Upload 17 files

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

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
+}

config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "Qwen/Qwen2.5-0.5B",
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.48.2",
+  "use_cache": true,
+  "use_mrope": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.48.2"
+}

latest

@@ -0,0 +1 @@
+global_step185

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a466322107f68f0a3c43f872627c4b7280c0ed1d3ebc7c839ab644da8de518d6
+size 988097824

rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3d22bceae33d8fa6a9e820d963f99d0f227ce7b3477aa4373ed2fbe3a43f885
+size 14960

rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ecd010a86fd576fcd29315fc3a11d52719bd00ffa884e166f0be898f8838904d
+size 14960

rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4dcaaaf463d3940cee4cc80fe5010ffb143ff956ebdf8094b13819faf9882797
+size 14960

scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fafb6eac3438bdded8cbd63e49dc9dc46e78320e402770e1f7a0035f1dae4984
+size 1064

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": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5eee858c5123a4279c3e1f7b81247343f356ac767940b2692a928ad929543214
+size 11422063

tokenizer_config.json

@@ -0,0 +1,209 @@
+{
+  "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 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 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": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "left",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

trainer_state.json

@@ -0,0 +1,2451 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.9813333333333332,
+  "eval_steps": 10,
+  "global_step": 186,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "completion_length": 535.1041831970215,
+      "epoch": 0.010666666666666666,
+      "grad_norm": 0.10447478294372559,
+      "kl": 0.0,
+      "learning_rate": 1.5789473684210525e-07,
+      "loss": -0.0025,
+      "reward": 0.031250000931322575,
+      "reward_std": 0.10825317353010178,
+      "rewards/accuracy_reward": 0.031250000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 1
+    },
+    {
+      "completion_length": 555.5937728881836,
+      "epoch": 0.021333333333333333,
+      "grad_norm": 0.05305163562297821,
+      "kl": 0.0,
+      "learning_rate": 3.157894736842105e-07,
+      "loss": 0.009,
+      "reward": 0.02083333395421505,
+      "reward_std": 0.04865618050098419,
+      "rewards/accuracy_reward": 0.02083333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 2
+    },
+    {
+      "completion_length": 510.41668701171875,
+      "epoch": 0.032,
+      "grad_norm": 0.0009671024745330215,
+      "kl": 0.00044345855712890625,
+      "learning_rate": 4.736842105263158e-07,
+      "loss": 0.0,
+      "reward": 0.0,
+      "reward_std": 0.0,
+      "rewards/accuracy_reward": 0.0,
+      "rewards/format_reward": 0.0,
+      "step": 3
+    },
+    {
+      "completion_length": 570.4166831970215,
+      "epoch": 0.042666666666666665,
+      "grad_norm": 0.06929787248373032,
+      "kl": 0.00043070316314697266,
+      "learning_rate": 6.31578947368421e-07,
+      "loss": -0.0023,
+      "reward": 0.02083333395421505,
+      "reward_std": 0.04865618050098419,
+      "rewards/accuracy_reward": 0.02083333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 4
+    },
+    {
+      "completion_length": 564.0833435058594,
+      "epoch": 0.05333333333333334,
+      "grad_norm": 0.03304513171315193,
+      "kl": 0.0004966259002685547,
+      "learning_rate": 7.894736842105263e-07,
+      "loss": 0.0287,
+      "reward": 0.010416666977107525,
+      "reward_std": 0.03608439117670059,
+      "rewards/accuracy_reward": 0.010416666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 5
+    },
+    {
+      "completion_length": 434.66668701171875,
+      "epoch": 0.064,
+      "grad_norm": 0.0016080428613349795,
+      "kl": 0.0005583763122558594,
+      "learning_rate": 9.473684210526316e-07,
+      "loss": 0.0,
+      "reward": 0.0,
+      "reward_std": 0.0,
+      "rewards/accuracy_reward": 0.0,
+      "rewards/format_reward": 0.0,
+      "step": 6
+    },
+    {
+      "completion_length": 617.7396087646484,
+      "epoch": 0.07466666666666667,
+      "grad_norm": 0.08006145060062408,
+      "kl": 0.000385284423828125,
+      "learning_rate": 1.1052631578947369e-06,
+      "loss": 0.0445,
+      "reward": 0.052083334885537624,
+      "reward_std": 0.13339675962924957,
+      "rewards/accuracy_reward": 0.052083334885537624,
+      "rewards/format_reward": 0.0,
+      "step": 7
+    },
+    {
+      "completion_length": 527.3333511352539,
+      "epoch": 0.08533333333333333,
+      "grad_norm": 0.0587327815592289,
+      "kl": 0.0003993511199951172,
+      "learning_rate": 1.263157894736842e-06,
+      "loss": 0.0507,
+      "reward": 0.031250000931322575,
+      "reward_std": 0.10825317353010178,
+      "rewards/accuracy_reward": 0.031250000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 8
+    },
+    {
+      "completion_length": 551.2083435058594,
+      "epoch": 0.096,
+      "grad_norm": 0.001157123246230185,
+      "kl": 0.0004718303680419922,
+      "learning_rate": 1.4210526315789473e-06,
+      "loss": 0.0,
+      "reward": 0.0,
+      "reward_std": 0.0,
+      "rewards/accuracy_reward": 0.0,
+      "rewards/format_reward": 0.0,
+      "step": 9
+    },
+    {
+      "completion_length": 486.7291793823242,
+      "epoch": 0.10666666666666667,
+      "grad_norm": 0.12886527180671692,
+      "kl": 0.00043964385986328125,
+      "learning_rate": 1.5789473684210526e-06,
+      "loss": -0.0082,
+      "reward": 0.08333333395421505,
+      "reward_std": 0.13744790107011795,
+      "rewards/accuracy_reward": 0.08333333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 10
+    },
+    {
+      "completion_length": 512.5000114440918,
+      "epoch": 0.11733333333333333,
+      "grad_norm": 0.10466578602790833,
+      "kl": 0.0008594989776611328,
+      "learning_rate": 1.736842105263158e-06,
+      "loss": 0.0013,
+      "reward": 0.031250000931322575,
+      "reward_std": 0.10825317353010178,
+      "rewards/accuracy_reward": 0.031250000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 11
+    },
+    {
+      "completion_length": 485.0000114440918,
+      "epoch": 0.128,
+      "grad_norm": 0.17413121461868286,
+      "kl": 0.0007164478302001953,
+      "learning_rate": 1.8947368421052632e-06,
+      "loss": -0.0509,
+      "reward": 0.031250000931322575,
+      "reward_std": 0.10825317353010178,
+      "rewards/accuracy_reward": 0.031250000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 12
+    },
+    {
+      "completion_length": 495.39584732055664,
+      "epoch": 0.13866666666666666,
+      "grad_norm": 0.0025116761680692434,
+      "kl": 0.0007200241088867188,
+      "learning_rate": 2.0526315789473687e-06,
+      "loss": 0.0,
+      "reward": 0.0,
+      "reward_std": 0.0,
+      "rewards/accuracy_reward": 0.0,
+      "rewards/format_reward": 0.0,
+      "step": 13
+    },
+    {
+      "completion_length": 579.197925567627,
+      "epoch": 0.14933333333333335,
+      "grad_norm": 0.10607028752565384,
+      "kl": 0.001117706298828125,
+      "learning_rate": 2.2105263157894738e-06,
+      "loss": 0.0997,
+      "reward": 0.05208333395421505,
+      "reward_std": 0.12870215997099876,
+      "rewards/accuracy_reward": 0.05208333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 14
+    },
+    {
+      "completion_length": 618.0729370117188,
+      "epoch": 0.16,
+      "grad_norm": 0.04636254161596298,
+      "kl": 0.00128936767578125,
+      "learning_rate": 2.368421052631579e-06,
+      "loss": -0.0278,
+      "reward": 0.010416666977107525,
+      "reward_std": 0.03608439117670059,
+      "rewards/accuracy_reward": 0.010416666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 15
+    },
+    {
+      "completion_length": 529.8020973205566,
+      "epoch": 0.17066666666666666,
+      "grad_norm": 0.12151255458593369,
+      "kl": 0.0032758712768554688,
+      "learning_rate": 2.526315789473684e-06,
+      "loss": 0.0198,
+      "reward": 0.0416666679084301,
+      "reward_std": 0.14433756470680237,
+      "rewards/accuracy_reward": 0.0416666679084301,
+      "rewards/format_reward": 0.0,
+      "step": 16
+    },
+    {
+      "completion_length": 583.6979370117188,
+      "epoch": 0.18133333333333335,
+      "grad_norm": 0.09600550681352615,
+      "kl": 0.0034093856811523438,
+      "learning_rate": 2.6842105263157895e-06,
+      "loss": 0.0187,
+      "reward": 0.0729166679084301,
+      "reward_std": 0.17735834047198296,
+      "rewards/accuracy_reward": 0.0729166679084301,
+      "rewards/format_reward": 0.0,
+      "step": 17
+    },
+    {
+      "completion_length": 526.9583473205566,
+      "epoch": 0.192,
+      "grad_norm": 0.09340275824069977,
+      "kl": 0.005596160888671875,
+      "learning_rate": 2.8421052631578946e-06,
+      "loss": 0.026,
+      "reward": 0.0416666679084301,
+      "reward_std": 0.12082496285438538,
+      "rewards/accuracy_reward": 0.0416666679084301,
+      "rewards/format_reward": 0.0,
+      "step": 18
+    },
+    {
+      "completion_length": 465.8645935058594,
+      "epoch": 0.20266666666666666,
+      "grad_norm": 0.08133361488580704,
+      "kl": 0.006252288818359375,
+      "learning_rate": 3e-06,
+      "loss": -0.0449,
+      "reward": 0.052083334885537624,
+      "reward_std": 0.13339675962924957,
+      "rewards/accuracy_reward": 0.052083334885537624,
+      "rewards/format_reward": 0.0,
+      "step": 19
+    },
+    {
+      "completion_length": 525.0937690734863,
+      "epoch": 0.21333333333333335,
+      "grad_norm": 0.1193581074476242,
+      "kl": 0.008209228515625,
+      "learning_rate": 2.9997345912364375e-06,
+      "loss": -0.0093,
+      "reward": 0.07291666883975267,
+      "reward_std": 0.16979892551898956,
+      "rewards/accuracy_reward": 0.07291666883975267,
+      "rewards/format_reward": 0.0,
