End of training

all_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.09253680892288685,
-    "train_runtime": 133428.0905,
-    "train_samples": 6,
+    "train_loss": 0.13233287669718266,
+    "train_runtime": 164161.8368,
+    "train_samples": 17,
     "train_samples_per_second": 0.001,
     "train_steps_per_second": 0.0
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dfc539f081d55c30d556844ce743afa9621005237355f33123dd09554b9e76cb
+oid sha256:2b1555ac512f53c15e1bd9d4da2105f74129d6491fcaf701e885d7bceb5246d0
 size 1976163472

train_results.json

@@ -1,8 +1,8 @@
 {
     "total_flos": 0.0,
-    "train_loss": 0.09253680892288685,
-    "train_runtime": 133428.0905,
-    "train_samples": 6,
+    "train_loss": 0.13233287669718266,
+    "train_runtime": 164161.8368,
+    "train_samples": 17,
     "train_samples_per_second": 0.001,
     "train_steps_per_second": 0.0
 }

trainer_state.json

@@ -2,7 +2,7 @@
   "best_global_step": null,
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 19.666666666666668,
+  "epoch": 4.9411764705882355,
   "eval_steps": 500,
   "global_step": 20,
   "is_hyper_param_search": false,
@@ -15,37 +15,37 @@
       "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": 585.5,
-      "completions/max_terminated_length": 585.5,
-      "completions/mean_length": 164.9166717529297,
-      "completions/mean_terminated_length": 164.9166717529297,
-      "completions/min_length": 14.0,
-      "completions/min_terminated_length": 14.0,
-      "epoch": 1.6666666666666665,
-      "grad_norm": 16.132144927978516,
+      "completions/clipped_ratio": 0.0625,
+      "completions/max_length": 628.0,
+      "completions/max_terminated_length": 293.0,
+      "completions/mean_length": 176.4375,
+      "completions/mean_terminated_length": 123.19642639160156,
+      "completions/min_length": 8.0,
+      "completions/min_terminated_length": 8.0,
+      "epoch": 0.47058823529411764,
+      "grad_norm": 51.56406784057617,
       "kl": 0.0,
       "learning_rate": 5e-07,
-      "loss": -0.1438,
-      "num_tokens": 6253.0,
-      "reward": 0.09254130534827709,
-      "reward_std": 0.03140645381063223,
-      "rewards/concensus_correctness_reward_func/mean": 0.0,
-      "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.0,
-      "rewards/consensus_reward_func/std": 0.0,
+      "loss": 0.3766,
+      "num_tokens": 6919.0,
+      "reward": 0.3104562256485224,
+      "reward_std": 0.32131396047770977,
+      "rewards/concensus_correctness_reward_func/mean": 0.03125,
+      "rewards/concensus_correctness_reward_func/std": 0.0883883461356163,
+      "rewards/consensus_reward_func/mean": 0.25,
+      "rewards/consensus_reward_func/std": 0.46291008591651917,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.08829130791127682,
-      "rewards/question_recreation_reward_func/std": 0.04877444030717015,
+      "rewards/question_recreation_reward_func/mean": 0.02920621819794178,
+      "rewards/question_recreation_reward_func/std": 0.023029986768960953,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.004249999765306711,
-      "rewards/xmlcount_reward_func/std": 0.01550806313753128,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 2
     },
     {
@@ -55,36 +55,36 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 247.0,
-      "completions/max_terminated_length": 247.0,
-      "completions/mean_length": 109.91666412353516,
-      "completions/mean_terminated_length": 109.91666412353516,
-      "completions/min_length": 7.5,
-      "completions/min_terminated_length": 7.5,
-      "epoch": 3.6666666666666665,
-      "grad_norm": 7.240252494812012,
-      "kl": 0.0010323685710318387,
+      "completions/max_length": 419.5,
+      "completions/max_terminated_length": 419.5,
+      "completions/mean_length": 305.375,
+      "completions/mean_terminated_length": 305.375,
+      "completions/min_length": 193.5,
+      "completions/min_terminated_length": 193.5,
+      "epoch": 0.9411764705882353,
+      "grad_norm": 11.458623886108398,
+      "kl": 0.0007662410207558423,
       "learning_rate": 4.864543104251586e-07,
-      "loss": 0.0967,
-      "num_tokens": 12726.0,
-      "reward": 0.1256694793701172,
-      "reward_std": 0.3006303310394287,
+      "loss": 0.0561,
+      "num_tokens": 14324.0,
+      "reward": 0.023023009300231934,
+      "reward_std": 0.01096764812245965,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.125,
-      "rewards/consensus_reward_func/std": 0.3535533845424652,
