rollouts: control-thinking/control-thinking-phase768/rollouts/raw_rollouts.rank-0.jsonl provenance

rollouts/control-thinking/control-thinking-phase768/rollouts/raw_rollouts.rank-0.jsonl.meta.json

@@ -0,0 +1,23 @@
+{
+ "source_repo": "arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1-runs",
+ "source_path": "control-thinking/control-thinking-phase768/rollouts/raw_rollouts.rank-0.jsonl",
+ "declared_bytes": 36236134037,
+ "streamed_bytes": 36236134037,
+ "sha256_streamed": "10a26f1e5ef52c6d2c2fd22b08288c8e98fe55629350dae6bfde2cf5482615f9",
+ "sha256_expected": "10a26f1e5ef52c6d2c2fd22b08288c8e98fe55629350dae6bfde2cf5482615f9",
+ "sha256_match": true,
+ "rows": 47104,
+ "dropped_fields": [
+  "completion_ids",
+  "log_extra",
+  "log_metric",
+  "trainer_state"
+ ],
+ "lines_seen": 47104,
+ "unparseable_lines": 0,
+ "blank_lines": 0,
+ "rows_per_rollout_audit": null,
+ "rows_match_audit": null,
+ "trainer_state": "TrainerState(epoch=0.041666666666666664, global_step=32, max_steps=768, logging_steps=1, eval_steps=500, save_steps=17, train_batch_size=4, num_train_epochs=1, num_input_tokens_seen=7748764, total_flos=0.0, log_history=[{'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.765625, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3606.0, 'completions/mean_length': 3669.421875, 'completions/mean_terminated_length': 2275.933349609375, 'completions/min_length': 1060.0, 'completions/min_terminated_length': 1060.0, 'entropy': 0.11464092414826155, 'epoch': 0.0013020833333333333, 'frac_reward_zero_std': 0.75, 'grad_norm': 5.118177432450466e-05, 'learning_rate': 1e-05, 'loss': -0.00012268195860087872, 'num_tokens': 294619.0, 'reward': 0.09375, 'reward/selected_zero_std_group_fraction': 0.5, 'reward/zero_std_group_fraction': 0.75, 'reward_components/channel_close_count': 0.234375, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.765625, 'reward_components/format_valid': 0.09375, 'reward_components/native_boundary_valid': 0.234375, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.234375, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.09375, 'reward_components/reward': 0.09375, 'reward_components/runs_correct': 0.1875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.09375, 'reward_std': 0.29378482699394226, 'rewards/reward_func/mean': 0.09375, 'rewards/reward_func/std': 0.29378482699394226, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.7739913463592529, 'sampling/importance_sampling_ratio/min': 6.604204116911205e-08, 'sampling/sampling_logp_difference/max': 5.984663963317871, 'sampling/sampling_logp_difference/mean': 0.017266588285565376, 'step': 1, 'step_time': 199.22395840508398}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.484375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3985.0, 'completions/mean_length': 3057.71875, 'completions/mean_terminated_length': 2082.36376953125, 'completions/min_length': 1040.0, 'completions/min_terminated_length': 1040.0, 'entropy': 0.07956582214683294, 'epoch': 0.0026041666666666665, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.006041001062840223, 'learning_rate': 1e-05, 'loss': 0.008527711033821106, 'num_tokens': 567881.0, 'reward': 0.46875, 'reward/selected_zero_std_group_fraction': 0.5, 'reward/zero_std_group_fraction': 0.75, 'reward_components/channel_close_count': 0.515625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.484375, 'reward_components/format_valid': 0.484375, 'reward_components/native_boundary_valid': 0.515625, 'reward_components/parser_json': 0.015625, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.5, 'reward_components/parser_unsafe': 0.03125, 'reward_components/parser_valid': 0.484375, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 0.609375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.46875, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.5096418857574463, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.357892990112305, 'sampling/sampling_logp_difference/mean': 0.015825413167476654, 'step': 2, 'step_time': 176.64610660797916}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.296875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3973.0, 'completions/mean_length': 2763.734375, 'completions/mean_terminated_length': 2201.22216796875, 'completions/min_length': 829.0, 'completions/min_terminated_length': 829.0, 'entropy': 0.14969482272863388, 'epoch': 0.00390625, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.003225702326744795, 'learning_rate': 1e-05, 'loss': -0.009826848283410072, 'num_tokens': 802104.0, 'reward': 0.515625, 'reward/selected_zero_std_group_fraction': 0.3333333333333333, 'reward/zero_std_group_fraction': 0.6666666666666666, 'reward_components/channel_close_count': 0.71875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.296875, 'reward_components/format_valid': 0.5625, 'reward_components/native_boundary_valid': 0.71875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.71875, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.5625, 'reward_components/reward': 0.515625, 'reward_components/runs_correct': 0.796875, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.515625, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.515625, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.3443005084991455, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.6105804443359375, 'sampling/sampling_logp_difference/mean': 0.018400561064481735, 'step': 3, 'step_time': 164.07066006807145}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.46875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4038.0, 'completions/mean_length': 3050.71875, 'completions/mean_terminated_length': 2128.411865234375, 'completions/min_length': 764.0, 'completions/min_terminated_length': 764.0, 'entropy': 0.1334073906764388, 'epoch': 0.005208333333333333, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.0029910786543041468, 'learning_rate': 1e-05, 'loss': -0.00887129083275795, 'num_tokens': 1074790.