+      "step": 20
+    },
+    {
+      "completion_length": 504.42709732055664,
+      "epoch": 0.224,
+      "grad_norm": 0.1320876032114029,
+      "kl": 0.011646270751953125,
+      "learning_rate": 2.9989384588681657e-06,
+      "loss": -0.027,
+      "reward": 0.13541666697710752,
+      "reward_std": 0.21443026512861252,
+      "rewards/accuracy_reward": 0.13541666697710752,
+      "rewards/format_reward": 0.0,
+      "step": 21
+    },
+    {
+      "completion_length": 566.2187690734863,
+      "epoch": 0.23466666666666666,
+      "grad_norm": 0.11368165165185928,
+      "kl": 0.012992382049560547,
+      "learning_rate": 2.9976118846291945e-06,
+      "loss": 0.0194,
+      "reward": 0.06250000186264515,
+      "reward_std": 0.19299374520778656,
+      "rewards/accuracy_reward": 0.06250000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 22
+    },
+    {
+      "completion_length": 457.18751525878906,
+      "epoch": 0.24533333333333332,
+      "grad_norm": 0.21248622238636017,
+      "kl": 0.023479461669921875,
+      "learning_rate": 2.995755337965428e-06,
+      "loss": 0.0218,
+      "reward": 0.11458333767950535,
+      "reward_std": 0.2338472306728363,
+      "rewards/accuracy_reward": 0.11458333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 23
+    },
+    {
+      "completion_length": 559.9166831970215,
+      "epoch": 0.256,
+      "grad_norm": 0.129845529794693,
+      "kl": 0.02849578857421875,
+      "learning_rate": 2.99336947586854e-06,
+      "loss": -0.035,
+      "reward": 0.12500000186264515,
+      "reward_std": 0.28254789859056473,
+      "rewards/accuracy_reward": 0.12500000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 24
+    },
+    {
+      "completion_length": 564.6979446411133,
+      "epoch": 0.26666666666666666,
+      "grad_norm": 0.10117102414369583,
+      "kl": 0.0623321533203125,
+      "learning_rate": 2.9904551426434754e-06,
+      "loss": 0.0704,
+      "reward": 0.1145833358168602,
+      "reward_std": 0.23890409246087074,
+      "rewards/accuracy_reward": 0.1145833358168602,
+      "rewards/format_reward": 0.0,
+      "step": 25
+    },
+    {
+      "completion_length": 605.0937690734863,
+      "epoch": 0.2773333333333333,
+      "grad_norm": 0.1382395625114441,
+      "kl": 0.037479400634765625,
+      "learning_rate": 2.987013369609672e-06,
+      "loss": -0.0203,
+      "reward": 0.10416666977107525,
+      "reward_std": 0.26679351180791855,
+      "rewards/accuracy_reward": 0.10416666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 26
+    },
+    {
+      "completion_length": 517.604190826416,
+      "epoch": 0.288,
+      "grad_norm": 0.21559005975723267,
+      "kl": 0.061981201171875,
+      "learning_rate": 2.983045374736096e-06,
+      "loss": 0.0576,
+      "reward": 0.14583333767950535,
+      "reward_std": 0.17174768447875977,
+      "rewards/accuracy_reward": 0.14583333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 27
+    },
+    {
+      "completion_length": 594.8958473205566,
+      "epoch": 0.2986666666666667,
+      "grad_norm": 0.1260552555322647,
+      "kl": 0.02332305908203125,
+      "learning_rate": 2.978552562210232e-06,
+      "loss": 0.0939,
+      "reward": 0.12500000279396772,
+      "reward_std": 0.25147588923573494,
+      "rewards/accuracy_reward": 0.12500000279396772,
+      "rewards/format_reward": 0.0,
+      "step": 28
+    },
+    {
+      "completion_length": 546.0312690734863,
+      "epoch": 0.30933333333333335,
+      "grad_norm": 0.12078198045492172,
+      "kl": 0.0504302978515625,
+      "learning_rate": 2.9735365219411707e-06,
+      "loss": -0.0113,
+      "reward": 0.18750001024454832,
+      "reward_std": 0.14910665899515152,
+      "rewards/accuracy_reward": 0.18750001024454832,
+      "rewards/format_reward": 0.0,
+      "step": 29
+    },
+    {
+      "completion_length": 469.4896049499512,
+      "epoch": 0.32,
+      "grad_norm": 0.16322949528694153,
+      "kl": 0.0821685791015625,
+      "learning_rate": 2.9679990289969723e-06,
+      "loss": -0.011,
+      "reward": 0.15625000186264515,
+      "reward_std": 0.2555270306766033,
+      "rewards/accuracy_reward": 0.15625000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 30
+    },
+    {
+      "completion_length": 556.3125152587891,
+      "epoch": 0.33066666666666666,
+      "grad_norm": 0.12340069562196732,
+      "kl": 0.03859710693359375,
+      "learning_rate": 2.9619420429765106e-06,
+      "loss": -0.0312,
+      "reward": 0.18750000093132257,
+      "reward_std": 0.2543407045304775,
+      "rewards/accuracy_reward": 0.18750000093132257,
+      "rewards/format_reward": 0.0,
+      "step": 31
+    },
+    {
+      "completion_length": 469.6770896911621,
+      "epoch": 0.3413333333333333,
+      "grad_norm": 0.1940806806087494,
+      "kl": 0.06516265869140625,
+      "learning_rate": 2.955367707316014e-06,
+      "loss": -0.0537,
+      "reward": 0.1979166716337204,
+      "reward_std": 0.2885033190250397,
+      "rewards/accuracy_reward": 0.1979166716337204,
+      "rewards/format_reward": 0.0,
+      "step": 32
+    },
+    {
+      "completion_length": 602.7396011352539,
+      "epoch": 0.352,
+      "grad_norm": 0.09024544060230255,
+      "kl": 0.02783203125,
+      "learning_rate": 2.948278348530547e-06,
+      "loss": 0.0558,
+      "reward": 0.12500000186264515,
+      "reward_std": 0.21124766021966934,
+      "rewards/accuracy_reward": 0.12500000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 33
+    },
+    {
+      "completion_length": 593.6770935058594,
+      "epoch": 0.3626666666666667,
+      "grad_norm": 0.08488824963569641,
+      "kl": 0.05257415771484375,
+      "learning_rate": 2.9406764753907096e-06,
+      "loss": 0.0303,
+      "reward": 0.07291666883975267,
+      "reward_std": 0.18205294013023376,
+      "rewards/accuracy_reward": 0.07291666883975267,
+      "rewards/format_reward": 0.0,
+      "step": 34
+    },
+    {
+      "completion_length": 574.5312728881836,
+      "epoch": 0.37333333333333335,
+      "grad_norm": 0.12150894105434418,
+      "kl": 0.03229522705078125,
+      "learning_rate": 2.9325647780348364e-06,
+      "loss": 0.0854,
+      "reward": 0.1041666679084301,
+      "reward_std": 0.23389172554016113,
+      "rewards/accuracy_reward": 0.1041666679084301,
+      "rewards/format_reward": 0.0,
+      "step": 35
+    },
+    {
+      "completion_length": 626.3750152587891,
+      "epoch": 0.384,
+      "grad_norm": 0.12855271995067596,
+      "kl": 0.02007293701171875,
+      "learning_rate": 2.9239461270170137e-06,
+      "loss": 0.0565,
+      "reward": 0.17708333674818277,
+      "reward_std": 0.3208988308906555,
+      "rewards/accuracy_reward": 0.17708333674818277,
+      "rewards/format_reward": 0.0,
+      "step": 36
+    },
+    {
+      "completion_length": 628.3437652587891,
+      "epoch": 0.39466666666666667,
+      "grad_norm": 0.14817272126674652,
+      "kl": 0.03232574462890625,
+      "learning_rate": 2.914823572291257e-06,
+      "loss": 0.0374,
+      "reward": 0.20833333767950535,
+      "reward_std": 0.2605394050478935,
+      "rewards/accuracy_reward": 0.20833333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 37
+    },
+    {
+      "completion_length": 580.7604331970215,
+      "epoch": 0.4053333333333333,
+      "grad_norm": 0.17469142377376556,
+      "kl": 0.0690155029296875,
+      "learning_rate": 2.9052003421321917e-06,
+      "loss": 0.038,
+      "reward": 0.0729166679084301,
+      "reward_std": 0.15384574607014656,
+      "rewards/accuracy_reward": 0.0729166679084301,
+      "rewards/format_reward": 0.0,
+      "step": 38
+    },
+    {
+      "completion_length": 609.6666831970215,
+      "epoch": 0.416,
+      "grad_norm": 0.16530297696590424,
+      "kl": 0.06507110595703125,
+      "learning_rate": 2.8950798419926422e-06,
+      "loss": 0.0371,
+      "reward": 0.22916667629033327,
+      "reward_std": 0.37515120208263397,
+      "rewards/accuracy_reward": 0.22916667629033327,
+      "rewards/format_reward": 0.0,
+      "step": 39
+    },
+    {
+      "completion_length": 589.1979293823242,
+      "epoch": 0.4266666666666667,
+      "grad_norm": 0.16562184691429138,
+      "kl": 0.0455322265625,
+      "learning_rate": 2.884465653298514e-06,
+      "loss": 0.1025,
+      "reward": 0.17708334047347307,
+      "reward_std": 0.2671857886016369,
+      "rewards/accuracy_reward": 0.17708334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 40
+    },
+    {
+      "completion_length": 597.1875190734863,
+      "epoch": 0.43733333333333335,
+      "grad_norm": 0.46513137221336365,
+      "kl": 0.05785369873046875,
+      "learning_rate": 2.8733615321814036e-06,
+      "loss": 0.1034,
+      "reward": 0.2291666753590107,
+      "reward_std": 0.339111290872097,
+      "rewards/accuracy_reward": 0.2291666753590107,
+      "rewards/format_reward": 0.0,
+      "step": 41
+    },
+    {
+      "completion_length": 553.8333435058594,
+      "epoch": 0.448,
+      "grad_norm": 0.12673671543598175,
+      "kl": 0.06561279296875,
+      "learning_rate": 2.8617714081493856e-06,
+      "loss": 0.0195,
+      "reward": 0.19791667815297842,
+      "reward_std": 0.28564153611660004,
+      "rewards/accuracy_reward": 0.19791667815297842,
+      "rewards/format_reward": 0.0,
+      "step": 42
+    },
+    {
+      "completion_length": 502.4479331970215,
+      "epoch": 0.45866666666666667,
+      "grad_norm": 0.20127542316913605,
+      "kl": 0.103759765625,
+      "learning_rate": 2.849699382696445e-06,
+      "loss": -0.0076,
+      "reward": 0.22916667256504297,
+      "reward_std": 0.2759315297007561,
+      "rewards/accuracy_reward": 0.22916667256504297,
+      "rewards/format_reward": 0.0,
+      "step": 43
+    },
+    {
+      "completion_length": 490.50000762939453,
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.231169655919075,
+      "kl": 0.0522918701171875,
+      "learning_rate": 2.8371497278510474e-06,
+      "loss": 0.1514,
+      "reward": 0.3645833432674408,
