Training in progress, step 95, checkpoint

.gitattributes

@@ -52,3 +52,4 @@ checkpoint-75/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 checkpoint-80/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 checkpoint-85/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 checkpoint-90/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+checkpoint-95/tokenizer.json filter=lfs diff=lfs merge=lfs -text

checkpoint-95/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

checkpoint-95/config.json

@@ -0,0 +1,30 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 6144,
+  "max_position_embeddings": 40960,
+  "max_window_layers": 28,
+  "model_type": "qwen3",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 28,
+  "num_key_value_heads": 8,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

checkpoint-95/generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 0.6,
+  "top_k": 20,
+  "top_p": 0.95,
+  "transformers_version": "4.51.3"
+}

checkpoint-95/global_step95/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f19695a0e344f8e5439ac7c1657e236518c202b64b5244c8379897ef34202c78
+size 2949562736

checkpoint-95/global_step95/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af99565b1953eb069ae6a656f95e44cf0ad2fc686525cfba5a8c6422da34be07
+size 2949565232

checkpoint-95/global_step95/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a74fd5125fd3f6970ae5656e4f2bb271e1196951b7f170c28f2ec8917dd4f4d
+size 2949565936

checkpoint-95/global_step95/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3af388ac91faec34268fdb06a5a7121ebf50b6cb0c87ca6f7eac9be77c68a125
+size 2949566064

checkpoint-95/global_step95/bf16_zero_pp_rank_4_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb24bab227cbd1427cbc0e3e409928a77f317652c115db728ce8586fe9d77121
+size 2949565808

checkpoint-95/global_step95/bf16_zero_pp_rank_5_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f89034eb0ccd9f5312ac5a0bb44ef113bd98764144df24d0bfe58751a90298e
+size 2949565936

checkpoint-95/global_step95/bf16_zero_pp_rank_6_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51664d7fa9519493b9d7ef97bd42a1cedd3115cb5c9652749fffe868b5c909ec
+size 2949565936

checkpoint-95/global_step95/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a836bc2ed6fe32aa19909271a800b73e2f239376c0d23b39ee5e85209c4b856
+size 3441239288

checkpoint-95/latest

@@ -0,0 +1 @@
+global_step95

checkpoint-95/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5140d73b1e57dcb42f562d40e576b4d0e16f80475345601c6c6cb89a2fcc21ef
+size 4063515640

checkpoint-95/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcbaded82e3003e400116c1b7c2149b7f621db376b8678c41ea2960f2fe7c893
+size 15664

checkpoint-95/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab8c553488476670847626ca0cd90a10a26aaada95055efb13560c976b864810
+size 15728

checkpoint-95/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:22755f67648e4909f702fa30d6e8e9a31ff6fdbe5eb81f65743a9fcc4b6af22f
+size 15728

checkpoint-95/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f5d06df2b7664668e4b6e1901ce5b2f22fde390f8a0be5dd22b1f278e84a77c
+size 15792

checkpoint-95/rng_state_4.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80d3f8775a096ab578258a2256310da955f77f14de94b0af1adb8a4c8483d5de
+size 15728

checkpoint-95/rng_state_5.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e642e41a37b00d42af084c8c0561e9ef84e0844a350b48e7b45b9e93265d266e
+size 15728

checkpoint-95/rng_state_6.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3546e4855436972ad46c20248d0199b1abcb2ef7abbb34b24c3e5d66336fe2df
+size 15728

checkpoint-95/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3caba2b763c105d8c99e12fe5d18e730d3a0d9f6e6b4c883814817fcb947a4b9
+size 1064

checkpoint-95/special_tokens_map.json

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

checkpoint-95/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67cc0080ffd7555f723f423c27cfef314e1ad9d335c8b79f465c5faba1ed478b
+size 11422821

checkpoint-95/tokenizer_config.json

@@ -0,0 +1,240 @@
+{
+  "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
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "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\\n' }}\n    {%- endif %}\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    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n        {%- set ns.multi_step_tool = false %}\n        {%- set ns.last_query_index = index %}\n    {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set content = message.content %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in message.content %}\n                {%- set content = message.content.split('</think>')[-1].lstrip('\\n') %}\n                {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {%- if loop.last or (not loop.last and reasoning_content) %}\n                {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n            {%- else %}\n                {{- '<|im_start|>' + message.role + '\\n' + content }}\n            {%- endif %}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if (loop.first and content) or (not loop.first) %}\n                    {{- '\\n' }}\n                {%- endif %}\n                {%- if tool_call.function %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n{\"name\": \"' }}\n                {{- tool_call.name }}\n                {{- '\", \"arguments\": ' }}\n                {%- if tool_call.arguments is string %}\n                    {{- tool_call.arguments }}\n                {%- else %}\n                    {{- tool_call.arguments | tojson }}\n                {%- endif %}\n                {{- '}\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.first 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    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

