.41168022977501195,
+      "grad_norm": 0.8334207534790039,
+      "learning_rate": 6.85474800434884e-06,
+      "loss": 0.4287,
+      "step": 215
+    },
+    {
+      "epoch": 0.41359502154140737,
+      "grad_norm": 0.7907735705375671,
+      "learning_rate": 6.8253012243790565e-06,
+      "loss": 0.4068,
+      "step": 216
+    },
+    {
+      "epoch": 0.4155098133078028,
+      "grad_norm": 0.8516376614570618,
+      "learning_rate": 6.795781217705436e-06,
+      "loss": 0.4531,
+      "step": 217
+    },
+    {
+      "epoch": 0.4174246050741982,
+      "grad_norm": 0.7719292640686035,
+      "learning_rate": 6.76618916859979e-06,
+      "loss": 0.3641,
+      "step": 218
+    },
+    {
+      "epoch": 0.41933939684059357,
+      "grad_norm": 0.769619882106781,
+      "learning_rate": 6.736526264224101e-06,
+      "loss": 0.389,
+      "step": 219
+    },
+    {
+      "epoch": 0.421254188606989,
+      "grad_norm": 0.8193688988685608,
+      "learning_rate": 6.706793694582892e-06,
+      "loss": 0.381,
+      "step": 220
+    },
+    {
+      "epoch": 0.4231689803733844,
+      "grad_norm": 0.8246089220046997,
+      "learning_rate": 6.676992652475487e-06,
+      "loss": 0.4105,
+      "step": 221
+    },
+    {
+      "epoch": 0.4250837721397798,
+      "grad_norm": 0.7855746746063232,
+      "learning_rate": 6.647124333448165e-06,
+      "loss": 0.3711,
+      "step": 222
+    },
+    {
+      "epoch": 0.4269985639061752,
+      "grad_norm": 0.8676920533180237,
+      "learning_rate": 6.617189935746191e-06,
+      "loss": 0.4071,
+      "step": 223
+    },
+    {
+      "epoch": 0.4289133556725706,
+      "grad_norm": 0.8358131647109985,
+      "learning_rate": 6.587190660265752e-06,
+      "loss": 0.425,
+      "step": 224
+    },
+    {
+      "epoch": 0.430828147438966,
+      "grad_norm": 0.8328396677970886,
+      "learning_rate": 6.55712771050577e-06,
+      "loss": 0.4156,
+      "step": 225
+    },
+    {
+      "epoch": 0.43274293920536144,
+      "grad_norm": 0.7680657505989075,
+      "learning_rate": 6.52700229251963e-06,
+      "loss": 0.393,
+      "step": 226
+    },
+    {
+      "epoch": 0.4346577309717568,
+      "grad_norm": 0.8950127959251404,
+      "learning_rate": 6.496815614866792e-06,
+      "loss": 0.4035,
+      "step": 227
+    },
+    {
+      "epoch": 0.4365725227381522,
+      "grad_norm": 0.8672182559967041,
+      "learning_rate": 6.466568888564303e-06,
+      "loss": 0.4581,
+      "step": 228
+    },
+    {
+      "epoch": 0.43848731450454764,
+      "grad_norm": 0.8580407500267029,
+      "learning_rate": 6.436263327038225e-06,
+      "loss": 0.4039,
+      "step": 229
+    },
+    {
+      "epoch": 0.44040210627094306,
+      "grad_norm": 0.8145782351493835,
+      "learning_rate": 6.405900146074941e-06,
+      "loss": 0.3873,
+      "step": 230
+    },
+    {
+      "epoch": 0.4423168980373384,
+      "grad_norm": 0.8403171896934509,
+      "learning_rate": 6.375480563772391e-06,
+      "loss": 0.4373,
+      "step": 231
+    },
+    {
+      "epoch": 0.44423168980373384,
+      "grad_norm": 0.7500009536743164,
+      "learning_rate": 6.3450058004912004e-06,
+      "loss": 0.3661,
+      "step": 232
+    },
+    {
+      "epoch": 0.44614648157012926,
+      "grad_norm": 0.8253698945045471,
+      "learning_rate": 6.314477078805724e-06,
+      "loss": 0.4005,
+      "step": 233
+    },
+    {
+      "epoch": 0.4480612733365247,
+      "grad_norm": 0.7586554884910583,
+      "learning_rate": 6.283895623454997e-06,
+      "loss": 0.3858,
+      "step": 234
+    },
+    {
+      "epoch": 0.44997606510292004,
+      "grad_norm": 0.8090248107910156,
+      "learning_rate": 6.2532626612936035e-06,
+      "loss": 0.4085,
+      "step": 235
+    },
+    {
+      "epoch": 0.45189085686931546,
+      "grad_norm": 0.8589561581611633,
+      "learning_rate": 6.2225794212424565e-06,
+      "loss": 0.4402,
+      "step": 236
+    },
+    {
+      "epoch": 0.4538056486357109,
+      "grad_norm": 0.8317797780036926,
+      "learning_rate": 6.191847134239496e-06,
+      "loss": 0.4,
+      "step": 237
+    },
+    {
+      "epoch": 0.4557204404021063,
+      "grad_norm": 0.8422788381576538,
+      "learning_rate": 6.161067033190311e-06,
+      "loss": 0.4026,
+      "step": 238
+    },
+    {
+      "epoch": 0.45763523216850166,
+      "grad_norm": 0.763214647769928,
+      "learning_rate": 6.130240352918675e-06,
+      "loss": 0.3976,
+      "step": 239
+    },
+    {
+      "epoch": 0.4595500239348971,
+      "grad_norm": 0.8719058036804199,
+      "learning_rate": 6.0993683301170046e-06,
+      "loss": 0.4356,
+      "step": 240
+    },
+    {
+      "epoch": 0.4614648157012925,
+      "grad_norm": 0.7786772847175598,
+      "learning_rate": 6.068452203296754e-06,
+      "loss": 0.3858,
+      "step": 241
+    },
+    {
+      "epoch": 0.4633796074676879,