+      "reward_std": 0.427349217236042,
+      "rewards/accuracy_reward": 0.3645833432674408,
+      "rewards/format_reward": 0.0,
+      "step": 44
+    },
+    {
+      "completion_length": 597.1875228881836,
+      "epoch": 0.48,
+      "grad_norm": 0.48629194498062134,
+      "kl": 0.05603790283203125,
+      "learning_rate": 2.8241268846643613e-06,
+      "loss": 0.0505,
+      "reward": 0.15625000279396772,
+      "reward_std": 0.22226297482848167,
+      "rewards/accuracy_reward": 0.15625000279396772,
+      "rewards/format_reward": 0.0,
+      "step": 45
+    },
+    {
+      "completion_length": 529.3333587646484,
+      "epoch": 0.49066666666666664,
+      "grad_norm": 0.14516475796699524,
+      "kl": 0.0585784912109375,
+      "learning_rate": 2.8106354616386633e-06,
+      "loss": 0.0372,
+      "reward": 0.3229166744276881,
+      "reward_std": 0.3968754820525646,
+      "rewards/accuracy_reward": 0.3229166744276881,
+      "rewards/format_reward": 0.0,
+      "step": 46
+    },
+    {
+      "completion_length": 586.8333549499512,
+      "epoch": 0.5013333333333333,
+      "grad_norm": 0.13861708343029022,
+      "kl": 0.0889892578125,
+      "learning_rate": 2.796680233096492e-06,
+      "loss": -0.0095,
+      "reward": 0.1770833358168602,
+      "reward_std": 0.21539149805903435,
+      "rewards/accuracy_reward": 0.1770833358168602,
+      "rewards/format_reward": 0.0,
+      "step": 47
+    },
+    {
+      "completion_length": 578.0312614440918,
+      "epoch": 0.512,
+      "grad_norm": 0.14956602454185486,
+      "kl": 0.1368408203125,
+      "learning_rate": 2.7822661374911177e-06,
+      "loss": 0.0516,
+      "reward": 0.2708333395421505,
+      "reward_std": 0.30700354650616646,
+      "rewards/accuracy_reward": 0.2708333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 48
+    },
+    {
+      "completion_length": 623.2916870117188,
+      "epoch": 0.5226666666666666,
+      "grad_norm": 0.18586260080337524,
+      "kl": 0.120880126953125,
+      "learning_rate": 2.7673982756589312e-06,
+      "loss": 0.0158,
+      "reward": 0.22916667070239782,
+      "reward_std": 0.3167550005018711,
+      "rewards/accuracy_reward": 0.22916667070239782,
+      "rewards/format_reward": 0.0,
+      "step": 49
+    },
+    {
+      "completion_length": 697.7291793823242,
+      "epoch": 0.5333333333333333,
+      "grad_norm": 6.404502868652344,
+      "kl": 0.319976806640625,
+      "learning_rate": 2.7520819090143655e-06,
+      "loss": 0.0299,
+      "reward": 0.1562500037252903,
+      "reward_std": 0.2504701688885689,
+      "rewards/accuracy_reward": 0.1562500037252903,
+      "rewards/format_reward": 0.0,
+      "step": 50
+    },
+    {
+      "completion_length": 584.5208435058594,
+      "epoch": 0.544,
+      "grad_norm": 0.21096980571746826,
+      "kl": 0.162750244140625,
+      "learning_rate": 2.736322457687999e-06,
+      "loss": -0.0173,
+      "reward": 0.19791667070239782,
+      "reward_std": 0.3315218612551689,
+      "rewards/accuracy_reward": 0.19791667070239782,
+      "rewards/format_reward": 0.0,
+      "step": 51
+    },
+    {
+      "completion_length": 646.1875152587891,
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.34582799673080444,
+      "kl": 0.1904296875,
+      "learning_rate": 2.720125498608487e-06,
+      "loss": 0.0753,
+      "reward": 0.10416666883975267,
+      "reward_std": 0.20281970128417015,
+      "rewards/accuracy_reward": 0.10416666883975267,
+      "rewards/format_reward": 0.0,
+      "step": 52
+    },
+    {
+      "completion_length": 722.7604370117188,
+      "epoch": 0.5653333333333334,
+      "grad_norm": 0.16253501176834106,
+      "kl": 0.1534423828125,
+      "learning_rate": 2.7034967635290054e-06,
+      "loss": 0.1153,
+      "reward": 0.23958333674818277,
+      "reward_std": 0.25935307890176773,
+      "rewards/accuracy_reward": 0.23958333674818277,
+      "rewards/format_reward": 0.0,
+      "step": 53
+    },
+    {
+      "completion_length": 713.0312652587891,
+      "epoch": 0.576,
+      "grad_norm": 0.1711745262145996,
+      "kl": 0.2108154296875,
+      "learning_rate": 2.6864421369989112e-06,
+      "loss": 0.0778,
+      "reward": 0.16666667070239782,
+      "reward_std": 0.27498848363757133,
+      "rewards/accuracy_reward": 0.16666667070239782,
+      "rewards/format_reward": 0.0,
+      "step": 54
+    },
+    {
+      "completion_length": 709.7916946411133,
+      "epoch": 0.5866666666666667,
+      "grad_norm": 0.21810775995254517,
+      "kl": 0.18017578125,
+      "learning_rate": 2.668967654281324e-06,
+      "loss": 0.0342,
+      "reward": 0.2395833469927311,
+      "reward_std": 0.254644013941288,
+      "rewards/accuracy_reward": 0.2395833469927311,
+      "rewards/format_reward": 0.0,
+      "step": 55
+    },
+    {
+      "completion_length": 673.8229293823242,
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.3457491099834442,
+      "kl": 0.2532958984375,
+      "learning_rate": 2.65107949921738e-06,
+      "loss": 0.1691,
+      "reward": 0.20833333674818277,
+      "reward_std": 0.3889237828552723,
+      "rewards/accuracy_reward": 0.20833333674818277,
+      "rewards/format_reward": 0.0,
+      "step": 56
+    },
+    {
+      "completion_length": 575.7500152587891,
+      "epoch": 0.608,
+      "grad_norm": 0.8664973974227905,
+      "kl": 0.773681640625,
+      "learning_rate": 2.632784002037904e-06,
+      "loss": 0.1391,
+      "reward": 0.3229166716337204,
+      "reward_std": 0.4173244684934616,
+      "rewards/accuracy_reward": 0.3229166716337204,
+      "rewards/format_reward": 0.0,
+      "step": 57
+    },
+    {
+      "completion_length": 687.2500152587891,
+      "epoch": 0.6186666666666667,
+      "grad_norm": 0.33583614230155945,
+      "kl": 0.62158203125,
+      "learning_rate": 2.6140876371232755e-06,
+      "loss": 0.051,
+      "reward": 0.17708334047347307,
+      "reward_std": 0.2671857886016369,
+      "rewards/accuracy_reward": 0.17708334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 58
+    },
+    {
+      "completion_length": 726.1875076293945,
+      "epoch": 0.6293333333333333,
+      "grad_norm": 2.140490770339966,
+      "kl": 1.1827392578125,
+      "learning_rate": 2.594997020712288e-06,
+      "loss": 0.043,
+      "reward": 0.08333333395421505,
+      "reward_std": 0.13744790107011795,
+      "rewards/accuracy_reward": 0.08333333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 59
+    },
+    {
+      "completion_length": 749.4895935058594,
+      "epoch": 0.64,
+      "grad_norm": 0.2817237377166748,
+      "kl": 0.57421875,
+      "learning_rate": 2.5755189085608046e-06,
+      "loss": 0.0747,
+      "reward": 0.09375000279396772,
+      "reward_std": 0.24196770787239075,
+      "rewards/accuracy_reward": 0.09375000279396772,
+      "rewards/format_reward": 0.0,
+      "step": 60
+    },
+    {
+      "completion_length": 735.6354370117188,
+      "epoch": 0.6506666666666666,
+      "grad_norm": 0.3483006954193115,
+      "kl": 0.52978515625,
+      "learning_rate": 2.5556601935510422e-06,
+      "loss": 0.0503,
+      "reward": 0.15625000186264515,
+      "reward_std": 0.2912936396896839,
+      "rewards/accuracy_reward": 0.15625000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 61
+    },
+    {
+      "completion_length": 692.6041793823242,
+      "epoch": 0.6613333333333333,
+      "grad_norm": 0.3014882206916809,
+      "kl": 0.6661376953125,
+      "learning_rate": 2.53542790325233e-06,
+      "loss": 0.0723,
+      "reward": 0.0416666679084301,
+      "reward_std": 0.06154575198888779,
+      "rewards/accuracy_reward": 0.0416666679084301,
+      "rewards/format_reward": 0.0,
+      "step": 62
+    },
+    {
+      "completion_length": 617.5937728881836,
+      "epoch": 0.672,
+      "grad_norm": 0.887993335723877,
+      "kl": 0.4677734375,
+      "learning_rate": 2.5148291974342035e-06,
+      "loss": 0.1084,
+      "reward": 0.322916672565043,
+      "reward_std": 0.41752326115965843,
+      "rewards/accuracy_reward": 0.322916672565043,
+      "rewards/format_reward": 0.0,
+      "step": 63
+    },
+    {
+      "completion_length": 653.7395973205566,
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.3804032802581787,
+      "kl": 0.9503173828125,
+      "learning_rate": 2.493871365532719e-06,
+      "loss": 0.0645,
+      "reward": 0.17708333767950535,
+      "reward_std": 0.2633597403764725,
+      "rewards/accuracy_reward": 0.17708333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 64
+    },
+    {
+      "completion_length": 558.3646011352539,
+      "epoch": 0.6933333333333334,
+      "grad_norm": 0.8641887903213501,
+      "kl": 0.542236328125,
+      "learning_rate": 2.4725618240708804e-06,
+      "loss": 0.1947,
+      "reward": 0.3125000037252903,
+      "reward_std": 0.3014373779296875,
+      "rewards/accuracy_reward": 0.3125000037252903,
+      "rewards/format_reward": 0.0,
+      "step": 65
+    },
+    {
+      "completion_length": 554.1041793823242,
+      "epoch": 0.704,
+      "grad_norm": 0.7259989380836487,
+      "kl": 1.319580078125,
+      "learning_rate": 2.4509081140340877e-06,
+      "loss": 0.0516,
+      "reward": 0.3020833423361182,
+      "reward_std": 0.33246491476893425,
+      "rewards/accuracy_reward": 0.3020833423361182,
+      "rewards/format_reward": 0.0,
+      "step": 66
+    },
+    {
+      "completion_length": 534.1146011352539,
+      "epoch": 0.7146666666666667,
+      "grad_norm": 0.7615457773208618,
+      "kl": 1.759307861328125,
+      "learning_rate": 2.428917898201553e-06,
+      "loss": 0.1123,
+      "reward": 0.2812500027939677,
+      "reward_std": 0.2837342247366905,
+      "rewards/accuracy_reward": 0.2812500027939677,
+      "rewards/format_reward": 0.0,
+      "step": 67
+    },
+    {
+      "completion_length": 536.3125152587891,
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.9113590121269226,
+      "kl": 0.890380859375,
+      "learning_rate": 2.4065989584346e-06,
+      "loss": 0.1048,
+      "reward": 0.22916667629033327,
