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/ryanzhang123-the-university-of-auckland/RLCR/runs/nzjcqerr) 
+[<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/ryanzhang123-the-university-of-auckland/RLCR/runs/7t6rinro) 
 
 
 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).

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7c5b20d1810ce201680570cc498981959856d7778eec726b87f509577af1748
+oid sha256:d56de5b2370feb4dbb509c230ac580149cec26e9aad73a718a4595f40806a488
 size 3554214752

trainer_state.json

@@ -3,7 +3,7 @@
   "best_model_checkpoint": null,
   "epoch": 2.9790660225442833,
   "eval_steps": 50,
-  "global_step": 228,
+  "global_step": 230,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -1055,24 +1055,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.00625,
-      "completions/max_length": 1361.2,
-      "completions/max_terminated_length": 1078.8,
-      "completions/mean_length": 117.1703125,
-      "completions/mean_terminated_length": 108.25378723144532,
-      "completions/min_length": 18.0,
-      "completions/min_terminated_length": 18.0,
-      "epoch": 2.42512077294686,
-      "grad_norm": 0.0008059889660216868,
+      "completions/clipped_ratio": 0.0125,
+      "completions/max_length": 1480.2,
+      "completions/max_terminated_length": 1096.6,
+      "completions/mean_length": 138.0765625,
+      "completions/mean_terminated_length": 120.40022125244141,
+      "completions/min_length": 21.4,
+      "completions/min_terminated_length": 21.4,
+      "epoch": 2.399355877616747,
+      "grad_norm": 0.002646031556650996,
       "learning_rate": 1e-06,
-      "loss": 0.0036,
-      "num_tokens": 21029281.0,
-      "reward": 0.49453125,
-      "reward_std": 0.015467960573732852,
+      "loss": 0.0047,
+      "num_tokens": 21079397.0,
+      "reward": 0.4921875,
+      "reward_std": 0.022097086161375047,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9890625,
+      "rewards/format_reward": 0.984375,
       "rewards/mean_confidence_reward": 0.0,
       "step": 185
     },
@@ -1082,24 +1082,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0109375,
-      "completions/max_length": 1209.8,
-      "completions/max_terminated_length": 1021.0,
-      "completions/mean_length": 127.6078125,
-      "completions/mean_terminated_length": 112.13069000244141,
-      "completions/min_length": 16.8,
-      "completions/min_terminated_length": 16.8,
-      "epoch": 2.4895330112721417,
-      "grad_norm": 0.001932392013259232,
+      "completions/clipped_ratio": 0.0015625,
+      "completions/max_length": 1132.6,
+      "completions/max_terminated_length": 983.0,
+      "completions/mean_length": 115.8234375,
+      "completions/mean_terminated_length": 113.60880889892579,
+      "completions/min_length": 17.4,
+      "completions/min_terminated_length": 17.4,
+      "epoch": 2.463768115942029,
+      "grad_norm": 0.005643462762236595,
       "learning_rate": 1e-06,
-      "loss": 0.0046,
-      "num_tokens": 21574886.0,
-      "reward": 0.49453125,
-      "reward_std": 0.015467960387468338,
+      "loss": 0.0008,
+      "num_tokens": 21628532.0,
+      "reward": 0.49765625,
+      "reward_std": 0.0066291259601712225,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9890625,
+      "rewards/format_reward": 0.9953125,
       "rewards/mean_confidence_reward": 0.0,
       "step": 190
     },
@@ -1109,24 +1109,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.00625,
-      "completions/max_length": 1176.8,
-      "completions/max_terminated_length": 840.4,
-      "completions/mean_length": 101.34375,
-      "completions/mean_terminated_length": 92.35088500976562,
-      "completions/min_length": 21.6,
-      "completions/min_terminated_length": 21.6,
-      "epoch": 2.5539452495974233,
-      "grad_norm": 0.0011432544561102986,
+      "completions/clipped_ratio": 0.009375,
+      "completions/max_length": 1536.0,
+      "completions/max_terminated_length": 988.6,
+      "completions/mean_length": 128.975,
+      "completions/mean_terminated_length": 115.63809661865234,
+      "completions/min_length": 18.4,
