rollouts: charter-direct-run2/charter-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl provenance

rollouts/charter-direct-run2/charter-direct-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": "charter-direct-run2/charter-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl",
+ "declared_bytes": 34213281212,
+ "streamed_bytes": 34213281212,
+ "sha256_streamed": "4d926922f849033c7ca13d9896e8abe8edb97bf79cfadddbc6293b29aac07c00",
+ "sha256_expected": "4d926922f849033c7ca13d9896e8abe8edb97bf79cfadddbc6293b29aac07c00",
+ "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=2281993, 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': 0.0, 'completions/max_length': 183.0, 'completions/max_terminated_length': 183.0, 'completions/mean_length': 44.515625, 'completions/mean_terminated_length': 44.515625, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.09303654730319977, 'epoch': 0.0013020833333333333, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.0500573068857193, 'learning_rate': 1e-05, 'loss': 0.0008304134244099259, 'num_tokens': 62433.0, 'reward': 0.5625, 'reward/selected_zero_std_group_fraction': 0.0, 'reward/zero_std_group_fraction': 0.5, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.8125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.171875, 'reward_components/parser_valid': 0.8125, 'reward_components/reward': 0.5625, 'reward_components/runs_correct': 1.15625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.6875, 'reward_std': 0.5, 'rewards/reward_func/mean': 0.5625, 'rewards/reward_func/std': 0.5, 'sampling/importance_sampling_ratio/max': 2.6301701068878174, 'sampling/importance_sampling_ratio/mean': 0.6844139695167542, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.227639675140381, 'sampling/sampling_logp_difference/mean': 0.04235227406024933, 'step': 1, 'step_time': 15.837826842907816}, {'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': 0.0, 'completions/max_length': 64.0, 'completions/max_terminated_length': 64.0, 'completions/mean_length': 27.609375, 'completions/mean_terminated_length': 27.609375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.039884528843685985, 'epoch': 0.0026041666666666665, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03617817908525467, 'learning_rate': 1e-05, 'loss': -0.0010686604073271155, 'num_tokens': 141576.0, 'reward': 0.65625, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.65625, 'reward_components/runs_correct': 0.875, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.703125, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.65625, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 1.764050841331482, 'sampling/importance_sampling_ratio/mean': 0.7105439305305481, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.889498710632324, 'sampling/sampling_logp_difference/mean': 0.06447670608758926, 'step': 2, 'step_time': 15.016032165964134}, {'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': 0.0, 'completions/max_length': 75.0, 'completions/max_terminated_length': 75.0, 'completions/mean_length': 34.484375, 'completions/mean_terminated_length': 34.484375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.14413694432005286, 'epoch': 0.00390625, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.053939398378133774, 'learning_rate': 1e-05, 'loss': -0.0024253963492810726, 'num_tokens': 200935.0, 'reward': 0.609375, 'reward/selected_zero_std_group_fraction': 0.3333333333333333, 'reward/zero_std_group_fraction': 0.6666666666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.859375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.859375, 'reward_components/reward': 0.609375, 'reward_components/runs_correct': 1.109375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.671875, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.609375, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 2.554532766342163, 'sampling/importance_sampling_ratio/mean': 0.7010618448257446, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.251251220703125, 'sampling/sampling_logp_difference/mean': 0.05141923204064369, 'step': 3, 'step_time': 11.858036200981587}, {'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': 0.0, 'completions/max_length': 158.0, 'completions/max_terminated_length': 158.0, 'completions/mean_length': 35.5625, 'completions/mean_terminated_length': 35.5625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.09382735099643469, 'epoch': 0.005208333333333333, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.03262415528297424, 'learning_rate': 1e-05, 'loss': -0.0012213281588628888, 'num_tokens': 280459.0, 'reward': 0.515625, 'reward/selected_zero_std_group_fraction': 0.4375, 'reward/zero_std_group_fraction': 0.71875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.515625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.484375, 'reward_components/parser_valid': 0.515625, 'reward_components/reward': 0.515625, 'reward_components/runs_correct': 0.78125, 'reward_components/runs_total': 1.625, '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': 2.200551748275757, 'sampling/importance_sampling_ratio/mean': 0.7684018611907959, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.2495975494384766, 'sampling/sampling_logp_difference/mean': 0.03753465414047241, 'step': 4, 'step_time': 15.941707275109366}, {'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': 0.0, 'completions/max_length': 305.0, 'completions/max_terminated_length': 305.0, 'completions/mean_length': 61.875, 'completions/mean_terminated_length': 