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/llm-m_wandb-weblab/Qwen-2.5-7B-Simple-RL/runs/g0yupti6) 
+[<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/llm-m_wandb-weblab/Qwen-2.5-7B-Simple-RL/runs/hj3gutx0) 
 
 
 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": 0.04530468287652936,
-    "train_runtime": 7393.0573,
+    "train_loss": 0.03837096903087764,
+    "train_runtime": 7408.2441,
     "train_samples": 7500,
-    "train_samples_per_second": 1.014,
+    "train_samples_per_second": 1.012,
     "train_steps_per_second": 0.008
 }

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:652c0062e313b080ded6dc118054b8355aab172335449c79e31298498dd175c8
+oid sha256:af644dc48cacfea1faf2db28e6ce35d933cc07cb5462d7ec17692cd8c6e2a1b7
 size 4877660776

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4edfb9f8e2fa11447f83d92c95eff786578ca63e8d1ab69acb942f4908bda027
+oid sha256:9199e8bc30eb8f7f38d53f72ba9a3d074a6164cab27201eae345195c0caf8ef5
 size 4932751008

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7bb7eff7159352c74e3ee948c581c4a6f610cd77282497b5c9c24f9f6255bcf8
+oid sha256:5984f38db6a80598781c5f071501c812765112897222034f2939cc7b151bc012
 size 4330865200

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64ff496d19efceac7d92a1fe726a17a8ac0a0fc60dd46de7e0556dc7de10ccd6
+oid sha256:70213b43816db78e11b0fd5254e922e87bca76cb67dd9ab9f92b95ed4615f799
 size 1089994880

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.04530468287652936,
-    "train_runtime": 7393.0573,
+    "train_loss": 0.03837096903087764,
+    "train_runtime": 7408.2441,
     "train_samples": 7500,
-    "train_samples_per_second": 1.014,
+    "train_samples_per_second": 1.012,
     "train_steps_per_second": 0.008
 }

