Model save

README.md

@@ -26,7 +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/quantumcgx-sjtu/huggingface/runs/1i280yqx) 
+[<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/quantumcgx-sjtu/huggingface/runs/q7992nd7) 
 
 
 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": 0.03796683862837481,
-    "train_runtime": 12237.2268,
+    "train_loss": 0.04025471127784715,
+    "train_runtime": 12294.0923,
     "train_samples": 7500,
-    "train_samples_per_second": 0.613,
+    "train_samples_per_second": 0.61,
     "train_steps_per_second": 0.005
 }

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5ca58a71519cc63005696ca2a2766476d42a1dc0611709f76d10eaa8d472f7cf
+oid sha256:c7f34a79b22d7146ceb012375790d8cf75921e2932374539663700d1a5724a4b
 size 4877660776

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:687be6592543c8cf505c6ebddffa69d98186a673f4542948749e9989677e4263
+oid sha256:79126fd3e643eaed94e367207aaf34c6a3dd1d38b711a309c2b457e4471f005b
 size 4932751008

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3356de55d4d190c4856b36206c3f124f0bf4f7d7f270d60fe5a4f6959a23d217
+oid sha256:34075134a56e830a924933f0728a0c255399dcee0324ab7cf63d2a3cc5c9ad29
 size 4330865200

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:555d5a4a5edc7b5e9c399086a329edc1a0101d01eac749df32cece6a0ad74c36
+oid sha256:9e186cd77e5c087ab5c6ef72af5c054e5abb46082077a8f0d81ce0d095400dbf
 size 1089994880

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.03796683862837481,
-    "train_runtime": 12237.2268,
+    "train_loss": 0.04025471127784715,
+    "train_runtime": 12294.0923,
     "train_samples": 7500,
-    "train_samples_per_second": 0.613,
+    "train_samples_per_second": 0.61,
     "train_steps_per_second": 0.005
 }