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.3125, 'reward/zero_std_group_fraction': 0.65625, 'reward_components/channel_close_count': 0.53125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.46875, 'reward_components/format_valid': 0.375, 'reward_components/native_boundary_valid': 0.53125, 'reward_components/parser_json': 0.015625, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.515625, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.375, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.546875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.3671875, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.359375, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 1.071272850036621, 'sampling/importance_sampling_ratio/mean': 0.5123374462127686, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.785722255706787, 'sampling/sampling_logp_difference/mean': 0.01855107583105564, 'step': 4, 'step_time': 178.95729677181225}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.4375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3831.0, 'completions/mean_length': 2997.921875, 'completions/mean_terminated_length': 2143.861083984375, 'completions/min_length': 950.0, 'completions/min_terminated_length': 950.0, 'entropy': 0.09467202052474022, 'epoch': 0.006510416666666667, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0023277432192116976, 'learning_rate': 1e-05, 'loss': -0.007675642147660255, 'num_tokens': 1331361.0, 'reward': 0.5625, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.5625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.4375, 'reward_components/format_valid': 0.5625, 'reward_components/native_boundary_valid': 0.5625, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.5625, 'reward_components/parser_unsafe': 0.0, 'reward_components/parser_valid': 0.5625, 'reward_components/reward': 0.5625, 'reward_components/runs_correct': 0.890625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.5625, 'reward_std': 0.5, 'rewards/reward_func/mean': 0.5625, 'rewards/reward_func/std': 0.5, 'sampling/importance_sampling_ratio/max': 2.8180346488952637, 'sampling/importance_sampling_ratio/mean': 0.4982297718524933, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.2166824340820312, 'sampling/sampling_logp_difference/mean': 0.013301491737365723, 'step': 5, 'step_time': 174.63352737308014}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.46875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4029.0, 'completions/mean_length': 3091.421875, 'completions/mean_terminated_length': 2205.029541015625, 'completions/min_length': 992.0, 'completions/min_terminated_length': 992.0, 'entropy': 0.0794667536392808, 'epoch': 0.0078125, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.00024385291908401996, 'learning_rate': 1e-05, 'loss': -0.0006516455323435366, 'num_tokens': 1596220.0, 'reward': 0.21875, 'reward/selected_zero_std_group_fraction': 0.2916666666666667, 'reward/zero_std_group_fraction': 0.6458333333333334, 'reward_components/channel_close_count': 0.546875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.46875, 'reward_components/format_valid': 0.21875, 'reward_components/native_boundary_valid': 0.546875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.546875, 'reward_components/parser_unsafe': 0.3125, 'reward_components/parser_valid': 0.21875, 'reward_components/reward': 0.21875, 'reward_components/runs_correct': 0.4375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.21875, 'reward_std': 0.4166666865348816, 'rewards/reward_func/mean': 0.21875, 'rewards/reward_func/std': 0.4166666865348816, 'sampling/importance_sampling_ratio/max': 1.2318096160888672, 'sampling/importance_sampling_ratio/mean': 0.5213725566864014, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.652525424957275, 'sampling/sampling_logp_difference/mean': 0.01827034167945385, 'step': 6, 'step_time': 176.67608823697083}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.328125, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4018.0, 'completions/mean_length': 2612.953125, 'completions/mean_terminated_length': 1888.6744384765625, 'completions/min_length': 510.0, 'completions/min_terminated_length': 510.0, 'entropy': 0.09812331292778254, 'epoch': 0.009114583333333334, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.007505455520004034, 'learning_rate': 1e-05, 'loss': -0.0021405669394880533, 'num_tokens': 1829241.0, 'reward': 0.53125, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.671875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.328125, 'reward_components/format_valid': 0.53125, 'reward_components/native_boundary_valid': 0.671875, 'reward_components/parser_json': 0.109375, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.5625, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.53125, 'reward_components/reward': 0.53125, 'reward_components/runs_correct': 0.828125, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.53125, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.53125, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.4033518135547638, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 9.102952003479004, 'sampling/sampling_logp_difference/mean': 0.013518767431378365, 'step': 7, 'step_time': 166.27224498591386}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.40625, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4045.0, 'completions/mean_length': 3009.421875, 'completions/mean_terminated_length': 2265.9736328125, 'completions/min_length': 805.0, 'completions/min_terminated_length': 805.0, 'entropy': 0.11144792195409536, 'epoch': 0.010416666666666666, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0028186386916786432, 'learning_rate': 1e-05, 'loss': 0.0031182863749563694, 'num_tokens': 2093076.