trainer_state.json

@@ -11,7 +11,7 @@
     {
       "completion_length": 616.6462249755859,
       "epoch": 0.017057569296375266,
-      "grad_norm": 1.7058640718460083,
+      "grad_norm": 1.7272135019302368,
       "kl": 0.0,
       "learning_rate": 5e-07,
       "loss": 0.0357,
@@ -22,161 +22,161 @@
       "step": 1
     },
     {
-      "completion_length": 607.0837326049805,
+      "completion_length": 611.7843494415283,
       "epoch": 0.08528784648187633,
-      "grad_norm": 0.3163924515247345,
-      "kl": 0.00047519803047180176,
+      "grad_norm": 0.3145679533481598,
+      "kl": 0.00017917156219482422,
       "learning_rate": 2.5e-06,
-      "loss": 0.045,
-      "reward": 0.6755022592842579,
-      "reward_std": 0.32974539790302515,
-      "rewards/accuracy_reward": 0.6623884197324514,
-      "rewards/format_reward": 0.013113839813740924,
+      "loss": 0.0441,
+      "reward": 0.6612723525613546,
+      "reward_std": 0.322592549957335,
+      "rewards/accuracy_reward": 0.6476004794239998,
+      "rewards/format_reward": 0.013671875844011083,
       "step": 5
     },
     {
-      "completion_length": 603.5386444091797,
+      "completion_length": 606.3022621154785,
       "epoch": 0.17057569296375266,
-      "grad_norm": 0.2700318992137909,
-      "kl": 0.0039319276809692385,
+      "grad_norm": 0.9135303497314453,
+      "kl": 0.002117562294006348,
       "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0542,
-      "reward": 0.7120536059141159,
-      "reward_std": 0.3251304797828197,
-      "rewards/accuracy_reward": 0.6888393148779869,
-      "rewards/format_reward": 0.02321428689174354,
+      "loss": 0.0486,
+      "reward": 0.7031250327825547,
+      "reward_std": 0.3129757303744555,
+      "rewards/accuracy_reward": 0.6863839581608773,
+      "rewards/format_reward": 0.01674107222352177,
       "step": 10
     },
     {
-      "completion_length": 586.4022621154785,
+      "completion_length": 589.8219047546387,
       "epoch": 0.255863539445629,
-      "grad_norm": 0.7752754092216492,
-      "kl": 0.02043609619140625,
+      "grad_norm": 0.9820574522018433,
+      "kl": 0.01598987579345703,
       "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.0499,
-      "reward": 0.9937500432133675,
-      "reward_std": 0.42666376307606696,
-      "rewards/accuracy_reward": 0.7332589581608773,
-      "rewards/format_reward": 0.2604910831898451,
+      "loss": 0.0522,
+      "reward": 0.9247768327593804,
+      "reward_std": 0.3940876517444849,
+      "rewards/accuracy_reward": 0.7486607536673546,
+      "rewards/format_reward": 0.1761160797206685,
       "step": 15
     },
     {
-      "completion_length": 604.2397590637207,
+      "completion_length": 623.1513641357421,
       "epoch": 0.3411513859275053,
-      "grad_norm": 0.5679246187210083,
-      "kl": 0.015423583984375,
+      "grad_norm": 0.30078014731407166,
+      "kl": 0.0150848388671875,
       "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.0598,
-      "reward": 1.2245536342263221,
-      "reward_std": 0.5004812128841877,
-      "rewards/accuracy_reward": 0.7361607432365418,
-      "rewards/format_reward": 0.48839287981390955,
+      "loss": 0.0499,
+      "reward": 1.178348271548748,
+      "reward_std": 0.4842135950922966,
+      "rewards/accuracy_reward": 0.7214286014437675,
+      "rewards/format_reward": 0.45691966339945794,
       "step": 20
     },
     {
-      "completion_length": 640.4687789916992,
+      "completion_length": 673.8540496826172,
       "epoch": 0.42643923240938164,
-      "grad_norm": 0.3916238844394684,
-      "kl": 0.02191925048828125,
+      "grad_norm": 0.27627527713775635,
+      "kl": 0.023541259765625,
       "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.0536,
-      "reward": 1.4299107789993286,
-      "reward_std": 0.46242206543684006,
-      "rewards/accuracy_reward": 0.7180803850293159,
-      "rewards/format_reward": 0.7118303880095482,
+      "loss": 0.0483,
+      "reward": 1.3975447028875352,
+      "reward_std": 0.46302561461925507,
+      "rewards/accuracy_reward": 0.7002232432365417,
+      "rewards/format_reward": 0.6973214596509933,
       "step": 25
     },
     {
-      "completion_length": 644.6181091308594,
+      "completion_length": 645.6444488525391,
       "epoch": 0.511727078891258,
-      "grad_norm": 0.19792738556861877,
-      "kl": 0.02194671630859375,
+      "grad_norm": 0.15549008548259735,
+      "kl": 0.01860809326171875,
       "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0471,
-      "reward": 1.5294643580913543,
-      "reward_std": 0.409032379090786,
-      "rewards/accuracy_reward": 0.7118303909897804,
-      "rewards/format_reward": 0.817633967101574,
+      "loss": 0.051,
+      "reward": 1.4763393461704255,
+      "reward_std": 0.4491942249238491,
+      "rewards/accuracy_reward": 0.7241071775555611,
+      "rewards/format_reward": 0.7522321805357933,
