Model save

README.md

@@ -1,5 +1,5 @@
 ---
-base_model: Qwen/Qwen2.5-Math-7B
+base_model: Qwen/Qwen2.5-7B
 library_name: transformers
 model_name: Qwen-2.5-7B-Simple-RL
 tags:
@@ -11,7 +11,7 @@ licence: license
 
 # Model Card for Qwen-2.5-7B-Simple-RL
 
-This model is a fine-tuned version of [Qwen/Qwen2.5-Math-7B](https://huggingface.co/Qwen/Qwen2.5-Math-7B).
+This model is a fine-tuned version of [Qwen/Qwen2.5-7B](https://huggingface.co/Qwen/Qwen2.5-7B).
 It has been trained using [TRL](https://github.com/huggingface/trl).
 
 ## Quick start
@@ -27,7 +27,7 @@ print(output["generated_text"])
 
 ## Training procedure
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jzhang-x-01/huggingface/runs/l1ih86fe) 
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jzhang-x-01/huggingface/runs/38wymkvl) 
 
 
 This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).

all_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.002154490305616204,
-    "train_runtime": 8228.6792,
+    "train_loss": 0.0010652160639474098,
+    "train_runtime": 7821.2028,
     "train_samples": 7500,
-    "train_samples_per_second": 0.911,
+    "train_samples_per_second": 0.959,
     "train_steps_per_second": 0.007
 }

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "Qwen/Qwen2.5-Math-7B",
+  "_name_or_path": "Qwen/Qwen2.5-7B",
   "architectures": [
     "Qwen2ForCausalLM"
   ],
@@ -10,7 +10,7 @@
   "hidden_size": 3584,
   "initializer_range": 0.02,
   "intermediate_size": 18944,
-  "max_position_embeddings": 4096,
+  "max_position_embeddings": 131072,
   "max_window_layers": 28,
   "model_type": "qwen2",
   "num_attention_heads": 28,
@@ -18,8 +18,8 @@
   "num_key_value_heads": 4,
   "rms_norm_eps": 1e-06,
   "rope_scaling": null,
-  "rope_theta": 10000,
-  "sliding_window": 4096,
+  "rope_theta": 1000000.0,
+  "sliding_window": 131072,
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.49.0",

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da4320c5ba6d132b13e0697de200379e6696bb453e297dd29a845b6cf0a5ad11
+oid sha256:971e4ddf8f15c63088fd84ae0519d21cefa81db02d1139bf166e145d3aa25c60
 size 4877660776

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb5180786ab5e7eb5c36cbfc486ba5d1a87942134f42b637939668da6e4b7b68
+oid sha256:80cd7a84698f060a90ab0265b51e4cf1a70c426458465b9eac5a921a40a7a8d7
 size 4932751008

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c5df4aabc965da16054c18701d906c2afcfea880192d0a5ac4c9759ba58ae05
+oid sha256:e0a4d9f859f988abe01b85a737bbadad6b40346139099ffd99afc810b99388c1
 size 4330865200

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aac5bbf08545411482d72b15e3233be68a3e299b0f204d9e15b514f68fa374df
+oid sha256:8ad0dc1af500970bf866d1f09e280121d246b4c2e1cd22a99357bfaaa1ecd6c6
 size 1089994880

tokenizer_config.json

@@ -195,7 +195,7 @@
     "<|video_pad|>"
   ],
   "bos_token": null,
-  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'Please reason step by step, and put your final answer within \\\\boxed{}.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nPlease reason step by step, and put your final answer within \\\\boxed{}.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\n# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n",
   "clean_up_tokenization_spaces": false,
   "eos_token": "<|endoftext|>",
   "errors": "replace",

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.002154490305616204,
-    "train_runtime": 8228.6792,
+    "train_loss": 0.0010652160639474098,
+    "train_runtime": 7821.2028,
     "train_samples": 7500,
-    "train_samples_per_second": 0.911,
+    "train_samples_per_second": 0.959,
     "train_steps_per_second": 0.007
 }