checkpoint-95/trainer_state.json

@@ -0,0 +1,2314 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.3767349636483807,
+  "eval_steps": 500,
+  "global_step": 95,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6785714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5080.0,
+      "completions/mean_length": 4529.693359375,
+      "completions/mean_terminated_length": 3283.49072265625,
+      "completions/min_length": 986.0,
+      "completions/min_terminated_length": 986.0,
+      "epoch": 0.0039656311962987445,
+      "grad_norm": 0.15875904262065887,
+      "learning_rate": 0.0,
+      "loss": 0.0215,
+      "num_tokens": 1587977.0,
+      "reward": 0.3303571343421936,
+      "reward_std": 0.19694826006889343,
+      "rewards/accuracy_reward/mean": 0.3303571343421936,
+      "rewards/accuracy_reward/std": 0.4710431694984436,
+      "step": 1
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6904761904761905,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5117.0,
+      "completions/mean_length": 4590.5654296875,
+      "completions/mean_terminated_length": 3409.519287109375,
+      "completions/min_length": 460.0,
+      "completions/min_terminated_length": 460.0,
+      "epoch": 0.007931262392597489,
+      "grad_norm": 0.17138810455799103,
+      "learning_rate": 3.846153846153846e-08,
+      "loss": 0.0399,
+      "num_tokens": 3197661.0,
+      "reward": 0.3154762089252472,
+      "reward_std": 0.21732226014137268,
+      "rewards/accuracy_reward/mean": 0.3154761791229248,
+      "rewards/accuracy_reward/std": 0.4653983414173126,
+      "step": 2
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6309523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5119.0,
+      "completions/mean_length": 4508.61328125,
+      "completions/mean_terminated_length": 3463.338623046875,
+      "completions/min_length": 889.0,
+      "completions/min_terminated_length": 889.0,
+      "epoch": 0.011896893588896233,
+      "grad_norm": 0.13771647214889526,
+      "learning_rate": 7.692307692307692e-08,
+      "loss": 0.0191,
+      "num_tokens": 4779191.0,
+      "reward": 0.4375,
+      "reward_std": 0.15370962023735046,
+      "rewards/accuracy_reward/mean": 0.4375,
+      "rewards/accuracy_reward/std": 0.4968182444572449,
+      "step": 3
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6547619047619048,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5057.0,
+      "completions/mean_length": 4419.59814453125,
+      "completions/mean_terminated_length": 3091.25,
+      "completions/min_length": 513.0,
+      "completions/min_terminated_length": 513.0,
+      "epoch": 0.015862524785194978,
+      "grad_norm": 0.19674311578273773,
+      "learning_rate": 1.1538461538461539e-07,
+      "loss": 0.0414,
+      "num_tokens": 6334034.0,
+      "reward": 0.3541666865348816,
+      "reward_std": 0.24031147360801697,
+      "rewards/accuracy_reward/mean": 0.3541666567325592,
+      "rewards/accuracy_reward/std": 0.4789733290672302,
+      "step": 4
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6845238095238095,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4470.1220703125,
+      "completions/mean_terminated_length": 3060.009521484375,
+      "completions/min_length": 521.0,
+      "completions/min_terminated_length": 521.0,
+      "epoch": 0.01982815598149372,
+      "grad_norm": 0.19068369269371033,
+      "learning_rate": 1.5384615384615385e-07,
+      "loss": 0.0247,
+      "num_tokens": 7905049.0,
+      "reward": 0.3452381193637848,
+      "reward_std": 0.19077685475349426,
+      "rewards/accuracy_reward/mean": 0.3452380895614624,
+      "rewards/accuracy_reward/std": 0.4761549234390259,
+      "step": 5
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6785714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5057.0,
+      "completions/mean_length": 4508.5712890625,
+      "completions/mean_terminated_length": 3217.77783203125,
+      "completions/min_length": 806.0,
+      "completions/min_terminated_length": 806.0,
+      "epoch": 0.023793787177792465,
+      "grad_norm": 0.1556510627269745,
+      "learning_rate": 1.9230769230769231e-07,
+      "loss": 0.0215,
+      "num_tokens": 9490465.0,
+      "reward": 0.324404776096344,
+      "reward_std": 0.17464300990104675,
+      "rewards/accuracy_reward/mean": 0.324404776096344,
+      "rewards/accuracy_reward/std": 0.4688502550125122,
+      "step": 6
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7113095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5108.0,
+      "completions/mean_length": 4571.119140625,
+      "completions/mean_terminated_length": 3218.721435546875,
+      "completions/min_length": 739.0,
+      "completions/min_terminated_length": 739.0,
+      "epoch": 0.02775941837409121,
+      "grad_norm": 0.17015115916728973,
+      "learning_rate": 2.3076923076923078e-07,
+      "loss": 0.033,
+      "num_tokens": 11093225.0,
+      "reward": 0.3273809850215912,
+      "reward_std": 0.15950332581996918,
+      "rewards/accuracy_reward/mean": 0.3273809552192688,
+      "rewards/accuracy_reward/std": 0.4699574410915375,
+      "step": 7
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7351190476190477,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5112.0,
+      "completions/mean_length": 4626.4912109375,
+      "completions/mean_terminated_length": 3256.865234375,
+      "completions/min_length": 785.0,
+      "completions/min_terminated_length": 785.0,
+      "epoch": 0.031725049570389956,
+      "grad_norm": 0.14421533048152924,
+      "learning_rate": 2.692307692307692e-07,
+      "loss": 0.0034,
+      "num_tokens": 12715148.0,
+      "reward": 0.2678571343421936,
+      "reward_std": 0.16237163543701172,
+      "rewards/accuracy_reward/mean": 0.2678571343421936,
+      "rewards/accuracy_reward/std": 0.443503201007843,
+      "step": 8
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6190476190476191,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5098.0,
+      "completions/mean_length": 4260.5654296875,
+      "completions/mean_terminated_length": 2863.984375,
+      "completions/min_length": 401.0,
+      "completions/min_terminated_length": 401.0,
+      "epoch": 0.0356906807666887,
+      "grad_norm": 0.1457386314868927,
+      "learning_rate": 3.076923076923077e-07,
+      "loss": 0.0147,
+      "num_tokens": 14213118.0,
+      "reward": 0.3988095223903656,
+      "reward_std": 0.17121532559394836,
+      "rewards/accuracy_reward/mean": 0.3988095223903656,
+      "rewards/accuracy_reward/std": 0.49038368463516235,
+      "step": 9
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6011904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5039.0,
+      "completions/mean_length": 4433.375,
+      "completions/mean_terminated_length": 3398.313232421875,
+      "completions/min_length": 1047.0,
+      "completions/min_terminated_length": 1047.0,
+      "epoch": 0.03965631196298744,
+      "grad_norm": 0.15511175990104675,
+      "learning_rate": 3.461538461538461e-07,
+      "loss": 0.016,
+      "num_tokens": 15773184.0,
+      "reward": 0.4047619104385376,
+      "reward_std": 0.183305025100708,
+      "rewards/accuracy_reward/mean": 0.4047619104385376,
+      "rewards/accuracy_reward/std": 0.49157798290252686,
+      "step": 10
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6607142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5068.0,
+      "completions/mean_length": 4363.806640625,
+      "completions/mean_terminated_length": 2891.21923828125,
+      "completions/min_length": 660.0,
+      "completions/min_terminated_length": 660.0,
+      "epoch": 0.043621943159286185,
+      "grad_norm": 0.15191057324409485,
+      "learning_rate": 3.8461538461538463e-07,
+      "loss": 0.0289,
+      "num_tokens": 17301943.0,
+      "reward": 0.3482142984867096,
+      "reward_std": 0.1535850614309311,
+      "rewards/accuracy_reward/mean": 0.3482142984867096,
+      "rewards/accuracy_reward/std": 0.47711488604545593,
+      "step": 11
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7023809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5111.0,
+      "completions/mean_length": 4614.21435546875,
+      "completions/mean_terminated_length": 3420.559814453125,
+      "completions/min_length": 884.0,
+      "completions/min_terminated_length": 884.0,
+      "epoch": 0.04758757435558493,
+      "grad_norm": 0.15457122027873993,
+      "learning_rate": 4.2307692307692304e-07,
+      "loss": 0.0168,
+      "num_tokens": 18919753.0,
+      "reward": 0.3095238208770752,
+      "reward_std": 0.15950334072113037,
+      "rewards/accuracy_reward/mean": 0.3095238208770752,
+      "rewards/accuracy_reward/std": 0.4629868268966675,
+      "step": 12
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.8244047619047619,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5068.0,
+      "completions/mean_length": 4856.9794921875,
+      "completions/mean_terminated_length": 3622.11865234375,
+      "completions/min_length": 1378.0,
+      "completions/min_terminated_length": 1378.0,
+      "epoch": 0.051553205551883675,
+      "grad_norm": 0.14849524199962616,
+      "learning_rate": 4.6153846153846156e-07,
+      "loss": 0.0164,
+      "num_tokens": 20626224.0,
+      "reward": 0.1815476268529892,
+      "reward_std": 0.144363671541214,
+      "rewards/accuracy_reward/mean": 0.181547611951828,
+      "rewards/accuracy_reward/std": 0.3860461115837097,
+      "step": 13
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6398809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5091.0,
+      "completions/mean_length": 4491.96142578125,
+      "completions/mean_terminated_length": 3376.024658203125,
+      "completions/min_length": 1230.0,
+      "completions/min_terminated_length": 1230.0,
+      "epoch": 0.05551883674818242,
+      "grad_norm": 0.1783030778169632,
+      "learning_rate": 5e-07,
+      "loss": 0.0229,
+      "num_tokens": 22203953.0,
+      "reward": 0.3839285671710968,
+      "reward_std": 0.22374281287193298,
+      "rewards/accuracy_reward/mean": 0.3839285671710968,