+      "reward_std": 0.3314773738384247,
+      "rewards/accuracy_reward": 0.22916667629033327,
+      "rewards/format_reward": 0.0,
+      "step": 68
+    },
+    {
+      "completion_length": 524.9895935058594,
+      "epoch": 0.736,
+      "grad_norm": 0.4094133973121643,
+      "kl": 0.77490234375,
+      "learning_rate": 2.3839591929228413e-06,
+      "loss": 0.0492,
+      "reward": 0.2708333423361182,
+      "reward_std": 0.41816671192646027,
+      "rewards/accuracy_reward": 0.2708333423361182,
+      "rewards/format_reward": 0.0,
+      "step": 69
+    },
+    {
+      "completion_length": 532.0625133514404,
+      "epoch": 0.7466666666666667,
+      "grad_norm": 2.149420976638794,
+      "kl": 0.80133056640625,
+      "learning_rate": 2.3610066133891706e-06,
+      "loss": 0.0904,
+      "reward": 0.260416672565043,
+      "reward_std": 0.35224414244294167,
+      "rewards/accuracy_reward": 0.260416672565043,
+      "rewards/format_reward": 0.0,
+      "step": 70
+    },
+    {
+      "completion_length": 639.7916870117188,
+      "epoch": 0.7573333333333333,
+      "grad_norm": 0.29616329073905945,
+      "kl": 0.350982666015625,
+      "learning_rate": 2.337749342254595e-06,
+      "loss": 0.0271,
+      "reward": 0.2708333395421505,
+      "reward_std": 0.2885478064417839,
+      "rewards/accuracy_reward": 0.2708333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 71
+    },
+    {
+      "completion_length": 584.4166831970215,
+      "epoch": 0.768,
+      "grad_norm": 0.42845651507377625,
+      "kl": 0.38970947265625,
+      "learning_rate": 2.3141956097638817e-06,
+      "loss": 0.0633,
+      "reward": 0.17708334047347307,
+      "reward_std": 0.3029524050652981,
+      "rewards/accuracy_reward": 0.17708334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 72
+    },
+    {
+      "completion_length": 543.7083435058594,
+      "epoch": 0.7786666666666666,
+      "grad_norm": 0.6256530284881592,
+      "kl": 0.306884765625,
+      "learning_rate": 2.2903537510730547e-06,
+      "loss": 0.1157,
+      "reward": 0.4583333535119891,
+      "reward_std": 0.4731736220419407,
+      "rewards/accuracy_reward": 0.4583333535119891,
+      "rewards/format_reward": 0.0,
+      "step": 73
+    },
+    {
+      "completion_length": 560.2812614440918,
+      "epoch": 0.7893333333333333,
+      "grad_norm": 0.5944956541061401,
+      "kl": 0.662353515625,
+      "learning_rate": 2.2662322032997616e-06,
+      "loss": 0.1079,
+      "reward": 0.19791667349636555,
+      "reward_std": 0.29912634938955307,
+      "rewards/accuracy_reward": 0.19791667349636555,
+      "rewards/format_reward": 0.0,
+      "step": 74
+    },
+    {
+      "completion_length": 699.5521011352539,
+      "epoch": 0.8,
+      "grad_norm": 0.554081380367279,
+      "kl": 0.7374267578125,
+      "learning_rate": 2.241839502537563e-06,
+      "loss": 0.0715,
+      "reward": 0.15625000651925802,
+      "reward_std": 0.22636231780052185,
+      "rewards/accuracy_reward": 0.15625000651925802,
+      "rewards/format_reward": 0.0,
+      "step": 75
+    },
+    {
+      "completion_length": 646.3437652587891,
+      "epoch": 0.8106666666666666,
+      "grad_norm": 1.3772621154785156,
+      "kl": 1.7783203125,
+      "learning_rate": 2.21718428083519e-06,
+      "loss": 0.0255,
+      "reward": 0.17708333767950535,
+      "reward_std": 0.28982681408524513,
+      "rewards/accuracy_reward": 0.17708333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 76
+    },
+    {
+      "completion_length": 632.2500152587891,
+      "epoch": 0.8213333333333334,
+      "grad_norm": 0.865872859954834,
+      "kl": 0.7518310546875,
+      "learning_rate": 2.1922752631418534e-06,
+      "loss": 0.035,
+      "reward": 0.2500000074505806,
+      "reward_std": 0.28194961696863174,
+      "rewards/accuracy_reward": 0.2500000074505806,
+      "rewards/format_reward": 0.0,
+      "step": 77
+    },
+    {
+      "completion_length": 702.2395935058594,
+      "epoch": 0.832,
+      "grad_norm": 0.4784860908985138,
+      "kl": 0.572998046875,
+      "learning_rate": 2.167121264219669e-06,
+      "loss": 0.0299,
+      "reward": 0.15625000186264515,
+      "reward_std": 0.23201442882418633,
+      "rewards/accuracy_reward": 0.15625000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 78
+    },
+    {
+      "completion_length": 673.2396087646484,
+      "epoch": 0.8426666666666667,
+      "grad_norm": 1.1683677434921265,
+      "kl": 0.584228515625,
+      "learning_rate": 2.1417311855243084e-06,
+      "loss": 0.1041,
+      "reward": 0.18750000465661287,
+      "reward_std": 0.29944413155317307,
+      "rewards/accuracy_reward": 0.18750000465661287,
+      "rewards/format_reward": 0.0,
+      "step": 79
+    },
+    {
+      "completion_length": 561.9270973205566,
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.7934542894363403,
+      "kl": 1.1396484375,
+      "learning_rate": 2.116114012054961e-06,
+      "loss": 0.1309,
+      "reward": 0.27083334047347307,
+      "reward_std": 0.38526491820812225,
+      "rewards/accuracy_reward": 0.27083334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 80
+    },
+    {
+      "completion_length": 486.2395935058594,
+      "epoch": 0.864,
+      "grad_norm": 1.099777340888977,
+      "kl": 1.2906494140625,
+      "learning_rate": 2.0902788091747436e-06,
+      "loss": 0.0744,
+      "reward": 0.22916667256504297,
+      "reward_std": 0.3494238071143627,
+      "rewards/accuracy_reward": 0.22916667256504297,
+      "rewards/format_reward": 0.0,
+      "step": 81
+    },
+    {
+      "completion_length": 573.5729293823242,
+      "epoch": 0.8746666666666667,
+      "grad_norm": 0.3417358994483948,
+      "kl": 0.5443115234375,
+      "learning_rate": 2.064234719402658e-06,
+      "loss": 0.1249,
+      "reward": 0.2812500037252903,
+      "reward_std": 0.30425771325826645,
+      "rewards/accuracy_reward": 0.2812500037252903,
+      "rewards/format_reward": 0.0,
+      "step": 82
+    },
+    {
+      "completion_length": 600.3646011352539,
+      "epoch": 0.8853333333333333,
+      "grad_norm": 0.3621276021003723,
+      "kl": 0.4794921875,
+      "learning_rate": 2.037990959178259e-06,
+      "loss": 0.0338,
+      "reward": 0.281250006519258,
+      "reward_std": 0.2800753600895405,
+      "rewards/accuracy_reward": 0.281250006519258,
+      "rewards/format_reward": 0.0,
+      "step": 83
+    },
+    {
+      "completion_length": 584.6458473205566,
+      "epoch": 0.896,
+      "grad_norm": 0.5323257446289062,
+      "kl": 0.467041015625,
+      "learning_rate": 2.0115568156001488e-06,
+      "loss": 0.0821,
+      "reward": 0.27083334513008595,
+      "reward_std": 0.3042132258415222,
+      "rewards/accuracy_reward": 0.27083334513008595,
+      "rewards/format_reward": 0.0,
+      "step": 84
+    },
+    {
+      "completion_length": 585.9166870117188,
+      "epoch": 0.9066666666666666,
+      "grad_norm": 0.467734694480896,
+      "kl": 0.3864898681640625,
+      "learning_rate": 1.984941643139478e-06,
+      "loss": 0.1324,
+      "reward": 0.3229166753590107,
+      "reward_std": 0.1980806216597557,
+      "rewards/accuracy_reward": 0.3229166753590107,
+      "rewards/format_reward": 0.0,
+      "step": 85
+    },
+    {
+      "completion_length": 619.7187652587891,
+      "epoch": 0.9173333333333333,
+      "grad_norm": 0.3862552344799042,
+      "kl": 0.523406982421875,
+      "learning_rate": 1.9581548603295995e-06,
+      "loss": 0.0401,
+      "reward": 0.19791667256504297,
+      "reward_std": 0.2671857886016369,
+      "rewards/accuracy_reward": 0.19791667256504297,
+      "rewards/format_reward": 0.0,
+      "step": 86
+    },
+    {
+      "completion_length": 613.7083511352539,
+      "epoch": 0.928,
+      "grad_norm": 0.27887099981307983,
+      "kl": 0.801605224609375,
+      "learning_rate": 1.9312059464330545e-06,
+      "loss": 0.0648,
+      "reward": 0.1875000037252903,
+      "reward_std": 0.2856829836964607,
+      "rewards/accuracy_reward": 0.1875000037252903,
+      "rewards/format_reward": 0.0,
+      "step": 87
+    },
+    {
+      "completion_length": 588.0521049499512,
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.4169585704803467,
+      "kl": 0.687591552734375,
+      "learning_rate": 1.9041044380870643e-06,
+      "loss": 0.0083,
+      "reward": 0.1979166716337204,
+      "reward_std": 0.32426992803812027,
+      "rewards/accuracy_reward": 0.1979166716337204,
+      "rewards/format_reward": 0.0,
+      "step": 88
+    },
+    {
+      "completion_length": 581.4687614440918,
+      "epoch": 0.9493333333333334,
+      "grad_norm": 0.6219002604484558,
+      "kl": 0.71435546875,
+      "learning_rate": 1.8768599259287235e-06,
+      "loss": 0.0248,
+      "reward": 0.26041667349636555,
+      "reward_std": 0.20783207565546036,
+      "rewards/accuracy_reward": 0.26041667349636555,
+      "rewards/format_reward": 0.0,
+      "step": 89
+    },
+    {
+      "completion_length": 576.2916831970215,
+      "epoch": 0.96,
+      "grad_norm": 0.5098808407783508,
+      "kl": 0.950927734375,
+      "learning_rate": 1.8494820512010797e-06,
+      "loss": 0.1095,
+      "reward": 0.2291666753590107,
+      "reward_std": 0.2797575779259205,
+      "rewards/accuracy_reward": 0.2291666753590107,
+      "rewards/format_reward": 0.0,
+      "step": 90
+    },
+    {
+      "completion_length": 624.4271087646484,
+      "epoch": 0.9706666666666667,
+      "grad_norm": 0.6733171343803406,
+      "kl": 0.8636474609375,
+      "learning_rate": 1.821980502341308e-06,
+      "loss": 0.0614,
+      "reward": 0.15625000651925802,
+      "reward_std": 0.2185296081006527,
+      "rewards/accuracy_reward": 0.15625000651925802,
+      "rewards/format_reward": 0.0,
+      "step": 91
+    },
+    {
+      "completion_length": 587.5208587646484,
+      "epoch": 0.9813333333333333,
+      "grad_norm": 1.0600099563598633,
+      "kl": 0.5804443359375,
+      "learning_rate": 1.7943650115521788e-06,
+      "loss": 0.1491,