+      "grad_norm": 0.7639849185943604,
+      "learning_rate": 6.0374932127387234e-06,
+      "loss": 0.3936,
+      "step": 242
+    },
+    {
+      "epoch": 0.4652943992340833,
+      "grad_norm": 0.8484832644462585,
+      "learning_rate": 6.006492600443301e-06,
+      "loss": 0.4006,
+      "step": 243
+    },
+    {
+      "epoch": 0.4672091910004787,
+      "grad_norm": 0.835634708404541,
+      "learning_rate": 5.975451610080643e-06,
+      "loss": 0.3826,
+      "step": 244
+    },
+    {
+      "epoch": 0.4691239827668741,
+      "grad_norm": 0.8132709860801697,
+      "learning_rate": 5.944371486940772e-06,
+      "loss": 0.3934,
+      "step": 245
+    },
+    {
+      "epoch": 0.47103877453326953,
+      "grad_norm": 0.93300861120224,
+      "learning_rate": 5.913253477883629e-06,
+      "loss": 0.4376,
+      "step": 246
+    },
+    {
+      "epoch": 0.4729535662996649,
+      "grad_norm": 0.8036344647407532,
+      "learning_rate": 5.882098831289044e-06,
+      "loss": 0.3892,
+      "step": 247
+    },
+    {
+      "epoch": 0.4748683580660603,
+      "grad_norm": 0.7971835732460022,
+      "learning_rate": 5.850908797006656e-06,
+      "loss": 0.399,
+      "step": 248
+    },
+    {
+      "epoch": 0.47678314983245573,
+      "grad_norm": 0.8496800065040588,
+      "learning_rate": 5.819684626305776e-06,
+      "loss": 0.4385,
+      "step": 249
+    },
+    {
+      "epoch": 0.47869794159885115,
+      "grad_norm": 0.8178644776344299,
+      "learning_rate": 5.788427571825186e-06,
+      "loss": 0.3946,
+      "step": 250
+    },
+    {
+      "epoch": 0.4806127333652465,
+      "grad_norm": 0.8840231895446777,
+      "learning_rate": 5.757138887522884e-06,
+      "loss": 0.4125,
+      "step": 251
+    },
+    {
+      "epoch": 0.48252752513164193,
+      "grad_norm": 0.831671416759491,
+      "learning_rate": 5.725819828625782e-06,
+      "loss": 0.413,
+      "step": 252
+    },
+    {
+      "epoch": 0.48444231689803735,
+      "grad_norm": 0.7894381880760193,
+      "learning_rate": 5.694471651579346e-06,
+      "loss": 0.399,
+      "step": 253
+    },
+    {
+      "epoch": 0.48635710866443277,
+      "grad_norm": 0.8008851408958435,
+      "learning_rate": 5.663095613997196e-06,
+      "loss": 0.3875,
+      "step": 254
+    },
+    {
+      "epoch": 0.48827190043082813,
+      "grad_norm": 0.7766909003257751,
+      "learning_rate": 5.631692974610647e-06,
+      "loss": 0.3755,
+      "step": 255
+    },
+    {
+      "epoch": 0.49018669219722355,
+      "grad_norm": 0.8729162812232971,
+      "learning_rate": 5.600264993218215e-06,
+      "loss": 0.4103,
+      "step": 256
+    },
+    {
+      "epoch": 0.49210148396361897,
+      "grad_norm": 0.8618749380111694,
+      "learning_rate": 5.568812930635076e-06,
+      "loss": 0.3963,
+      "step": 257
+    },
+    {
+      "epoch": 0.4940162757300144,
+      "grad_norm": 0.8580813407897949,
+      "learning_rate": 5.537338048642487e-06,
+      "loss": 0.4353,
+      "step": 258
+    },
+    {
+      "epoch": 0.49593106749640975,
+      "grad_norm": 0.8418040871620178,
+      "learning_rate": 5.505841609937162e-06,
+      "loss": 0.3801,
+      "step": 259
+    },
+    {
+      "epoch": 0.49784585926280517,
+      "grad_norm": 0.7449955940246582,
+      "learning_rate": 5.474324878080623e-06,
+      "loss": 0.3355,
+      "step": 260
+    },
+    {
+      "epoch": 0.4997606510292006,
+      "grad_norm": 0.8238184452056885,
+      "learning_rate": 5.4427891174485014e-06,
+      "loss": 0.3868,
+      "step": 261
+    },
+    {
+      "epoch": 0.501675442795596,
+      "grad_norm": 0.8558638691902161,
+      "learning_rate": 5.41123559317982e-06,
+      "loss": 0.4149,
+      "step": 262
+    },
+    {
+      "epoch": 0.5035902345619914,
+      "grad_norm": 0.836510956287384,
+      "learning_rate": 5.379665571126232e-06,
+      "loss": 0.378,
+      "step": 263
+    },
+    {
+      "epoch": 0.5055050263283868,
+      "grad_norm": 0.7436261177062988,
+      "learning_rate": 5.348080317801244e-06,
+      "loss": 0.3672,
+      "step": 264
+    },
+    {
+      "epoch": 0.5074198180947822,
+      "grad_norm": 0.897144079208374,
+      "learning_rate": 5.316481100329408e-06,
+      "loss": 0.4309,
+      "step": 265
+    },
+    {
+      "epoch": 0.5093346098611776,
+      "grad_norm": 0.8236995935440063,
+      "learning_rate": 5.284869186395478e-06,
+      "loss": 0.4164,
+      "step": 266
+    },
+    {
+      "epoch": 0.511249401627573,
+      "grad_norm": 0.8249524831771851,
+      "learning_rate": 5.253245844193564e-06,
+      "loss": 0.408,
+      "step": 267
+    },
+    {
+      "epoch": 0.5131641933939685,
+      "grad_norm": 0.8149250745773315,
+      "learning_rate": 5.22161234237625e-06,
+      "loss": 0.4014,
+      "step": 268
+    },
+    {
+      "epoch": 0.5150789851603638,