+      "rewards/consensus_reward_func/mean": 0.0,
+      "rewards/consensus_reward_func/std": 0.0,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.020169468596577644,
-      "rewards/question_recreation_reward_func/std": 0.015208072261884809,
+      "rewards/question_recreation_reward_func/mean": 0.023023009300231934,
+      "rewards/question_recreation_reward_func/std": 0.015222020447254181,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": -0.019500000402331352,
-      "rewards/xmlcount_reward_func/std": 0.2519115321338177,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 4
     },
     {
@@ -94,20 +94,20 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 318.5,
-      "completions/max_terminated_length": 318.5,
-      "completions/mean_length": 110.41666793823242,
-      "completions/mean_terminated_length": 110.41666793823242,
-      "completions/min_length": 5.0,
-      "completions/min_terminated_length": 5.0,
-      "epoch": 5.666666666666667,
-      "grad_norm": 12.744539260864258,
-      "kl": 0.001115468876378145,
+      "completions/max_length": 367.0,
+      "completions/max_terminated_length": 367.0,
+      "completions/mean_length": 114.125,
+      "completions/mean_terminated_length": 114.125,
+      "completions/min_length": 8.0,
+      "completions/min_terminated_length": 8.0,
+      "epoch": 1.4705882352941178,
+      "grad_norm": 44.851749420166016,
+      "kl": 0.0016121815278893337,
       "learning_rate": 4.472851273490984e-07,
-      "loss": 0.2975,
-      "num_tokens": 18890.0,
-      "reward": 0.03514278866350651,
-      "reward_std": 0.029419854748994112,
+      "loss": -0.0794,
+      "num_tokens": 20246.0,
+      "reward": 0.17371663451194763,
+      "reward_std": 0.0570271871984005,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
       "rewards/consensus_reward_func/mean": 0.0,
@@ -116,14 +116,14 @@
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.01989278756082058,
-      "rewards/question_recreation_reward_func/std": 0.01218721829354763,
+      "rewards/question_recreation_reward_func/mean": 0.17371663451194763,
+      "rewards/question_recreation_reward_func/std": 0.06652013398706913,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.015250000171363354,
-      "rewards/xmlcount_reward_func/std": 0.04313351586461067,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 6
     },
     {
@@ -133,36 +133,36 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 330.5,
-      "completions/max_terminated_length": 330.5,
-      "completions/mean_length": 93.75000190734863,
-      "completions/mean_terminated_length": 93.75000190734863,
-      "completions/min_length": 6.0,
-      "completions/min_terminated_length": 6.0,
-      "epoch": 7.666666666666667,
-      "grad_norm": 24.797382354736328,
-      "kl": 0.0010699216581997462,
+      "completions/max_length": 305.0,
+      "completions/max_terminated_length": 305.0,
+      "completions/mean_length": 152.5625,
+      "completions/mean_terminated_length": 152.5625,
+      "completions/min_length": 42.0,
+      "completions/min_terminated_length": 42.0,
+      "epoch": 1.9411764705882353,
+      "grad_norm": 9.940235137939453,
+      "kl": 0.000744940982258413,
       "learning_rate": 3.867370395306068e-07,
-      "loss": 0.3365,
-      "num_tokens": 24139.0,
-      "reward": 0.2419363632798195,
-      "reward_std": 0.2218155935406685,
+      "loss": 0.0534,
+      "num_tokens": 27557.0,
+      "reward": 0.02366686426103115,
+      "reward_std": 0.008607423398643732,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.125,
-      "rewards/consensus_reward_func/std": 0.3535533845424652,
+      "rewards/consensus_reward_func/mean": 0.0,
+      "rewards/consensus_reward_func/std": 0.0,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.07118633016943932,
-      "rewards/question_recreation_reward_func/std": 0.04484980972483754,
+      "rewards/question_recreation_reward_func/mean": 0.023666863329708576,
+      "rewards/question_recreation_reward_func/std": 0.017653593327850103,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.04575000051409006,
-      "rewards/xmlcount_reward_func/std": 0.09960854053497314,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 8
     },
     {
@@ -172,20 +172,20 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 243.0,
-      "completions/max_terminated_length": 243.0,
-      "completions/mean_length": 128.16666412353516,