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.21875, 'reward/zero_std_group_fraction': 0.609375, 'reward_components/channel_close_count': 0.59375, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.40625, 'reward_components/format_valid': 0.5, 'reward_components/native_boundary_valid': 0.59375, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.59375, 'reward_components/parser_unsafe': 0.09375, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.921875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.5, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 1.4967069625854492, 'sampling/importance_sampling_ratio/mean': 0.48360365629196167, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.9243125915527344, 'sampling/sampling_logp_difference/mean': 0.015845099464058876, 'step': 8, 'step_time': 176.7978098159656}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.53125, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3904.0, 'completions/mean_length': 3180.3125, 'completions/mean_terminated_length': 2142.533447265625, 'completions/min_length': 722.0, 'completions/min_terminated_length': 722.0, 'entropy': 0.11426979303359985, 'epoch': 0.01171875, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.00018392878700979054, 'learning_rate': 1e-05, 'loss': -0.0003445656329859048, 'num_tokens': 2377064.0, 'reward': 0.453125, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.46875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.53125, 'reward_components/format_valid': 0.453125, 'reward_components/native_boundary_valid': 0.46875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.46875, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.453125, 'reward_components/reward': 0.453125, 'reward_components/runs_correct': 0.734375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.453125, 'reward_std': 0.501733124256134, 'rewards/reward_func/mean': 0.453125, 'rewards/reward_func/std': 0.501733124256134, 'sampling/importance_sampling_ratio/max': 1.9269592761993408, 'sampling/importance_sampling_ratio/mean': 0.6507316827774048, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.993579626083374, 'sampling/sampling_logp_difference/mean': 0.015155397355556488, 'step': 9, 'step_time': 183.32763514318503}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.359375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4057.0, 'completions/mean_length': 2577.953125, 'completions/mean_terminated_length': 1726.3658447265625, 'completions/min_length': 426.0, 'completions/min_terminated_length': 426.0, 'entropy': 0.14673618786036968, 'epoch': 0.013020833333333334, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.000542615307494998, 'learning_rate': 1e-05, 'loss': -0.001980179687961936, 'num_tokens': 2596197.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.3, 'reward/zero_std_group_fraction': 0.65, 'reward_components/channel_close_count': 0.640625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.359375, 'reward_components/format_valid': 0.359375, 'reward_components/native_boundary_valid': 0.640625, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.640625, 'reward_components/parser_unsafe': 0.28125, 'reward_components/parser_valid': 0.359375, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.46875, 'reward_components/runs_total': 1.125, 'reward_components/semantic_correct': 0.359375, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.359375, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 2.0611624717712402, 'sampling/importance_sampling_ratio/mean': 0.46213120222091675, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.446122407913208, 'sampling/sampling_logp_difference/mean': 0.017191965132951736, 'step': 10, 'step_time': 157.8165815101238}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.5, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3812.0, 'completions/mean_length': 2687.515625, 'completions/mean_terminated_length': 1279.03125, 'completions/min_length': 447.0, 'completions/min_terminated_length': 447.0, 'entropy': 0.17564184218645096, 'epoch': 0.014322916666666666, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.01946909911930561, 'learning_rate': 1e-05, 'loss': -0.003934633452445269, 'num_tokens': 2827910.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.3181818181818182, 'reward/zero_std_group_fraction': 0.6590909090909091, 'reward_components/channel_close_count': 0.5, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.5, 'reward_components/format_valid': 0.359375, 'reward_components/native_boundary_valid': 0.5, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.5, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.359375, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.390625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.359375, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.359375, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 1.251930594444275, 'sampling/importance_sampling_ratio/mean': 0.6004859805107117, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 17.22214698791504, 'sampling/sampling_logp_difference/mean': 0.020521041005849838, 'step': 11, 'step_time': 161.99184783909004}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.421875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3902.0, 'completions/mean_length': 2724.078125, 'completions/mean_terminated_length': 1722.946044921875, 'completions/min_length': 694.0, 'completions/min_terminated_length': 694.0, 'entropy': 0.08522575721144676, 'epoch': 0.015625, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.0013244427973404527, 'learning_rate': 1e-05, 'loss': -0.0021439625415951014, 'num_tokens': 3082123.0, 'reward': 0.53125, 'reward/selected_zero_std_group_fraction': 0.3333333333333333, 'reward/zero_std_group_fraction': 0.6666666666666666, 'reward_components/channel_close_count': 0.578125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.421875, 'reward_components/format_valid': 0.53125, 'reward_components/native_boundary_valid': 0.578125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.578125, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.53125, 'reward_components/reward': 0.53125, 'reward_components/runs_correct': 0.65625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.53125, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.53125, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 1.247983455657959, 'sampling/importance_sampling_ratio/mean': 0.5126374959945679, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.970280647277832, 