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/no0v40a9) 
+[<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/7uoij859) 
 
 
 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:d66cb65a1041f8fda1a53da903ef1c492178657071c945a0890410b1df4818ca
+oid sha256:5cbb52a1207ba56f8b0e8fe6458e0e811a1a2e6ecb80ef1af9e299fd57178083
 size 3554214752

trainer_state.json

@@ -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": 0.0,
-      "completions/max_length": 1218.0,
-      "completions/max_terminated_length": 1218.0,
-      "completions/mean_length": 109.996875,
-      "completions/mean_terminated_length": 109.996875,
-      "completions/min_length": 17.2,
-      "completions/min_terminated_length": 17.2,
+      "completions/clipped_ratio": 0.0078125,
+      "completions/max_length": 1392.2,
+      "completions/max_terminated_length": 805.2,
+      "completions/mean_length": 115.64375,
+      "completions/mean_terminated_length": 104.5155517578125,
+      "completions/min_length": 18.2,
+      "completions/min_terminated_length": 18.2,
       "epoch": 2.399355877616747,
-      "grad_norm": 0.0019124990794807673,
+      "grad_norm": 0.011998126283288002,
       "learning_rate": 1e-06,
-      "loss": 0.0006,
-      "num_tokens": 21060786.0,
-      "reward": 0.49921875,
-      "reward_std": 0.0022097086533904076,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0081,
+      "num_tokens": 21064400.0,
+      "reward": 0.59453125,
+      "reward_std": 0.18795942664146423,
+      "rewards/accuracy_reward": 0.1984375,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9984375,
+      "rewards/format_reward": 0.990625,
       "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": 0.0015625,
-      "completions/max_length": 949.8,
-      "completions/max_terminated_length": 824.4,
-      "completions/mean_length": 105.5375,
-      "completions/mean_terminated_length": 103.31301727294922,
-      "completions/min_length": 17.0,
-      "completions/min_terminated_length": 17.0,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 868.6,
+      "completions/max_terminated_length": 868.6,
+      "completions/mean_length": 99.8234375,
+      "completions/mean_terminated_length": 99.8234375,
+      "completions/min_length": 16.8,
+      "completions/min_terminated_length": 16.8,
       "epoch": 2.463768115942029,
-      "grad_norm": 0.0021424347069114447,
+      "grad_norm": 0.01248540636152029,
       "learning_rate": 1e-06,
-      "loss": 0.0002,
-      "num_tokens": 21602698.0,
-      "reward": 0.4984375,
-      "reward_std": 0.004419417306780815,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0108,
+      "num_tokens": 21602655.0,
+      "reward": 0.61171875,
+      "reward_std": 0.19059042632579803,
+      "rewards/accuracy_reward": 0.225,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.996875,
+      "rewards/format_reward": 0.9984375,
       "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": 1301.0,
-      "completions/max_terminated_length": 1011.4,
-      "completions/mean_length": 117.728125,
-      "completions/mean_terminated_length": 108.78486022949218,
-      "completions/min_length": 16.8,
-      "completions/min_terminated_length": 16.8,
+      "completions/clipped_ratio": 0.0015625,
+      "completions/max_length": 674.2,
+      "completions/max_terminated_length": 518.6,
+      "completions/mean_length": 41.5515625,
+      "completions/mean_terminated_length": 39.206139373779294,
+      "completions/min_length": 16.4,
+      "completions/min_terminated_length": 16.4,
       "epoch": 2.528180354267311,
-      "grad_norm": 0.0,
+      "grad_norm": 0.01337147131562233,
       "learning_rate": 1e-06,
-      "loss": 0.0028,
-      "num_tokens": 22198220.0,
-      "reward": 0.4953125,
-      "reward_std": 0.01020579319447279,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0071,