-      "completions/mean_terminated_length": 128.16666412353516,
-      "completions/min_length": 32.0,
-      "completions/min_terminated_length": 32.0,
-      "epoch": 9.666666666666666,
-      "grad_norm": 26.717540740966797,
-      "kl": 0.001343091771559557,
+      "completions/max_length": 509.0,
+      "completions/max_terminated_length": 509.0,
+      "completions/mean_length": 250.5,
+      "completions/mean_terminated_length": 250.5,
+      "completions/min_length": 64.5,
+      "completions/min_terminated_length": 64.5,
+      "epoch": 2.4705882352941178,
+      "grad_norm": 5.178226470947266,
+      "kl": 0.0007074162567732856,
       "learning_rate": 3.1137137178519977e-07,
-      "loss": -0.0579,
-      "num_tokens": 30080.0,
-      "reward": 0.07627854682505131,
-      "reward_std": 0.056225333362817764,
+      "loss": 0.199,
+      "num_tokens": 35661.0,
+      "reward": 0.08728898875415325,
+      "reward_std": 0.025852903723716736,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
       "rewards/consensus_reward_func/mean": 0.0,
@@ -194,14 +194,14 @@
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.06927854334935546,
-      "rewards/question_recreation_reward_func/std": 0.03935649152845144,
+      "rewards/question_recreation_reward_func/mean": 0.08728898782283068,
+      "rewards/question_recreation_reward_func/std": 0.03699528565630317,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.007000000216066837,
-      "rewards/xmlcount_reward_func/std": 0.05189000070095062,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 10
     },
     {
@@ -211,36 +211,36 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 150.0,
-      "completions/max_terminated_length": 150.0,
-      "completions/mean_length": 54.58333396911621,
-      "completions/mean_terminated_length": 54.58333396911621,
-      "completions/min_length": 9.0,
-      "completions/min_terminated_length": 9.0,
-      "epoch": 11.666666666666666,
-      "grad_norm": 18.737913131713867,
-      "kl": 0.00392270221709623,
+      "completions/max_length": 304.5,
+      "completions/max_terminated_length": 304.5,
+      "completions/mean_length": 166.875,
+      "completions/mean_terminated_length": 166.875,
+      "completions/min_length": 82.5,
+      "completions/min_terminated_length": 82.5,
+      "epoch": 2.9411764705882355,
+      "grad_norm": 9.689749717712402,
+      "kl": 0.0008264059497378184,
       "learning_rate": 2.2935516363191693e-07,
-      "loss": 0.1934,
-      "num_tokens": 34949.0,
-      "reward": 0.29006822407245636,
-      "reward_std": 0.22484304383397102,
+      "loss": 0.2115,
+      "num_tokens": 42270.0,
+      "reward": 0.20029596518725157,
+      "reward_std": 0.19888783944770694,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.0,
-      "rewards/consensus_reward_func/std": 0.0,
+      "rewards/consensus_reward_func/mean": 0.125,
+      "rewards/consensus_reward_func/std": 0.3535533845424652,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
-      "rewards/final_correctness_reward_func/mean": 0.125,
-      "rewards/final_correctness_reward_func/std": 0.3535533845424652,
-      "rewards/question_recreation_reward_func/mean": 0.09619322419166565,
-      "rewards/question_recreation_reward_func/std": 0.05533384159207344,
+      "rewards/final_correctness_reward_func/mean": 0.0,
+      "rewards/final_correctness_reward_func/std": 0.0,
+      "rewards/question_recreation_reward_func/mean": 0.07529596518725157,
+      "rewards/question_recreation_reward_func/std": 0.03689309814944863,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.06887500174343586,
-      "rewards/xmlcount_reward_func/std": 0.11247996985912323,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 12
     },
     {
@@ -250,20 +250,20 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 333.5,
-      "completions/max_terminated_length": 333.5,
-      "completions/mean_length": 98.33333206176758,
-      "completions/mean_terminated_length": 98.33333206176758,
-      "completions/min_length": 4.5,
-      "completions/min_terminated_length": 4.5,
-      "epoch": 13.666666666666666,
-      "grad_norm": 15.76174259185791,
-      "kl": 0.0010197372721449938,
+      "completions/max_length": 426.5,
+      "completions/max_terminated_length": 426.5,
+      "completions/mean_length": 159.5,
+      "completions/mean_terminated_length": 159.5,
+      "completions/min_length": 2.0,
+      "completions/min_terminated_length": 2.0,