'sampling/sampling_logp_difference/mean': 0.014084962196648121, 'step': 12, 'step_time': 167.04041529202368}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.328125, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3936.0, 'completions/mean_length': 2455.140625, 'completions/mean_terminated_length': 1653.7906494140625, 'completions/min_length': 611.0, 'completions/min_terminated_length': 611.0, 'entropy': 0.13437950238585472, 'epoch': 0.016927083333333332, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0015296951169148088, 'learning_rate': 1e-05, 'loss': -0.0026422361843287945, 'num_tokens': 3293524.0, 'reward': 0.46875, 'reward/selected_zero_std_group_fraction': 0.3076923076923077, 'reward/zero_std_group_fraction': 0.6538461538461539, 'reward_components/channel_close_count': 0.671875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.328125, 'reward_components/format_valid': 0.546875, 'reward_components/native_boundary_valid': 0.671875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.671875, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.546875, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 0.59375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.46875, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.9824352264404297, 'sampling/importance_sampling_ratio/mean': 0.4512457251548767, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 10.88036823272705, 'sampling/sampling_logp_difference/mean': 0.015610691159963608, 'step': 13, 'step_time': 157.10858551599085}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.234375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3142.0, 'completions/mean_length': 2129.546875, 'completions/mean_terminated_length': 1527.5714111328125, 'completions/min_length': 465.0, 'completions/min_terminated_length': 465.0, 'entropy': 0.07177048269659281, 'epoch': 0.018229166666666668, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.0021202724892646074, 'learning_rate': 1e-05, 'loss': -0.008725866675376892, 'num_tokens': 3508663.0, 'reward': 0.65625, 'reward/selected_zero_std_group_fraction': 0.30357142857142855, 'reward/zero_std_group_fraction': 0.6517857142857143, 'reward_components/channel_close_count': 0.765625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.234375, 'reward_components/format_valid': 0.65625, 'reward_components/native_boundary_valid': 0.765625, 'reward_components/parser_json': 0.015625, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.75, 'reward_components/parser_unsafe': 0.109375, 'reward_components/parser_valid': 0.65625, 'reward_components/reward': 0.65625, 'reward_components/runs_correct': 0.890625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.65625, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.65625, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 2.031008720397949, 'sampling/importance_sampling_ratio/mean': 0.43108054995536804, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 13.054469108581543, 'sampling/sampling_logp_difference/mean': 0.014828605577349663, 'step': 14, 'step_time': 152.63235335214995}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.421875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3681.0, 'completions/mean_length': 2637.625, 'completions/mean_terminated_length': 1573.4053955078125, 'completions/min_length': 687.0, 'completions/min_terminated_length': 687.0, 'entropy': 0.10809349920600653, 'epoch': 0.01953125, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.01583734340965748, 'learning_rate': 1e-05, 'loss': -0.0400751456618309, 'num_tokens': 3735583.0, 'reward': 0.578125, 'reward/selected_zero_std_group_fraction': 0.3, 'reward/zero_std_group_fraction': 0.65, 'reward_components/channel_close_count': 0.578125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.421875, 'reward_components/format_valid': 0.578125, 'reward_components/native_boundary_valid': 0.578125, 'reward_components/parser_json': 0.09375, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.484375, 'reward_components/parser_unsafe': 0.0, 'reward_components/parser_valid': 0.578125, 'reward_components/reward': 0.578125, 'reward_components/runs_correct': 0.703125, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.578125, 'reward_std': 0.49776285886764526, 'rewards/reward_func/mean': 0.578125, 'rewards/reward_func/std': 0.49776285886764526, 'sampling/importance_sampling_ratio/max': 2.6422102451324463, 'sampling/importance_sampling_ratio/mean': 0.6450210809707642, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 10.47492504119873, 'sampling/sampling_logp_difference/mean': 0.014680149964988232, 'step': 15, 'step_time': 158.13342921703588}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.265625, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3296.0, 'completions/mean_length': 2196.59375, 'completions/mean_terminated_length': 1509.574462890625, 'completions/min_length': 405.0, 'completions/min_terminated_length': 405.0, 'entropy': 0.0970615679398179, 'epoch': 0.020833333333333332, 'frac_reward_zero_std': 0.375, 'grad_norm': 0.0014741194900125265, 'learning_rate': 1e-05, 'loss': -0.00523255905136466, 'num_tokens': 3945221.0, 'reward': 0.609375, 'reward/selected_zero_std_group_fraction': 0.28125, 'reward/zero_std_group_fraction': 0.6328125, 'reward_components/channel_close_count': 0.734375, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.265625, 'reward_components/format_valid': 0.609375, 'reward_components/native_boundary_valid': 0.734375, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.734375, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.609375, 'reward_components/reward': 0.609375, 'reward_components/runs_correct': 0.90625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.609375, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.609375, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 1.4811298847198486, 'sampling/importance_sampling_ratio/mean': 0.35813361406326294, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.191697597503662, 'sampling/sampling_logp_difference/mean': 0.01573760434985161, 'step': 16, 'step_time': 