Model save

README.md

@@ -1,10 +1,8 @@
 ---
-datasets: DigitalLearningGmbH/MATH-lighteval
 library_name: transformers
 model_name: qwen25-Math-7b-3epoch
 tags:
 - generated_from_trainer
-- open-r1
 - trl
 - grpo
 licence: license
@@ -12,7 +10,7 @@ licence: license
 
 # Model Card for qwen25-Math-7b-3epoch
 
-This model is a fine-tuned version of [None](https://huggingface.co/None) on the [DigitalLearningGmbH/MATH-lighteval](https://huggingface.co/datasets/DigitalLearningGmbH/MATH-lighteval) dataset.
+This model is a fine-tuned version of [None](https://huggingface.co/None).
 It has been trained using [TRL](https://github.com/huggingface/trl).
 
 ## Quick start
@@ -28,7 +26,7 @@ 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/1105645918-bit/huggingface/runs/4lyg3uex) 
+[<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/1105645918-bit/huggingface/runs/cj1e0d7i) 
 
 
 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": 340917.58753671293,
-    "train_runtime": 20435.4937,
+    "train_loss": 822442.3794969221,
+    "train_runtime": 20401.2733,
     "train_samples": 7500,
-    "train_samples_per_second": 0.367,
+    "train_samples_per_second": 0.368,
     "train_steps_per_second": 0.003
 }

config.json

@@ -23,7 +23,7 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.49.0",
-  "use_cache": true,
+  "use_cache": false,
   "use_mrope": false,
   "use_sliding_window": false,
   "vocab_size": 152064

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41520b81a9c3d56c53546892aa71f706b8205615b76b191808f1df22b519f73e
+oid sha256:95d0e9f55dafb4fba4b654413ccac0799d5ecc0923de9f38a639afa7bd3c35cb
 size 4877660776

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:510bbaf6488bc84d772d6d484d2ac5f0e8c4ec64e787e8e30765ce6910d62057
+oid sha256:bb0e9121a24ad88917e86451c50cf72a4815fd67da828d5a8e32fc0074599ea9
 size 4932751008

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6936226b04fd0568523668398ec3d4de37a13f92ae88bcd78979881ad780694
+oid sha256:493b1971e3661a1c734c54feba073f3bbc8cab483062d1b2748f13009a6d8c3e
 size 4330865200

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d4c3350dea854c7937e01daf970a5458ec24c855b8ea029a52ea5ffda5fd2398
+oid sha256:3cb01982f51589497085b5c0f412bc6cccef24d303a0f8f4658acf5c05bde000
 size 1089994880

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 340917.58753671293,
-    "train_runtime": 20435.4937,
+    "train_loss": 822442.3794969221,
+    "train_runtime": 20401.2733,
     "train_samples": 7500,
-    "train_samples_per_second": 0.367,
+    "train_samples_per_second": 0.368,
     "train_steps_per_second": 0.003
 }