+      "grad_norm": 0.8144116401672363,
+      "learning_rate": 5.189969950003697e-06,
+      "loss": 0.4013,
+      "step": 269
+    },
+    {
+      "epoch": 0.5169937769267592,
+      "grad_norm": 0.8883258700370789,
+      "learning_rate": 5.158319936492736e-06,
+      "loss": 0.4262,
+      "step": 270
+    },
+    {
+      "epoch": 0.5189085686931546,
+      "grad_norm": 0.7828565239906311,
+      "learning_rate": 5.12666357156594e-06,
+      "loss": 0.3877,
+      "step": 271
+    },
+    {
+      "epoch": 0.52082336045955,
+      "grad_norm": 0.8680728673934937,
+      "learning_rate": 5.0950021252006845e-06,
+      "loss": 0.4288,
+      "step": 272
+    },
+    {
+      "epoch": 0.5227381522259454,
+      "grad_norm": 0.8556041121482849,
+      "learning_rate": 5.063336867578201e-06,
+      "loss": 0.4397,
+      "step": 273
+    },
+    {
+      "epoch": 0.5246529439923409,
+      "grad_norm": 0.83294278383255,
+      "learning_rate": 5.0316690690326175e-06,
+      "loss": 0.3855,
+      "step": 274
+    },
+    {
+      "epoch": 0.5265677357587363,
+      "grad_norm": 0.8231511116027832,
+      "learning_rate": 5e-06,
+      "loss": 0.3712,
+      "step": 275
+    },
+    {
+      "epoch": 0.5284825275251317,
+      "grad_norm": 0.8193358182907104,
+      "learning_rate": 4.9683309309673825e-06,
+      "loss": 0.3833,
+      "step": 276
+    },
+    {
+      "epoch": 0.530397319291527,
+      "grad_norm": 0.7521802186965942,
+      "learning_rate": 4.936663132421801e-06,
+      "loss": 0.3668,
+      "step": 277
+    },
+    {
+      "epoch": 0.5323121110579224,
+      "grad_norm": 0.7623764872550964,
+      "learning_rate": 4.904997874799316e-06,
+      "loss": 0.3825,
+      "step": 278
+    },
+    {
+      "epoch": 0.5342269028243178,
+      "grad_norm": 0.8101343512535095,
+      "learning_rate": 4.873336428434062e-06,
+      "loss": 0.3864,
+      "step": 279
+    },
+    {
+      "epoch": 0.5361416945907133,
+      "grad_norm": 0.8006393909454346,
+      "learning_rate": 4.841680063507265e-06,
+      "loss": 0.4227,
+      "step": 280
+    },
+    {
+      "epoch": 0.5380564863571087,
+      "grad_norm": 0.8386540412902832,
+      "learning_rate": 4.8100300499963045e-06,
+      "loss": 0.4131,
+      "step": 281
+    },
+    {
+      "epoch": 0.5399712781235041,
+      "grad_norm": 0.7718150019645691,
+      "learning_rate": 4.778387657623751e-06,
+      "loss": 0.3768,
+      "step": 282
+    },
+    {
+      "epoch": 0.5418860698898995,
+      "grad_norm": 0.8466550707817078,
+      "learning_rate": 4.746754155806437e-06,
+      "loss": 0.4512,
+      "step": 283
+    },
+    {
+      "epoch": 0.5438008616562949,
+      "grad_norm": 0.8129767775535583,
+      "learning_rate": 4.715130813604522e-06,
+      "loss": 0.3961,
+      "step": 284
+    },
+    {
+      "epoch": 0.5457156534226902,
+      "grad_norm": 0.8341888189315796,
+      "learning_rate": 4.683518899670594e-06,
+      "loss": 0.3919,
+      "step": 285
+    },
+    {
+      "epoch": 0.5476304451890857,
+      "grad_norm": 0.8497266173362732,
+      "learning_rate": 4.651919682198756e-06,
+      "loss": 0.394,
+      "step": 286
+    },
+    {
+      "epoch": 0.5495452369554811,
+      "grad_norm": 0.8146570920944214,
+      "learning_rate": 4.62033442887377e-06,
+      "loss": 0.3989,
+      "step": 287
+    },
+    {
+      "epoch": 0.5514600287218765,
+      "grad_norm": 0.806999921798706,
+      "learning_rate": 4.588764406820181e-06,
+      "loss": 0.4288,
+      "step": 288
+    },
+    {
+      "epoch": 0.5533748204882719,
+      "grad_norm": 0.7479027509689331,
+      "learning_rate": 4.5572108825515e-06,
+      "loss": 0.3616,
+      "step": 289
+    },
+    {
+      "epoch": 0.5552896122546673,
+      "grad_norm": 0.7727148532867432,
+      "learning_rate": 4.5256751219193784e-06,
+      "loss": 0.3902,
+      "step": 290
+    },
+    {
+      "epoch": 0.5572044040210627,
+      "grad_norm": 0.8226932883262634,
+      "learning_rate": 4.49415839006284e-06,
+      "loss": 0.4038,
+      "step": 291
+    },
+    {
+      "epoch": 0.5591191957874582,
+      "grad_norm": 0.8253747224807739,
+      "learning_rate": 4.462661951357515e-06,
+      "loss": 0.4052,
+      "step": 292
+    },
+    {
+      "epoch": 0.5610339875538535,
+      "grad_norm": 0.8564761877059937,
+      "learning_rate": 4.431187069364927e-06,
+      "loss": 0.4099,
+      "step": 293
+    },
+    {
+      "epoch": 0.5629487793202489,
+      "grad_norm": 0.812239408493042,
+      "learning_rate": 4.3997350067817866e-06,
+      "loss": 0.3939,
+      "step": 294
+    },
+    {
+      "epoch": 0.5648635710866443,
+      "grad_norm": 0.8617464303970337,
+      "learning_rate": 4.368307025389355e-06,
+      "loss": 0.4189,
+      "step": 295
+    },
+    {
+      "epoch": 0.5667783628530397,
+      "grad_norm": 0.8013660907745361,