+      "num_tokens": 22149424.0,
+      "reward": 0.65859375,
+      "reward_std": 0.2084730714559555,
+      "rewards/accuracy_reward": 0.31875,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.990625,
+      "rewards/format_reward": 0.9984375,
       "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": 0.003125,
-      "completions/max_length": 1346.4,
-      "completions/max_terminated_length": 1101.8,
-      "completions/mean_length": 119.5875,
-      "completions/mean_terminated_length": 115.14135131835937,
-      "completions/min_length": 16.6,
-      "completions/min_terminated_length": 16.6,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 65.8,
+      "completions/max_terminated_length": 65.8,
+      "completions/mean_length": 18.2390625,
+      "completions/mean_terminated_length": 18.2390625,
+      "completions/min_length": 16.0,
+      "completions/min_terminated_length": 16.0,
       "epoch": 2.5925925925925926,
-      "grad_norm": 0.0009198391926474869,
+      "grad_norm": 0.012995072640478611,
       "learning_rate": 1e-06,
-      "loss": 0.0009,
-      "num_tokens": 22731476.0,
-      "reward": 0.49765625,
-      "reward_std": 0.0066291259601712225,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0017,
+      "num_tokens": 22617817.0,
+      "reward": 0.71640625,
+      "reward_std": 0.20620474815368653,
+      "rewards/accuracy_reward": 0.4328125,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9953125,
+      "rewards/format_reward": 1.0,
       "rewards/mean_confidence_reward": 0.0,
       "step": 200
     },
     {
       "epoch": 2.5925925925925926,
-      "eval_completions/clipped_ratio": 0.006340579710144928,
-      "eval_completions/max_length": 389.9130434782609,
-      "eval_completions/max_terminated_length": 328.77536231884056,
-      "eval_completions/mean_length": 117.83695652173913,
-      "eval_completions/mean_terminated_length": 108.77691534291144,
-      "eval_completions/min_length": 33.07971014492754,
-      "eval_completions/min_terminated_length": 33.07971014492754,
+      "eval_completions/clipped_ratio": 0.0,
+      "eval_completions/max_length": 19.565217391304348,
+      "eval_completions/max_terminated_length": 19.565217391304348,
+      "eval_completions/mean_length": 17.383152173913043,
+      "eval_completions/mean_terminated_length": 17.383152173913043,
+      "eval_completions/min_length": 16.89855072463768,
+      "eval_completions/min_terminated_length": 16.89855072463768,
       "eval_loss": 0.0,
-      "eval_num_tokens": 22731476.0,
-      "eval_reward": 0.4963768115942029,
-      "eval_reward_std": 0.010247924189636673,
-      "eval_rewards/accuracy_reward": 0.0,
+      "eval_num_tokens": 22617817.0,
+      "eval_reward": 0.7391304347826086,
+      "eval_reward_std": 0.1717898635112721,
+      "eval_rewards/accuracy_reward": 0.4791666666666667,
       "eval_rewards/brier_reward": 0.0,
       "eval_rewards/confidence_one_or_zero": 0.0,
-      "eval_rewards/format_reward": 0.9927536231884058,
+      "eval_rewards/format_reward": 0.9990942028985508,
       "eval_rewards/mean_confidence_reward": 0.0,
-      "eval_runtime": 342.0823,
-      "eval_samples_per_second": 0.403,
-      "eval_steps_per_second": 0.053,
+      "eval_runtime": 37.0255,
+      "eval_samples_per_second": 3.727,
+      "eval_steps_per_second": 0.486,
       "step": 200
     },
     {
@@ -1186,21 +1186,21 @@
       "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": 1091.6,
-      "completions/max_terminated_length": 1001.0,
-      "completions/mean_length": 112.4625,
-      "completions/mean_terminated_length": 110.25907897949219,
-      "completions/min_length": 17.0,
-      "completions/min_terminated_length": 17.0,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 145.0,
+      "completions/max_terminated_length": 145.0,
+      "completions/mean_length": 18.1015625,