151.0304883638164}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.25, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3899.0, 'completions/mean_length': 2266.140625, 'completions/mean_terminated_length': 1656.1875, 'completions/min_length': 499.0, 'completions/min_terminated_length': 499.0, 'entropy': 0.14947441406548023, 'epoch': 0.022135416666666668, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0339917354285717, 'learning_rate': 1e-05, 'loss': -0.018607715144753456, 'num_tokens': 4146766.0, 'reward': 0.53125, 'reward/selected_zero_std_group_fraction': 0.0, 'reward/zero_std_group_fraction': 0.5, 'reward_components/channel_close_count': 0.75, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.25, 'reward_components/format_valid': 0.546875, 'reward_components/native_boundary_valid': 0.75, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.75, 'reward_components/parser_unsafe': 0.171875, 'reward_components/parser_valid': 0.546875, 'reward_components/reward': 0.53125, 'reward_components/runs_correct': 0.640625, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.53125, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.53125, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.923262357711792, 'sampling/importance_sampling_ratio/mean': 0.4647424817085266, 'sampling/importance_sampling_ratio/min': 3.455766506288427e-10, 'sampling/sampling_logp_difference/max': 20.009214401245117, 'sampling/sampling_logp_difference/mean': 0.01694958284497261, 'step': 17, 'step_time': 165.81407095189206}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.234375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3917.0, 'completions/mean_length': 2185.03125, 'completions/mean_terminated_length': 1600.040771484375, 'completions/min_length': 656.0, 'completions/min_terminated_length': 656.0, 'entropy': 0.10460947267711163, 'epoch': 0.0234375, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.00048010883620008826, 'learning_rate': 1e-05, 'loss': -0.0010868092067539692, 'num_tokens': 4349648.0, 'reward': 0.65625, 'reward/selected_zero_std_group_fraction': 0.125, 'reward/zero_std_group_fraction': 0.5625, 'reward_components/channel_close_count': 0.765625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.234375, 'reward_components/format_valid': 0.6875, 'reward_components/native_boundary_valid': 0.765625, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.640625, 'reward_components/parser_unsafe': 0.0625, 'reward_components/parser_valid': 0.6875, 'reward_components/reward': 0.65625, 'reward_components/runs_correct': 0.90625, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.671875, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.65625, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 1.7622472047805786, 'sampling/importance_sampling_ratio/mean': 0.3137361705303192, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 12.053547859191895, 'sampling/sampling_logp_difference/mean': 0.015630807727575302, 'step': 18, 'step_time': 148.41101996705402}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.28125, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3912.0, 'completions/mean_length': 2387.78125, 'completions/mean_terminated_length': 1719.347900390625, 'completions/min_length': 517.0, 'completions/min_terminated_length': 517.0, 'entropy': 0.14741394761949778, 'epoch': 0.024739583333333332, 'frac_reward_zero_std': 0.25, 'grad_norm': 0.006646530237048864, 'learning_rate': 1e-05, 'loss': -0.003316863439977169, 'num_tokens': 4581698.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.08333333333333333, 'reward/zero_std_group_fraction': 0.4583333333333333, 'reward_components/channel_close_count': 0.71875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.28125, 'reward_components/format_valid': 0.515625, 'reward_components/native_boundary_valid': 0.71875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.71875, 'reward_components/parser_unsafe': 0.203125, 'reward_components/parser_valid': 0.515625, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.6875, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.5, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 1.0671097040176392, 'sampling/importance_sampling_ratio/mean': 0.3408268988132477, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 9.353429794311523, 'sampling/sampling_logp_difference/mean': 0.01983843371272087, 'step': 19, 'step_time': 161.5163500001654}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.21875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3815.0, 'completions/mean_length': 2250.984375, 'completions/mean_terminated_length': 1734.3800048828125, 'completions/min_length': 501.0, 'completions/min_terminated_length': 501.0, 'entropy': 0.12713784631341696, 'epoch': 0.026041666666666668, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0006752036861144006, 'learning_rate': 1e-05, 'loss': -0.0017371552530676126, 'num_tokens': 4791745.0, 'reward': 0.71875, 'reward/selected_zero_std_group_fraction': 0.0625, 'reward/zero_std_group_fraction': 0.46875, 'reward_components/channel_close_count': 0.78125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.21875, 'reward_components/format_valid': 0.78125, 'reward_components/native_boundary_valid': 0.78125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.78125, 'reward_components/parser_unsafe': 0.0, 'reward_components/parser_valid': 0.78125, 'reward_components/reward': 0.71875, 'reward_components/runs_correct': 1.140625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.71875, 'reward_std': 0.4531635046005249, 'rewards/reward_func/mean': 0.71875, 'rewards/reward_func/std': 0.4531635046005249, 'sampling/importance_sampling_ratio/max': 1.779668927192688, 'sampling/importance_sampling_ratio/mean': 0.3335704505443573, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.5490639209747314, 'sampling/sampling_logp_difference/mean': 0.014014334417879581, 'step': 20, 'step_time': 149.13405402109493}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.1875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3537.0, 'completions/mean_length': 2004.859375, 'completions/mean_terminated_length': 1522.28857421875, 'completions/min_length': 771.0, 'completions/min_terminated_length': 771.0, 'entropy': 0.10058648930862546, 'epoch': 0.02734375, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 4971000.