+      "rewards/accuracy_reward/std": 0.48706620931625366,
+      "step": 14
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6309523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5107.0,
+      "completions/mean_length": 4464.40771484375,
+      "completions/mean_terminated_length": 3343.556396484375,
+      "completions/min_length": 1600.0,
+      "completions/min_terminated_length": 1600.0,
+      "epoch": 0.059484467944481166,
+      "grad_norm": 0.15186423063278198,
+      "learning_rate": 5.384615384615384e-07,
+      "loss": 0.0218,
+      "num_tokens": 23771476.0,
+      "reward": 0.386904776096344,
+      "reward_std": 0.17159302532672882,
+      "rewards/accuracy_reward/mean": 0.386904776096344,
+      "rewards/accuracy_reward/std": 0.4877679646015167,
+      "step": 15
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6666666666666667,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5112.0,
+      "completions/mean_length": 4450.630859375,
+      "completions/mean_terminated_length": 3111.89306640625,
+      "completions/min_length": 753.0,
+      "completions/min_terminated_length": 753.0,
+      "epoch": 0.06345009914077991,
+      "grad_norm": 0.1414450705051422,
+      "learning_rate": 5.769230769230768e-07,
+      "loss": 0.0157,
+      "num_tokens": 25332360.0,
+      "reward": 0.386904776096344,
+      "reward_std": 0.16062207520008087,
+      "rewards/accuracy_reward/mean": 0.386904776096344,
+      "rewards/accuracy_reward/std": 0.48776793479919434,
+      "step": 16
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6696428571428572,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5099.0,
+      "completions/mean_length": 4384.7529296875,
+      "completions/mean_terminated_length": 2894.387451171875,
+      "completions/min_length": 1048.0,
+      "completions/min_terminated_length": 1048.0,
+      "epoch": 0.06741573033707865,
+      "grad_norm": 0.13619941473007202,
+      "learning_rate": 6.153846153846154e-07,
+      "loss": 0.0175,
+      "num_tokens": 26878273.0,
+      "reward": 0.3273809552192688,
+      "reward_std": 0.13819226622581482,
+      "rewards/accuracy_reward/mean": 0.3273809552192688,
+      "rewards/accuracy_reward/std": 0.4699574410915375,
+      "step": 17
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6815476190476191,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5099.0,
+      "completions/mean_length": 4545.28271484375,
+      "completions/mean_terminated_length": 3315.2802734375,
+      "completions/min_length": 839.0,
+      "completions/min_terminated_length": 839.0,
+      "epoch": 0.0713813615333774,
+      "grad_norm": 0.13691294193267822,
+      "learning_rate": 6.538461538461538e-07,
+      "loss": 0.0151,
+      "num_tokens": 28476366.0,
+      "reward": 0.348214328289032,
+      "reward_std": 0.17850549519062042,
+      "rewards/accuracy_reward/mean": 0.3482142984867096,
+      "rewards/accuracy_reward/std": 0.47711488604545593,
+      "step": 18
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6696428571428572,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5079.0,
+      "completions/mean_length": 4489.875,
+      "completions/mean_terminated_length": 3212.5947265625,
+      "completions/min_length": 737.0,
+      "completions/min_terminated_length": 737.0,
+      "epoch": 0.07534699272967614,
+      "grad_norm": 0.16069504618644714,
+      "learning_rate": 6.923076923076922e-07,
+      "loss": 0.0153,
+      "num_tokens": 30051060.0,
+      "reward": 0.3035714328289032,
+      "reward_std": 0.2042384147644043,
+      "rewards/accuracy_reward/mean": 0.3035714328289032,
+      "rewards/accuracy_reward/std": 0.4604855179786682,
+      "step": 19
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.625,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5111.0,
+      "completions/mean_length": 4455.880859375,
+      "completions/mean_terminated_length": 3349.01611328125,
+      "completions/min_length": 926.0,
+      "completions/min_terminated_length": 926.0,
+      "epoch": 0.07931262392597488,
+      "grad_norm": 0.15025638043880463,
+      "learning_rate": 7.307692307692307e-07,
+      "loss": 0.0268,
+      "num_tokens": 31615238.0,
+      "reward": 0.3988095223903656,
+      "reward_std": 0.1662341058254242,
+      "rewards/accuracy_reward/mean": 0.3988095223903656,
+      "rewards/accuracy_reward/std": 0.49038371443748474,
+      "step": 20
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.8273809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4765.625,
+      "completions/mean_terminated_length": 3067.06884765625,
+      "completions/min_length": 720.0,
+      "completions/min_terminated_length": 720.0,
+      "epoch": 0.08327825512227363,
+      "grad_norm": 0.13238537311553955,
+      "learning_rate": 7.692307692307693e-07,
+      "loss": 0.0069,
+      "num_tokens": 33284558.0,
+      "reward": 0.1577381044626236,
+      "reward_std": 0.1326516568660736,
+      "rewards/accuracy_reward/mean": 0.1577380895614624,
+      "rewards/accuracy_reward/std": 0.3650388717651367,
+      "step": 21
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6190476190476191,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5085.0,
+      "completions/mean_length": 4539.9345703125,
+      "completions/mean_terminated_length": 3597.328125,
+      "completions/min_length": 1152.0,
+      "completions/min_terminated_length": 1152.0,
+      "epoch": 0.08724388631857237,
+      "grad_norm": 0.1624104231595993,
+      "learning_rate": 8.076923076923077e-07,
+      "loss": 0.0234,
+      "num_tokens": 34875460.0,
+      "reward": 0.3958333432674408,
+      "reward_std": 0.16928410530090332,
+      "rewards/accuracy_reward/mean": 0.3958333432674408,
+      "rewards/accuracy_reward/std": 0.4897582530975342,
+      "step": 22
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6398809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5091.0,
+      "completions/mean_length": 4520.8095703125,
+      "completions/mean_terminated_length": 3456.132080078125,
+      "completions/min_length": 1125.0,
+      "completions/min_terminated_length": 1125.0,
+      "epoch": 0.09120951751487112,
+      "grad_norm": 0.14350676536560059,
+      "learning_rate": 8.461538461538461e-07,
+      "loss": 0.0207,
+      "num_tokens": 36463128.0,
+      "reward": 0.3541666865348816,
+      "reward_std": 0.14336946606636047,
+      "rewards/accuracy_reward/mean": 0.3541666567325592,
+      "rewards/accuracy_reward/std": 0.4789733290672302,
+      "step": 23
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6488095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5051.0,
+      "completions/mean_length": 4480.482421875,
+      "completions/mean_terminated_length": 3299.0,
+      "completions/min_length": 806.0,
+      "completions/min_terminated_length": 806.0,
+      "epoch": 0.09517514871116986,
+      "grad_norm": 0.16741891205310822,
+      "learning_rate": 8.846153846153846e-07,
+      "loss": 0.0353,
+      "num_tokens": 38034894.0,
+      "reward": 0.4047619104385376,
+      "reward_std": 0.20037592947483063,
+      "rewards/accuracy_reward/mean": 0.4047619104385376,
+      "rewards/accuracy_reward/std": 0.49157795310020447,
+      "step": 24
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7202380952380952,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5061.0,
+      "completions/mean_length": 4567.5419921875,
+      "completions/mean_terminated_length": 3145.255126953125,
+      "completions/min_length": 1458.0,
+      "completions/min_terminated_length": 1458.0,
+      "epoch": 0.0991407799074686,
+      "grad_norm": 0.11739571392536163,
+      "learning_rate": 9.230769230769231e-07,
+      "loss": 0.0118,
+      "num_tokens": 39638744.0,
+      "reward": 0.306547611951828,
+      "reward_std": 0.12878917157649994,
+      "rewards/accuracy_reward/mean": 0.306547611951828,
+      "rewards/accuracy_reward/std": 0.46174749732017517,
+      "step": 25
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6636904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5087.0,
+      "completions/mean_length": 4468.39013671875,
+      "completions/mean_terminated_length": 3182.468994140625,
+      "completions/min_length": 532.0,
+      "completions/min_terminated_length": 532.0,
+      "epoch": 0.10310641110376735,
+      "grad_norm": 0.143392875790596,
+      "learning_rate": 9.615384615384615e-07,
+      "loss": 0.0239,
+      "num_tokens": 41207245.0,
+      "reward": 0.380952388048172,
+      "reward_std": 0.1844237744808197,
+      "rewards/accuracy_reward/mean": 0.380952388048172,
+      "rewards/accuracy_reward/std": 0.48634520173072815,
+      "step": 26
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6339285714285714,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5111.0,
+      "completions/mean_length": 4317.0654296875,
+      "completions/mean_terminated_length": 2926.61767578125,
+      "completions/min_length": 574.0,
+      "completions/min_terminated_length": 574.0,
+      "epoch": 0.10707204230006609,
+      "grad_norm": 0.18580368161201477,
+      "learning_rate": 1e-06,
+      "loss": 0.0182,
+      "num_tokens": 42723047.0,
+      "reward": 0.3630952537059784,
+      "reward_std": 0.22242805361747742,
+      "rewards/accuracy_reward/mean": 0.363095223903656,
+      "rewards/accuracy_reward/std": 0.48160919547080994,
+      "step": 27
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6607142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5115.0,
+      "completions/mean_length": 4533.1337890625,
+      "completions/mean_terminated_length": 3390.28955078125,
+      "completions/min_length": 741.0,
+      "completions/min_terminated_length": 741.0,
+      "epoch": 0.11103767349636484,
+      "grad_norm": 0.1611405462026596,
+      "learning_rate": 9.95575221238938e-07,
+      "loss": 0.0383,
+      "num_tokens": 44312342.0,
+      "reward": 0.3392857313156128,
+      "reward_std": 0.22529634833335876,
+      "rewards/accuracy_reward/mean": 0.3392857015132904,
+      "rewards/accuracy_reward/std": 0.47417306900024414,