+      "epoch": 3.4705882352941178,
+      "grad_norm": 8.103471755981445,
+      "kl": 0.0017809901619330049,
       "learning_rate": 1.4957614383675767e-07,
-      "loss": 0.1083,
-      "num_tokens": 40602.0,
-      "reward": 0.11224918067455292,
-      "reward_std": 0.035229440312832594,
+      "loss": 0.0366,
+      "num_tokens": 48918.0,
+      "reward": 0.01776011288166046,
+      "reward_std": 0.006043280474841595,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
       "rewards/consensus_reward_func/mean": 0.0,
@@ -272,14 +272,14 @@
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.09249918535351753,
-      "rewards/question_recreation_reward_func/std": 0.02914312807843089,
+      "rewards/question_recreation_reward_func/mean": 0.01776011334732175,
+      "rewards/question_recreation_reward_func/std": 0.017151668202131987,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.019749999977648258,
-      "rewards/xmlcount_reward_func/std": 0.08688061870634556,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 14
     },
     {
@@ -289,36 +289,36 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 240.0,
-      "completions/max_terminated_length": 240.0,
-      "completions/mean_length": 68.41666507720947,
-      "completions/mean_terminated_length": 68.41666507720947,
-      "completions/min_length": 4.5,
-      "completions/min_terminated_length": 4.5,
-      "epoch": 15.666666666666666,
-      "grad_norm": 37.723182678222656,
-      "kl": 0.0013724988639296498,
+      "completions/max_length": 217.5,
+      "completions/max_terminated_length": 217.5,
+      "completions/mean_length": 112.6875,
+      "completions/mean_terminated_length": 112.6875,
+      "completions/min_length": 52.5,
+      "completions/min_terminated_length": 52.5,
+      "epoch": 3.9411764705882355,
+      "grad_norm": 8.5344877243042,
+      "kl": 0.0009168514661723748,
       "learning_rate": 8.067960709356478e-08,
-      "loss": -0.1065,
-      "num_tokens": 45868.0,
-      "reward": 0.1889171525835991,
-      "reward_std": 0.28485801815986633,
+      "loss": 0.2618,
+      "num_tokens": 55559.0,
+      "reward": 0.14910611137747765,
+      "reward_std": 0.18657799949869514,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.0,
-      "rewards/consensus_reward_func/std": 0.0,
+      "rewards/consensus_reward_func/mean": 0.125,
+      "rewards/consensus_reward_func/std": 0.3535533845424652,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
-      "rewards/final_correctness_reward_func/mean": 0.125,
-      "rewards/final_correctness_reward_func/std": 0.3535533845424652,
-      "rewards/question_recreation_reward_func/mean": 0.01604215893894434,
-      "rewards/question_recreation_reward_func/std": 0.010827061953023076,
+      "rewards/final_correctness_reward_func/mean": 0.0,
+      "rewards/final_correctness_reward_func/std": 0.0,
+      "rewards/question_recreation_reward_func/mean": 0.02410610392689705,
+      "rewards/question_recreation_reward_func/std": 0.02181409765034914,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.04787500109523535,
-      "rewards/xmlcount_reward_func/std": 0.13023310527205467,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 16
     },
     {
@@ -328,36 +328,36 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 294.5,
-      "completions/max_terminated_length": 294.5,
-      "completions/mean_length": 94.5,
-      "completions/mean_terminated_length": 94.5,
-      "completions/min_length": 4.0,
-      "completions/min_terminated_length": 4.0,
-      "epoch": 17.666666666666668,
-      "grad_norm": 10.869548797607422,
-      "kl": 0.0006961950239201542,
+      "completions/max_length": 281.5,
+      "completions/max_terminated_length": 281.5,
+      "completions/mean_length": 118.0,
+      "completions/mean_terminated_length": 118.0,
+      "completions/min_length": 18.5,
+      "completions/min_terminated_length": 18.5,
+      "epoch": 4.470588235294118,
+      "grad_norm": 19.964595794677734,
+      "kl": 0.0011214540027140174,
       "learning_rate": 3.013156219837776e-08,
-      "loss": 0.2133,
-      "num_tokens": 52010.0,
-      "reward": 0.15816733241081238,
-      "reward_std": 0.04277936555445194,
+      "loss": 0.2282,
+      "num_tokens": 61543.0,
+      "reward": 0.32675740122795105,
+      "reward_std": 0.044070890406146646,
       "rewards/concensus_correctness_reward_func/mean": 0.0,
       "rewards/concensus_correctness_reward_func/std": 0.0,
