Model save

README.md

@@ -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/gongc1995-city-university-of-hong-kong/huggingface/runs/tw88l9f9) 
+[<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/gongc1995-city-university-of-hong-kong/huggingface/runs/o4ou28au) 
 
 
 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": 0.04267983935002623,
-    "train_runtime": 7907.107,
+    "train_loss": 0.04428564981910689,
+    "train_runtime": 7921.619,
     "train_samples": 7500,
-    "train_samples_per_second": 0.949,
+    "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.007
 }

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8da2c323738051ed0b639c6c6c3cabc3a7186aa2833f3340294351c8e4a4abeb
+oid sha256:8cb74df1f51c7b5cf6f92df96b571f86a5d24c12666a82626c80881f07f102a7
 size 4877660776

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9be10ba534848b1d7712651d761f47d9857ce2deafdf659b3e64780412dab1ae
+oid sha256:1562089ab4298c8e3ab1d9b5112aa468472695bc876edbaa0d3da760884391b4
 size 4932751008

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ab0d56ac8ae6378b286f76fdbd6b9d2c2194938d9d386ce97316567d20706da4
+oid sha256:97ddbf5e2431a28323e4f5261064c74d112b605caf7bc43ae5eb36599756d68b
 size 4330865200

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:95098f0abc1df599d0b888c5de8f72d3da1a83efe178a52514487eefbaea1fbb
+oid sha256:012dcf0ed4f1423109622326588a9ee01e2288cffd2ee73564633813291ea143
 size 1089994880

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.04267983935002623,
-    "train_runtime": 7907.107,
+    "train_loss": 0.04428564981910689,
+    "train_runtime": 7921.619,
     "train_samples": 7500,
-    "train_samples_per_second": 0.949,
+    "train_samples_per_second": 0.947,
     "train_steps_per_second": 0.007
 }