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.575, 'reward_components/channel_close_count': 0.8125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.1875, 'reward_components/format_valid': 0.5, 'reward_components/native_boundary_valid': 0.8125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.8125, 'reward_components/parser_unsafe': 0.3125, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.625, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.5, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 2.6407172679901123, 'sampling/importance_sampling_ratio/mean': 0.33751243352890015, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 24.552444458007812, 'sampling/sampling_logp_difference/mean': 0.016265638172626495, 'step': 21, 'step_time': 142.5607630573213}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.3125, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 4083.0, 'completions/mean_length': 2677.203125, 'completions/mean_terminated_length': 2032.2955322265625, 'completions/min_length': 828.0, 'completions/min_terminated_length': 828.0, 'entropy': 0.09389743208885193, 'epoch': 0.028645833333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.0005022212862968445, 'learning_rate': 1e-05, 'loss': -0.0016652453923597932, 'num_tokens': 5210373.0, 'reward': 0.65625, 'reward/selected_zero_std_group_fraction': 0.2916666666666667, 'reward/zero_std_group_fraction': 0.6041666666666666, 'reward_components/channel_close_count': 0.6875, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.3125, 'reward_components/format_valid': 0.65625, 'reward_components/native_boundary_valid': 0.6875, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.6875, 'reward_components/parser_unsafe': 0.03125, 'reward_components/parser_valid': 0.65625, 'reward_components/reward': 0.65625, 'reward_components/runs_correct': 1.3125, 'reward_components/runs_total': 2.0, 'reward_components/semantic_correct': 0.65625, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.65625, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 2.26767897605896, 'sampling/importance_sampling_ratio/mean': 0.47983866930007935, 'sampling/importance_sampling_ratio/min': 5.547015825868584e-05, 'sampling/sampling_logp_difference/max': 2.6758079528808594, 'sampling/sampling_logp_difference/mean': 0.011174303479492664, 'step': 22, 'step_time': 167.99802872806322}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.46875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3728.0, 'completions/mean_length': 2589.140625, 'completions/mean_terminated_length': 1259.558837890625, 'completions/min_length': 474.0, 'completions/min_terminated_length': 474.0, 'entropy': 0.13536358252167702, 'epoch': 0.029947916666666668, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.000490852864459157, 'learning_rate': 1e-05, 'loss': -0.0008431912865489721, 'num_tokens': 5445518.0, 'reward': 0.390625, 'reward/selected_zero_std_group_fraction': 0.2857142857142857, 'reward/zero_std_group_fraction': 0.6071428571428571, 'reward_components/channel_close_count': 0.53125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.46875, 'reward_components/format_valid': 0.390625, 'reward_components/native_boundary_valid': 0.53125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.53125, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.390625, 'reward_components/reward': 0.390625, 'reward_components/runs_correct': 0.421875, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.390625, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.390625, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 1.7340031862258911, 'sampling/importance_sampling_ratio/mean': 0.6195780038833618, 'sampling/importance_sampling_ratio/min': 6.511048923130147e-06, 'sampling/sampling_logp_difference/max': 4.814200401306152, 'sampling/sampling_logp_difference/mean': 0.015022839419543743, 'step': 23, 'step_time': 164.37902907596435}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.484375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3903.0, 'completions/mean_length': 3020.859375, 'completions/mean_terminated_length': 2010.87890625, 'completions/min_length': 795.0, 'completions/min_terminated_length': 795.0, 'entropy': 0.12369069922715425, 'epoch': 0.03125, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.0035190910566598177, 'learning_rate': 1e-05, 'loss': -0.0007988286088220775, 'num_tokens': 5706821.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.28125, 'reward/zero_std_group_fraction': 0.609375, 'reward_components/channel_close_count': 0.515625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.484375, 'reward_components/format_valid': 0.5, 'reward_components/native_boundary_valid': 0.515625, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.515625, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.890625, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.5, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 2.5913801193237305, 'sampling/importance_sampling_ratio/mean': 0.5817782878875732, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 20.54939079284668, 'sampling/sampling_logp_difference/mean': 0.016912346705794334, 'step': 24, 'step_time': 173.83616784703918}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.1875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3603.0, 'completions/mean_length': 2381.0, 'completions/mean_terminated_length': 1985.2308349609375, 'completions/min_length': 964.0, 'completions/min_terminated_length': 964.0, 'entropy': 0.09716686978936195, 'epoch': 0.032552083333333336, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.019776809960603714, 'learning_rate': 1e-05, 'loss': -0.040387801826000214, 'num_tokens': 5938181.0, 'reward': 0.765625, 'reward/selected_zero_std_group_fraction': 