+      "step": 28
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6339285714285714,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5068.0,
+      "completions/mean_length": 4274.41064453125,
+      "completions/mean_terminated_length": 2810.097412109375,
+      "completions/min_length": 790.0,
+      "completions/min_terminated_length": 790.0,
+      "epoch": 0.11500330469266358,
+      "grad_norm": 0.13915002346038818,
+      "learning_rate": 9.91150442477876e-07,
+      "loss": 0.0167,
+      "num_tokens": 45815672.0,
+      "reward": 0.3511904776096344,
+      "reward_std": 0.1395641267299652,
+      "rewards/accuracy_reward/mean": 0.3511904776096344,
+      "rewards/accuracy_reward/std": 0.4780542850494385,
+      "step": 29
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5892857142857143,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5108.0,
+      "completions/mean_length": 4291.70849609375,
+      "completions/mean_terminated_length": 3103.289794921875,
+      "completions/min_length": 938.0,
+      "completions/min_terminated_length": 938.0,
+      "epoch": 0.11896893588896233,
+      "grad_norm": 0.18397149443626404,
+      "learning_rate": 9.867256637168142e-07,
+      "loss": 0.0263,
+      "num_tokens": 47322384.0,
+      "reward": 0.4226190447807312,
+      "reward_std": 0.22417759895324707,
+      "rewards/accuracy_reward/mean": 0.4226190447807312,
+      "rewards/accuracy_reward/std": 0.49471262097358704,
+      "step": 30
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7380952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5099.0,
+      "completions/mean_length": 4753.0595703125,
+      "completions/mean_terminated_length": 3718.95458984375,
+      "completions/min_length": 1595.0,
+      "completions/min_terminated_length": 1595.0,
+      "epoch": 0.12293456708526107,
+      "grad_norm": 0.17866523563861847,
+      "learning_rate": 9.82300884955752e-07,
+      "loss": 0.0282,
+      "num_tokens": 48988364.0,
+      "reward": 0.3422619104385376,
+      "reward_std": 0.24853865802288055,
+      "rewards/accuracy_reward/mean": 0.3422619104385376,
+      "rewards/accuracy_reward/std": 0.4751743972301483,
+      "step": 31
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6130952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4394.5654296875,
+      "completions/mean_terminated_length": 3245.03076171875,
+      "completions/min_length": 557.0,
+      "completions/min_terminated_length": 557.0,
+      "epoch": 0.12690019828155982,
+      "grad_norm": 0.1624467670917511,
+      "learning_rate": 9.778761061946902e-07,
+      "loss": 0.0359,
+      "num_tokens": 50531388.0,
+      "reward": 0.4017857313156128,
+      "reward_std": 0.18523626029491425,
+      "rewards/accuracy_reward/mean": 0.4017857015132904,
+      "rewards/accuracy_reward/std": 0.4909902513027191,
+      "step": 32
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6547619047619048,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5109.0,
+      "completions/mean_length": 4470.70263671875,
+      "completions/mean_terminated_length": 3239.27587890625,
+      "completions/min_length": 473.0,
+      "completions/min_terminated_length": 473.0,
+      "epoch": 0.13086582947785855,
+      "grad_norm": 0.1461624801158905,
+      "learning_rate": 9.734513274336282e-07,
+      "loss": 0.0233,
+      "num_tokens": 52098884.0,
+      "reward": 0.3601190447807312,
+      "reward_std": 0.1533319354057312,
+      "rewards/accuracy_reward/mean": 0.3601190447807312,
+      "rewards/accuracy_reward/std": 0.4807506501674652,
+      "step": 33
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6279761904761905,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5065.0,
+      "completions/mean_length": 4372.806640625,
+      "completions/mean_terminated_length": 3111.544189453125,
+      "completions/min_length": 620.0,
+      "completions/min_terminated_length": 620.0,
+      "epoch": 0.1348314606741573,
+      "grad_norm": 0.15794773399829865,
+      "learning_rate": 9.690265486725663e-07,
+      "loss": 0.022,
+      "num_tokens": 53640597.0,
+      "reward": 0.3571428656578064,
+      "reward_std": 0.21744683384895325,
+      "rewards/accuracy_reward/mean": 0.3571428656578064,
+      "rewards/accuracy_reward/std": 0.47987204790115356,
+      "step": 34
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6696428571428572,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5044.0,
+      "completions/mean_length": 4431.982421875,
+      "completions/mean_terminated_length": 3037.351318359375,
+      "completions/min_length": 837.0,
+      "completions/min_terminated_length": 837.0,
+      "epoch": 0.13879709187045605,
+      "grad_norm": 0.23040273785591125,
+      "learning_rate": 9.646017699115042e-07,
+      "loss": 0.0174,
+      "num_tokens": 55195623.0,
+      "reward": 0.3571428656578064,
+      "reward_std": 0.16237162053585052,
+      "rewards/accuracy_reward/mean": 0.3571428656578064,
+      "rewards/accuracy_reward/std": 0.47987204790115356,
+      "step": 35
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6517857142857143,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5069.0,
+      "completions/mean_length": 4549.818359375,
+      "completions/mean_terminated_length": 3482.5556640625,
+      "completions/min_length": 848.0,
+      "completions/min_terminated_length": 848.0,
+      "epoch": 0.1427627230667548,
+      "grad_norm": 0.15275774896144867,
+      "learning_rate": 9.601769911504426e-07,
+      "loss": 0.0147,
+      "num_tokens": 56794820.0,
+      "reward": 0.3720238208770752,
+      "reward_std": 0.18274565041065216,
+      "rewards/accuracy_reward/mean": 0.3720238208770752,
+      "rewards/accuracy_reward/std": 0.48406556248664856,
+      "step": 36
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6785714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5108.0,
+      "completions/mean_length": 4404.52685546875,
+      "completions/mean_terminated_length": 2894.083251953125,
+      "completions/min_length": 820.0,
+      "completions/min_terminated_length": 820.0,
+      "epoch": 0.14672835426305353,
+      "grad_norm": 0.15565535426139832,
+      "learning_rate": 9.557522123893805e-07,
+      "loss": 0.0246,
+      "num_tokens": 58342355.0,
+      "reward": 0.2797619104385376,
+      "reward_std": 0.14916320145130157,
+      "rewards/accuracy_reward/mean": 0.2797619104385376,
+      "rewards/accuracy_reward/std": 0.44955164194107056,
+      "step": 37
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6607142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5105.0,
+      "completions/mean_length": 4408.27685546875,
+      "completions/mean_terminated_length": 3022.28955078125,
+      "completions/min_length": 768.0,
+      "completions/min_terminated_length": 768.0,
+      "epoch": 0.15069398545935228,
+      "grad_norm": 0.15097402036190033,
+      "learning_rate": 9.513274336283185e-07,
+      "loss": 0.0153,
+      "num_tokens": 59888900.0,
+      "reward": 0.3392857313156128,
+      "reward_std": 0.176574245095253,
+      "rewards/accuracy_reward/mean": 0.3392857015132904,
+      "rewards/accuracy_reward/std": 0.47417306900024414,
+      "step": 38
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6785714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5068.0,
+      "completions/mean_length": 4313.22900390625,
+      "completions/mean_terminated_length": 2610.04638671875,
+      "completions/min_length": 407.0,
+      "completions/min_terminated_length": 407.0,
+      "epoch": 0.15465961665565103,
+      "grad_norm": 0.12911847233772278,
+      "learning_rate": 9.469026548672566e-07,
+      "loss": 0.0222,
+      "num_tokens": 61408285.0,
+      "reward": 0.3601190745830536,
+      "reward_std": 0.11570531129837036,
+      "rewards/accuracy_reward/mean": 0.3601190447807312,
+      "rewards/accuracy_reward/std": 0.4807506501674652,
+      "step": 39
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5416666666666667,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5072.0,
+      "completions/mean_length": 4226.119140625,
+      "completions/mean_terminated_length": 3169.71435546875,
+      "completions/min_length": 814.0,
+      "completions/min_terminated_length": 814.0,
+      "epoch": 0.15862524785194976,
+      "grad_norm": 0.1661224365234375,
+      "learning_rate": 9.424778761061947e-07,
+      "loss": 0.0308,
+      "num_tokens": 62893463.0,
+      "reward": 0.4970238208770752,
+      "reward_std": 0.17464299499988556,
+      "rewards/accuracy_reward/mean": 0.4970238208770752,
+      "rewards/accuracy_reward/std": 0.5007368326187134,
+      "step": 40
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5535714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5114.0,
+      "completions/mean_length": 4209.2412109375,
+      "completions/mean_terminated_length": 3079.900146484375,
+      "completions/min_length": 662.0,
+      "completions/min_terminated_length": 662.0,
+      "epoch": 0.1625908790482485,
+      "grad_norm": 0.14029008150100708,
+      "learning_rate": 9.380530973451328e-07,
+      "loss": 0.0097,
+      "num_tokens": 64374356.0,
+      "reward": 0.4404762089252472,
+      "reward_std": 0.1558939814567566,
+      "rewards/accuracy_reward/mean": 0.4404761791229248,
+      "rewards/accuracy_reward/std": 0.497184693813324,
+      "step": 41
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7291666666666667,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5120.0,
+      "completions/mean_length": 4692.96142578125,
+      "completions/mean_terminated_length": 3543.241943359375,
+      "completions/min_length": 683.0,
+      "completions/min_terminated_length": 683.0,
+      "epoch": 0.16655651024454726,
+      "grad_norm": 0.14169327914714813,
+      "learning_rate": 9.336283185840708e-07,
+      "loss": 0.0194,
+      "num_tokens": 66024259.0,
+      "reward": 0.3125,
+      "reward_std": 0.18772687017917633,
+      "rewards/accuracy_reward/mean": 0.3125,
+      "rewards/accuracy_reward/std": 0.46420371532440186,