+      "learning_rate": 4.336904386002805e-06,
+      "loss": 0.3853,
+      "step": 296
+    },
+    {
+      "epoch": 0.5686931546194351,
+      "grad_norm": 0.8232107162475586,
+      "learning_rate": 4.3055283484206565e-06,
+      "loss": 0.4228,
+      "step": 297
+    },
+    {
+      "epoch": 0.5706079463858306,
+      "grad_norm": 0.7978037595748901,
+      "learning_rate": 4.27418017137422e-06,
+      "loss": 0.3745,
+      "step": 298
+    },
+    {
+      "epoch": 0.572522738152226,
+      "grad_norm": 0.786128044128418,
+      "learning_rate": 4.2428611124771184e-06,
+      "loss": 0.3708,
+      "step": 299
+    },
+    {
+      "epoch": 0.5744375299186214,
+      "grad_norm": 0.7764582633972168,
+      "learning_rate": 4.211572428174816e-06,
+      "loss": 0.3617,
+      "step": 300
+    },
+    {
+      "epoch": 0.5763523216850167,
+      "grad_norm": 0.8071704506874084,
+      "learning_rate": 4.180315373694225e-06,
+      "loss": 0.4013,
+      "step": 301
+    },
+    {
+      "epoch": 0.5782671134514121,
+      "grad_norm": 0.8064809441566467,
+      "learning_rate": 4.149091202993345e-06,
+      "loss": 0.3593,
+      "step": 302
+    },
+    {
+      "epoch": 0.5801819052178075,
+      "grad_norm": 0.8199291229248047,
+      "learning_rate": 4.11790116871096e-06,
+      "loss": 0.4174,
+      "step": 303
+    },
+    {
+      "epoch": 0.582096696984203,
+      "grad_norm": 0.9001505374908447,
+      "learning_rate": 4.086746522116372e-06,
+      "loss": 0.4535,
+      "step": 304
+    },
+    {
+      "epoch": 0.5840114887505984,
+      "grad_norm": 0.7408649325370789,
+      "learning_rate": 4.055628513059231e-06,
+      "loss": 0.3874,
+      "step": 305
+    },
+    {
+      "epoch": 0.5859262805169938,
+      "grad_norm": 0.8281508684158325,
+      "learning_rate": 4.02454838991936e-06,
+      "loss": 0.3783,
+      "step": 306
+    },
+    {
+      "epoch": 0.5878410722833892,
+      "grad_norm": 0.8515765070915222,
+      "learning_rate": 3.993507399556699e-06,
+      "loss": 0.4306,
+      "step": 307
+    },
+    {
+      "epoch": 0.5897558640497846,
+      "grad_norm": 0.8016983270645142,
+      "learning_rate": 3.962506787261278e-06,
+      "loss": 0.3988,
+      "step": 308
+    },
+    {
+      "epoch": 0.59167065581618,
+      "grad_norm": 0.7859688997268677,
+      "learning_rate": 3.931547796703245e-06,
+      "loss": 0.3801,
+      "step": 309
+    },
+    {
+      "epoch": 0.5935854475825754,
+      "grad_norm": 0.8068103194236755,
+      "learning_rate": 3.900631669882996e-06,
+      "loss": 0.3925,
+      "step": 310
+    },
+    {
+      "epoch": 0.5955002393489708,
+      "grad_norm": 0.811817467212677,
+      "learning_rate": 3.869759647081326e-06,
+      "loss": 0.3693,
+      "step": 311
+    },
+    {
+      "epoch": 0.5974150311153662,
+      "grad_norm": 0.7820096611976624,
+      "learning_rate": 3.83893296680969e-06,
+      "loss": 0.3837,
+      "step": 312
+    },
+    {
+      "epoch": 0.5993298228817616,
+      "grad_norm": 0.7839851379394531,
+      "learning_rate": 3.8081528657605045e-06,
+      "loss": 0.3762,
+      "step": 313
+    },
+    {
+      "epoch": 0.601244614648157,
+      "grad_norm": 0.7672075033187866,
+      "learning_rate": 3.7774205787575455e-06,
+      "loss": 0.3872,
+      "step": 314
+    },
+    {
+      "epoch": 0.6031594064145525,
+      "grad_norm": 0.8048450946807861,
+      "learning_rate": 3.7467373387063973e-06,
+      "loss": 0.4239,
+      "step": 315
+    },
+    {
+      "epoch": 0.6050741981809479,
+      "grad_norm": 0.8440662622451782,
+      "learning_rate": 3.7161043765450044e-06,
+      "loss": 0.4095,
+      "step": 316
+    },
+    {
+      "epoch": 0.6069889899473432,
+      "grad_norm": 0.8100243210792542,
+      "learning_rate": 3.685522921194276e-06,
+      "loss": 0.3772,
+      "step": 317
+    },
+    {
+      "epoch": 0.6089037817137386,
+      "grad_norm": 0.8134592175483704,
+      "learning_rate": 3.6549941995088012e-06,
+      "loss": 0.3983,
+      "step": 318
+    },
+    {
+      "epoch": 0.610818573480134,
+      "grad_norm": 0.8278323411941528,
+      "learning_rate": 3.62451943622761e-06,
+      "loss": 0.4249,
+      "step": 319
+    },
+    {
+      "epoch": 0.6127333652465294,
+      "grad_norm": 0.7204639315605164,
+      "learning_rate": 3.5940998539250614e-06,
+      "loss": 0.3538,
+      "step": 320
+    },
+    {
+      "epoch": 0.6146481570129249,
+      "grad_norm": 0.7791761159896851,
+      "learning_rate": 3.5637366729617766e-06,
+      "loss": 0.3741,
+      "step": 321
+    },
+    {
+      "epoch": 0.6165629487793203,
+      "grad_norm": 0.8510886430740356,
+      "learning_rate": 3.5334311114356983e-06,
+      "loss": 0.4194,
+      "step": 322
+    },
+    {
+      "epoch": 0.6184777405457157,
+      "grad_norm": 0.823306679725647,