+      "completions/min_terminated_length": 18.4,
+      "epoch": 2.528180354267311,
+      "grad_norm": 0.0020768363028764725,
       "learning_rate": 1e-06,
-      "loss": 0.0027,
-      "num_tokens": 22083498.0,
-      "reward": 0.496875,
-      "reward_std": 0.00883883461356163,
+      "loss": 0.0033,
+      "num_tokens": 22231892.0,
+      "reward": 0.49453125,
+      "reward_std": 0.015467960573732852,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.99375,
+      "rewards/format_reward": 0.9890625,
       "rewards/mean_confidence_reward": 0.0,
       "step": 195
     },
@@ -1136,48 +1136,48 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0109375,
-      "completions/max_length": 1407.4,
-      "completions/max_terminated_length": 902.4,
-      "completions/mean_length": 109.025,
-      "completions/mean_terminated_length": 93.25486755371094,
-      "completions/min_length": 16.8,
-      "completions/min_terminated_length": 16.8,
-      "epoch": 2.6183574879227054,
-      "grad_norm": 0.0014830685686320066,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 1536.0,
+      "completions/max_terminated_length": 1052.8,
+      "completions/mean_length": 134.6171875,
+      "completions/mean_terminated_length": 123.58267517089844,
+      "completions/min_length": 23.4,
+      "completions/min_terminated_length": 23.4,
+      "epoch": 2.5925925925925926,
+      "grad_norm": 0.0007894107839092612,
       "learning_rate": 1e-06,
-      "loss": 0.0037,
-      "num_tokens": 22645682.0,
-      "reward": 0.49375,
-      "reward_std": 0.01767766922712326,
+      "loss": 0.0029,
+      "num_tokens": 22775407.0,
+      "reward": 0.49609375,
+      "reward_std": 0.011048543266952038,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9875,
+      "rewards/format_reward": 0.9921875,
       "rewards/mean_confidence_reward": 0.0,
       "step": 200
     },
     {
-      "epoch": 2.6183574879227054,
+      "epoch": 2.5925925925925926,
       "eval_completions/clipped_ratio": 0.005434782608695652,
-      "eval_completions/max_length": 327.9130434782609,
-      "eval_completions/max_terminated_length": 296.4130434782609,
-      "eval_completions/mean_length": 100.78985507246377,
-      "eval_completions/mean_terminated_length": 95.00241558793662,
-      "eval_completions/min_length": 31.869565217391305,
-      "eval_completions/min_terminated_length": 31.869565217391305,
+      "eval_completions/max_length": 440.4855072463768,
+      "eval_completions/max_terminated_length": 395.9927536231884,
+      "eval_completions/mean_length": 130.15851449275362,
+      "eval_completions/mean_terminated_length": 122.56198280445044,
+      "eval_completions/min_length": 38.231884057971016,
+      "eval_completions/min_terminated_length": 38.231884057971016,
       "eval_loss": 0.0,
-      "eval_num_tokens": 22645682.0,
-      "eval_reward": 0.4963768115942029,
-      "eval_reward_std": 0.009363153490467348,
+      "eval_num_tokens": 22775407.0,
+      "eval_reward": 0.49682971014492755,
+      "eval_reward_std": 0.008966933665932089,
       "eval_rewards/accuracy_reward": 0.0,
       "eval_rewards/brier_reward": 0.0,
       "eval_rewards/confidence_one_or_zero": 0.0,
-      "eval_rewards/format_reward": 0.9927536231884058,
+      "eval_rewards/format_reward": 0.9936594202898551,
       "eval_rewards/mean_confidence_reward": 0.0,
-      "eval_runtime": 264.3356,
-      "eval_samples_per_second": 0.522,
-      "eval_steps_per_second": 0.068,
+      "eval_runtime": 370.7818,
+      "eval_samples_per_second": 0.372,
+      "eval_steps_per_second": 0.049,
       "step": 200
     },
     {
@@ -1186,24 +1186,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0015625,
-      "completions/max_length": 906.8,
-      "completions/max_terminated_length": 767.6,
-      "completions/mean_length": 94.16875,
-      "completions/mean_terminated_length": 91.92613220214844,
-      "completions/min_length": 17.6,
-      "completions/min_terminated_length": 17.6,
-      "epoch": 2.682769726247987,
-      "grad_norm": 0.0008456017822027206,
+      "completions/clipped_ratio": 0.00625,