+      "reward": 0.281250006519258,
+      "reward_std": 0.2633597403764725,
+      "rewards/accuracy_reward": 0.281250006519258,
+      "rewards/format_reward": 0.0,
+      "step": 92
+    },
+    {
+      "completion_length": 635.9375152587891,
+      "epoch": 0.992,
+      "grad_norm": 0.22917693853378296,
+      "kl": 0.6182861328125,
+      "learning_rate": 1.766645351358047e-06,
+      "loss": 0.085,
+      "reward": 0.2395833395421505,
+      "reward_std": 0.24681130424141884,
+      "rewards/accuracy_reward": 0.2395833395421505,
+      "rewards/format_reward": 0.0,
+      "step": 93
+    },
+    {
+      "completion_length": 693.3055826822916,
+      "epoch": 1.0,
+      "grad_norm": 0.22917693853378296,
+      "kl": 1.232421875,
+      "learning_rate": 1.7388313311465638e-06,
+      "loss": 0.07,
+      "reward": 0.22222223257025084,
+      "reward_std": 0.2980559766292572,
+      "rewards/accuracy_reward": 0.22222223257025084,
+      "rewards/format_reward": 0.0,
+      "step": 94
+    },
+    {
+      "completion_length": 568.6562652587891,
+      "epoch": 1.0106666666666666,
+      "grad_norm": 0.39828255772590637,
+      "kl": 1.0860595703125,
+      "learning_rate": 1.7109327936973479e-06,
+      "loss": 0.1009,
+      "reward": 0.22916667349636555,
+      "reward_std": 0.2341650128364563,
+      "rewards/accuracy_reward": 0.22916667349636555,
+      "rewards/format_reward": 0.0,
+      "step": 95
+    },
+    {
+      "completion_length": 568.7083549499512,
+      "epoch": 1.0213333333333334,
+      "grad_norm": 0.3863210380077362,
+      "kl": 0.499755859375,
+      "learning_rate": 1.6829596116988381e-06,
+      "loss": 0.0364,
+      "reward": 0.43750001676380634,
+      "reward_std": 0.44166112691164017,
+      "rewards/accuracy_reward": 0.43750001676380634,
+      "rewards/format_reward": 0.0,
+      "step": 96
+    },
+    {
+      "completion_length": 576.3958549499512,
+      "epoch": 1.032,
+      "grad_norm": 0.621791660785675,
+      "kl": 0.8270263671875,
+      "learning_rate": 1.6549216842545649e-06,
+      "loss": -0.0011,
+      "reward": 0.18750000465661287,
+      "reward_std": 0.35948578268289566,
+      "rewards/accuracy_reward": 0.18750000465661287,
+      "rewards/format_reward": 0.0,
+      "step": 97
+    },
+    {
+      "completion_length": 618.697925567627,
+      "epoch": 1.0426666666666666,
+      "grad_norm": 0.27708661556243896,
+      "kl": 0.358612060546875,
+      "learning_rate": 1.6268289333800686e-06,
+      "loss": 0.1116,
+      "reward": 0.41666668001562357,
+      "reward_std": 0.35948577523231506,
+      "rewards/accuracy_reward": 0.41666668001562357,
+      "rewards/format_reward": 0.0,
+      "step": 98
+    },
+    {
+      "completion_length": 602.5625152587891,
+      "epoch": 1.0533333333333332,
+      "grad_norm": 1.2479599714279175,
+      "kl": 0.67333984375,
+      "learning_rate": 1.598691300491716e-06,
+      "loss": 0.1111,
+      "reward": 0.33333334513008595,
+      "reward_std": 0.28194961696863174,
+      "rewards/accuracy_reward": 0.33333334513008595,
+      "rewards/format_reward": 0.0,
+      "step": 99
+    },
+    {
+      "completion_length": 664.4062576293945,
+      "epoch": 1.064,
+      "grad_norm": 0.31423419713974,
+      "kl": 0.4830322265625,
+      "learning_rate": 1.5705187428886465e-06,
+      "loss": 0.0806,
+      "reward": 0.3437500139698386,
+      "reward_std": 0.3402976170182228,
+      "rewards/accuracy_reward": 0.3437500139698386,
+      "rewards/format_reward": 0.0,
+      "step": 100
+    },
+    {
+      "completion_length": 620.0312652587891,
+      "epoch": 1.0746666666666667,
+      "grad_norm": 0.2487853467464447,
+      "kl": 0.7161865234375,
+      "learning_rate": 1.542321230229099e-06,
+      "loss": 0.1072,
+      "reward": 0.36458333767950535,
+      "reward_std": 0.35291389375925064,
+      "rewards/accuracy_reward": 0.36458333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 101
+    },
+    {
+      "completion_length": 554.6875152587891,
+      "epoch": 1.0853333333333333,
+      "grad_norm": 0.3372454047203064,
+      "kl": 0.9202880859375,
+      "learning_rate": 1.5141087410023688e-06,
+      "loss": 0.1058,
+      "reward": 0.27083334140479565,
+      "reward_std": 0.3755134642124176,
+      "rewards/accuracy_reward": 0.27083334140479565,
+      "rewards/format_reward": 0.0,
+      "step": 102
+    },
+    {
+      "completion_length": 661.9583587646484,
+      "epoch": 1.096,
+      "grad_norm": 0.3665924668312073,
+      "kl": 0.68231201171875,
+      "learning_rate": 1.485891258997632e-06,
+      "loss": 0.071,
+      "reward": 0.20833333767950535,
+      "reward_std": 0.29638051614165306,
+      "rewards/accuracy_reward": 0.20833333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 103
+    },
+    {
+      "completion_length": 570.8541793823242,
+      "epoch": 1.1066666666666667,
+      "grad_norm": 0.4261507987976074,
+      "kl": 0.9154052734375,
+      "learning_rate": 1.4576787697709012e-06,
+      "loss": 0.0776,
+      "reward": 0.27083334047347307,
+      "reward_std": 0.4203435964882374,
+      "rewards/accuracy_reward": 0.27083334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 104
+    },
+    {
+      "completion_length": 570.9791793823242,
+      "epoch": 1.1173333333333333,
+      "grad_norm": 1.0135829448699951,
+      "kl": 0.7562255859375,
+      "learning_rate": 1.429481257111354e-06,
+      "loss": 0.1382,
+      "reward": 0.5000000102445483,
+      "reward_std": 0.39492369070649147,
+      "rewards/accuracy_reward": 0.5000000102445483,
+      "rewards/format_reward": 0.0,
+      "step": 105
+    },
+    {
+      "completion_length": 485.27084732055664,
+      "epoch": 1.1280000000000001,
+      "grad_norm": 0.5100330710411072,
+      "kl": 0.30596923828125,
+      "learning_rate": 1.401308699508284e-06,
+      "loss": 0.139,
+      "reward": 0.5729166828095913,
+      "reward_std": 0.3619955964386463,
+      "rewards/accuracy_reward": 0.5729166828095913,
+      "rewards/format_reward": 0.0,
+      "step": 106
+    },
+    {
+      "completion_length": 745.677116394043,
+      "epoch": 1.1386666666666667,
+      "grad_norm": 0.6321167349815369,
+      "kl": 0.743896484375,
+      "learning_rate": 1.3731710666199315e-06,
+      "loss": 0.1902,
+      "reward": 0.29166667349636555,
+      "reward_std": 0.39369287714362144,
+      "rewards/accuracy_reward": 0.29166667349636555,
+      "rewards/format_reward": 0.0,
+      "step": 107
+    },
+    {
+      "completion_length": 683.0833587646484,
+      "epoch": 1.1493333333333333,
+      "grad_norm": 0.19921176135540009,
+      "kl": 0.631439208984375,
+      "learning_rate": 1.3450783157454354e-06,
+      "loss": 0.0708,
+      "reward": 0.3437500102445483,
+      "reward_std": 0.29825781285762787,
+      "rewards/accuracy_reward": 0.3437500102445483,
+      "rewards/format_reward": 0.0,
+      "step": 108
+    },
+    {
+      "completion_length": 543.0416831970215,
+      "epoch": 1.16,
+      "grad_norm": 0.653483510017395,
+      "kl": 0.3345947265625,
+      "learning_rate": 1.317040388301162e-06,
+      "loss": 0.2456,
+      "reward": 0.45833335164934397,
+      "reward_std": 0.34686949849128723,
+      "rewards/accuracy_reward": 0.45833335164934397,
+      "rewards/format_reward": 0.0,
+      "step": 109
+    },
+    {
+      "completion_length": 568.3125114440918,
+      "epoch": 1.1706666666666667,
+      "grad_norm": 1.6327720880508423,
+      "kl": 0.2816009521484375,
+      "learning_rate": 1.289067206302653e-06,
+      "loss": 0.0651,
+      "reward": 0.45833334885537624,
+      "reward_std": 0.3930976279079914,
+      "rewards/accuracy_reward": 0.45833334885537624,
+      "rewards/format_reward": 0.0,
+      "step": 110
+    },
+    {
+      "completion_length": 613.6250114440918,
+      "epoch": 1.1813333333333333,
+      "grad_norm": 0.16312555968761444,
+      "kl": 0.50360107421875,
+      "learning_rate": 1.2611686688534365e-06,
+      "loss": 0.0214,
+      "reward": 0.19791667349636555,
+      "reward_std": 0.27467070892453194,
+      "rewards/accuracy_reward": 0.19791667349636555,
+      "rewards/format_reward": 0.0,
+      "step": 111
+    },
+    {
+      "completion_length": 548.9583587646484,
+      "epoch": 1.192,
+      "grad_norm": 0.5374172329902649,
+      "kl": 0.38507080078125,
+      "learning_rate": 1.2333546486419533e-06,
+      "loss": 0.0715,
+      "reward": 0.42708334047347307,
+      "reward_std": 0.3080092594027519,
+      "rewards/accuracy_reward": 0.42708334047347307,
+      "rewards/format_reward": 0.0,
+      "step": 112
+    },
+    {
+      "completion_length": 542.7083549499512,
+      "epoch": 1.2026666666666666,
+      "grad_norm": 0.6037431955337524,
+      "kl": 0.6873931884765625,
+      "learning_rate": 1.2056349884478213e-06,
+      "loss": 0.1474,
+      "reward": 0.4375000149011612,
+      "reward_std": 0.35419290140271187,
+      "rewards/accuracy_reward": 0.4375000149011612,
+      "rewards/format_reward": 0.0,
+      "step": 113
+    },
+    {
+      "completion_length": 563.3645935058594,
+      "epoch": 1.2133333333333334,
+      "grad_norm": 0.7209116220474243,
+      "kl": 0.3206329345703125,
+      "learning_rate": 1.178019497658692e-06,
+      "loss": 0.0499,
+      "reward": 0.3854166744276881,
+      "reward_std": 0.31488074362277985,
+      "rewards/accuracy_reward": 0.3854166744276881,
+      "rewards/format_reward": 0.0,
+      "step": 114
+    },
+    {
+      "completion_length": 641.1041831970215,
+      "epoch": 1.224,
+      "grad_norm": 0.1957225352525711,
+      "kl": 0.39288330078125,
+      "learning_rate": 1.1505179487989203e-06,
+      "loss": 0.038,
+      "reward": 0.30208333395421505,
+      "reward_std": 0.2357477955520153,
+      "rewards/accuracy_reward": 0.30208333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 115
+    },
+    {
+      "completion_length": 507.43751525878906,
+      "epoch": 1.2346666666666666,
+      "grad_norm": 0.23401759564876556,
+      "kl": 0.369598388671875,