       "step": 30
     },
     {
-      "completion_length": 637.4656509399414,
+      "completion_length": 638.6131942749023,
       "epoch": 0.5970149253731343,
-      "grad_norm": 0.17083711922168732,
-      "kl": 0.01954345703125,
+      "grad_norm": 0.14530694484710693,
+      "kl": 0.018096923828125,
       "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0447,
-      "reward": 1.5256697118282319,
-      "reward_std": 0.4126044109463692,
-      "rewards/accuracy_reward": 0.7125000350177288,
-      "rewards/format_reward": 0.8131696805357933,
+      "loss": 0.0398,
+      "reward": 1.5024554222822188,
+      "reward_std": 0.41087508872151374,
+      "rewards/accuracy_reward": 0.7180803894996644,
+      "rewards/format_reward": 0.7843750357627869,
       "step": 35
     },
     {
-      "completion_length": 624.153377532959,
+      "completion_length": 644.3710052490235,
       "epoch": 0.6823027718550106,
-      "grad_norm": 0.17209741473197937,
-      "kl": 0.019207763671875,
+      "grad_norm": 0.21266834437847137,
+      "kl": 0.02000579833984375,
       "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0367,
-      "reward": 1.577232214808464,
-      "reward_std": 0.39118890166282655,
-      "rewards/accuracy_reward": 0.7370536088943481,
-      "rewards/format_reward": 0.8401786178350449,
+      "loss": 0.03,
+      "reward": 1.574776855111122,
+      "reward_std": 0.3656768787652254,
+      "rewards/accuracy_reward": 0.7404018193483353,
+      "rewards/format_reward": 0.8343750387430191,
       "step": 40
     },
     {
-      "completion_length": 646.4712326049805,
+      "completion_length": 660.5759185791015,
       "epoch": 0.767590618336887,
-      "grad_norm": 0.20322322845458984,
-      "kl": 0.01898193359375,
+      "grad_norm": 0.19263683259487152,
+      "kl": 0.021728515625,
       "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0347,
-      "reward": 1.5633929252624512,
-      "reward_std": 0.405256462097168,
-      "rewards/accuracy_reward": 0.7270089596509933,
-      "rewards/format_reward": 0.8363839671015739,
+      "loss": 0.0311,
+      "reward": 1.5725447177886962,
+      "reward_std": 0.3758295692503452,
+      "rewards/accuracy_reward": 0.7430803939700127,
+      "rewards/format_reward": 0.829464316368103,
       "step": 45
     },
     {
-      "completion_length": 655.3091796875,
+      "completion_length": 651.923243713379,
       "epoch": 0.8528784648187633,
-      "grad_norm": 0.2283862978219986,
-      "kl": 0.02166290283203125,
+      "grad_norm": 0.24893562495708466,
+      "kl": 0.023681640625,
       "learning_rate": 1.718159615201853e-07,
-      "loss": 0.0315,
-      "reward": 1.5857143521308898,
-      "reward_std": 0.3743378147482872,
-      "rewards/accuracy_reward": 0.729241107404232,
-      "rewards/format_reward": 0.8564732536673546,
+      "loss": 0.028,
+      "reward": 1.5631697088479997,
+      "reward_std": 0.37856253087520597,
+      "rewards/accuracy_reward": 0.7366071760654449,
+      "rewards/format_reward": 0.8265625387430191,
       "step": 50
     },
     {
-      "completion_length": 653.1265914916992,
+      "completion_length": 644.7035942077637,
       "epoch": 0.9381663113006397,
-      "grad_norm": 0.1660618782043457,
-      "kl": 0.0220703125,
+      "grad_norm": 0.13216549158096313,
+      "kl": 0.0205230712890625,
       "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0359,
-      "reward": 1.6171875655651093,
-      "reward_std": 0.3830992128700018,
-      "rewards/accuracy_reward": 0.750669676065445,
-      "rewards/format_reward": 0.866517898440361,
+      "loss": 0.0207,
+      "reward": 1.619196492433548,
+      "reward_std": 0.3612879175692797,
+      "rewards/accuracy_reward": 0.7718750357627868,
+      "rewards/format_reward": 0.847321467101574,
       "step": 55
     },
     {
-      "completion_length": 648.9576187133789,
+      "completion_length": 639.9285990397135,
       "epoch": 0.9893390191897654,
-      "kl": 0.020642598470052082,
-      "reward": 1.6212798357009888,
-      "reward_std": 0.356278399626414,
-      "rewards/accuracy_reward": 0.7392113457123438,
-      "rewards/format_reward": 0.8820684949556986,
+      "kl": 0.049023946126302086,
+      "reward": 1.606026868025462,
+      "reward_std": 0.3578349606444438,
+      "rewards/accuracy_reward": 0.750000037252903,
+      "rewards/format_reward": 0.8560268258055052,
       "step": 58,
       "total_flos": 0.0,
-      "train_loss": 0.04530468287652936,
-      "train_runtime": 7393.0573,
-      "train_samples_per_second": 1.014,
+      "train_loss": 0.03837096903087764,
+      "train_runtime": 7408.2441,
+      "train_samples_per_second": 1.012,
       "train_steps_per_second": 0.008
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81b69755f9e49e96cbf0db7213abcb6630493e4de25c94eece388f9df70aedb8
+oid sha256:bbe8e983bf6c683a0ff792c7a303d2dcce24e72b228492b9f40c8d8311cdf1c7
 size 7928