+      "completions/max_length": 1263.6,
+      "completions/max_terminated_length": 911.0,
+      "completions/mean_length": 116.421875,
+      "completions/mean_terminated_length": 107.53674011230468,
+      "completions/min_length": 16.8,
+      "completions/min_terminated_length": 16.8,
+      "epoch": 2.6570048309178746,
+      "grad_norm": 0.00095662398962304,
       "learning_rate": 1e-06,
-      "loss": 0.001,
-      "num_tokens": 23170518.0,
-      "reward": 0.4984375,
-      "reward_std": 0.004419417306780815,
+      "loss": 0.0029,
+      "num_tokens": 23355853.0,
+      "reward": 0.49609375,
+      "reward_std": 0.011048543266952038,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.996875,
+      "rewards/format_reward": 0.9921875,
       "rewards/mean_confidence_reward": 0.0,
       "step": 205
     },
@@ -1213,24 +1213,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0015625,
-      "completions/max_length": 1083.0,
-      "completions/max_terminated_length": 987.2,
-      "completions/mean_length": 100.8828125,
-      "completions/mean_terminated_length": 98.64908905029297,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 1519.0,
+      "completions/max_terminated_length": 1180.4,
+      "completions/mean_length": 142.6796875,
+      "completions/mean_terminated_length": 131.7208999633789,
       "completions/min_length": 17.4,
       "completions/min_terminated_length": 17.4,
-      "epoch": 2.7471819645732687,
-      "grad_norm": 0.0,
+      "epoch": 2.7214170692431563,
+      "grad_norm": 0.00048346660332754254,
       "learning_rate": 1e-06,
-      "loss": 0.0018,
-      "num_tokens": 23710739.0,
-      "reward": 0.49609375,
-      "reward_std": 0.011048543080687524,
+      "loss": 0.0025,
+      "num_tokens": 23956464.0,
+      "reward": 0.4953125,
+      "reward_std": 0.013258251920342445,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9921875,
+      "rewards/format_reward": 0.990625,
       "rewards/mean_confidence_reward": 0.0,
       "step": 210
     },
@@ -1240,24 +1240,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.003125,
-      "completions/max_length": 1102.0,
-      "completions/max_terminated_length": 789.8,
-      "completions/mean_length": 95.671875,
-      "completions/mean_terminated_length": 91.14502868652343,
-      "completions/min_length": 17.6,
-      "completions/min_terminated_length": 17.6,
-      "epoch": 2.8115942028985508,
-      "grad_norm": 0.0005856005591340363,
+      "completions/clipped_ratio": 0.00625,
+      "completions/max_length": 1356.6,
+      "completions/max_terminated_length": 878.0,
+      "completions/mean_length": 116.9703125,
+      "completions/mean_terminated_length": 108.06337432861328,
+      "completions/min_length": 21.4,
+      "completions/min_terminated_length": 21.4,
+      "epoch": 2.785829307568438,
+      "grad_norm": 0.0006744141574017704,
       "learning_rate": 1e-06,
-      "loss": 0.0012,
-      "num_tokens": 24266489.0,
-      "reward": 0.4984375,
-      "reward_std": 0.004419417306780815,
+      "loss": 0.0026,
+      "num_tokens": 24487189.0,
+      "reward": 0.496875,
+      "reward_std": 0.00883883461356163,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.996875,
+      "rewards/format_reward": 0.99375,
       "rewards/mean_confidence_reward": 0.0,
       "step": 215
     },
@@ -1267,24 +1267,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0125,
-      "completions/max_length": 1196.8,
-      "completions/max_terminated_length": 874.6,
-      "completions/mean_length": 112.078125,
-      "completions/mean_terminated_length": 95.37962341308594,
-      "completions/min_length": 16.2,
-      "completions/min_terminated_length": 16.2,
-      "epoch": 2.8760064412238324,
-      "grad_norm": 0.30407413840293884,
+      "completions/clipped_ratio": 0.0046875,
+      "completions/max_length": 1141.2,
+      "completions/max_terminated_length": 915.0,
+      "completions/mean_length": 123.528125,
+      "completions/mean_terminated_length": 116.93916931152344,
+      "completions/min_length": 22.4,
+      "completions/min_terminated_length": 22.4,