+      "learning_rate": 1.123140074071277e-06,
+      "loss": 0.0856,
+      "reward": 0.4479166818782687,
+      "reward_std": 0.47189461439847946,
+      "rewards/accuracy_reward": 0.4479166818782687,
+      "rewards/format_reward": 0.0,
+      "step": 116
+    },
+    {
+      "completion_length": 629.677116394043,
+      "epoch": 1.2453333333333334,
+      "grad_norm": 0.4538763463497162,
+      "kl": 0.764892578125,
+      "learning_rate": 1.0958955619129364e-06,
+      "loss": -0.0148,
+      "reward": 0.09375000279396772,
+      "reward_std": 0.1569838561117649,
+      "rewards/accuracy_reward": 0.09375000279396772,
+      "rewards/format_reward": 0.0,
+      "step": 117
+    },
+    {
+      "completion_length": 507.5937614440918,
+      "epoch": 1.256,
+      "grad_norm": 0.48793312907218933,
+      "kl": 0.26214599609375,
+      "learning_rate": 1.068794053566946e-06,
+      "loss": 0.0633,
+      "reward": 0.5000000102445483,
+      "reward_std": 0.2546139992773533,
+      "rewards/accuracy_reward": 0.5000000102445483,
+      "rewards/format_reward": 0.0,
+      "step": 118
+    },
+    {
+      "completion_length": 633.2291946411133,
+      "epoch": 1.2666666666666666,
+      "grad_norm": 0.45710209012031555,
+      "kl": 0.2803955078125,
+      "learning_rate": 1.0418451396704008e-06,
+      "loss": 0.0682,
+      "reward": 0.2395833395421505,
+      "reward_std": 0.22601452842354774,
+      "rewards/accuracy_reward": 0.2395833395421505,
+      "rewards/format_reward": 0.0,
+      "step": 119
+    },
+    {
+      "completion_length": 471.57293701171875,
+      "epoch": 1.2773333333333334,
+      "grad_norm": 0.4903794229030609,
+      "kl": 0.21380615234375,
+      "learning_rate": 1.0150583568605221e-06,
+      "loss": 0.1971,
+      "reward": 0.5833333507180214,
+      "reward_std": 0.4508173316717148,
+      "rewards/accuracy_reward": 0.5833333507180214,
+      "rewards/format_reward": 0.0,
+      "step": 120
+    },
+    {
+      "completion_length": 668.8020935058594,
+      "epoch": 1.288,
+      "grad_norm": 0.42050671577453613,
+      "kl": 0.277130126953125,
+      "learning_rate": 9.884431843998509e-07,
+      "loss": 0.1015,
+      "reward": 0.34375000558793545,
+      "reward_std": 0.2768445611000061,
+      "rewards/accuracy_reward": 0.34375000558793545,
+      "rewards/format_reward": 0.0,
+      "step": 121
+    },
+    {
+      "completion_length": 497.5000228881836,
+      "epoch": 1.2986666666666666,
+      "grad_norm": 0.38155919313430786,
+      "kl": 0.33343505859375,
+      "learning_rate": 9.620090408217412e-07,
+      "loss": 0.0934,
+      "reward": 0.593750013038516,
+      "reward_std": 0.3603988066315651,
+      "rewards/accuracy_reward": 0.593750013038516,
+      "rewards/format_reward": 0.0,
+      "step": 122
+    },
+    {
+      "completion_length": 613.1666870117188,
+      "epoch": 1.3093333333333335,
+      "grad_norm": 2.1156508922576904,
+      "kl": 0.541748046875,
+      "learning_rate": 9.357652805973417e-07,
+      "loss": 0.0729,
+      "reward": 0.2291666716337204,
+      "reward_std": 0.24179892987012863,
+      "rewards/accuracy_reward": 0.2291666716337204,
+      "rewards/format_reward": 0.0,
+      "step": 123
+    },
+    {
+      "completion_length": 648.8229331970215,
+      "epoch": 1.32,
+      "grad_norm": 0.4442153871059418,
+      "kl": 0.259307861328125,
+      "learning_rate": 9.097211908252569e-07,
+      "loss": 0.0742,
+      "reward": 0.41666668467223644,
+      "reward_std": 0.3726486414670944,
+      "rewards/accuracy_reward": 0.41666668467223644,
+      "rewards/format_reward": 0.0,
+      "step": 124
+    },
+    {
+      "completion_length": 593.2708511352539,
+      "epoch": 1.3306666666666667,
+      "grad_norm": 0.5561993718147278,
+      "kl": 0.417083740234375,
+      "learning_rate": 8.838859879450389e-07,
+      "loss": 0.1325,
+      "reward": 0.36458334140479565,
+      "reward_std": 0.4144596457481384,
+      "rewards/accuracy_reward": 0.36458334140479565,
+      "rewards/format_reward": 0.0,
+      "step": 125
+    },
+    {
+      "completion_length": 622.2083511352539,
+      "epoch": 1.3413333333333333,
+      "grad_norm": 0.29431644082069397,
+      "kl": 0.33892822265625,
+      "learning_rate": 8.582688144756916e-07,
+      "loss": -0.0189,
+      "reward": 0.3229166781529784,
+      "reward_std": 0.2596263736486435,
+      "rewards/accuracy_reward": 0.3229166781529784,
+      "rewards/format_reward": 0.0,
+      "step": 126
+    },
+    {
+      "completion_length": 553.5104370117188,
+      "epoch": 1.3519999999999999,
+      "grad_norm": 0.35601741075515747,
+      "kl": 0.4847412109375,
+      "learning_rate": 8.328787357803312e-07,
+      "loss": 0.0578,
+      "reward": 0.21875000186264515,
+      "reward_std": 0.3450811840593815,
+      "rewards/accuracy_reward": 0.21875000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 127
+    },
+    {
+      "completion_length": 556.8958396911621,
+      "epoch": 1.3626666666666667,
+      "grad_norm": 0.2601093351840973,
+      "kl": 0.44805908203125,
+      "learning_rate": 8.077247368581468e-07,
+      "loss": 0.0774,
+      "reward": 0.43750000931322575,
+      "reward_std": 0.2957107648253441,
+      "rewards/accuracy_reward": 0.43750000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 128
+    },
+    {
+      "completion_length": 671.4896011352539,
+      "epoch": 1.3733333333333333,
+      "grad_norm": 0.3361031711101532,
+      "kl": 0.432769775390625,
+      "learning_rate": 7.828157191648098e-07,
+      "loss": 0.0428,
+      "reward": 0.07291666977107525,
+      "reward_std": 0.13653486967086792,
+      "rewards/accuracy_reward": 0.07291666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 129
+    },
+    {
+      "completion_length": 533.1354370117188,
+      "epoch": 1.384,
+      "grad_norm": 0.4666847288608551,
+      "kl": 0.45465087890625,
+      "learning_rate": 7.581604974624371e-07,
+      "loss": 0.1254,
+      "reward": 0.3750000027939677,
+      "reward_std": 0.21944263949990273,
+      "rewards/accuracy_reward": 0.3750000027939677,
+      "rewards/format_reward": 0.0,
+      "step": 130
+    },
+    {
+      "completion_length": 493.0937614440918,
+      "epoch": 1.3946666666666667,
+      "grad_norm": 0.47746792435646057,
+      "kl": 0.385833740234375,
+      "learning_rate": 7.337677967002385e-07,
+      "loss": 0.0853,
+      "reward": 0.4166666781529784,
+      "reward_std": 0.3755134642124176,
+      "rewards/accuracy_reward": 0.4166666781529784,
+      "rewards/format_reward": 0.0,
+      "step": 131
+    },
+    {
+      "completion_length": 625.0104331970215,
+      "epoch": 1.4053333333333333,
+      "grad_norm": 0.8601658940315247,
+      "kl": 0.3341064453125,
+      "learning_rate": 7.096462489269454e-07,
+      "loss": 0.1341,
+      "reward": 0.28125000931322575,
+      "reward_std": 0.32490549236536026,
+      "rewards/accuracy_reward": 0.28125000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 132
+    },
+    {
+      "completion_length": 595.4583473205566,
+      "epoch": 1.416,
+      "grad_norm": 0.45403310656547546,
+      "kl": 0.3372802734375,
+      "learning_rate": 6.858043902361181e-07,
+      "loss": 0.0947,
+      "reward": 0.32291667629033327,
+      "reward_std": 0.4055767357349396,
+      "rewards/accuracy_reward": 0.32291667629033327,
+      "rewards/format_reward": 0.0,
+      "step": 133
+    },
+    {
+      "completion_length": 478.6666717529297,
+      "epoch": 1.4266666666666667,
+      "grad_norm": 1.0305501222610474,
+      "kl": 0.44744873046875,
+      "learning_rate": 6.622506577454054e-07,
+      "loss": 0.2416,
+      "reward": 0.5416666809469461,
+      "reward_std": 0.3745673857629299,
+      "rewards/accuracy_reward": 0.5416666809469461,
+      "rewards/format_reward": 0.0,
+      "step": 134
+    },
+    {
+      "completion_length": 591.9895973205566,
+      "epoch": 1.4373333333333334,
+      "grad_norm": 1.416900873184204,
+      "kl": 0.714111328125,
+      "learning_rate": 6.389933866108296e-07,
+      "loss": 0.1579,
+      "reward": 0.3020833395421505,
+      "reward_std": 0.42134930938482285,
+      "rewards/accuracy_reward": 0.3020833395421505,
+      "rewards/format_reward": 0.0,
+      "step": 135
+    },
+    {
+      "completion_length": 578.6146011352539,
+      "epoch": 1.448,
+      "grad_norm": 0.6322238445281982,
+      "kl": 1.034423828125,
+      "learning_rate": 6.160408070771595e-07,
+      "loss": 0.1152,
+      "reward": 0.18750000651925802,
+      "reward_std": 0.28289569541811943,
+      "rewards/accuracy_reward": 0.18750000651925802,
+      "rewards/format_reward": 0.0,
+      "step": 136
+    },
+    {
+      "completion_length": 669.4375228881836,
+      "epoch": 1.4586666666666668,
+      "grad_norm": 1.085103154182434,
+      "kl": 1.0677490234375,
+      "learning_rate": 5.934010415653995e-07,
+      "loss": 0.0643,
+      "reward": 0.25000001210719347,
+      "reward_std": 0.3393845856189728,
+      "rewards/accuracy_reward": 0.25000001210719347,
+      "rewards/format_reward": 0.0,
+      "step": 137
+    },
+    {
+      "completion_length": 716.8229293823242,
+      "epoch": 1.4693333333333334,
+      "grad_norm": 0.7747601270675659,
+      "kl": 1.439208984375,
+      "learning_rate": 5.710821017984476e-07,
+      "loss": 0.1124,
+      "reward": 0.21875001024454832,
+      "reward_std": 0.22636231780052185,
+      "rewards/accuracy_reward": 0.21875001024454832,
+      "rewards/format_reward": 0.0,
+      "step": 138
+    },
+    {
+      "completion_length": 531.7916793823242,
+      "epoch": 1.48,
+      "grad_norm": 1.2720061540603638,
+      "kl": 0.59039306640625,
+      "learning_rate": 5.490918859659122e-07,
+      "loss": 0.1796,
+      "reward": 0.38541667722165585,
+      "reward_std": 0.3657289706170559,
+      "rewards/accuracy_reward": 0.38541667722165585,
+      "rewards/format_reward": 0.0,
+      "step": 139
+    },
+    {
+      "completion_length": 592.1666831970215,