+      "step": 42
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6934523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5097.0,
+      "completions/mean_length": 4603.607421875,
+      "completions/mean_terminated_length": 3435.456298828125,
+      "completions/min_length": 1193.0,
+      "completions/min_terminated_length": 1193.0,
+      "epoch": 0.17052214144084601,
+      "grad_norm": 0.127787783741951,
+      "learning_rate": 9.292035398230088e-07,
+      "loss": 0.0194,
+      "num_tokens": 67639465.0,
+      "reward": 0.324404776096344,
+      "reward_std": 0.1418730616569519,
+      "rewards/accuracy_reward/mean": 0.324404776096344,
+      "rewards/accuracy_reward/std": 0.4688502550125122,
+      "step": 43
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5773809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5077.0,
+      "completions/mean_length": 4377.84814453125,
+      "completions/mean_terminated_length": 3363.92236328125,
+      "completions/min_length": 855.0,
+      "completions/min_terminated_length": 855.0,
+      "epoch": 0.17448777263714474,
+      "grad_norm": 0.1367839127779007,
+      "learning_rate": 9.247787610619469e-07,
+      "loss": 0.0085,
+      "num_tokens": 69180496.0,
+      "reward": 0.386904776096344,
+      "reward_std": 0.1542689949274063,
+      "rewards/accuracy_reward/mean": 0.386904776096344,
+      "rewards/accuracy_reward/std": 0.48776793479919434,
+      "step": 44
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5952380952380952,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5114.0,
+      "completions/mean_length": 4334.4462890625,
+      "completions/mean_terminated_length": 3179.220703125,
+      "completions/min_length": 705.0,
+      "completions/min_terminated_length": 705.0,
+      "epoch": 0.1784534038334435,
+      "grad_norm": 0.16425253450870514,
+      "learning_rate": 9.203539823008849e-07,
+      "loss": 0.0325,
+      "num_tokens": 70702426.0,
+      "reward": 0.449404776096344,
+      "reward_std": 0.19445765018463135,
+      "rewards/accuracy_reward/mean": 0.449404776096344,
+      "rewards/accuracy_reward/std": 0.49817541241645813,
+      "step": 45
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5863095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5071.0,
+      "completions/mean_length": 4327.92578125,
+      "completions/mean_terminated_length": 3205.345458984375,
+      "completions/min_length": 478.0,
+      "completions/min_terminated_length": 478.0,
+      "epoch": 0.18241903502974224,
+      "grad_norm": 0.17978708446025848,
+      "learning_rate": 9.159292035398229e-07,
+      "loss": 0.024,
+      "num_tokens": 72223461.0,
+      "reward": 0.3958333432674408,
+      "reward_std": 0.24031147360801697,
+      "rewards/accuracy_reward/mean": 0.3958333432674408,
+      "rewards/accuracy_reward/std": 0.4897582530975342,
+      "step": 46
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6607142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5082.0,
+      "completions/mean_length": 4581.931640625,
+      "completions/mean_terminated_length": 3534.114013671875,
+      "completions/min_length": 1192.0,
+      "completions/min_terminated_length": 1192.0,
+      "epoch": 0.18638466622604097,
+      "grad_norm": 0.12531961500644684,
+      "learning_rate": 9.11504424778761e-07,
+      "loss": 0.0157,
+      "num_tokens": 73829620.0,
+      "reward": 0.392857164144516,
+      "reward_std": 0.15028195083141327,
+      "rewards/accuracy_reward/mean": 0.3928571343421936,
+      "rewards/accuracy_reward/std": 0.4891138970851898,
+      "step": 47
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5952380952380952,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4333.232421875,
+      "completions/mean_terminated_length": 3176.220703125,
+      "completions/min_length": 561.0,
+      "completions/min_terminated_length": 561.0,
+      "epoch": 0.19035029742233972,
+      "grad_norm": 0.16320611536502838,
+      "learning_rate": 9.07079646017699e-07,
+      "loss": 0.0232,
+      "num_tokens": 75351616.0,
+      "reward": 0.455357164144516,
+      "reward_std": 0.180001899600029,
+      "rewards/accuracy_reward/mean": 0.4553571343421936,
+      "rewards/accuracy_reward/std": 0.49874576926231384,
+      "step": 48
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.7410714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5116.0,
+      "completions/mean_length": 4707.73828125,
+      "completions/mean_terminated_length": 3527.816162109375,
+      "completions/min_length": 1823.0,
+      "completions/min_terminated_length": 1823.0,
+      "epoch": 0.19431592861863847,
+      "grad_norm": 0.1277765929698944,
+      "learning_rate": 9.026548672566371e-07,
+      "loss": 0.0119,
+      "num_tokens": 77003658.0,
+      "reward": 0.2916666865348816,
+      "reward_std": 0.1556408554315567,
+      "rewards/accuracy_reward/mean": 0.2916666567325592,
+      "rewards/accuracy_reward/std": 0.45520758628845215,
+      "step": 49
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6011904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5114.0,
+      "completions/mean_length": 4386.875,
+      "completions/mean_terminated_length": 3281.71630859375,
+      "completions/min_length": 1135.0,
+      "completions/min_terminated_length": 1135.0,
+      "epoch": 0.1982815598149372,
+      "grad_norm": 0.14270982146263123,
+      "learning_rate": 8.982300884955752e-07,
+      "loss": 0.0208,
+      "num_tokens": 78546252.0,
+      "reward": 0.3958333432674408,
+      "reward_std": 0.15894398093223572,
+      "rewards/accuracy_reward/mean": 0.3958333432674408,
+      "rewards/accuracy_reward/std": 0.4897582530975342,
+      "step": 50
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6011904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5087.0,
+      "completions/mean_length": 4292.96142578125,
+      "completions/mean_terminated_length": 3046.231201171875,
+      "completions/min_length": 996.0,
+      "completions/min_terminated_length": 996.0,
+      "epoch": 0.20224719101123595,
+      "grad_norm": 0.12100622802972794,
+      "learning_rate": 8.938053097345132e-07,
+      "loss": 0.0128,
+      "num_tokens": 80055173.0,
+      "reward": 0.4345238208770752,
+      "reward_std": 0.1268579363822937,
+      "rewards/accuracy_reward/mean": 0.4345238208770752,
+      "rewards/accuracy_reward/std": 0.49643364548683167,
+      "step": 51
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6309523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5117.0,
+      "completions/mean_length": 4264.794921875,
+      "completions/mean_terminated_length": 2802.669189453125,
+      "completions/min_length": 344.0,
+      "completions/min_terminated_length": 344.0,
+      "epoch": 0.2062128222075347,
+      "grad_norm": 0.14565984904766083,
+      "learning_rate": 8.893805309734513e-07,
+      "loss": 0.0132,
+      "num_tokens": 81555452.0,
+      "reward": 0.4226190447807312,
+      "reward_std": 0.19077685475349426,
+      "rewards/accuracy_reward/mean": 0.4226190447807312,
+      "rewards/accuracy_reward/std": 0.49471262097358704,
+      "step": 52
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6696428571428572,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5107.0,
+      "completions/mean_length": 4495.76513671875,
+      "completions/mean_terminated_length": 3230.423583984375,
+      "completions/min_length": 834.0,
+      "completions/min_terminated_length": 834.0,
+      "epoch": 0.21017845340383345,
+      "grad_norm": 0.13897976279258728,
+      "learning_rate": 8.849557522123894e-07,
+      "loss": 0.0275,
+      "num_tokens": 83133445.0,
+      "reward": 0.324404776096344,
+      "reward_std": 0.15645335614681244,
+      "rewards/accuracy_reward/mean": 0.324404776096344,
+      "rewards/accuracy_reward/std": 0.4688502252101898,
+      "step": 53
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5535714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5110.0,
+      "completions/mean_length": 4314.41357421875,
+      "completions/mean_terminated_length": 3315.48681640625,
+      "completions/min_length": 1070.0,
+      "completions/min_terminated_length": 1070.0,
+      "epoch": 0.21414408460013218,
+      "grad_norm": 0.17090049386024475,
+      "learning_rate": 8.805309734513274e-07,
+      "loss": 0.0261,
+      "num_tokens": 84645452.0,
+      "reward": 0.4375,
+      "reward_std": 0.22074994444847107,
+      "rewards/accuracy_reward/mean": 0.4375,
+      "rewards/accuracy_reward/std": 0.4968182444572449,
+      "step": 54
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6130952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5099.0,
+      "completions/mean_length": 4196.64306640625,
+      "completions/mean_terminated_length": 2733.476806640625,
+      "completions/min_length": 555.0,
+      "completions/min_terminated_length": 555.0,
+      "epoch": 0.21810971579643093,
+      "grad_norm": 0.10867369920015335,
+      "learning_rate": 8.761061946902655e-07,
+      "loss": 0.0086,
+      "num_tokens": 86127560.0,
+      "reward": 0.380952388048172,
+      "reward_std": 0.11327183246612549,
+      "rewards/accuracy_reward/mean": 0.380952388048172,
+      "rewards/accuracy_reward/std": 0.48634520173072815,
+      "step": 55
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6488095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5092.0,
+      "completions/mean_length": 4437.10107421875,
+      "completions/mean_terminated_length": 3175.474609375,
+      "completions/min_length": 1082.0,
+      "completions/min_terminated_length": 1082.0,
+      "epoch": 0.22207534699272968,
+      "grad_norm": 0.16882988810539246,
+      "learning_rate": 8.716814159292035e-07,
+      "loss": 0.0308,
+      "num_tokens": 87688512.0,
+      "reward": 0.386904776096344,
+      "reward_std": 0.23202714323997498,
+      "rewards/accuracy_reward/mean": 0.386904776096344,
+      "rewards/accuracy_reward/std": 0.48776793479919434,
+      "step": 56