trainer_state.json

@@ -12,7 +12,7 @@
       "clip_ratio": 0.0,
       "completion_length": 594.6719055175781,
       "epoch": 0.017057569296375266,
-      "grad_norm": 0.3894534707069397,
+      "grad_norm": 0.39137744903564453,
       "kl": 0.0,
       "learning_rate": 5e-07,
       "loss": -0.0015,
@@ -24,172 +24,172 @@
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 606.3172702789307,
+      "completion_length": 607.5834541320801,
       "epoch": 0.08528784648187633,
-      "grad_norm": 0.3337987959384918,
-      "kl": 0.00029712915420532227,
+      "grad_norm": 1.0777415037155151,
+      "kl": 0.00038692355155944824,
       "learning_rate": 2.5e-06,
-      "loss": 0.0289,
-      "reward": 0.5892857406288385,
-      "reward_std": 0.3550556660629809,
-      "rewards/accuracy_reward": 0.5892857406288385,
+      "loss": 0.0259,
+      "reward": 0.5962611874565482,
+      "reward_std": 0.36695866473019123,
+      "rewards/accuracy_reward": 0.5962611874565482,
       "rewards/format_reward": 0.0,
       "step": 5
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 610.3469032287597,
+      "completion_length": 606.7803886413574,
       "epoch": 0.17057569296375266,
-      "grad_norm": 0.4427908658981323,
-      "kl": 0.015810632705688478,
+      "grad_norm": 1.1746565103530884,
+      "kl": 0.010793495178222656,
       "learning_rate": 2.956412726139078e-06,
-      "loss": 0.0597,
-      "reward": 0.6875000283122062,
-      "reward_std": 0.3029041964560747,
-      "rewards/accuracy_reward": 0.6875000283122062,
+      "loss": 0.048,
+      "reward": 0.6758928880095482,
+      "reward_std": 0.30583293717354537,
+      "rewards/accuracy_reward": 0.6758928880095482,
       "rewards/format_reward": 0.0,
       "step": 10
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 599.6522605895996,
+      "completion_length": 602.4096214294434,
       "epoch": 0.255863539445629,
-      "grad_norm": 0.33527594804763794,
-      "kl": 0.005640411376953125,
+      "grad_norm": 0.1635478436946869,
+      "kl": 0.008692169189453125,
       "learning_rate": 2.7836719084521715e-06,
-      "loss": 0.0684,
-      "reward": 0.7642857536673546,
-      "reward_std": 0.213128218986094,
-      "rewards/accuracy_reward": 0.7642857536673546,
+      "loss": 0.0619,
+      "reward": 0.755357176065445,
+      "reward_std": 0.22135366648435592,
+      "rewards/accuracy_reward": 0.755357176065445,
       "rewards/format_reward": 0.0,
       "step": 15
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 590.715650177002,
+      "completion_length": 593.5116325378418,
       "epoch": 0.3411513859275053,
-      "grad_norm": 0.08924940973520279,
-      "kl": 0.006121444702148438,
+      "grad_norm": 0.3694714605808258,
+      "kl": 0.00950794219970703,
       "learning_rate": 2.4946839873611927e-06,
-      "loss": 0.057,
-      "reward": 0.7723214641213417,
-      "reward_std": 0.19404921345412732,
-      "rewards/accuracy_reward": 0.7723214641213417,
+      "loss": 0.0543,
+      "reward": 0.7654018238186836,
+      "reward_std": 0.2084361480548978,
+      "rewards/accuracy_reward": 0.7654018238186836,
       "rewards/format_reward": 0.0,
       "step": 20
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 595.2143127441407,
+      "completion_length": 603.0078392028809,
       "epoch": 0.42643923240938164,
-      "grad_norm": 0.2607589662075043,
-      "kl": 0.0135772705078125,
+      "grad_norm": 0.21624431014060974,
+      "kl": 0.004351234436035157,
       "learning_rate": 2.1156192081791355e-06,
-      "loss": 0.0401,
-      "reward": 0.7671875298023224,
-      "reward_std": 0.18023715307936072,
-      "rewards/accuracy_reward": 0.7671875298023224,
+      "loss": 0.0513,
+      "reward": 0.7598214641213417,
+      "reward_std": 0.19899208266288043,
+      "rewards/accuracy_reward": 0.7598214641213417,
       "rewards/format_reward": 0.0,
       "step": 25
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 598.6375213623047,
+      "completion_length": 605.0727897644043,
       "epoch": 0.511727078891258,
-      "grad_norm": 0.22541268169879913,
-      "kl": 0.007162857055664063,
+      "grad_norm": 0.26695239543914795,
+      "kl": 0.005370330810546875,
       "learning_rate": 1.6808050203829845e-06,
-      "loss": 0.0346,
-      "reward": 0.7511161044239998,
-      "reward_std": 0.17934485003352166,
-      "rewards/accuracy_reward": 0.7511161044239998,
+      "loss": 0.0455,
+      "reward": 0.7566964611411094,
+      "reward_std": 0.18958494029939174,
+      "rewards/accuracy_reward": 0.7566964611411094,
       "rewards/format_reward": 0.0,
       "step": 30