-      "rewards/consensus_reward_func/mean": 0.0,
-      "rewards/consensus_reward_func/std": 0.0,
+      "rewards/consensus_reward_func/mean": 0.25,
+      "rewards/consensus_reward_func/std": 0.46291008591651917,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
       "rewards/final_correctness_reward_func/mean": 0.0,
       "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.12766733393073082,
-      "rewards/question_recreation_reward_func/std": 0.053603312000632286,
+      "rewards/question_recreation_reward_func/mean": 0.07675742218270898,
+      "rewards/question_recreation_reward_func/std": 0.03963730251416564,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.030500000342726707,
-      "rewards/xmlcount_reward_func/std": 0.05647502467036247,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 18
     },
     {
@@ -367,52 +367,52 @@
       "clip_ratio/low_min": 0.0,
       "clip_ratio/region_mean": 0.0,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 366.0,
-      "completions/max_terminated_length": 366.0,
-      "completions/mean_length": 85.33333587646484,
-      "completions/mean_terminated_length": 85.33333587646484,
-      "completions/min_length": 4.5,
-      "completions/min_terminated_length": 4.5,
-      "epoch": 19.666666666666668,
-      "grad_norm": 27.261966705322266,
-      "kl": 0.0019625511777121574,
+      "completions/max_length": 221.0,
+      "completions/max_terminated_length": 221.0,
+      "completions/mean_length": 119.3125,
+      "completions/mean_terminated_length": 119.3125,
+      "completions/min_length": 62.5,
+      "completions/min_terminated_length": 62.5,
+      "epoch": 4.9411764705882355,
+      "grad_norm": 14.930213928222656,
+      "kl": 0.0009697647255961783,
       "learning_rate": 3.4096741493194193e-09,
-      "loss": -0.012,
-      "num_tokens": 57319.0,
-      "reward": 0.21800990775227547,
-      "reward_std": 0.24578243133146316,
-      "rewards/concensus_correctness_reward_func/mean": 0.04774999991059303,
-      "rewards/concensus_correctness_reward_func/std": 0.13505738973617554,
-      "rewards/consensus_reward_func/mean": 0.125,
-      "rewards/consensus_reward_func/std": 0.3535533845424652,
+      "loss": -0.0206,
+      "num_tokens": 67694.0,
+      "reward": 0.2407371997833252,
+      "reward_std": 0.20068720169365406,
+      "rewards/concensus_correctness_reward_func/mean": 0.0,
+      "rewards/concensus_correctness_reward_func/std": 0.0,
+      "rewards/consensus_reward_func/mean": 0.0,
+      "rewards/consensus_reward_func/std": 0.0,
       "rewards/cumulative_reward_2/mean": 0.0,
       "rewards/cumulative_reward_2/std": 0.0,
-      "rewards/final_correctness_reward_func/mean": 0.0,
-      "rewards/final_correctness_reward_func/std": 0.0,
-      "rewards/question_recreation_reward_func/mean": 0.014759915880858898,
-      "rewards/question_recreation_reward_func/std": 0.005782891297712922,
+      "rewards/final_correctness_reward_func/mean": 0.125,
+      "rewards/final_correctness_reward_func/std": 0.3535533845424652,
+      "rewards/question_recreation_reward_func/mean": 0.11573721468448639,
+      "rewards/question_recreation_reward_func/std": 0.07402005046606064,
       "rewards/soft_format_reward_func/mean": 0.0,
       "rewards/soft_format_reward_func/std": 0.0,
       "rewards/strict_format_reward_func/mean": 0.0,
       "rewards/strict_format_reward_func/std": 0.0,
-      "rewards/xmlcount_reward_func/mean": 0.030500000342726707,
-      "rewards/xmlcount_reward_func/std": 0.05647502467036247,
+      "rewards/xmlcount_reward_func/mean": 0.0,
+      "rewards/xmlcount_reward_func/std": 0.0,
       "step": 20
     },
     {
-      "epoch": 19.666666666666668,
+      "epoch": 4.9411764705882355,
       "step": 20,
       "total_flos": 0.0,
-      "train_loss": 0.09253680892288685,
-      "train_runtime": 133428.0905,
+      "train_loss": 0.13233287669718266,
+      "train_runtime": 164161.8368,
       "train_samples_per_second": 0.001,
       "train_steps_per_second": 0.0
     }
   ],
   "logging_steps": 2,
   "max_steps": 20,
-  "num_input_tokens_seen": 57319,
-  "num_train_epochs": 20,
+  "num_input_tokens_seen": 67694,
+  "num_train_epochs": 5,
   "save_steps": 25,
   "stateful_callbacks": {
     "TrainerControl": {

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:620dcea1f14346a2fbd9e648d69f599772d688654d6de6ac407a35553b204b96
+oid sha256:7e2e3bc5392965d5dafa468948526080a63b0daeacc65feb1086c19428f6ca9d
 size 6929