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5684523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5064.0,
+      "completions/mean_length": 4294.3603515625,
+      "completions/mean_terminated_length": 3206.793212890625,
+      "completions/min_length": 1172.0,
+      "completions/min_terminated_length": 1172.0,
+      "epoch": 0.2260409781890284,
+      "grad_norm": 0.1328190416097641,
+      "learning_rate": 8.672566371681415e-07,
+      "loss": 0.0188,
+      "num_tokens": 89198407.0,
+      "reward": 0.455357164144516,
+      "reward_std": 0.15432612597942352,
+      "rewards/accuracy_reward/mean": 0.4553571343421936,
+      "rewards/accuracy_reward/std": 0.49874576926231384,
+      "step": 57
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5386904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5113.0,
+      "completions/mean_length": 4218.33056640625,
+      "completions/mean_terminated_length": 3165.412841796875,
+      "completions/min_length": 560.0,
+      "completions/min_terminated_length": 560.0,
+      "epoch": 0.23000660938532716,
+      "grad_norm": 0.1562289297580719,
+      "learning_rate": 8.628318584070797e-07,
+      "loss": 0.0262,
+      "num_tokens": 90682978.0,
+      "reward": 0.4315476417541504,
+      "reward_std": 0.18772687017917633,
+      "rewards/accuracy_reward/mean": 0.431547611951828,
+      "rewards/accuracy_reward/std": 0.4960308074951172,
+      "step": 58
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6190476190476191,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5069.0,
+      "completions/mean_length": 4421.03857421875,
+      "completions/mean_terminated_length": 3285.2265625,
+      "completions/min_length": 596.0,
+      "completions/min_terminated_length": 596.0,
+      "epoch": 0.2339722405816259,
+      "grad_norm": 0.148189514875412,
+      "learning_rate": 8.584070796460177e-07,
+      "loss": 0.0319,
+      "num_tokens": 92234375.0,
+      "reward": 0.392857164144516,
+      "reward_std": 0.20810091495513916,
+      "rewards/accuracy_reward/mean": 0.3928571343421936,
+      "rewards/accuracy_reward/std": 0.4891138970851898,
+      "step": 59
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6160714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5103.0,
+      "completions/mean_length": 4161.0,
+      "completions/mean_terminated_length": 2622.139404296875,
+      "completions/min_length": 415.0,
+      "completions/min_terminated_length": 415.0,
+      "epoch": 0.23793787177792466,
+      "grad_norm": 0.14123181998729706,
+      "learning_rate": 8.539823008849557e-07,
+      "loss": 0.0407,
+      "num_tokens": 93702269.0,
+      "reward": 0.3720238208770752,
+      "reward_std": 0.18137377500534058,
+      "rewards/accuracy_reward/mean": 0.3720238208770752,
+      "rewards/accuracy_reward/std": 0.48406559228897095,
+      "step": 60
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6011904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5076.0,
+      "completions/mean_length": 4358.625,
+      "completions/mean_terminated_length": 3210.880615234375,
+      "completions/min_length": 658.0,
+      "completions/min_terminated_length": 658.0,
+      "epoch": 0.2419035029742234,
+      "grad_norm": 0.14963048696517944,
+      "learning_rate": 8.495575221238938e-07,
+      "loss": 0.0024,
+      "num_tokens": 95235605.0,
+      "reward": 0.3571428954601288,
+      "reward_std": 0.1869143843650818,
+      "rewards/accuracy_reward/mean": 0.3571428656578064,
+      "rewards/accuracy_reward/std": 0.47987204790115356,
+      "step": 61
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.4732142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5119.0,
+      "completions/mean_length": 4099.69970703125,
+      "completions/mean_terminated_length": 3183.158203125,
+      "completions/min_length": 989.0,
+      "completions/min_terminated_length": 989.0,
+      "epoch": 0.24586913417052214,
+      "grad_norm": 0.16955052316188812,
+      "learning_rate": 8.451327433628318e-07,
+      "loss": 0.0343,
+      "num_tokens": 96683388.0,
+      "reward": 0.5357142686843872,
+      "reward_std": 0.19290408492088318,
+      "rewards/accuracy_reward/mean": 0.5357142686843872,
+      "rewards/accuracy_reward/std": 0.4994666576385498,
+      "step": 62
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5565476190476191,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5111.0,
+      "completions/mean_length": 4184.46435546875,
+      "completions/mean_terminated_length": 3010.335693359375,
+      "completions/min_length": 621.0,
+      "completions/min_terminated_length": 621.0,
+      "epoch": 0.2498347653668209,
+      "grad_norm": 0.14555992186069489,
+      "learning_rate": 8.407079646017698e-07,
+      "loss": 0.0175,
+      "num_tokens": 98153430.0,
+      "reward": 0.4672619104385376,
+      "reward_std": 0.1731465756893158,
+      "rewards/accuracy_reward/mean": 0.4672619104385376,
+      "rewards/accuracy_reward/std": 0.49967116117477417,
+      "step": 63
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5625,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5046.0,
+      "completions/mean_length": 4144.9287109375,
+      "completions/mean_terminated_length": 2891.26513671875,
+      "completions/min_length": 653.0,
+      "completions/min_terminated_length": 653.0,
+      "epoch": 0.25380039656311965,
+      "grad_norm": 0.15088406205177307,
+      "learning_rate": 8.362831858407079e-07,
+      "loss": 0.0261,
+      "num_tokens": 99614730.0,
+      "reward": 0.455357164144516,
+      "reward_std": 0.19694827497005463,
+      "rewards/accuracy_reward/mean": 0.4553571343421936,
+      "rewards/accuracy_reward/std": 0.49874576926231384,
+      "step": 64
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5238095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5110.0,
+      "completions/mean_length": 4130.64892578125,
+      "completions/mean_terminated_length": 3042.362548828125,
+      "completions/min_length": 564.0,
+      "completions/min_terminated_length": 564.0,
+      "epoch": 0.25776602775941837,
+      "grad_norm": 0.16884246468544006,
+      "learning_rate": 8.318584070796459e-07,
+      "loss": 0.0273,
+      "num_tokens": 101069588.0,
+      "reward": 0.4732142984867096,
+      "reward_std": 0.2261088490486145,
+      "rewards/accuracy_reward/mean": 0.4732142984867096,
+      "rewards/accuracy_reward/std": 0.5000266432762146,
+      "step": 65
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5833333333333333,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5115.0,
+      "completions/mean_length": 4317.08935546875,
+      "completions/mean_terminated_length": 3193.014404296875,
+      "completions/min_length": 826.0,
+      "completions/min_terminated_length": 826.0,
+      "epoch": 0.2617316589557171,
+      "grad_norm": 0.14746442437171936,
+      "learning_rate": 8.274336283185839e-07,
+      "loss": 0.0208,
+      "num_tokens": 102587822.0,
+      "reward": 0.4583333432674408,
+      "reward_std": 0.17906485497951508,
+      "rewards/accuracy_reward/mean": 0.4583333432674408,
+      "rewards/accuracy_reward/std": 0.49900397658348083,
+      "step": 66
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5982142857142857,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5097.0,
+      "completions/mean_length": 4235.64306640625,
+      "completions/mean_terminated_length": 2918.933349609375,
+      "completions/min_length": 659.0,
+      "completions/min_terminated_length": 659.0,
+      "epoch": 0.2656972901520159,
+      "grad_norm": 0.13867241144180298,
+      "learning_rate": 8.230088495575221e-07,
+      "loss": 0.0153,
+      "num_tokens": 104079980.0,
+      "reward": 0.392857164144516,
+      "reward_std": 0.1869143843650818,
+      "rewards/accuracy_reward/mean": 0.3928571343421936,
+      "rewards/accuracy_reward/std": 0.4891138970851898,
+      "step": 67
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5386904761904762,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5087.0,
+      "completions/mean_length": 4176.443359375,
+      "completions/mean_terminated_length": 3074.61279296875,
+      "completions/min_length": 661.0,
+      "completions/min_terminated_length": 661.0,
+      "epoch": 0.2696629213483146,
+      "grad_norm": 0.17491590976715088,
+      "learning_rate": 8.185840707964602e-07,
+      "loss": 0.0416,
+      "num_tokens": 105553423.0,
+      "reward": 0.4464285671710968,
+      "reward_std": 0.2120879590511322,
+      "rewards/accuracy_reward/mean": 0.4464285671710968,
+      "rewards/accuracy_reward/std": 0.4978632628917694,
+      "step": 68
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5505952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4295.85107421875,
+      "completions/mean_terminated_length": 3286.132568359375,
+      "completions/min_length": 1242.0,
+      "completions/min_terminated_length": 1242.0,
+      "epoch": 0.2736285525446133,
+      "grad_norm": 0.14897483587265015,
+      "learning_rate": 8.141592920353983e-07,
+      "loss": 0.0193,
+      "num_tokens": 107066105.0,
+      "reward": 0.4672619104385376,
+      "reward_std": 0.20118845999240875,
+      "rewards/accuracy_reward/mean": 0.4672619104385376,
+      "rewards/accuracy_reward/std": 0.49967116117477417,
+      "step": 69
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5327380952380952,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5031.0,
+      "completions/mean_length": 4196.72314453125,
+      "completions/mean_terminated_length": 3144.070068359375,
+      "completions/min_length": 1143.0,
+      "completions/min_terminated_length": 1143.0,
+      "epoch": 0.2775941837409121,
+      "grad_norm": 0.129190132021904,
+      "learning_rate": 8.097345132743363e-07,
+      "loss": 0.0133,
+      "num_tokens": 108538820.0,
+      "reward": 0.4375,
+      "reward_std": 0.1522131860256195,
+      "rewards/accuracy_reward/mean": 0.4375,
+      "rewards/accuracy_reward/std": 0.4968182444572449,
+      "step": 70
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5595238095238095,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5093.0,