+      "completions/mean_terminated_length": 18.1015625,
+      "completions/min_length": 16.0,
+      "completions/min_terminated_length": 16.0,
       "epoch": 2.6570048309178746,
-      "grad_norm": 0.0014365437673404813,
+      "grad_norm": 0.006278031039983034,
       "learning_rate": 1e-06,
-      "loss": 0.0008,
-      "num_tokens": 23308748.0,
-      "reward": 0.49921875,
-      "reward_std": 0.0022097086533904076,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0015,
+      "num_tokens": 23134698.0,
+      "reward": 0.78046875,
+      "reward_std": 0.14939574152231216,
+      "rewards/accuracy_reward": 0.5625,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
       "rewards/format_reward": 0.9984375,
@@ -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": 0.00625,
-      "completions/max_length": 1477.8,
-      "completions/max_terminated_length": 1071.2,
-      "completions/mean_length": 117.846875,
-      "completions/mean_terminated_length": 108.93180236816406,
-      "completions/min_length": 16.0,
-      "completions/min_terminated_length": 16.0,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 20.2,
+      "completions/max_terminated_length": 20.2,
+      "completions/mean_length": 17.04375,
+      "completions/mean_terminated_length": 17.04375,
+      "completions/min_length": 16.2,
+      "completions/min_terminated_length": 16.2,
       "epoch": 2.7214170692431563,
-      "grad_norm": 0.0007683393196202815,
+      "grad_norm": 0.007409343961626291,
       "learning_rate": 1e-06,
-      "loss": 0.004,
-      "num_tokens": 23892826.0,
-      "reward": 0.49375,
-      "reward_std": 0.017677669040858747,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0002,
+      "num_tokens": 23654262.0,
+      "reward": 0.72890625,
+      "reward_std": 0.049137798324227336,
+      "rewards/accuracy_reward": 0.4578125,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9875,
+      "rewards/format_reward": 1.0,
       "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": 0.0046875,
-      "completions/max_length": 1254.2,
-      "completions/max_terminated_length": 974.2,
-      "completions/mean_length": 111.115625,
-      "completions/mean_terminated_length": 104.36334838867188,
-      "completions/min_length": 16.8,
-      "completions/min_terminated_length": 16.8,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 19.2,
+      "completions/max_terminated_length": 19.2,
+      "completions/mean_length": 17.01875,
+      "completions/mean_terminated_length": 17.01875,
+      "completions/min_length": 17.0,
+      "completions/min_terminated_length": 17.0,
       "epoch": 2.785829307568438,
-      "grad_norm": 0.0009448982309550047,
+      "grad_norm": 0.0,
       "learning_rate": 1e-06,
-      "loss": 0.0029,
-      "num_tokens": 24419164.0,
-      "reward": 0.49609375,
-      "reward_std": 0.011048543266952038,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0,
+      "num_tokens": 24120378.0,
+      "reward": 0.77421875,
+      "reward_std": 0.0022097086533904076,
+      "rewards/accuracy_reward": 0.5484375,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9921875,
+      "rewards/format_reward": 1.0,
       "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": 0.00625,
-      "completions/max_length": 1252.0,
-      "completions/max_terminated_length": 985.6,
-      "completions/mean_length": 103.759375,
-      "completions/mean_terminated_length": 94.76344604492188,
-      "completions/min_length": 16.4,
-      "completions/min_terminated_length": 16.4,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 18.2,
+      "completions/max_terminated_length": 18.2,
+      "completions/mean_length": 17.00625,
+      "completions/mean_terminated_length": 17.00625,
+      "completions/min_length": 16.6,
+      "completions/min_terminated_length": 16.6,