trainer_state.json

@@ -10,186 +10,186 @@
   "log_history": [
     {
       "clip_ratio": 0.0,
-      "completion_length": 611.475471496582,
+      "completion_length": 607.8761520385742,
       "epoch": 0.017057569296375266,
-      "grad_norm": 0.3608338534832001,
+      "grad_norm": 0.21163204312324524,
       "kl": 0.0,
       "learning_rate": 5e-07,
-      "loss": 0.0485,
+      "loss": 0.0347,
       "reward": 0.7064732536673546,
-      "reward_std": 0.29637950100004673,
-      "rewards/accuracy_reward": 0.6975446715950966,
-      "rewards/format_reward": 0.008928571944124997,
+      "reward_std": 0.2907280754297972,
+      "rewards/accuracy_reward": 0.6964286044239998,
+      "rewards/format_reward": 0.010044643422588706,
       "step": 1
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 603.289924621582,
+      "completion_length": 598.3354082107544,
       "epoch": 0.08528784648187633,
-      "grad_norm": 0.24193009734153748,
-      "kl": 0.00037926435470581055,
+      "grad_norm": 0.22014343738555908,
+      "kl": 0.00016763806343078613,
       "learning_rate": 2.5e-06,
-      "loss": 0.0419,
-      "reward": 0.6629464607685804,
-      "reward_std": 0.3164387275464833,
-      "rewards/accuracy_reward": 0.6540178805589676,
-      "rewards/format_reward": 0.008928571856813505,
+      "loss": 0.0443,
+      "reward": 0.6668527107685804,
+      "reward_std": 0.30428835609927773,
+      "rewards/accuracy_reward": 0.6609933376312256,
+      "rewards/format_reward": 0.005859375291038305,
       "step": 5
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 602.1417655944824,
+      "completion_length": 602.3207862854003,
       "epoch": 0.17057569296375266,
-      "grad_norm": 0.21082398295402527,
-      "kl": 0.0039100170135498045,
+      "grad_norm": 0.1867409348487854,
+      "kl": 0.006086993217468262,
       "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0591,
-      "reward": 0.6986607477068901,
-      "reward_std": 0.29480395577847957,
-      "rewards/accuracy_reward": 0.6926339596509934,
-      "rewards/format_reward": 0.006026786076836288,
+      "loss": 0.0561,
+      "reward": 0.6991071760654449,
+      "reward_std": 0.28411166220903394,
+      "rewards/accuracy_reward": 0.6957589611411095,
+      "rewards/format_reward": 0.0033482144586741923,
       "step": 10
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 597.1254707336426,
+      "completion_length": 591.975025177002,
       "epoch": 0.255863539445629,
-      "grad_norm": 0.11069315671920776,
-      "kl": 0.010307502746582032,
+      "grad_norm": 0.272446870803833,
+      "kl": 0.006195259094238281,
       "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.0681,
-      "reward": 0.7524553880095481,
-      "reward_std": 0.22803126517683267,
-      "rewards/accuracy_reward": 0.7508928924798965,
-      "rewards/format_reward": 0.0015625000698491931,
+      "loss": 0.0687,
+      "reward": 0.7625000327825546,
+      "reward_std": 0.21445324290543794,
+      "rewards/accuracy_reward": 0.7611607477068901,
+      "rewards/format_reward": 0.0013392857741564511,
       "step": 15
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 586.2643089294434,
+      "completion_length": 580.5134178161621,
       "epoch": 0.3411513859275053,
-      "grad_norm": 0.5713587999343872,
-      "kl": 0.00367889404296875,
+      "grad_norm": 0.4638945162296295,
+      "kl": 0.0028181076049804688,
       "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.058,
-      "reward": 0.7647321790456771,
-      "reward_std": 0.21236355928704143,
-      "rewards/accuracy_reward": 0.7631696805357933,
-      "rewards/format_reward": 0.0015625000698491931,
+      "loss": 0.0489,
+      "reward": 0.7725446745753288,
+      "reward_std": 0.1972122782841325,
+      "rewards/accuracy_reward": 0.7714286059141159,
+      "rewards/format_reward": 0.0011160714784637094,
       "step": 20
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 596.1701164245605,
+      "completion_length": 586.8551612854004,
       "epoch": 0.42643923240938164,
-      "grad_norm": 0.2316303700208664,
-      "kl": 0.007324790954589844,
+      "grad_norm": 0.13729101419448853,
+      "kl": 0.0033502578735351562,
       "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.0476,
-      "reward": 0.7662946805357933,
-      "reward_std": 0.19612560272216797,
-      "rewards/accuracy_reward": 0.761383967101574,
-      "rewards/format_reward": 0.004910714505240321,
+      "loss": 0.0425,
+      "reward": 0.7703125372529029,
+      "reward_std": 0.19164941012859343,
+      "rewards/accuracy_reward": 0.769866107404232,
+      "rewards/format_reward": 0.00044642859138548373,
       "step": 25
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 599.6569450378418,
+      "completion_length": 589.1917663574219,
       "epoch": 0.511727078891258,
-      "grad_norm": 23.480575561523438,
-      "kl": 0.018339920043945312,
+      "grad_norm": 0.10644034296274185,
+      "kl": 0.004234695434570312,
       "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0385,
-      "reward": 0.7616071805357933,
-      "reward_std": 0.21710715610533954,
-      "rewards/accuracy_reward": 0.7415178880095482,
-      "rewards/format_reward": 0.02008928672876209,
+      "loss": 0.034,