+      "epoch": 1.4906666666666666,
+      "grad_norm": 0.8733373284339905,
+      "kl": 0.7615966796875,
+      "learning_rate": 5.2743817592912e-07,
+      "loss": 0.119,
+      "reward": 0.27083334140479565,
+      "reward_std": 0.2904551103711128,
+      "rewards/accuracy_reward": 0.27083334140479565,
+      "rewards/format_reward": 0.0,
+      "step": 140
+    },
+    {
+      "completion_length": 509.00001525878906,
+      "epoch": 1.5013333333333332,
+      "grad_norm": 0.5739810466766357,
+      "kl": 0.782470703125,
+      "learning_rate": 5.061286344672812e-07,
+      "loss": 0.2072,
+      "reward": 0.4270833460614085,
+      "reward_std": 0.3695550188422203,
+      "rewards/accuracy_reward": 0.4270833460614085,
+      "rewards/format_reward": 0.0,
+      "step": 141
+    },
+    {
+      "completion_length": 632.4166793823242,
+      "epoch": 1.512,
+      "grad_norm": 0.5129138231277466,
+      "kl": 0.471435546875,
+      "learning_rate": 4.85170802565797e-07,
+      "loss": 0.0212,
+      "reward": 0.2083333395421505,
+      "reward_std": 0.30546221882104874,
+      "rewards/accuracy_reward": 0.2083333395421505,
+      "rewards/format_reward": 0.0,
+      "step": 142
+    },
+    {
+      "completion_length": 675.6354293823242,
+      "epoch": 1.5226666666666666,
+      "grad_norm": 0.6775838136672974,
+      "kl": 1.49658203125,
+      "learning_rate": 4.645720967476709e-07,
+      "loss": 0.1271,
+      "reward": 0.16666666977107525,
+      "reward_std": 0.3798602595925331,
+      "rewards/accuracy_reward": 0.16666666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 143
+    },
+    {
+      "completion_length": 599.3333473205566,
+      "epoch": 1.5333333333333332,
+      "grad_norm": 0.7139673233032227,
+      "kl": 1.02490234375,
+      "learning_rate": 4.443398064489581e-07,
+      "loss": 0.1713,
+      "reward": 0.32291667629033327,
+      "reward_std": 0.4055767357349396,
+      "rewards/accuracy_reward": 0.32291667629033327,
+      "rewards/format_reward": 0.0,
+      "step": 144
+    },
+    {
+      "completion_length": 571.4687652587891,
+      "epoch": 1.544,
+      "grad_norm": 0.5391110181808472,
+      "kl": 0.8048095703125,
+      "learning_rate": 4.244810914391956e-07,
+      "loss": 0.0891,
+      "reward": 0.3020833348855376,
+      "reward_std": 0.2281883805990219,
+      "rewards/accuracy_reward": 0.3020833348855376,
+      "rewards/format_reward": 0.0,
+      "step": 145
+    },
+    {
+      "completion_length": 616.6250267028809,
+      "epoch": 1.5546666666666666,
+      "grad_norm": 0.23279787600040436,
+      "kl": 0.5025634765625,
+      "learning_rate": 4.050029792877119e-07,
+      "loss": 0.1257,
+      "reward": 0.17708333767950535,
+      "reward_std": 0.22828106209635735,
+      "rewards/accuracy_reward": 0.17708333767950535,
+      "rewards/format_reward": 0.0,
+      "step": 146
+    },
+    {
+      "completion_length": 546.5104331970215,
+      "epoch": 1.5653333333333332,
+      "grad_norm": 0.29375898838043213,
+      "kl": 0.83935546875,
+      "learning_rate": 3.8591236287672435e-07,
+      "loss": 0.161,
+      "reward": 0.40625001210719347,
+      "reward_std": 0.44266683980822563,
+      "rewards/accuracy_reward": 0.40625001210719347,
+      "rewards/format_reward": 0.0,
+      "step": 147
+    },
+    {
+      "completion_length": 602.1250190734863,
+      "epoch": 1.576,
+      "grad_norm": 0.38167208433151245,
+      "kl": 0.68280029296875,
+      "learning_rate": 3.6721599796209593e-07,
+      "loss": 0.0954,
+      "reward": 0.2916666688397527,
+      "reward_std": 0.3333779387176037,
+      "rewards/accuracy_reward": 0.2916666688397527,
+      "rewards/format_reward": 0.0,
+      "step": 148
+    },
+    {
+      "completion_length": 528.9479370117188,
+      "epoch": 1.5866666666666667,
+      "grad_norm": 0.2693575322628021,
+      "kl": 0.5400390625,
+      "learning_rate": 3.489205007826196e-07,
+      "loss": 0.1159,
+      "reward": 0.4166666818782687,
+      "reward_std": 0.31082959473133087,
+      "rewards/accuracy_reward": 0.4166666818782687,
+      "rewards/format_reward": 0.0,
+      "step": 149
+    },
+    {
+      "completion_length": 623.6250152587891,
+      "epoch": 1.5973333333333333,
+      "grad_norm": 0.4524155557155609,
+      "kl": 0.508544921875,
+      "learning_rate": 3.3103234571867633e-07,
+      "loss": 0.1098,
+      "reward": 0.22916666977107525,
+      "reward_std": 0.22326868772506714,
+      "rewards/accuracy_reward": 0.22916666977107525,
+      "rewards/format_reward": 0.0,
+      "step": 150
+    },
+    {
+      "completion_length": 543.322925567627,
+      "epoch": 1.608,
+      "grad_norm": 0.3804190456867218,
+      "kl": 0.6044921875,
+      "learning_rate": 3.135578630010894e-07,
+      "loss": 0.139,
+      "reward": 0.3020833469927311,
+      "reward_std": 0.2718803808093071,
+      "rewards/accuracy_reward": 0.3020833469927311,
+      "rewards/format_reward": 0.0,
+      "step": 151
+    },
+    {
+      "completion_length": 605.7708473205566,
+      "epoch": 1.6186666666666667,
+      "grad_norm": 0.2589690685272217,
+      "kl": 0.49359130859375,
+      "learning_rate": 2.965032364709947e-07,
+      "loss": 0.1063,
+      "reward": 0.3020833367481828,
+      "reward_std": 0.3481447994709015,
+      "rewards/accuracy_reward": 0.3020833367481828,
+      "rewards/format_reward": 0.0,
+      "step": 152
+    },
+    {
+      "completion_length": 596.3541793823242,
+      "epoch": 1.6293333333333333,
+      "grad_norm": 0.31066620349884033,
+      "kl": 0.743408203125,
+      "learning_rate": 2.798745013915132e-07,
+      "loss": 0.1456,
+      "reward": 0.3333333497866988,
+      "reward_std": 0.3522741571068764,
+      "rewards/accuracy_reward": 0.3333333497866988,
+      "rewards/format_reward": 0.0,
+      "step": 153
+    },
+    {
+      "completion_length": 630.0312614440918,
+      "epoch": 1.6400000000000001,
+      "grad_norm": 0.30095869302749634,
+      "kl": 0.455352783203125,
+      "learning_rate": 2.636775423120007e-07,
+      "loss": 0.0601,
+      "reward": 0.27083334140479565,
+      "reward_std": 0.2546139992773533,
+      "rewards/accuracy_reward": 0.27083334140479565,
+      "rewards/format_reward": 0.0,
+      "step": 154
+    },
+    {
+      "completion_length": 626.2812652587891,
+      "epoch": 1.6506666666666665,
+      "grad_norm": 0.4043548107147217,
+      "kl": 0.3909912109375,
+      "learning_rate": 2.479180909856347e-07,
+      "loss": 0.0545,
+      "reward": 0.18750000465661287,
+      "reward_std": 0.28787805512547493,
+      "rewards/accuracy_reward": 0.18750000465661287,
+      "rewards/format_reward": 0.0,
+      "step": 155
+    },
+    {
+      "completion_length": 611.9791870117188,
+      "epoch": 1.6613333333333333,
+      "grad_norm": 0.22528155148029327,
+      "kl": 0.381103515625,
+      "learning_rate": 2.3260172434106913e-07,
+      "loss": 0.0343,
+      "reward": 0.2708333386108279,
+      "reward_std": 0.19899365305900574,
+      "rewards/accuracy_reward": 0.2708333386108279,
+      "rewards/format_reward": 0.0,
+      "step": 156
+    },
+    {
+      "completion_length": 681.979175567627,
+      "epoch": 1.6720000000000002,
+      "grad_norm": 0.22992321848869324,
+      "kl": 0.5267333984375,
+      "learning_rate": 2.177338625088826e-07,
+      "loss": 0.2209,
+      "reward": 0.3020833395421505,
+      "reward_std": 0.33629096299409866,
+      "rewards/accuracy_reward": 0.3020833395421505,
+      "rewards/format_reward": 0.0,
+      "step": 157
+    },
+    {
+      "completion_length": 550.0833511352539,
+      "epoch": 1.6826666666666665,
+      "grad_norm": 1.0294519662857056,
+      "kl": 0.4268798828125,
+      "learning_rate": 2.0331976690350844e-07,
+      "loss": 0.0676,
+      "reward": 0.30208334885537624,
+      "reward_std": 0.2103346362709999,
+      "rewards/accuracy_reward": 0.30208334885537624,
+      "rewards/format_reward": 0.0,
+      "step": 158
+    },
+    {
+      "completion_length": 541.1458396911621,
+      "epoch": 1.6933333333333334,
+      "grad_norm": 0.44960758090019226,
+      "kl": 0.64111328125,
+      "learning_rate": 1.893645383613371e-07,
+      "loss": 0.1856,
+      "reward": 0.4583333423361182,
+      "reward_std": 0.3755134642124176,
+      "rewards/accuracy_reward": 0.4583333423361182,
+      "rewards/format_reward": 0.0,
+      "step": 159
+    },
+    {
+      "completion_length": 585.1354293823242,
+      "epoch": 1.704,
+      "grad_norm": 0.2397642582654953,
+      "kl": 0.6611480712890625,
+      "learning_rate": 1.7587311533563887e-07,
+      "loss": 0.1011,
+      "reward": 0.34375000186264515,
+      "reward_std": 0.3033146671950817,
+      "rewards/accuracy_reward": 0.34375000186264515,
+      "rewards/format_reward": 0.0,
+      "step": 160
+    },
+    {
+      "completion_length": 618.8854446411133,
+      "epoch": 1.7146666666666666,
+      "grad_norm": 0.30194056034088135,
+      "kl": 0.41375732421875,
+      "learning_rate": 1.6285027214895227e-07,
+      "loss": 0.0246,
+      "reward": 0.21875000558793545,
+      "reward_std": 0.32490549236536026,
+      "rewards/accuracy_reward": 0.21875000558793545,
+      "rewards/format_reward": 0.0,
+      "step": 161
+    },
+    {
+      "completion_length": 480.3020896911621,
+      "epoch": 1.7253333333333334,
+      "grad_norm": 0.3240108788013458,
+      "kl": 0.6272735595703125,
+      "learning_rate": 1.503006173035552e-07,
+      "loss": 0.0577,
+      "reward": 0.3020833348855376,
+      "reward_std": 0.21069689840078354,
+      "rewards/accuracy_reward": 0.3020833348855376,
+      "rewards/format_reward": 0.0,
+      "step": 162
+    },
+    {
+      "completion_length": 655.1562614440918,
+      "epoch": 1.736,
+      "grad_norm": 0.16485954821109772,
+      "kl": 0.4539642333984375,
+      "learning_rate": 1.3822859185061453e-07,
+      "loss": 0.0482,
+      "reward": 0.34375000558793545,
+      "reward_std": 0.3551059402525425,
+      "rewards/accuracy_reward": 0.34375000558793545,