+      "learning_rate": 3.5031843851332105e-06,
+      "loss": 0.4171,
+      "step": 323
+    },
+    {
+      "epoch": 0.6203925323121111,
+      "grad_norm": 0.8224660158157349,
+      "learning_rate": 3.472997707480372e-06,
+      "loss": 0.3959,
+      "step": 324
+    },
+    {
+      "epoch": 0.6223073240785064,
+      "grad_norm": 0.876380980014801,
+      "learning_rate": 3.4428722894942313e-06,
+      "loss": 0.4203,
+      "step": 325
+    },
+    {
+      "epoch": 0.6242221158449018,
+      "grad_norm": 0.8525562286376953,
+      "learning_rate": 3.4128093397342508e-06,
+      "loss": 0.4404,
+      "step": 326
+    },
+    {
+      "epoch": 0.6261369076112973,
+      "grad_norm": 0.7982501983642578,
+      "learning_rate": 3.3828100642538097e-06,
+      "loss": 0.3801,
+      "step": 327
+    },
+    {
+      "epoch": 0.6280516993776927,
+      "grad_norm": 0.779940664768219,
+      "learning_rate": 3.352875666551838e-06,
+      "loss": 0.3674,
+      "step": 328
+    },
+    {
+      "epoch": 0.6299664911440881,
+      "grad_norm": 0.7782029509544373,
+      "learning_rate": 3.323007347524515e-06,
+      "loss": 0.4004,
+      "step": 329
+    },
+    {
+      "epoch": 0.6318812829104835,
+      "grad_norm": 0.7265905737876892,
+      "learning_rate": 3.2932063054171108e-06,
+      "loss": 0.3854,
+      "step": 330
+    },
+    {
+      "epoch": 0.6337960746768789,
+      "grad_norm": 0.7720629572868347,
+      "learning_rate": 3.2634737357758994e-06,
+      "loss": 0.3828,
+      "step": 331
+    },
+    {
+      "epoch": 0.6357108664432743,
+      "grad_norm": 0.7668139338493347,
+      "learning_rate": 3.2338108314002102e-06,
+      "loss": 0.3815,
+      "step": 332
+    },
+    {
+      "epoch": 0.6376256582096697,
+      "grad_norm": 0.7281336784362793,
+      "learning_rate": 3.204218782294565e-06,
+      "loss": 0.3678,
+      "step": 333
+    },
+    {
+      "epoch": 0.6395404499760651,
+      "grad_norm": 0.7879943251609802,
+      "learning_rate": 3.174698775620947e-06,
+      "loss": 0.3949,
+      "step": 334
+    },
+    {
+      "epoch": 0.6414552417424605,
+      "grad_norm": 0.8132904171943665,
+      "learning_rate": 3.145251995651162e-06,
+      "loss": 0.4011,
+      "step": 335
+    },
+    {
+      "epoch": 0.6433700335088559,
+      "grad_norm": 0.7830522656440735,
+      "learning_rate": 3.1158796237193444e-06,
+      "loss": 0.3785,
+      "step": 336
+    },
+    {
+      "epoch": 0.6452848252752513,
+      "grad_norm": 0.7736674547195435,
+      "learning_rate": 3.0865828381745515e-06,
+      "loss": 0.3817,
+      "step": 337
+    },
+    {
+      "epoch": 0.6471996170416467,
+      "grad_norm": 0.7734500765800476,
+      "learning_rate": 3.0573628143334986e-06,
+      "loss": 0.3623,
+      "step": 338
+    },
+    {
+      "epoch": 0.6491144088080422,
+      "grad_norm": 0.7965434193611145,
+      "learning_rate": 3.0282207244334084e-06,
+      "loss": 0.3722,
+      "step": 339
+    },
+    {
+      "epoch": 0.6510292005744376,
+      "grad_norm": 0.7860566973686218,
+      "learning_rate": 2.999157737584971e-06,
+      "loss": 0.3757,
+      "step": 340
+    },
+    {
+      "epoch": 0.6529439923408329,
+      "grad_norm": 0.8249862790107727,
+      "learning_rate": 2.970175019725465e-06,
+      "loss": 0.4184,
+      "step": 341
+    },
+    {
+      "epoch": 0.6548587841072283,
+      "grad_norm": 0.9115679860115051,
+      "learning_rate": 2.94127373357196e-06,
+      "loss": 0.4364,
+      "step": 342
+    },
+    {
+      "epoch": 0.6567735758736237,
+      "grad_norm": 0.8391016125679016,
+      "learning_rate": 2.912455038574686e-06,
+      "loss": 0.3973,
+      "step": 343
+    },
+    {
+      "epoch": 0.6586883676400191,
+      "grad_norm": 0.7986558079719543,
+      "learning_rate": 2.88372009087051e-06,
+      "loss": 0.4069,
+      "step": 344
+    },
+    {
+      "epoch": 0.6606031594064146,
+      "grad_norm": 0.7595184445381165,
+      "learning_rate": 2.8550700432365647e-06,
+      "loss": 0.3619,
+      "step": 345
+    },
+    {
+      "epoch": 0.66251795117281,
+      "grad_norm": 0.907979428768158,
+      "learning_rate": 2.8265060450439887e-06,
+      "loss": 0.4441,
+      "step": 346
+    },
+    {
+      "epoch": 0.6644327429392054,
+      "grad_norm": 0.801184356212616,
+      "learning_rate": 2.7980292422118282e-06,
+      "loss": 0.3791,
+      "step": 347
+    },
+    {
+      "epoch": 0.6663475347056008,
+      "grad_norm": 0.8330556154251099,
+      "learning_rate": 2.769640777161063e-06,
+      "loss": 0.3936,
+      "step": 348
+    },
+    {
+      "epoch": 0.6682623264719961,
+      "grad_norm": 0.8160421252250671,
+      "learning_rate": 2.7413417887687644e-06,
+      "loss": 0.4238,
+      "step": 349
+    },
+    {
+      "epoch": 0.6701771182383915,
+      "grad_norm": 0.8634615540504456,