0.3055555555555556, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.8125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.1875, 'reward_components/format_valid': 0.765625, 'reward_components/native_boundary_valid': 0.8125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.8125, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.765625, 'reward_components/reward': 0.765625, 'reward_components/runs_correct': 1.5, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.765625, 'reward_std': 0.42695629596710205, 'rewards/reward_func/mean': 0.765625, 'rewards/reward_func/std': 0.42695629596710205, 'sampling/importance_sampling_ratio/max': 2.621847629547119, 'sampling/importance_sampling_ratio/mean': 0.3370857536792755, 'sampling/importance_sampling_ratio/min': 9.827195057710014e-09, 'sampling/sampling_logp_difference/max': 15.487744331359863, 'sampling/sampling_logp_difference/mean': 0.011683451943099499, 'step': 25, 'step_time': 154.03122657514177}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.4375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3714.0, 'completions/mean_length': 2818.5, 'completions/mean_terminated_length': 1907.361083984375, 'completions/min_length': 678.0, 'completions/min_terminated_length': 678.0, 'entropy': 0.13596281968057156, 'epoch': 0.033854166666666664, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.006161811761558056, 'learning_rate': 1e-05, 'loss': -0.007145062554627657, 'num_tokens': 6175205.0, 'reward': 0.3125, 'reward/selected_zero_std_group_fraction': 0.275, 'reward/zero_std_group_fraction': 0.6125, 'reward_components/channel_close_count': 0.578125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.421875, 'reward_components/format_valid': 0.3125, 'reward_components/native_boundary_valid': 0.5625, 'reward_components/parser_json': 0.109375, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.453125, 'reward_components/parser_unsafe': 0.25, 'reward_components/parser_valid': 0.3125, 'reward_components/reward': 0.3125, 'reward_components/runs_correct': 0.375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.3125, 'reward_std': 0.467176616191864, 'rewards/reward_func/mean': 0.3125, 'rewards/reward_func/std': 0.467176616191864, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.4888877868652344, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.106573581695557, 'sampling/sampling_logp_difference/mean': 0.016054607927799225, 'step': 26, 'step_time': 165.6563846381614}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.390625, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3895.0, 'completions/mean_length': 2764.3125, 'completions/mean_terminated_length': 1910.666748046875, 'completions/min_length': 677.0, 'completions/min_terminated_length': 677.0, 'entropy': 0.11075056064873934, 'epoch': 0.03515625, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.0015133855631574988, 'learning_rate': 1e-05, 'loss': -0.001269573695026338, 'num_tokens': 6427961.0, 'reward': 0.375, 'reward/selected_zero_std_group_fraction': 0.2727272727272727, 'reward/zero_std_group_fraction': 0.6136363636363636, 'reward_components/channel_close_count': 0.609375, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.390625, 'reward_components/format_valid': 0.375, 'reward_components/native_boundary_valid': 0.609375, 'reward_components/parser_json': 0.109375, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.5, 'reward_components/parser_unsafe': 0.234375, 'reward_components/parser_valid': 0.375, 'reward_components/reward': 0.375, 'reward_components/runs_correct': 0.515625, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.375, 'reward_std': 0.48795005679130554, 'rewards/reward_func/mean': 0.375, 'rewards/reward_func/std': 0.48795005679130554, 'sampling/importance_sampling_ratio/max': 1.0, 'sampling/importance_sampling_ratio/mean': 0.4292636513710022, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.178304195404053, 'sampling/sampling_logp_difference/mean': 0.016621198505163193, 'step': 27, 'step_time': 170.2280247291783}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.421875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3760.0, 'completions/mean_length': 2842.953125, 'completions/mean_terminated_length': 1928.567626953125, 'completions/min_length': 409.0, 'completions/min_terminated_length': 409.0, 'entropy': 0.12097175605595112, 'epoch': 0.036458333333333336, 'frac_reward_zero_std': 0.375, 'grad_norm': 0.001287449849769473, 'learning_rate': 1e-05, 'loss': -0.004199882969260216, 'num_tokens': 6677622.0, 'reward': 0.515625, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.59375, 'reward_components/channel_close_count': 0.578125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.421875, 'reward_components/format_valid': 0.53125, 'reward_components/native_boundary_valid': 0.578125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.578125, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.53125, 'reward_components/reward': 0.515625, 'reward_components/runs_correct': 0.90625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.5234375, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.515625, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 1.4424225091934204, 'sampling/importance_sampling_ratio/mean': 0.4959092438220978, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 8.406841278076172, 'sampling/sampling_logp_difference/mean': 0.013908586464822292, 'step': 28, 'step_time': 170.2047199900262}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.46875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3758.0, 'completions/mean_length': 2673.703125, 'completions/mean_terminated_length': 1418.7353515625, 'completions/min_length': 677.0, 'completions/min_terminated_length': 677.0, 'entropy': 0.12914463598281145, 'epoch': 0.037760416666666664, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.00100526527967304, 'learning_rate': 1e-05, 'loss': -0.0018889743369072676, 'num_tokens': 6922979.0, 'reward': 0.46875, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.5961538461538461, 'reward_components/channel_close_count': 0.53125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.46875, 