+      "rewards/format_reward": 0.0,
+      "step": 163
+    },
+    {
+      "completion_length": 551.5833549499512,
+      "epoch": 1.7466666666666666,
+      "grad_norm": 0.33394524455070496,
+      "kl": 0.3665771484375,
+      "learning_rate": 1.2663846781859673e-07,
+      "loss": 0.0863,
+      "reward": 0.43750001303851604,
+      "reward_std": 0.2447880432009697,
+      "rewards/accuracy_reward": 0.43750001303851604,
+      "rewards/format_reward": 0.0,
+      "step": 164
+    },
+    {
+      "completion_length": 633.9687728881836,
+      "epoch": 1.7573333333333334,
+      "grad_norm": 0.5149164795875549,
+      "kl": 0.45458984375,
+      "learning_rate": 1.1553434670148605e-07,
+      "loss": 0.0203,
+      "reward": 0.1666666716337204,
+      "reward_std": 0.18463725596666336,
+      "rewards/accuracy_reward": 0.1666666716337204,
+      "rewards/format_reward": 0.0,
+      "step": 165
+    },
+    {
+      "completion_length": 551.354175567627,
+      "epoch": 1.768,
+      "grad_norm": 0.5073640942573547,
+      "kl": 0.592529296875,
+      "learning_rate": 1.049201580073576e-07,
+      "loss": 0.1616,
+      "reward": 0.3125000111758709,
+      "reward_std": 0.35565972700715065,
+      "rewards/accuracy_reward": 0.3125000111758709,
+      "rewards/format_reward": 0.0,
+      "step": 166
+    },
+    {
+      "completion_length": 652.8750228881836,
+      "epoch": 1.7786666666666666,
+      "grad_norm": 0.38101547956466675,
+      "kl": 0.4947509765625,
+      "learning_rate": 9.479965786780847e-08,
+      "loss": 0.0703,
+      "reward": 0.28125000838190317,
+      "reward_std": 0.31957534328103065,
+      "rewards/accuracy_reward": 0.28125000838190317,
+      "rewards/format_reward": 0.0,
+      "step": 167
+    },
+    {
+      "completion_length": 664.114616394043,
+      "epoch": 1.7893333333333334,
+      "grad_norm": 0.29761403799057007,
+      "kl": 0.537353515625,
+      "learning_rate": 8.517642770874318e-08,
+      "loss": 0.0996,
+      "reward": 0.4062500111758709,
+      "reward_std": 0.2633597403764725,
+      "rewards/accuracy_reward": 0.4062500111758709,
+      "rewards/format_reward": 0.0,
+      "step": 168
+    },
+    {
+      "completion_length": 566.6146087646484,
+      "epoch": 1.8,
+      "grad_norm": 0.5472803115844727,
+      "kl": 0.47998046875,
+      "learning_rate": 7.60538729829861e-08,
+      "loss": 0.1289,
+      "reward": 0.36458334885537624,
+      "reward_std": 0.27501849830150604,
+      "rewards/accuracy_reward": 0.36458334885537624,
+      "rewards/format_reward": 0.0,
+      "step": 169
+    },
+    {
+      "completion_length": 566.8750228881836,
+      "epoch": 1.8106666666666666,
+      "grad_norm": 0.4136505722999573,
+      "kl": 0.71148681640625,
+      "learning_rate": 6.743522196516388e-08,
+      "loss": 0.1041,
+      "reward": 0.21875000838190317,
+      "reward_std": 0.2671857886016369,
+      "rewards/accuracy_reward": 0.21875000838190317,
+      "rewards/format_reward": 0.0,
+      "step": 170
+    },
+    {
+      "completion_length": 616.1458587646484,
+      "epoch": 1.8213333333333335,
+      "grad_norm": 0.42460325360298157,
+      "kl": 0.53466796875,
+      "learning_rate": 5.932352460929052e-08,
+      "loss": 0.1008,
+      "reward": 0.2604166688397527,
+      "reward_std": 0.2951196953654289,
+      "rewards/accuracy_reward": 0.2604166688397527,
+      "rewards/format_reward": 0.0,
+      "step": 171
+    },
+    {
+      "completion_length": 569.2083435058594,
+      "epoch": 1.8319999999999999,
+      "grad_norm": 0.3544805347919464,
+      "kl": 0.1856689453125,
+      "learning_rate": 5.172165146945296e-08,
+      "loss": 0.1447,
+      "reward": 0.3645833395421505,
+      "reward_std": 0.359878059476614,
+      "rewards/accuracy_reward": 0.3645833395421505,
+      "rewards/format_reward": 0.0,
+      "step": 172
+    },
+    {
+      "completion_length": 539.2812652587891,
+      "epoch": 1.8426666666666667,
+      "grad_norm": 0.19653554260730743,
+      "kl": 0.3319091796875,
+      "learning_rate": 4.463229268398611e-08,
+      "loss": 0.1411,
+      "reward": 0.3125000139698386,
+      "reward_std": 0.3444414474070072,
+      "rewards/accuracy_reward": 0.3125000139698386,
+      "rewards/format_reward": 0.0,
+      "step": 173
+    },
+    {
+      "completion_length": 485.46876525878906,
+      "epoch": 1.8533333333333335,
+      "grad_norm": 0.3276241421699524,
+      "kl": 0.443328857421875,
+      "learning_rate": 3.805795702348941e-08,
+      "loss": 0.0521,
+      "reward": 0.41666667722165585,
+      "reward_std": 0.3042132332921028,
+      "rewards/accuracy_reward": 0.41666667722165585,
+      "rewards/format_reward": 0.0,
+      "step": 174
+    },
+    {
+      "completion_length": 619.3125076293945,
+      "epoch": 1.8639999999999999,
+      "grad_norm": 0.6277369856834412,
+      "kl": 0.71484375,
+      "learning_rate": 3.200097100302812e-08,
+      "loss": 0.0719,
+      "reward": 0.17708333488553762,
+      "reward_std": 0.21069689840078354,
+      "rewards/accuracy_reward": 0.17708333488553762,
+      "rewards/format_reward": 0.0,
+      "step": 175
+    },
+    {
+      "completion_length": 574.572940826416,
+      "epoch": 1.8746666666666667,
+      "grad_norm": 0.5137563347816467,
+      "kl": 0.8292236328125,
+      "learning_rate": 2.6463478058829426e-08,
+      "loss": 0.0123,
+      "reward": 0.11458333861082792,
+      "reward_std": 0.1569838561117649,
+      "rewards/accuracy_reward": 0.11458333861082792,
+      "rewards/format_reward": 0.0,
+      "step": 176
+    },
+    {
+      "completion_length": 495.9583511352539,
+      "epoch": 1.8853333333333333,
+      "grad_norm": 0.7497807145118713,
+      "kl": 0.44970703125,
+      "learning_rate": 2.144743778976782e-08,
+      "loss": 0.0592,
+      "reward": 0.43750000931322575,
+      "reward_std": 0.3692372292280197,
+      "rewards/accuracy_reward": 0.43750000931322575,
+      "rewards/format_reward": 0.0,
+      "step": 177
+    },
+    {
+      "completion_length": 630.9271049499512,
+      "epoch": 1.896,
+      "grad_norm": 0.29420143365859985,
+      "kl": 0.35113525390625,
+      "learning_rate": 1.6954625263904177e-08,
+      "loss": 0.1362,
+      "reward": 0.312500006519258,
+      "reward_std": 0.37610871344804764,
+      "rewards/accuracy_reward": 0.312500006519258,
+      "rewards/format_reward": 0.0,
+      "step": 178
+    },
+    {
+      "completion_length": 545.3750114440918,
+      "epoch": 1.9066666666666667,
+      "grad_norm": 0.38894397020339966,
+      "kl": 0.6158599853515625,
+      "learning_rate": 1.2986630390328214e-08,
+      "loss": 0.0661,
+      "reward": 0.44791667722165585,
+      "reward_std": 0.31957533583045006,
+      "rewards/accuracy_reward": 0.44791667722165585,
+      "rewards/format_reward": 0.0,
+      "step": 179
+    },
+    {
+      "completion_length": 590.3020973205566,
+      "epoch": 1.9173333333333333,
+      "grad_norm": 0.2821875810623169,
+      "kl": 0.53765869140625,
+      "learning_rate": 9.54485735652455e-09,
+      "loss": 0.0518,
+      "reward": 0.3125000102445483,
+      "reward_std": 0.18237072229385376,
+      "rewards/accuracy_reward": 0.3125000102445483,
+      "rewards/format_reward": 0.0,
+      "step": 180
+    },
+    {
+      "completion_length": 635.7187690734863,
+      "epoch": 1.928,
+      "grad_norm": 0.24759963154792786,
+      "kl": 0.51806640625,
+      "learning_rate": 6.630524131460037e-09,
+      "loss": 0.0553,
+      "reward": 0.23958333674818277,
+      "reward_std": 0.24263745918869972,
+      "rewards/accuracy_reward": 0.23958333674818277,
+      "rewards/format_reward": 0.0,
+      "step": 181
+    },
+    {
+      "completion_length": 558.8333377838135,
+      "epoch": 1.9386666666666668,
+      "grad_norm": 0.2948830723762512,
+      "kl": 0.4090576171875,
+      "learning_rate": 4.244662034572022e-09,
+      "loss": 0.0335,
+      "reward": 0.29166667722165585,
+      "reward_std": 0.298287820070982,
+      "rewards/accuracy_reward": 0.29166667722165585,
+      "rewards/format_reward": 0.0,
+      "step": 182
+    },
+    {
+      "completion_length": 596.3646049499512,
+      "epoch": 1.9493333333333334,
+      "grad_norm": 0.23650887608528137,
+      "kl": 0.533660888671875,
+      "learning_rate": 2.3881153708059656e-09,
+      "loss": 0.1084,
+      "reward": 0.3229166669771075,
+      "reward_std": 0.2319217473268509,
+      "rewards/accuracy_reward": 0.3229166669771075,
+      "rewards/format_reward": 0.0,
+      "step": 183
+    },
+    {
+      "completion_length": 461.8020896911621,
+      "epoch": 1.96,
+      "grad_norm": 0.697127640247345,
+      "kl": 0.49359130859375,
+      "learning_rate": 1.0615411318344315e-09,
+      "loss": 0.1386,
+      "reward": 0.4895833460614085,
+      "reward_std": 0.45709774643182755,
+      "rewards/accuracy_reward": 0.4895833460614085,
+      "rewards/format_reward": 0.0,
+      "step": 184
+    },
+    {
+      "completion_length": 533.010425567627,
+      "epoch": 1.9706666666666668,
+      "grad_norm": 0.9241780042648315,
+      "kl": 0.367156982421875,
+      "learning_rate": 2.6540876356256906e-10,
+      "loss": 0.1469,
+      "reward": 0.5104166744276881,
+      "reward_std": 0.3491060249507427,
+      "rewards/accuracy_reward": 0.5104166744276881,
+      "rewards/format_reward": 0.0,
+      "step": 185
+    },
+    {
+      "completion_length": 609.9062728881836,
+      "epoch": 1.9813333333333332,
+      "grad_norm": 0.3692702353000641,
+      "kl": 0.4107666015625,
+      "learning_rate": 0.0,
+      "loss": 0.1591,
+      "reward": 0.31250001303851604,
+      "reward_std": 0.34694400057196617,
+      "rewards/accuracy_reward": 0.31250001303851604,
+      "rewards/format_reward": 0.0,
+      "step": 186
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 186,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7aafec21dede675d85d68d8609623d0ef4748bc3194704cb64502b4163145136
+size 7544

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)