+      "learning_rate": 2.713133412322424e-06,
+      "loss": 0.3979,
+      "step": 350
+    },
+    {
+      "epoch": 0.672091910004787,
+      "grad_norm": 0.8537343144416809,
+      "learning_rate": 2.6850167794743966e-06,
+      "loss": 0.4244,
+      "step": 351
+    },
+    {
+      "epoch": 0.6740067017711824,
+      "grad_norm": 0.7988967895507812,
+      "learning_rate": 2.6569930181965e-06,
+      "loss": 0.3854,
+      "step": 352
+    },
+    {
+      "epoch": 0.6759214935375778,
+      "grad_norm": 0.7857111096382141,
+      "learning_rate": 2.629063252734775e-06,
+      "loss": 0.3956,
+      "step": 353
+    },
+    {
+      "epoch": 0.6778362853039732,
+      "grad_norm": 0.8199499249458313,
+      "learning_rate": 2.601228603564368e-06,
+      "loss": 0.4084,
+      "step": 354
+    },
+    {
+      "epoch": 0.6797510770703686,
+      "grad_norm": 0.8062904477119446,
+      "learning_rate": 2.573490187344596e-06,
+      "loss": 0.3897,
+      "step": 355
+    },
+    {
+      "epoch": 0.681665868836764,
+      "grad_norm": 0.7628870606422424,
+      "learning_rate": 2.545849116874132e-06,
+      "loss": 0.3457,
+      "step": 356
+    },
+    {
+      "epoch": 0.6835806606031594,
+      "grad_norm": 0.7267881631851196,
+      "learning_rate": 2.5183065010463813e-06,
+      "loss": 0.3646,
+      "step": 357
+    },
+    {
+      "epoch": 0.6854954523695548,
+      "grad_norm": 0.7799604535102844,
+      "learning_rate": 2.490863444804976e-06,
+      "loss": 0.3586,
+      "step": 358
+    },
+    {
+      "epoch": 0.6874102441359502,
+      "grad_norm": 0.8678038120269775,
+      "learning_rate": 2.4635210490994648e-06,
+      "loss": 0.4128,
+      "step": 359
+    },
+    {
+      "epoch": 0.6893250359023456,
+      "grad_norm": 0.7644969820976257,
+      "learning_rate": 2.436280410841128e-06,
+      "loss": 0.3695,
+      "step": 360
+    },
+    {
+      "epoch": 0.691239827668741,
+      "grad_norm": 0.8090047240257263,
+      "learning_rate": 2.409142622858992e-06,
+      "loss": 0.3784,
+      "step": 361
+    },
+    {
+      "epoch": 0.6931546194351365,
+      "grad_norm": 0.7928206324577332,
+      "learning_rate": 2.3821087738559674e-06,
+      "loss": 0.3739,
+      "step": 362
+    },
+    {
+      "epoch": 0.6950694112015319,
+      "grad_norm": 0.7725571990013123,
+      "learning_rate": 2.3551799483651894e-06,
+      "loss": 0.3761,
+      "step": 363
+    },
+    {
+      "epoch": 0.6969842029679273,
+      "grad_norm": 0.8900219202041626,
+      "learning_rate": 2.3283572267064963e-06,
+      "loss": 0.4545,
+      "step": 364
+    },
+    {
+      "epoch": 0.6988989947343226,
+      "grad_norm": 0.7737187743186951,
+      "learning_rate": 2.3016416849431023e-06,
+      "loss": 0.4071,
+      "step": 365
+    },
+    {
+      "epoch": 0.700813786500718,
+      "grad_norm": 0.8239535093307495,
+      "learning_rate": 2.275034394838413e-06,
+      "loss": 0.4067,
+      "step": 366
+    },
+    {
+      "epoch": 0.7027285782671134,
+      "grad_norm": 0.8072943687438965,
+      "learning_rate": 2.2485364238130435e-06,
+      "loss": 0.4037,
+      "step": 367
+    },
+    {
+      "epoch": 0.7046433700335089,
+      "grad_norm": 0.7579889893531799,
+      "learning_rate": 2.2221488349019903e-06,
+      "loss": 0.3587,
+      "step": 368
+    },
+    {
+      "epoch": 0.7065581617999043,
+      "grad_norm": 0.8152466416358948,
+      "learning_rate": 2.1958726867119785e-06,
+      "loss": 0.4252,
+      "step": 369
+    },
+    {
+      "epoch": 0.7084729535662997,
+      "grad_norm": 0.7977010607719421,
+      "learning_rate": 2.169709033379004e-06,
+      "loss": 0.3942,
+      "step": 370
+    },
+    {
+      "epoch": 0.7103877453326951,
+      "grad_norm": 0.81230628490448,
+      "learning_rate": 2.1436589245260375e-06,
+      "loss": 0.4117,
+      "step": 371
+    },
+    {
+      "epoch": 0.7123025370990905,
+      "grad_norm": 0.8957315683364868,
+      "learning_rate": 2.1177234052209208e-06,
+      "loss": 0.4393,
+      "step": 372
+    },
+    {
+      "epoch": 0.7142173288654858,
+      "grad_norm": 0.8771640658378601,
+      "learning_rate": 2.09190351593443e-06,
+      "loss": 0.4254,
+      "step": 373
+    },
+    {
+      "epoch": 0.7161321206318813,
+      "grad_norm": 0.7827796339988708,
+      "learning_rate": 2.066200292498553e-06,
+      "loss": 0.389,
+      "step": 374
+    },
+    {
+      "epoch": 0.7180469123982767,
+      "grad_norm": 0.7941052317619324,
+      "learning_rate": 2.040614766064913e-06,
+      "loss": 0.3842,
+      "step": 375
+    },
+    {
+      "epoch": 0.7199617041646721,
+      "grad_norm": 0.8104212284088135,
+      "learning_rate": 2.0151479630634225e-06,
+      "loss": 0.4149,
+      "step": 376
+    },
+    {
+      "epoch": 0.7218764959310675,
+      "grad_norm": 0.7843247056007385,