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 586.3620780944824,
+      "completion_length": 592.4431076049805,
       "epoch": 0.5970149253731343,
-      "grad_norm": 0.12512558698654175,
-      "kl": 0.013501358032226563,
+      "grad_norm": 0.12381298094987869,
+      "kl": 0.004366302490234375,
       "learning_rate": 1.2296174432791415e-06,
-      "loss": 0.0307,
-      "reward": 0.7457589611411095,
-      "reward_std": 0.18167408434674143,
-      "rewards/accuracy_reward": 0.7457589611411095,
+      "loss": 0.0353,
+      "reward": 0.747098246216774,
+      "reward_std": 0.18222492672502993,
+      "rewards/accuracy_reward": 0.747098246216774,
       "rewards/format_reward": 0.0,
       "step": 35
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 578.3451164245605,
+      "completion_length": 583.5830627441406,
       "epoch": 0.6823027718550106,
-      "grad_norm": 0.15629476308822632,
-      "kl": 0.00733642578125,
+      "grad_norm": 0.15292133390903473,
+      "kl": 0.003928375244140625,
       "learning_rate": 8.029152419343472e-07,
-      "loss": 0.0295,
-      "reward": 0.7725446775555611,
-      "reward_std": 0.16549305748194457,
-      "rewards/accuracy_reward": 0.7725446775555611,
+      "loss": 0.0394,
+      "reward": 0.7667411059141159,
+      "reward_std": 0.1841550744138658,
+      "rewards/accuracy_reward": 0.7667411059141159,
       "rewards/format_reward": 0.0,
       "step": 40
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 590.9895362854004,
+      "completion_length": 599.4895347595215,
       "epoch": 0.767590618336887,
-      "grad_norm": 0.11738639324903488,
-      "kl": 0.016034698486328124,
+      "grad_norm": 0.2122126966714859,
+      "kl": 0.004204940795898437,
       "learning_rate": 4.3933982822017883e-07,
-      "loss": 0.0317,
-      "reward": 0.7482143223285675,
-      "reward_std": 0.1866074649617076,
-      "rewards/accuracy_reward": 0.7482143223285675,
+      "loss": 0.0305,
+      "reward": 0.7553571775555611,
+      "reward_std": 0.1827343128155917,
+      "rewards/accuracy_reward": 0.7553571775555611,
       "rewards/format_reward": 0.0,
       "step": 45
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 595.0875236511231,
+      "completion_length": 596.6477951049804,
       "epoch": 0.8528784648187633,
-      "grad_norm": 0.5059227347373962,
-      "kl": 0.01042022705078125,
+      "grad_norm": 0.10254240781068802,
+      "kl": 0.007993698120117188,
       "learning_rate": 1.718159615201853e-07,
-      "loss": 0.0294,
-      "reward": 0.7500000312924385,
-      "reward_std": 0.18064689869061112,
-      "rewards/accuracy_reward": 0.7500000312924385,
+      "loss": 0.0327,
+      "reward": 0.7506696820259094,
+      "reward_std": 0.19283948028460146,
+      "rewards/accuracy_reward": 0.7506696820259094,
       "rewards/format_reward": 0.0,
       "step": 50
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 581.5457832336426,
+      "completion_length": 590.7453369140625,
       "epoch": 0.9381663113006397,
-      "grad_norm": 0.13856537640094757,
-      "kl": 0.018731689453125,
+      "grad_norm": 0.22994418442249298,
+      "kl": 0.00672760009765625,
       "learning_rate": 2.4570139579284723e-08,
-      "loss": 0.0236,
-      "reward": 0.7825893223285675,
-      "reward_std": 0.18381581027060748,
-      "rewards/accuracy_reward": 0.7825893223285675,
+      "loss": 0.0323,
+      "reward": 0.7854911103844643,
+      "reward_std": 0.18603479415178298,
+      "rewards/accuracy_reward": 0.7854911103844643,
       "rewards/format_reward": 0.0,
       "step": 55
     },
     {
       "clip_ratio": 0.0,
-      "completion_length": 580.0205459594727,
+      "completion_length": 583.9811820983887,
       "epoch": 0.9893390191897654,
-      "kl": 0.008316675821940104,
-      "reward": 0.7719494452079138,
-      "reward_std": 0.17934072421242794,
-      "rewards/accuracy_reward": 0.7719494452079138,
+      "kl": 0.004169464111328125,
+      "reward": 0.7682292014360428,
+      "reward_std": 0.18959872238337994,
+      "rewards/accuracy_reward": 0.7682292014360428,
       "rewards/format_reward": 0.0,
       "step": 58,
       "total_flos": 0.0,
-      "train_loss": 0.03796683862837481,
-      "train_runtime": 12237.2268,
-      "train_samples_per_second": 0.613,
+      "train_loss": 0.04025471127784715,
+      "train_runtime": 12294.0923,
+      "train_samples_per_second": 0.61,
       "train_steps_per_second": 0.005
     }
   ],

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d860344276bc120dcbed788ba38d2cb700cfc46846b3acf5a999df20e714ff13
-size 8056
+oid sha256:4c755ecba0508c9f40824f0fbf0d2922ab2bc7d3e57c93d27c3e5b830f38b179
+size 7992