trainer_state.json

@@ -12,13 +12,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.017057569296375266,
-      "grad_norm": 0.09757333248853683,
+      "grad_norm": 0.09009335935115814,
       "kl": 0.0,
       "learning_rate": 5e-07,
       "loss": 0.0,
-      "reward": 0.20982143841683865,
-      "reward_std": 0.3133961074054241,
-      "rewards/accuracy_reward": 0.20982143841683865,
+      "reward": 0.2321428693830967,
+      "reward_std": 0.3139922395348549,
+      "rewards/accuracy_reward": 0.2321428693830967,
       "rewards/format_reward": 0.0,
       "step": 1
     },
@@ -26,13 +26,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.08528784648187633,
-      "grad_norm": 0.09003066271543503,
-      "kl": 0.00010742247104644775,
+      "grad_norm": 0.07332542538642883,
+      "kl": 9.924173355102539e-05,
       "learning_rate": 2.5e-06,
       "loss": 0.0,
-      "reward": 0.23046876164153218,
-      "reward_std": 0.3150851926766336,
-      "rewards/accuracy_reward": 0.23046876164153218,
+      "reward": 0.22935269074514508,
+      "reward_std": 0.30691308877430856,
+      "rewards/accuracy_reward": 0.22935269074514508,
       "rewards/format_reward": 0.0,
       "step": 5
     },
@@ -40,13 +40,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.17057569296375266,
-      "grad_norm": 29.32818031311035,
-      "kl": 0.004455947875976562,
+      "grad_norm": 0.0727960467338562,
+      "kl": 0.0027547836303710937,
       "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0002,
-      "reward": 0.44888395071029663,
-      "reward_std": 0.33523476012051107,
-      "rewards/accuracy_reward": 0.44888395071029663,
+      "loss": 0.0001,
+      "reward": 0.4187500193715096,
+      "reward_std": 0.33326763547956945,
+      "rewards/accuracy_reward": 0.4187500193715096,
       "rewards/format_reward": 0.0,
       "step": 10
     },
@@ -54,13 +54,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.255863539445629,
-      "grad_norm": 0.0765397846698761,
-      "kl": 0.010036468505859375,
+      "grad_norm": 0.06532283872365952,
+      "kl": 0.17901992797851562,
       "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.0004,
-      "reward": 0.7040178865194321,
-      "reward_std": 0.20274759102612733,
-      "rewards/accuracy_reward": 0.7040178865194321,
+      "loss": 0.0072,
+      "reward": 0.7390625298023223,
+      "reward_std": 0.17226867079734803,
+      "rewards/accuracy_reward": 0.7390625298023223,
       "rewards/format_reward": 0.0,
       "step": 15
     },
@@ -68,13 +68,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.3411513859275053,
-      "grad_norm": 0.06571836024522781,
-      "kl": 0.01729736328125,
+      "grad_norm": 0.08123523741960526,
+      "kl": 0.020281982421875,
       "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.0007,
-      "reward": 0.7073661059141159,
-      "reward_std": 0.18286588205955923,
-      "rewards/accuracy_reward": 0.7073661059141159,
+      "loss": 0.0008,
+      "reward": 0.7357143208384513,
+      "reward_std": 0.16036716918461025,
+      "rewards/accuracy_reward": 0.7357143208384513,
       "rewards/format_reward": 0.0,
       "step": 20
     },
@@ -82,13 +82,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.42643923240938164,
-      "grad_norm": 0.08071992546319962,
-      "kl": 0.012841796875,
+      "grad_norm": 98653831168.0,
+      "kl": 209715200.01299286,
       "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.0005,
-      "reward": 0.7102678894996644,
-      "reward_std": 0.18361264672130345,
-      "rewards/accuracy_reward": 0.7102678894996644,
+      "loss": 8388608.0,
+      "reward": 0.731919676065445,
+      "reward_std": 0.15736169517040252,
+      "rewards/accuracy_reward": 0.731919676065445,
       "rewards/format_reward": 0.0,
       "step": 25
     },
@@ -96,13 +96,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.511727078891258,
-      "grad_norm": 0.056903496384620667,
-      "kl": 0.01241302490234375,
+      "grad_norm": 13029.787109375,
+      "kl": 67.636962890625,
       "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0005,
-      "reward": 0.691517885029316,
-      "reward_std": 0.18483126908540726,
-      "rewards/accuracy_reward": 0.691517885029316,
+      "loss": 2.7023,
+      "reward": 0.7205357491970062,
+      "reward_std": 0.1578733256086707,