+      "learning_rate": 1.9898009051610847e-06,
+      "loss": 0.3723,
+      "step": 377
+    },
+    {
+      "epoch": 0.7237912876974629,
+      "grad_norm": 0.7872182130813599,
+      "learning_rate": 1.964574609221026e-06,
+      "loss": 0.394,
+      "step": 378
+    },
+    {
+      "epoch": 0.7257060794638583,
+      "grad_norm": 0.7840164303779602,
+      "learning_rate": 1.9394700872616856e-06,
+      "loss": 0.3603,
+      "step": 379
+    },
+    {
+      "epoch": 0.7276208712302538,
+      "grad_norm": 0.8055151700973511,
+      "learning_rate": 1.914488346416229e-06,
+      "loss": 0.3999,
+      "step": 380
+    },
+    {
+      "epoch": 0.7295356629966491,
+      "grad_norm": 0.8342682123184204,
+      "learning_rate": 1.8896303888921313e-06,
+      "loss": 0.4221,
+      "step": 381
+    },
+    {
+      "epoch": 0.7314504547630445,
+      "grad_norm": 0.7286022305488586,
+      "learning_rate": 1.8648972119309854e-06,
+      "loss": 0.3635,
+      "step": 382
+    },
+    {
+      "epoch": 0.7333652465294399,
+      "grad_norm": 0.839953601360321,
+      "learning_rate": 1.8402898077684806e-06,
+      "loss": 0.4145,
+      "step": 383
+    },
+    {
+      "epoch": 0.7352800382958353,
+      "grad_norm": 0.8201612830162048,
+      "learning_rate": 1.815809163594609e-06,
+      "loss": 0.4102,
+      "step": 384
+    },
+    {
+      "epoch": 0.7371948300622307,
+      "grad_norm": 0.8541584610939026,
+      "learning_rate": 1.7914562615140507e-06,
+      "loss": 0.4286,
+      "step": 385
+    },
+    {
+      "epoch": 0.7391096218286262,
+      "grad_norm": 0.7880145907402039,
+      "learning_rate": 1.7672320785067871e-06,
+      "loss": 0.3997,
+      "step": 386
+    },
+    {
+      "epoch": 0.7410244135950216,
+      "grad_norm": 0.8472251296043396,
+      "learning_rate": 1.74313758638889e-06,
+      "loss": 0.3665,
+      "step": 387
+    },
+    {
+      "epoch": 0.742939205361417,
+      "grad_norm": 0.8149013519287109,
+      "learning_rate": 1.7191737517735513e-06,
+      "loss": 0.3781,
+      "step": 388
+    },
+    {
+      "epoch": 0.7448539971278123,
+      "grad_norm": 0.7976882457733154,
+      "learning_rate": 1.6953415360322972e-06,
+      "loss": 0.3826,
+      "step": 389
+    },
+    {
+      "epoch": 0.7467687888942077,
+      "grad_norm": 0.7928311824798584,
+      "learning_rate": 1.6716418952564145e-06,
+      "loss": 0.3658,
+      "step": 390
+    },
+    {
+      "epoch": 0.7486835806606031,
+      "grad_norm": 0.7540788650512695,
+      "learning_rate": 1.648075780218607e-06,
+      "loss": 0.3496,
+      "step": 391
+    },
+    {
+      "epoch": 0.7505983724269986,
+      "grad_norm": 0.887441873550415,
+      "learning_rate": 1.6246441363348453e-06,
+      "loss": 0.39,
+      "step": 392
+    },
+    {
+      "epoch": 0.752513164193394,
+      "grad_norm": 0.7524732351303101,
+      "learning_rate": 1.6013479036264358e-06,
+      "loss": 0.373,
+      "step": 393
+    },
+    {
+      "epoch": 0.7544279559597894,
+      "grad_norm": 0.7478716969490051,
+      "learning_rate": 1.57818801668232e-06,
+      "loss": 0.3375,
+      "step": 394
+    },
+    {
+      "epoch": 0.7563427477261848,
+      "grad_norm": 0.8160089254379272,
+      "learning_rate": 1.555165404621567e-06,
+      "loss": 0.4017,
+      "step": 395
+    },
+    {
+      "epoch": 0.7582575394925802,
+      "grad_norm": 0.9428865909576416,
+      "learning_rate": 1.532280991056116e-06,
+      "loss": 0.452,
+      "step": 396
+    },
+    {
+      "epoch": 0.7601723312589755,
+      "grad_norm": 0.7749152779579163,
+      "learning_rate": 1.5095356940537053e-06,
+      "loss": 0.388,
+      "step": 397
+    },
+    {
+      "epoch": 0.762087123025371,
+      "grad_norm": 0.8464208841323853,
+      "learning_rate": 1.4869304261010586e-06,
+      "loss": 0.4002,
+      "step": 398
+    },
+    {
+      "epoch": 0.7640019147917664,
+      "grad_norm": 0.8656209707260132,
+      "learning_rate": 1.4644660940672628e-06,
+      "loss": 0.4426,
+      "step": 399
+    },
+    {
+      "epoch": 0.7659167065581618,
+      "grad_norm": 0.7765882015228271,
+      "learning_rate": 1.4421435991674e-06,
+      "loss": 0.3765,
+      "step": 400
+    },
+    {
+      "epoch": 0.7678314983245572,
+      "grad_norm": 0.7365686893463135,
+      "learning_rate": 1.4199638369263858e-06,
+      "loss": 0.3235,
+      "step": 401
+    },
+    {
+      "epoch": 0.7697462900909526,
+      "grad_norm": 0.7246995568275452,
+      "learning_rate": 1.3979276971430406e-06,
+      "loss": 0.3426,
+      "step": 402
+    },
+    {
+      "epoch": 0.771661081857348,
+      "grad_norm": 0.761492133140564,
+      "learning_rate": 1.3760360638544012e-06,
+      "loss": 0.3938,
+      "step": 403
+    },
+    {
+      "epoch": 0.7735758736237435,
+      "grad_norm": 0.7741461992263794,
+      "learning_rate": 1.3542898153002453e-06,