'reward_components/format_valid': 0.515625, 'reward_components/native_boundary_valid': 0.53125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.53125, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.515625, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 0.5, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.46875, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 1.7147115468978882, 'sampling/importance_sampling_ratio/mean': 0.527077317237854, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.426498889923096, 'sampling/sampling_logp_difference/mean': 0.015505732037127018, 'step': 29, 'step_time': 167.61777491483372}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.21875, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3917.0, 'completions/mean_length': 2608.796875, 'completions/mean_terminated_length': 2192.3798828125, 'completions/min_length': 729.0, 'completions/min_terminated_length': 729.0, 'entropy': 0.11506192665547132, 'epoch': 0.0390625, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.0010873839491978288, 'learning_rate': 1e-05, 'loss': 0.0022077029570937157, 'num_tokens': 7167766.0, 'reward': 0.5625, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.5982142857142857, 'reward_components/channel_close_count': 0.78125, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.21875, 'reward_components/format_valid': 0.5625, 'reward_components/native_boundary_valid': 0.78125, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.78125, 'reward_components/parser_unsafe': 0.21875, 'reward_components/parser_valid': 0.5625, 'reward_components/reward': 0.5625, 'reward_components/runs_correct': 1.09375, 'reward_components/runs_total': 1.75, 'reward_components/semantic_correct': 0.5625, 'reward_std': 0.5, 'rewards/reward_func/mean': 0.5625, 'rewards/reward_func/std': 0.5, 'sampling/importance_sampling_ratio/max': 2.3820912837982178, 'sampling/importance_sampling_ratio/mean': 0.33479487895965576, 'sampling/importance_sampling_ratio/min': 4.2296561808541355e-09, 'sampling/sampling_logp_difference/max': 17.130220413208008, 'sampling/sampling_logp_difference/mean': 0.014050661586225033, 'step': 30, 'step_time': 168.35775536298752}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.109375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3767.0, 'completions/mean_length': 1807.546875, 'completions/mean_terminated_length': 1526.5087890625, 'completions/min_length': 586.0, 'completions/min_terminated_length': 586.0, 'entropy': 0.0875889181625098, 'epoch': 0.040364583333333336, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.00391038553789258, 'learning_rate': 1e-05, 'loss': -0.013786152005195618, 'num_tokens': 7346873.0, 'reward': 0.875, 'reward/selected_zero_std_group_fraction': 0.26666666666666666, 'reward/zero_std_group_fraction': 0.6083333333333333, 'reward_components/channel_close_count': 0.90625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.109375, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 0.90625, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.90625, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.875, 'reward_components/runs_correct': 1.203125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.875, 'reward_std': 0.3333333432674408, 'rewards/reward_func/mean': 0.875, 'rewards/reward_func/std': 0.3333333432674408, 'sampling/importance_sampling_ratio/max': 2.9470725059509277, 'sampling/importance_sampling_ratio/mean': 0.42610615491867065, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.992341041564941, 'sampling/sampling_logp_difference/mean': 0.014681197702884674, 'step': 31, 'step_time': 143.634825425921}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.109375, 'completions/max_length': 4096.0, 'completions/max_terminated_length': 3941.0, 'completions/mean_length': 2077.484375, 'completions/mean_terminated_length': 1829.5965576171875, 'completions/min_length': 618.0, 'completions/min_terminated_length': 618.0, 'entropy': 0.1405503125861287, 'epoch': 0.041666666666666664, 'frac_reward_zero_std': 0.375, 'grad_norm': 0.000579166051466018, 'learning_rate': 1e-05, 'loss': -0.0024773390032351017, 'num_tokens': 7536216.0, 'reward': 0.703125, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.59375, 'reward_components/channel_close_count': 0.890625, 'reward_components/channel_open_count': 1.0, 'reward_components/completion_truncated': 0.109375, 'reward_components/format_valid': 0.71875, 'reward_components/native_boundary_valid': 0.890625, 'reward_components/parser_json': 0.09375, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.796875, 'reward_components/parser_unsafe': 0.171875, 'reward_components/parser_valid': 0.71875, 'reward_components/reward': 0.703125, 'reward_components/runs_correct': 0.9375, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.703125, 'reward_std': 0.4604927599430084, 'rewards/reward_func/mean': 0.703125, 'rewards/reward_func/std': 0.4604927599430084, 'sampling/importance_sampling_ratio/max': 2.4597437381744385, 'sampling/importance_sampling_ratio/mean': 0.2013377547264099, 'sampling/importance_sampling_ratio/min': 6.18775159149515e-13, 'sampling/sampling_logp_difference/max': 11.649497985839844, 'sampling/sampling_logp_difference/mean': 0.016497939825057983, 'step': 32, 'step_time': 147.15360558091197}], best_metric=None, best_global_step=None, best_model_checkpoint=None, is_local_process_zero=True, is_world_process_zero=True, is_hyper_param_search=False, trial_name=None, trial_params=None, stateful_callbacks={'TrainerControl': {'args': {'should_epoch_stop': False, 'should_evaluate': False, 'should_log': False, 'should_save': True, 'should_training_stop': True}, 'attributes': {}}})",
+ "note": "completion_ids removed; regenerate by tokenising completion_raw_text with the gemma4-26b-a4b tokenizer (verified exact on 221 records). log_extra/log_metric were bound-method reprs. Fetched as parallel byte ranges because one connection to this repo sustains only 1.4 MiB/s, and because the Hub's declared size for some files disagrees with their stored bytes."
+}