+      "completions/mean_length": 4183.77392578125,
+      "completions/mean_terminated_length": 2994.513671875,
+      "completions/min_length": 904.0,
+      "completions/min_terminated_length": 904.0,
+      "epoch": 0.28155981493721083,
+      "grad_norm": 0.12832552194595337,
+      "learning_rate": 8.053097345132743e-07,
+      "loss": 0.0138,
+      "num_tokens": 110013352.0,
+      "reward": 0.4107142984867096,
+      "reward_std": 0.13856995105743408,
+      "rewards/accuracy_reward/mean": 0.4107142984867096,
+      "rewards/accuracy_reward/std": 0.49269720911979675,
+      "step": 71
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6309523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5076.0,
+      "completions/mean_length": 4328.02392578125,
+      "completions/mean_terminated_length": 2974.0,
+      "completions/min_length": 915.0,
+      "completions/min_terminated_length": 915.0,
+      "epoch": 0.2855254461335096,
+      "grad_norm": 0.13086427748203278,
+      "learning_rate": 8.008849557522124e-07,
+      "loss": 0.019,
+      "num_tokens": 111535848.0,
+      "reward": 0.3392857313156128,
+      "reward_std": 0.14741365611553192,
+      "rewards/accuracy_reward/mean": 0.3392857015132904,
+      "rewards/accuracy_reward/std": 0.47417306900024414,
+      "step": 72
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6130952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5081.0,
+      "completions/mean_length": 4410.1220703125,
+      "completions/mean_terminated_length": 3285.238525390625,
+      "completions/min_length": 1054.0,
+      "completions/min_terminated_length": 1054.0,
+      "epoch": 0.28949107732980833,
+      "grad_norm": 0.15406593680381775,
+      "learning_rate": 7.964601769911504e-07,
+      "loss": 0.0249,
+      "num_tokens": 113081927.0,
+      "reward": 0.4017857313156128,
+      "reward_std": 0.15894395112991333,
+      "rewards/accuracy_reward/mean": 0.4017857015132904,
+      "rewards/accuracy_reward/std": 0.4909902513027191,
+      "step": 73
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6071428571428572,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5110.0,
+      "completions/mean_length": 4256.94970703125,
+      "completions/mean_terminated_length": 2923.14404296875,
+      "completions/min_length": 415.0,
+      "completions/min_terminated_length": 415.0,
+      "epoch": 0.29345670852610706,
+      "grad_norm": 0.1666187047958374,
+      "learning_rate": 7.920353982300884e-07,
+      "loss": 0.0374,
+      "num_tokens": 114581910.0,
+      "reward": 0.392857164144516,
+      "reward_std": 0.18193314969539642,
+      "rewards/accuracy_reward/mean": 0.3928571343421936,
+      "rewards/accuracy_reward/std": 0.4891138970851898,
+      "step": 74
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5892857142857143,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5115.0,
+      "completions/mean_length": 4196.20556640625,
+      "completions/mean_terminated_length": 2870.760986328125,
+      "completions/min_length": 660.0,
+      "completions/min_terminated_length": 660.0,
+      "epoch": 0.29742233972240584,
+      "grad_norm": 0.2190464437007904,
+      "learning_rate": 7.876106194690266e-07,
+      "loss": 0.0129,
+      "num_tokens": 116061381.0,
+      "reward": 0.3898809552192688,
+      "reward_std": 0.22361823916435242,
+      "rewards/accuracy_reward/mean": 0.3898809552192688,
+      "rewards/accuracy_reward/std": 0.4884504973888397,
+      "step": 75
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5863095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5119.0,
+      "completions/mean_length": 4346.5478515625,
+      "completions/mean_terminated_length": 3250.35986328125,
+      "completions/min_length": 957.0,
+      "completions/min_terminated_length": 957.0,
+      "epoch": 0.30138797091870456,
+      "grad_norm": 0.16212566196918488,
+      "learning_rate": 7.831858407079646e-07,
+      "loss": 0.0288,
+      "num_tokens": 117593053.0,
+      "reward": 0.3839285671710968,
+      "reward_std": 0.2445516437292099,
+      "rewards/accuracy_reward/mean": 0.3839285671710968,
+      "rewards/accuracy_reward/std": 0.4870661795139313,
+      "step": 76
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.47916666666666663,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5112.0,
+      "completions/mean_length": 4076.761962890625,
+      "completions/mean_terminated_length": 3116.98291015625,
+      "completions/min_length": 550.0,
+      "completions/min_terminated_length": 550.0,
+      "epoch": 0.3053536021150033,
+      "grad_norm": 0.1575162708759308,
+      "learning_rate": 7.787610619469026e-07,
+      "loss": 0.0344,
+      "num_tokens": 119032235.0,
+      "reward": 0.523809552192688,
+      "reward_std": 0.2109692096710205,
+      "rewards/accuracy_reward/mean": 0.523809552192688,
+      "rewards/accuracy_reward/std": 0.5001776814460754,
+      "step": 77
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6041666666666667,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5111.0,
+      "completions/mean_length": 4402.21728515625,
+      "completions/mean_terminated_length": 3306.654296875,
+      "completions/min_length": 1262.0,
+      "completions/min_terminated_length": 1262.0,
+      "epoch": 0.30931923331130207,
+      "grad_norm": 0.1132589727640152,
+      "learning_rate": 7.743362831858407e-07,
+      "loss": 0.0181,
+      "num_tokens": 120581874.0,
+      "reward": 0.3452381193637848,
+      "reward_std": 0.12536151707172394,
+      "rewards/accuracy_reward/mean": 0.3452380895614624,
+      "rewards/accuracy_reward/std": 0.47615495324134827,
+      "step": 78
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5091.0,
+      "completions/mean_length": 4005.196533203125,
+      "completions/mean_terminated_length": 2890.392822265625,
+      "completions/min_length": 665.0,
+      "completions/min_terminated_length": 665.0,
+      "epoch": 0.3132848645076008,
+      "grad_norm": 0.15795782208442688,
+      "learning_rate": 7.699115044247787e-07,
+      "loss": 0.0294,
+      "num_tokens": 121993464.0,
+      "reward": 0.4910714328289032,
+      "reward_std": 0.20903794467449188,
+      "rewards/accuracy_reward/mean": 0.4910714328289032,
+      "rewards/accuracy_reward/std": 0.5006658434867859,
+      "step": 79
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.49404761904761907,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5079.0,
+      "completions/mean_length": 4157.0029296875,
+      "completions/mean_terminated_length": 3216.664794921875,
+      "completions/min_length": 1245.0,
+      "completions/min_terminated_length": 1245.0,
+      "epoch": 0.3172504957038995,
+      "grad_norm": 0.14931175112724304,
+      "learning_rate": 7.654867256637167e-07,
+      "loss": 0.0226,
+      "num_tokens": 123457405.0,
+      "reward": 0.5148809552192688,
+      "reward_std": 0.17713361978530884,
+      "rewards/accuracy_reward/mean": 0.5148809552192688,
+      "rewards/accuracy_reward/std": 0.5005238652229309,
+      "step": 80
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5119047619047619,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5093.0,
+      "completions/mean_length": 4194.931640625,
+      "completions/mean_terminated_length": 3224.73779296875,
+      "completions/min_length": 548.0,
+      "completions/min_terminated_length": 548.0,
+      "epoch": 0.3212161269001983,
+      "grad_norm": 0.1550244241952896,
+      "learning_rate": 7.610619469026548e-07,
+      "loss": 0.0346,
+      "num_tokens": 124935476.0,
+      "reward": 0.4464285671710968,
+      "reward_std": 0.17233407497406006,
+      "rewards/accuracy_reward/mean": 0.4464285671710968,
+      "rewards/accuracy_reward/std": 0.4978632628917694,
+      "step": 81
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5863095238095238,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5011.0,
+      "completions/mean_length": 4212.47900390625,
+      "completions/mean_terminated_length": 2926.28076171875,
+      "completions/min_length": 844.0,
+      "completions/min_terminated_length": 844.0,
+      "epoch": 0.325181758096497,
+      "grad_norm": 0.14946086704730988,
+      "learning_rate": 7.566371681415928e-07,
+      "loss": 0.0229,
+      "num_tokens": 126419683.0,
+      "reward": 0.4107142984867096,
+      "reward_std": 0.1857956349849701,
+      "rewards/accuracy_reward/mean": 0.4107142984867096,
+      "rewards/accuracy_reward/std": 0.49269723892211914,
+      "step": 82
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5208333333333333,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5070.0,
+      "completions/mean_length": 4200.857421875,
+      "completions/mean_terminated_length": 3201.788818359375,
+      "completions/min_length": 969.0,
+      "completions/min_terminated_length": 969.0,
+      "epoch": 0.32914738929279574,
+      "grad_norm": 0.13284137845039368,
+      "learning_rate": 7.522123893805308e-07,
+      "loss": 0.0103,
+      "num_tokens": 127899775.0,
+      "reward": 0.4166666865348816,
+      "reward_std": 0.17233408987522125,
+      "rewards/accuracy_reward/mean": 0.4166666567325592,
+      "rewards/accuracy_reward/std": 0.4937419593334198,
+      "step": 83
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5505952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5084.0,
+      "completions/mean_length": 4249.08935546875,
+      "completions/mean_terminated_length": 3182.079345703125,
+      "completions/min_length": 827.0,
+      "completions/min_terminated_length": 827.0,
+      "epoch": 0.3331130204890945,
+      "grad_norm": 0.13733071088790894,
+      "learning_rate": 7.477876106194691e-07,
+      "loss": 0.0273,
+      "num_tokens": 129401689.0,
+      "reward": 0.4702380895614624,
+      "reward_std": 0.17507781088352203,
+      "rewards/accuracy_reward/mean": 0.4702380895614624,
+      "rewards/accuracy_reward/std": 0.4998578727245331,
+      "step": 84
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.4910714285714286,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5106.0,