+      "loss": 0.3781,
+      "step": 404
+    },
+    {
+      "epoch": 0.7754906653901388,
+      "grad_norm": 0.8266602158546448,
+      "learning_rate": 1.3326898238878716e-06,
+      "loss": 0.393,
+      "step": 405
+    },
+    {
+      "epoch": 0.7774054571565342,
+      "grad_norm": 0.8508222103118896,
+      "learning_rate": 1.3112369561570842e-06,
+      "loss": 0.3935,
+      "step": 406
+    },
+    {
+      "epoch": 0.7793202489229296,
+      "grad_norm": 0.7993992567062378,
+      "learning_rate": 1.2899320727454472e-06,
+      "loss": 0.3647,
+      "step": 407
+    },
+    {
+      "epoch": 0.781235040689325,
+      "grad_norm": 0.7505406737327576,
+      "learning_rate": 1.2687760283537414e-06,
+      "loss": 0.3869,
+      "step": 408
+    },
+    {
+      "epoch": 0.7831498324557205,
+      "grad_norm": 0.817995548248291,
+      "learning_rate": 1.2477696717116878e-06,
+      "loss": 0.4346,
+      "step": 409
+    },
+    {
+      "epoch": 0.7850646242221159,
+      "grad_norm": 0.7967957854270935,
+      "learning_rate": 1.226913845543895e-06,
+      "loss": 0.3821,
+      "step": 410
+    },
+    {
+      "epoch": 0.7869794159885113,
+      "grad_norm": 0.8469412922859192,
+      "learning_rate": 1.2062093865360458e-06,
+      "loss": 0.4482,
+      "step": 411
+    },
+    {
+      "epoch": 0.7888942077549067,
+      "grad_norm": 0.7760953903198242,
+      "learning_rate": 1.1856571253013393e-06,
+      "loss": 0.3687,
+      "step": 412
+    },
+    {
+      "epoch": 0.790808999521302,
+      "grad_norm": 0.8256305456161499,
+      "learning_rate": 1.1652578863471664e-06,
+      "loss": 0.4058,
+      "step": 413
+    },
+    {
+      "epoch": 0.7927237912876974,
+      "grad_norm": 0.7707417607307434,
+      "learning_rate": 1.145012488042026e-06,
+      "loss": 0.3646,
+      "step": 414
+    },
+    {
+      "epoch": 0.7946385830540929,
+      "grad_norm": 0.9702382683753967,
+      "learning_rate": 1.1249217425827063e-06,
+      "loss": 0.4735,
+      "step": 415
+    },
+    {
+      "epoch": 0.7965533748204883,
+      "grad_norm": 0.8380041718482971,
+      "learning_rate": 1.1049864559616885e-06,
+      "loss": 0.406,
+      "step": 416
+    },
+    {
+      "epoch": 0.7984681665868837,
+      "grad_norm": 0.7933167815208435,
+      "learning_rate": 1.0852074279348234e-06,
+      "loss": 0.3765,
+      "step": 417
+    },
+    {
+      "epoch": 0.8003829583532791,
+      "grad_norm": 0.7869566082954407,
+      "learning_rate": 1.0655854519892367e-06,
+      "loss": 0.3806,
+      "step": 418
+    },
+    {
+      "epoch": 0.8022977501196745,
+      "grad_norm": 0.7924370169639587,
+      "learning_rate": 1.046121315311508e-06,
+      "loss": 0.3953,
+      "step": 419
+    },
+    {
+      "epoch": 0.8042125418860699,
+      "grad_norm": 0.7889004945755005,
+      "learning_rate": 1.0268157987560773e-06,
+      "loss": 0.3996,
+      "step": 420
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 522,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 105,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.2094028312097587e+17,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-420/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66ffb9355835ef1e788c4b73fef523b9594f231dd9bde187500393bdf3018899
+size 10936

checkpoint-420/zero_to_fp32.py

@@ -0,0 +1,674 @@
+#!/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 . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import json
+from tqdm import tqdm
+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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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 _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+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() if _has_callable(shape, 'numel') else math.prod(shape)
+            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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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 tqdm(param_shapes.items(), desc='Gathering Sharded Weights'):
+        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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    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_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        state_dict_split = split_torch_state_dict_into_shards(state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard = {tensor: state_dict[tensor].contiguous() for tensor in tensors}
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard, output_path)
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+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_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

checkpoint-522/added_tokens.json

@@ -0,0 +1,24 @@
+{
+  "</tool_call>": 151658,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}