+      "epoch": 2.85024154589372,
+      "grad_norm": 0.0016728757182136178,
       "learning_rate": 1e-06,
-      "loss": 0.0049,
-      "num_tokens": 24831699.0,
-      "reward": 0.4921875,
-      "reward_std": 0.022097086533904076,
+      "loss": 0.002,
+      "num_tokens": 25022543.0,
+      "reward": 0.4953125,
+      "reward_std": 0.01325825173407793,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.984375,
+      "rewards/format_reward": 0.990625,
       "rewards/mean_confidence_reward": 0.0,
       "step": 220
     },
@@ -1294,24 +1294,24 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.0046875,
-      "completions/max_length": 1315.4,
-      "completions/max_terminated_length": 829.2,
-      "completions/mean_length": 89.290625,
-      "completions/mean_terminated_length": 82.46375427246093,
-      "completions/min_length": 17.8,
-      "completions/min_terminated_length": 17.8,
-      "epoch": 2.940418679549114,
-      "grad_norm": 0.0008350868010893464,
+      "completions/clipped_ratio": 0.009375,
+      "completions/max_length": 1212.0,
+      "completions/max_terminated_length": 884.8,
+      "completions/mean_length": 133.2875,
+      "completions/mean_terminated_length": 120.0623291015625,
+      "completions/min_length": 19.8,
+      "completions/min_terminated_length": 19.8,
+      "epoch": 2.9146537842190017,
+      "grad_norm": 0.0,
       "learning_rate": 1e-06,
-      "loss": 0.0016,
-      "num_tokens": 25352725.0,
-      "reward": 0.49765625,
-      "reward_std": 0.0066291259601712225,
+      "loss": 0.0028,
+      "num_tokens": 25553799.0,
+      "reward": 0.4953125,
+      "reward_std": 0.009863808192312718,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9953125,
+      "rewards/format_reward": 0.990625,
       "rewards/mean_confidence_reward": 0.0,
       "step": 225
     },
@@ -1321,33 +1321,40 @@
       "clip_ratio/low_mean": 0.0,
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
-      "completions/clipped_ratio": 0.005208333333333333,
-      "completions/max_length": 1218.0,
-      "completions/max_terminated_length": 1015.3333333333334,
-      "completions/mean_length": 110.06788380940755,
-      "completions/mean_terminated_length": 102.60350290934245,
-      "completions/min_length": 17.666666666666668,
-      "completions/min_terminated_length": 17.666666666666668,
+      "completions/clipped_ratio": 0.0046875,
+      "completions/max_length": 1162.8,
+      "completions/max_terminated_length": 894.4,
+      "completions/mean_length": 114.178125,
+      "completions/mean_terminated_length": 107.45401611328126,
+      "completions/min_length": 17.8,
+      "completions/min_terminated_length": 17.8,
       "epoch": 2.9790660225442833,
-      "num_tokens": 25682129.0,
-      "reward": 0.4947916666666667,
-      "reward_std": 0.014731391022602717,
+      "grad_norm": 0.0010454885195940733,
+      "learning_rate": 1e-06,
+      "loss": 0.003,
+      "num_tokens": 26101145.0,
+      "reward": 0.49609375,
+      "reward_std": 0.011048543266952038,
       "rewards/accuracy_reward": 0.0,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9895833333333334,
+      "rewards/format_reward": 0.9921875,
       "rewards/mean_confidence_reward": 0.0,
-      "step": 228,
+      "step": 230
+    },
+    {
+      "epoch": 2.9790660225442833,
+      "step": 230,
       "total_flos": 0.0,
-      "train_loss": 0.004439977674720515,
-      "train_runtime": 9596.7863,
-      "train_samples_per_second": 0.388,
-      "train_steps_per_second": 0.024
+      "train_loss": 0.0005941401033297829,
+      "train_runtime": 2082.8626,
+      "train_samples_per_second": 1.789,
+      "train_steps_per_second": 0.111
     }
   ],
   "logging_steps": 5,
   "max_steps": 231,
-  "num_input_tokens_seen": 25682129,
+  "num_input_tokens_seen": 26101145,
   "num_train_epochs": 3,
   "save_steps": 60,
   "stateful_callbacks": {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebfce4e612c23e2d7436653cfb2ec514c2a9b4250a4d15cd7a1d5c1d11f212fa
+oid sha256:e4552b4f72af82a334750ed040fd1a3e4be960cc8d5a42cdf1d5cfa672c2bc4c
 size 7992