+      "completions/mean_length": 4004.46435546875,
+      "completions/mean_terminated_length": 2928.0703125,
+      "completions/min_length": 1096.0,
+      "completions/min_terminated_length": 1096.0,
+      "epoch": 0.33707865168539325,
+      "grad_norm": 0.15919451415538788,
+      "learning_rate": 7.433628318584071e-07,
+      "loss": 0.0298,
+      "num_tokens": 130810045.0,
+      "reward": 0.523809552192688,
+      "reward_std": 0.2436145693063736,
+      "rewards/accuracy_reward/mean": 0.523809552192688,
+      "rewards/accuracy_reward/std": 0.5001776814460754,
+      "step": 85
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5029761904761905,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5080.0,
+      "completions/mean_length": 4173.9970703125,
+      "completions/mean_terminated_length": 3216.664794921875,
+      "completions/min_length": 967.0,
+      "completions/min_terminated_length": 967.0,
+      "epoch": 0.34104428288169203,
+      "grad_norm": 0.1184137687087059,
+      "learning_rate": 7.389380530973452e-07,
+      "loss": 0.0222,
+      "num_tokens": 132278778.0,
+      "reward": 0.4107142984867096,
+      "reward_std": 0.13432978093624115,
+      "rewards/accuracy_reward/mean": 0.4107142984867096,
+      "rewards/accuracy_reward/std": 0.49269723892211914,
+      "step": 86
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5505952380952381,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5088.0,
+      "completions/mean_length": 4156.34814453125,
+      "completions/mean_terminated_length": 2975.71533203125,
+      "completions/min_length": 756.0,
+      "completions/min_terminated_length": 756.0,
+      "epoch": 0.34500991407799075,
+      "grad_norm": 0.14675813913345337,
+      "learning_rate": 7.345132743362832e-07,
+      "loss": 0.0169,
+      "num_tokens": 133740399.0,
+      "reward": 0.4196428656578064,
+      "reward_std": 0.19171389937400818,
+      "rewards/accuracy_reward/mean": 0.4196428656578064,
+      "rewards/accuracy_reward/std": 0.4942364990711212,
+      "step": 87
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5625,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5083.0,
+      "completions/mean_length": 4250.58935546875,
+      "completions/mean_terminated_length": 3132.775390625,
+      "completions/min_length": 747.0,
+      "completions/min_terminated_length": 747.0,
+      "epoch": 0.3489755452742895,
+      "grad_norm": 0.14409150183200836,
+      "learning_rate": 7.300884955752212e-07,
+      "loss": 0.0216,
+      "num_tokens": 135235611.0,
+      "reward": 0.4375,
+      "reward_std": 0.19557639956474304,
+      "rewards/accuracy_reward/mean": 0.4375,
+      "rewards/accuracy_reward/std": 0.4968182444572449,
+      "step": 88
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5625,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5116.0,
+      "completions/mean_length": 4277.58056640625,
+      "completions/mean_terminated_length": 3194.46923828125,
+      "completions/min_length": 1027.0,
+      "completions/min_terminated_length": 1027.0,
+      "epoch": 0.35294117647058826,
+      "grad_norm": 0.1559421569108963,
+      "learning_rate": 7.256637168141593e-07,
+      "loss": 0.028,
+      "num_tokens": 136744128.0,
+      "reward": 0.4672619104385376,
+      "reward_std": 0.21165312826633453,
+      "rewards/accuracy_reward/mean": 0.4672619104385376,
+      "rewards/accuracy_reward/std": 0.49967116117477417,
+      "step": 89
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.49404761904761907,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 4992.0,
+      "completions/mean_length": 4141.375,
+      "completions/mean_terminated_length": 3185.776611328125,
+      "completions/min_length": 1130.0,
+      "completions/min_terminated_length": 1130.0,
+      "epoch": 0.356906807666887,
+      "grad_norm": 0.15738657116889954,
+      "learning_rate": 7.212389380530973e-07,
+      "loss": 0.0324,
+      "num_tokens": 138202512.0,
+      "reward": 0.523809552192688,
+      "reward_std": 0.22242805361747742,
+      "rewards/accuracy_reward/mean": 0.523809552192688,
+      "rewards/accuracy_reward/std": 0.5001776814460754,
+      "step": 90
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.4464285714285714,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5045.0,
+      "completions/mean_length": 4064.4345703125,
+      "completions/mean_terminated_length": 3213.172119140625,
+      "completions/min_length": 1205.0,
+      "completions/min_terminated_length": 1205.0,
+      "epoch": 0.3608724388631857,
+      "grad_norm": 0.14963069558143616,
+      "learning_rate": 7.168141592920353e-07,
+      "loss": 0.0308,
+      "num_tokens": 139632914.0,
+      "reward": 0.5476190447807312,
+      "reward_std": 0.19925718009471893,
+      "rewards/accuracy_reward/mean": 0.5476190447807312,
+      "rewards/accuracy_reward/std": 0.49846959114074707,
+      "step": 91
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5148809523809523,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5093.0,
+      "completions/mean_length": 4204.84228515625,
+      "completions/mean_terminated_length": 3233.539794921875,
+      "completions/min_length": 453.0,
+      "completions/min_terminated_length": 453.0,
+      "epoch": 0.3648380700594845,
+      "grad_norm": 0.1522281914949417,
+      "learning_rate": 7.123893805309734e-07,
+      "loss": 0.0193,
+      "num_tokens": 141112113.0,
+      "reward": 0.4702380895614624,
+      "reward_std": 0.19900403916835785,
+      "rewards/accuracy_reward/mean": 0.4702380895614624,
+      "rewards/accuracy_reward/std": 0.4998578727245331,
+      "step": 92
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.6041666666666667,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5067.0,
+      "completions/mean_length": 4348.01806640625,
+      "completions/mean_terminated_length": 3169.7294921875,
+      "completions/min_length": 1256.0,
+      "completions/min_terminated_length": 1256.0,
+      "epoch": 0.3688037012557832,
+      "grad_norm": 0.13845135271549225,
+      "learning_rate": 7.079646017699115e-07,
+      "loss": 0.0263,
+      "num_tokens": 142646997.0,
+      "reward": 0.3541666865348816,
+      "reward_std": 0.19956344366073608,
+      "rewards/accuracy_reward/mean": 0.3541666567325592,
+      "rewards/accuracy_reward/std": 0.4789733290672302,
+      "step": 93
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.49702380952380953,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5071.0,
+      "completions/mean_length": 3997.58642578125,
+      "completions/mean_terminated_length": 2888.45556640625,
+      "completions/min_length": 804.0,
+      "completions/min_terminated_length": 804.0,
+      "epoch": 0.37276933245208194,
+      "grad_norm": 0.18246114253997803,
+      "learning_rate": 7.035398230088495e-07,
+      "loss": 0.0314,
+      "num_tokens": 144056318.0,
+      "reward": 0.4166666865348816,
+      "reward_std": 0.257200688123703,
+      "rewards/accuracy_reward/mean": 0.4166666567325592,
+      "rewards/accuracy_reward/std": 0.4937419593334198,
+      "step": 94
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completions/clipped_ratio": 0.5059523809523809,
+      "completions/max_length": 5120.0,
+      "completions/max_terminated_length": 5114.0,
+      "completions/mean_length": 3974.881103515625,
+      "completions/mean_terminated_length": 2802.16845703125,
+      "completions/min_length": 364.0,
+      "completions/min_terminated_length": 364.0,
+      "epoch": 0.3767349636483807,
+      "grad_norm": 0.16523684561252594,
+      "learning_rate": 6.991150442477876e-07,
+      "loss": 0.0262,
+      "num_tokens": 145457896.0,
+      "reward": 0.4583333432674408,
+      "reward_std": 0.19077686965465546,
+      "rewards/accuracy_reward/mean": 0.4583333432674408,
+      "rewards/accuracy_reward/std": 0.49900397658348083,
+      "step": 95
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 252,
+  "num_input_tokens_seen": 145457896,
+  "num_train_epochs": 1,
+  "save_steps": 5,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 8,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-95/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18fe4c922e827abdfb06bcea04558c4590fdf389e4d33ba21d07d0f187bd54a5
+size 7800

checkpoint-95/zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/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 gc
+import json
+import numpy as np
+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, weights_only=False)
+
+        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 tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # 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}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] 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")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+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 = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * 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
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[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}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        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 to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=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
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    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.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    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")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+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,
+                                                          lazy_mode=True)
+
+    # 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")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_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 by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # 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)