+      "rewards/accuracy_reward": 0.7205357491970062,
       "rewards/format_reward": 0.0,
       "step": 30
     },
@@ -110,13 +110,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.5970149253731343,
-      "grad_norm": 0.07405390590429306,
-      "kl": 0.010370635986328125,
+      "grad_norm": 0.39150699973106384,
+      "kl": 2.8228424072265623,
       "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0004,
-      "reward": 0.6821428924798966,
-      "reward_std": 0.19620487112551926,
-      "rewards/accuracy_reward": 0.6821428924798966,
+      "loss": 0.1129,
+      "reward": 0.7111607402563095,
+      "reward_std": 0.16545333191752434,
+      "rewards/accuracy_reward": 0.7111607402563095,
       "rewards/format_reward": 0.0,
       "step": 35
     },
@@ -124,13 +124,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.6823027718550106,
-      "grad_norm": 0.06446688622236252,
-      "kl": 0.03368301391601562,
+      "grad_norm": 91.16580963134766,
+      "kl": 28502426.75253143,
       "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0014,
-      "reward": 0.6912946753203869,
-      "reward_std": 0.1812687825411558,
-      "rewards/accuracy_reward": 0.6912946753203869,
+      "loss": 1146119.4,
+      "reward": 0.7203125312924386,
+      "reward_std": 0.16900291871279477,
+      "rewards/accuracy_reward": 0.7203125312924386,
       "rewards/format_reward": 0.0,
       "step": 40
     },
@@ -138,13 +138,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.767590618336887,
-      "grad_norm": 0.06644736230373383,
-      "kl": 0.011377716064453125,
+      "grad_norm": 552.0439453125,
+      "kl": 116332.38398895264,
       "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0005,
-      "reward": 0.688169676065445,
-      "reward_std": 0.18324931319803,
-      "rewards/accuracy_reward": 0.688169676065445,
+      "loss": 4652.1254,
+      "reward": 0.7176339611411094,
+      "reward_std": 0.1728838672861457,
+      "rewards/accuracy_reward": 0.7176339611411094,
       "rewards/format_reward": 0.0,
       "step": 45
     },
@@ -152,13 +152,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.8528784648187633,
-      "grad_norm": 73763676160.0,
-      "kl": 99405360.00854492,
+      "grad_norm": 0.14540241658687592,
+      "kl": 801.6095245361328,
       "learning_rate": 1.718159615201853e-07,
-      "loss": 3954644.0,
-      "reward": 0.6783482417464256,
-      "reward_std": 0.18882247591391205,
-      "rewards/accuracy_reward": 0.6783482417464256,
+      "loss": 32.0004,
+      "reward": 0.7058035969734192,
+      "reward_std": 0.1750278390944004,
+      "rewards/accuracy_reward": 0.7058035969734192,
       "rewards/format_reward": 0.0,
       "step": 50
     },
@@ -166,13 +166,13 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.9381663113006397,
-      "grad_norm": 0.6496619582176208,
-      "kl": 0.010959625244140625,
+      "grad_norm": 4.781561851501465,
+      "kl": 22835.21565475464,
       "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0004,
-      "reward": 0.7225446715950966,
-      "reward_std": 0.17012470280751585,
-      "rewards/accuracy_reward": 0.7225446715950966,
+      "loss": 914.4983,
+      "reward": 0.7439732491970062,
+      "reward_std": 0.16849460527300836,
+      "rewards/accuracy_reward": 0.7439732491970062,
       "rewards/format_reward": 0.0,
       "step": 55
     },
@@ -180,16 +180,16 @@
       "clip_ratio": 0.0,
       "completion_length": 1024.0,
       "epoch": 0.9893390191897654,
-      "kl": 0.009393056233723959,
-      "reward": 0.6990327723324299,
-      "reward_std": 0.184214538273712,
-      "rewards/accuracy_reward": 0.6990327723324299,
+      "kl": 1372.5967407226562,
+      "reward": 0.7388393208384514,
+      "reward_std": 0.16757233006258807,
+      "rewards/accuracy_reward": 0.7388393208384514,
       "rewards/format_reward": 0.0,
       "step": 58,
       "total_flos": 0.0,
-      "train_loss": 340917.58753671293,
-      "train_runtime": 20435.4937,
-      "train_samples_per_second": 0.367,
+      "train_loss": 822442.3794969221,
+      "train_runtime": 20401.2733,
+      "train_samples_per_second": 0.368,
       "train_steps_per_second": 0.003
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:701b5383a7c085d106ec2fbb38b885b899773de758141018d0e779bcfbde47d7
+oid sha256:249ffb335e211dcd81001eb25e5b0037e40f4ebe8607c3ac99fb99e7d7750469
 size 8184