trainer_state.json

@@ -9,174 +9,174 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "completion_length": 606.5457763671875,
+      "completion_length": 473.94757080078125,
       "epoch": 0.017057569296375266,
-      "grad_norm": 5.351304250136567,
+      "grad_norm": 4.064376038455483,
       "kl": 0.0,
       "learning_rate": 5e-07,
       "loss": 0.0,
-      "reward": 0.6216518133878708,
-      "reward_std": 0.3478058036416769,
-      "rewards/accuracy_reward": 0.6216518133878708,
+      "reward": 0.1729910783469677,
+      "reward_std": 0.27430381812155247,
+      "rewards/accuracy_reward": 0.1729910783469677,
       "rewards/format_reward": 0.0,
       "step": 1
     },
     {
-      "completion_length": 607.6144218444824,
+      "completion_length": 485.79745292663574,
       "epoch": 0.08528784648187633,
-      "grad_norm": 11.35565102634353,
-      "kl": 0.00033593177795410156,
+      "grad_norm": 1.1351739582457168,
+      "kl": 0.0002506077289581299,
       "learning_rate": 2.5e-06,
       "loss": 0.0,
-      "reward": 0.6026785979047418,
-      "reward_std": 0.3518537702038884,
-      "rewards/accuracy_reward": 0.6026785979047418,
-      "rewards/format_reward": 0.0,
+      "reward": 0.13504464970901608,
+      "reward_std": 0.21868262230418622,
+      "rewards/accuracy_reward": 0.1347656319849193,
+      "rewards/format_reward": 0.00027901786961592734,
       "step": 5
     },
     {
-      "completion_length": 622.7357368469238,
+      "completion_length": 568.8616348266602,
       "epoch": 0.17057569296375266,
-      "grad_norm": 1.9080536354821958,
-      "kl": 0.004211950302124024,
+      "grad_norm": 0.5575129470407404,
+      "kl": 0.015078353881835937,
       "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0002,
-      "reward": 0.6660714566707611,
-      "reward_std": 0.31057145930826663,
-      "rewards/accuracy_reward": 0.6660714566707611,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0006,
+      "reward": 0.3587053745985031,
+      "reward_std": 0.3362843289971352,
+      "rewards/accuracy_reward": 0.35848215967416763,
+      "rewards/format_reward": 0.00022321429569274187,
       "step": 10
     },
     {
-      "completion_length": 631.8491333007812,
+      "completion_length": 584.8319480895996,
       "epoch": 0.255863539445629,
-      "grad_norm": 0.20168403792141756,
-      "kl": 0.007109832763671875,
+      "grad_norm": 0.2277859906277476,
+      "kl": 0.02491455078125,
       "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.0003,
-      "reward": 0.7424107462167739,
-      "reward_std": 0.2543599892407656,
-      "rewards/accuracy_reward": 0.7424107462167739,
+      "loss": 0.001,
+      "reward": 0.638169676065445,
+      "reward_std": 0.24866362921893598,
+      "rewards/accuracy_reward": 0.638169676065445,
       "rewards/format_reward": 0.0,
       "step": 15
     },
     {
-      "completion_length": 622.2803810119628,
+      "completion_length": 535.3618560791016,
       "epoch": 0.3411513859275053,
-      "grad_norm": 0.495428185088676,
-      "kl": 0.02277984619140625,
+      "grad_norm": 0.23887369640814476,
+      "kl": 0.0245208740234375,
       "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.0009,
-      "reward": 0.7529018178582192,
-      "reward_std": 0.23195756375789642,
-      "rewards/accuracy_reward": 0.7529018178582192,
-      "rewards/format_reward": 0.0,
+      "loss": 0.001,
+      "reward": 0.6774553880095482,
+      "reward_std": 0.2348104739561677,
+      "rewards/accuracy_reward": 0.6767857447266579,
+      "rewards/format_reward": 0.0006696428870782256,
       "step": 20
     },
     {
-      "completion_length": 629.6663253784179,
+      "completion_length": 542.0183288574219,
       "epoch": 0.42643923240938164,
-      "grad_norm": 0.2077076072198444,
-      "kl": 0.0070018768310546875,
+      "grad_norm": 0.3014825446189458,
+      "kl": 0.02587432861328125,
       "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.0003,
-      "reward": 0.7497768178582191,
-      "reward_std": 0.22464404683560132,
-      "rewards/accuracy_reward": 0.7497768178582191,
-      "rewards/format_reward": 0.0,
+      "loss": 0.001,
+      "reward": 0.6584821730852127,
+      "reward_std": 0.22855298295617105,
+      "rewards/accuracy_reward": 0.657589316368103,
+      "rewards/format_reward": 0.0008928571827709675,
       "step": 25
     },
     {
-      "completion_length": 628.7462333679199,
+      "completion_length": 536.8770370483398,
       "epoch": 0.511727078891258,
-      "grad_norm": 223.6982144990451,
-      "kl": 0.3876213073730469,
+      "grad_norm": 0.21529686209177487,
+      "kl": 0.0219482421875,
       "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0155,
-      "reward": 0.7379464656114578,
-      "reward_std": 0.21265463214367628,
-      "rewards/accuracy_reward": 0.7379464656114578,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0009,
+      "reward": 0.6495536014437675,
+      "reward_std": 0.2183901211246848,