+      "reward": 0.7531250387430191,
+      "reward_std": 0.18948373831808568,
+      "rewards/accuracy_reward": 0.7511161044239998,
+      "rewards/format_reward": 0.002008928661234677,
       "step": 30
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 584.1323913574219,
+      "completion_length": 578.0815017700195,
       "epoch": 0.5970149253731343,
-      "grad_norm": 0.18863163888454437,
-      "kl": 0.012631607055664063,
+      "grad_norm": 0.09403481334447861,
+      "kl": 0.004046249389648438,
       "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0322,
-      "reward": 0.7970982536673545,
-      "reward_std": 0.2728703811764717,
-      "rewards/accuracy_reward": 0.7404018193483353,
-      "rewards/format_reward": 0.056696431199088695,
+      "loss": 0.0411,
+      "reward": 0.7562500327825546,
+      "reward_std": 0.18806953858584166,
+      "rewards/accuracy_reward": 0.7495536088943482,
+      "rewards/format_reward": 0.006696428894065321,
       "step": 35
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 575.9098495483398,
+      "completion_length": 564.9935516357422,
       "epoch": 0.6823027718550106,
-      "grad_norm": 0.19670729339122772,
-      "kl": 0.008755111694335937,
+      "grad_norm": 0.22079423069953918,
+      "kl": 0.00465545654296875,
       "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0332,
-      "reward": 0.8162946745753288,
-      "reward_std": 0.2680498031899333,
-      "rewards/accuracy_reward": 0.7600446820259095,
-      "rewards/format_reward": 0.056250002863816914,
+      "loss": 0.0326,
+      "reward": 0.7790178969502449,
+      "reward_std": 0.18834841772913932,
+      "rewards/accuracy_reward": 0.7680803954601287,
+      "rewards/format_reward": 0.010937500628642738,
       "step": 40
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 590.6379707336425,
+      "completion_length": 582.787744140625,
       "epoch": 0.767590618336887,
-      "grad_norm": 0.377120703458786,
-      "kl": 0.013896942138671875,
+      "grad_norm": 0.16064484417438507,
+      "kl": 0.005054092407226563,
       "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0296,
-      "reward": 0.8042411118745804,
-      "reward_std": 0.3068106513470411,
-      "rewards/accuracy_reward": 0.7363839611411095,
-      "rewards/format_reward": 0.06785714637953788,
+      "loss": 0.0423,
+      "reward": 0.7665178939700127,
+      "reward_std": 0.22603920232504607,
+      "rewards/accuracy_reward": 0.7470982491970062,
+      "rewards/format_reward": 0.019419643795117736,
       "step": 45
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 589.7989082336426,
+      "completion_length": 580.5013664245605,
       "epoch": 0.8528784648187633,
-      "grad_norm": 0.6106008887290955,
-      "kl": 0.013324737548828125,
+      "grad_norm": 0.18539777398109436,
+      "kl": 0.00699310302734375,
       "learning_rate": 1.718159615201853e-07,
-      "loss": 0.0311,
-      "reward": 0.8252232551574707,
-      "reward_std": 0.30773016847670076,
-      "rewards/accuracy_reward": 0.7352678969502449,
-      "rewards/format_reward": 0.08995536118745803,
+      "loss": 0.0301,
+      "reward": 0.7915178954601287,
+      "reward_std": 0.2188779940828681,
+      "rewards/accuracy_reward": 0.755357176065445,
+      "rewards/format_reward": 0.036160716065205635,
       "step": 50
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 581.0469017028809,
+      "completion_length": 580.8031463623047,
       "epoch": 0.9381663113006397,
-      "grad_norm": 0.17101669311523438,
-      "kl": 0.017275238037109376,
+      "grad_norm": 0.2504195272922516,
+      "kl": 0.009212875366210937,
       "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0434,
-      "reward": 0.8486607581377029,
-      "reward_std": 0.3164860598742962,
-      "rewards/accuracy_reward": 0.7607143178582192,
-      "rewards/format_reward": 0.08794643317814917,
+      "loss": 0.038,
+      "reward": 0.8113839596509933,
+      "reward_std": 0.2401964722201228,
+      "rewards/accuracy_reward": 0.7790178954601288,
+      "rewards/format_reward": 0.03236607303842902,
       "step": 55
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 572.9547259012858,
+      "completion_length": 567.5023994445801,
       "epoch": 0.9893390191897654,
-      "kl": 0.011492411295572916,
-      "reward": 0.8437500322858492,
-      "reward_std": 0.3019534188012282,
-      "rewards/accuracy_reward": 0.7477678855260214,
-      "rewards/format_reward": 0.0959821470702688,
+      "kl": 0.011606852213541666,
+      "reward": 0.8020833730697632,
+      "reward_std": 0.23279494047164917,
+      "rewards/accuracy_reward": 0.7678571765621504,
+      "rewards/format_reward": 0.03422619208383063,
       "step": 58,
       "total_flos": 0.0,
-      "train_loss": 0.04267983935002623,
-      "train_runtime": 7907.107,
-      "train_samples_per_second": 0.949,
+      "train_loss": 0.04428564981910689,
+      "train_runtime": 7921.619,
+      "train_samples_per_second": 0.947,
       "train_steps_per_second": 0.007
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3af35ec78fd740b874622981c9d0bef9078fe3b060a798afef75bd2b09a2a4ce
+oid sha256:6c472b231fb3fda1371ef81b463d7f886a525b30b76aebce96a985135998daf2
 size 7992