       "epoch": 2.85024154589372,
-      "grad_norm": 0.0,
+      "grad_norm": 0.003187602385878563,
       "learning_rate": 1e-06,
-      "loss": 0.0022,
-      "num_tokens": 24941226.0,
-      "reward": 0.496875,
-      "reward_std": 0.00883883461356163,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0,
+      "num_tokens": 24586918.0,
+      "reward": 0.77421875,
+      "reward_std": 0.0022097086533904076,
+      "rewards/accuracy_reward": 0.5484375,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.99375,
+      "rewards/format_reward": 1.0,
       "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": 1380.2,
-      "completions/max_terminated_length": 939.6,
-      "completions/mean_length": 97.8796875,
-      "completions/mean_terminated_length": 91.06793060302735,
-      "completions/min_length": 16.4,
-      "completions/min_terminated_length": 16.4,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 17.0,
+      "completions/max_terminated_length": 17.0,
+      "completions/mean_length": 16.9984375,
+      "completions/mean_terminated_length": 16.9984375,
+      "completions/min_length": 16.8,
+      "completions/min_terminated_length": 16.8,
       "epoch": 2.9146537842190017,
       "grad_norm": 0.0,
       "learning_rate": 1e-06,
-      "loss": 0.0017,
-      "num_tokens": 25449181.0,
-      "reward": 0.49609375,
-      "reward_std": 0.011048543080687524,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0,
+      "num_tokens": 25043109.0,
+      "reward": 0.83046875,
+      "reward_std": 0.0022097086533904076,
+      "rewards/accuracy_reward": 0.6609375,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.9921875,
+      "rewards/format_reward": 1.0,
       "rewards/mean_confidence_reward": 0.0,
       "step": 225
     },
@@ -1321,24 +1321,24 @@
       "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": 1033.6,
-      "completions/max_terminated_length": 831.8,
-      "completions/mean_length": 78.1453125,
-      "completions/mean_terminated_length": 75.84559478759766,
-      "completions/min_length": 16.6,
-      "completions/min_terminated_length": 16.6,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 17.6,
+      "completions/max_terminated_length": 17.6,
+      "completions/mean_length": 17.0046875,
+      "completions/mean_terminated_length": 17.0046875,
+      "completions/min_length": 17.0,
+      "completions/min_terminated_length": 17.0,
       "epoch": 2.9790660225442833,
-      "grad_norm": 0.0008915381622500718,
+      "grad_norm": 0.0,
       "learning_rate": 1e-06,
-      "loss": 0.0005,
-      "num_tokens": 25972826.0,
-      "reward": 0.4984375,
-      "reward_std": 0.004419417306780815,
-      "rewards/accuracy_reward": 0.0,
+      "loss": 0.0,
+      "num_tokens": 25527624.0,
+      "reward": 0.76875,
+      "reward_std": 0.0,
+      "rewards/accuracy_reward": 0.5375,
       "rewards/brier_reward": 0.0,
       "rewards/confidence_one_or_zero": 0.0,
-      "rewards/format_reward": 0.996875,
+      "rewards/format_reward": 1.0,
       "rewards/mean_confidence_reward": 0.0,
       "step": 230
     },
@@ -1346,15 +1346,15 @@
       "epoch": 2.9790660225442833,
       "step": 230,
       "total_flos": 0.0,
-      "train_loss": 0.00036267064835714255,
-      "train_runtime": 2152.1672,
-      "train_samples_per_second": 1.731,
-      "train_steps_per_second": 0.107
+      "train_loss": 0.0006395614726225967,
+      "train_runtime": 1235.3467,
+      "train_samples_per_second": 3.016,
+      "train_steps_per_second": 0.187
     }
   ],
   "logging_steps": 5,
   "max_steps": 231,
-  "num_input_tokens_seen": 25972826,
+  "num_input_tokens_seen": 25527624,
   "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:cf06cc29d6059820188588aba3f689e5b3c53c90771469a03c4e2de494c9e2a4
+oid sha256:d621838afb400aedd4f5bc11f1106f7c3393295c516e1afd83b5de543331a2c3
 size 7992