+      "rewards/accuracy_reward": 0.6488839566707612,
+      "rewards/format_reward": 0.0006696428870782256,
       "step": 30
     },
     {
-      "completion_length": 614.8966758728027,
+      "completion_length": 528.6895355224609,
       "epoch": 0.5970149253731343,
-      "grad_norm": 2.3665634575263748,
-      "kl": 0.005752182006835938,
+      "grad_norm": 0.16077481890595569,
+      "kl": 0.0267181396484375,
       "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0002,
-      "reward": 0.7314732424914837,
-      "reward_std": 0.19922155924141408,
-      "rewards/accuracy_reward": 0.7314732424914837,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0011,
+      "reward": 0.6448661029338837,
+      "reward_std": 0.2145981529727578,
+      "rewards/accuracy_reward": 0.6441964589059352,
+      "rewards/format_reward": 0.0006696428870782256,
       "step": 35
     },
     {
-      "completion_length": 603.5955581665039,
+      "completion_length": 514.0154296875,
       "epoch": 0.6823027718550106,
-      "grad_norm": 0.41861750967507205,
-      "kl": 0.007678604125976563,
+      "grad_norm": 0.8368130250058291,
+      "kl": 0.02160186767578125,
       "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0003,
-      "reward": 0.7553571775555611,
-      "reward_std": 0.20080968737602234,
-      "rewards/accuracy_reward": 0.7553571775555611,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0009,
+      "reward": 0.6649553939700127,
+      "reward_std": 0.20347560551017524,
+      "rewards/accuracy_reward": 0.6647321790456772,
+      "rewards/format_reward": 0.00022321429569274187,
       "step": 40
     },
     {
-      "completion_length": 619.6895401000977,
+      "completion_length": 528.3629722595215,
       "epoch": 0.767590618336887,
-      "grad_norm": 0.5526088012366769,
-      "kl": 0.007487106323242188,
+      "grad_norm": 0.3144943865256562,
+      "kl": 0.0201324462890625,
       "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0003,
-      "reward": 0.725000037252903,
-      "reward_std": 0.23073445297777653,
-      "rewards/accuracy_reward": 0.725000037252903,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0008,
+      "reward": 0.6553571708500385,
+      "reward_std": 0.21330095445737243,
+      "rewards/accuracy_reward": 0.6542410992085934,
+      "rewards/format_reward": 0.0011160714784637094,
       "step": 45
     },
     {
-      "completion_length": 613.0727882385254,
+      "completion_length": 526.5335037231446,
       "epoch": 0.8528784648187633,
-      "grad_norm": 1.1048247702884233,
-      "kl": 0.00660247802734375,
+      "grad_norm": 2.72821016556171,
+      "kl": 0.027325439453125,
       "learning_rate": 1.718159615201853e-07,
-      "loss": 0.0003,
-      "reward": 0.7332589656114579,
-      "reward_std": 0.19824975840747355,
-      "rewards/accuracy_reward": 0.7332589656114579,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0011,
+      "reward": 0.6439732402563095,
+      "reward_std": 0.22680639270693065,
+      "rewards/accuracy_reward": 0.6437500268220901,
+      "rewards/format_reward": 0.00022321429569274187,
       "step": 50
     },
     {
-      "completion_length": 608.9152099609375,
+      "completion_length": 523.5705612182617,
       "epoch": 0.9381663113006397,
-      "grad_norm": 0.688326952525324,
-      "kl": 0.006847000122070313,
+      "grad_norm": 0.41062600235384844,
+      "kl": 0.0195343017578125,
       "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0003,
-      "reward": 0.7656250342726707,
-      "reward_std": 0.21631054822355508,
-      "rewards/accuracy_reward": 0.7656250342726707,
-      "rewards/format_reward": 0.0,
+      "loss": 0.0008,
+      "reward": 0.6685268178582191,
+      "reward_std": 0.23179882913827896,
+      "rewards/accuracy_reward": 0.6676339596509934,
+      "rewards/format_reward": 0.0008928571827709675,
       "step": 55
     },
     {
-      "completion_length": 603.1421381632487,
+      "completion_length": 517.845261891683,
       "epoch": 0.9893390191897654,
-      "kl": 0.0070037841796875,
-      "reward": 0.751860149204731,
-      "reward_std": 0.20624662407984337,
-      "rewards/accuracy_reward": 0.751860149204731,
-      "rewards/format_reward": 0.0,
+      "kl": 0.021814982096354168,
+      "reward": 0.6592262188593546,
+      "reward_std": 0.20016193420936665,
+      "rewards/accuracy_reward": 0.6588541939854622,
+      "rewards/format_reward": 0.0003720238261545698,
       "step": 58,
       "total_flos": 0.0,
-      "train_loss": 0.002154490305616204,
-      "train_runtime": 8228.6792,
-      "train_samples_per_second": 0.911,
+      "train_loss": 0.0010652160639474098,
+      "train_runtime": 7821.2028,
+      "train_samples_per_second": 0.959,
       "train_steps_per_second": 0.007
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:095c9ce79540846bb0717018dbf2b9b1f9126d535c416a479f2bd2e2f09039f3
+oid sha256:e9c97f93968e70f80d79d3367db73181a5a13edc4aad2638a46a6504eb5664cf
 size 7928