61.875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.17257735319435596, 'epoch': 0.006510416666666667, 'frac_reward_zero_std': 0.375, 'grad_norm': 0.0520184263586998, 'learning_rate': 1e-05, 'loss': 0.004880460910499096, 'num_tokens': 348931.0, 'reward': 0.59375, 'reward/selected_zero_std_group_fraction': 0.35, 'reward/zero_std_group_fraction': 0.65, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.71875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.015625, 'reward_components/parser_natural': 0.984375, 'reward_components/parser_unsafe': 0.28125, 'reward_components/parser_valid': 0.71875, 'reward_components/reward': 0.59375, 'reward_components/runs_correct': 1.0625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.65625, 'reward_std': 0.49501484632492065, 'rewards/reward_func/mean': 0.59375, 'rewards/reward_func/std': 0.49501484632492065, 'sampling/importance_sampling_ratio/max': 2.3793582916259766, 'sampling/importance_sampling_ratio/mean': 0.7962399125099182, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.201749801635742, 'sampling/sampling_logp_difference/mean': 0.04359675943851471, 'step': 5, 'step_time': 17.497907764976844}, {'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': 0.0, 'completions/max_length': 150.0, 'completions/max_terminated_length': 150.0, 'completions/mean_length': 37.359375, 'completions/mean_terminated_length': 37.359375, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.0586118622450158, 'epoch': 0.0078125, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.012652957811951637, 'learning_rate': 1e-05, 'loss': -0.0004533091851044446, 'num_tokens': 418138.0, 'reward': 0.484375, 'reward/selected_zero_std_group_fraction': 0.4166666666666667, 'reward/zero_std_group_fraction': 0.6875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.640625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.359375, 'reward_components/parser_valid': 0.640625, 'reward_components/reward': 0.484375, 'reward_components/runs_correct': 0.859375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.4921875, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.484375, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 2.558795213699341, 'sampling/importance_sampling_ratio/mean': 0.8477057814598083, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.1883955001831055, 'sampling/sampling_logp_difference/mean': 0.024992981925606728, 'step': 6, 'step_time': 14.179431472206488}, {'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': 0.0, 'completions/max_length': 77.0, 'completions/max_terminated_length': 77.0, 'completions/mean_length': 32.34375, 'completions/mean_terminated_length': 32.34375, 'completions/min_length': 14.0, 'completions/min_terminated_length': 14.0, 'entropy': 0.11707124533131719, 'epoch': 0.009114583333333334, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.01990974321961403, 'learning_rate': 1e-05, 'loss': -0.004664935637265444, 'num_tokens': 485808.0, 'reward': 0.546875, 'reward/selected_zero_std_group_fraction': 0.42857142857142855, 'reward/zero_std_group_fraction': 0.6964285714285714, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.96875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.03125, 'reward_components/parser_valid': 0.96875, 'reward_components/reward': 0.546875, 'reward_components/runs_correct': 1.109375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.6796875, 'reward_std': 0.501733124256134, 'rewards/reward_func/mean': 0.546875, 'rewards/reward_func/std': 0.501733124256134, 'sampling/importance_sampling_ratio/max': 2.880894660949707, 'sampling/importance_sampling_ratio/mean': 0.8921729326248169, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.3147506713867188, 'sampling/sampling_logp_difference/mean': 0.028512215241789818, 'step': 7, 'step_time': 12.624557440984063}, {'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': 0.0, 'completions/max_length': 81.0, 'completions/max_terminated_length': 81.0, 'completions/mean_length': 30.796875, 'completions/mean_terminated_length': 30.796875, 'completions/min_length': 10.0, 'completions/min_terminated_length': 10.0, 'entropy': 0.07878091558814049, 'epoch': 0.010416666666666666, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 558819.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.5, 'reward/zero_std_group_fraction': 0.734375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.75, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.25, 'reward_components/parser_valid': 0.75, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 1.125, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.625, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 2.129298686981201, 'sampling/importance_sampling_ratio/mean': 0.930873453617096, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.603330612182617, 'sampling/sampling_logp_difference/mean': 0.030032750219106674, 'step': 8, 'step_time': 13.811415073229}, {'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': 0.0625, 'completions/max_length': 512.0, 'completions/max_terminated_length': 427.0, 'completions/mean_length': 69.3125, 'completions/mean_terminated_length': 39.80000305175781, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.07240454899147153, 'epoch': 0.01171875, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.05933514982461929, 'learning_rate': 1e-05, 'loss': -0.001345925615169108, 'num_tokens': 643511.0, 'reward': 0.640625, 'reward/selected_zero_std_group_fraction': 0.5, 'reward/zero_std_group_fraction': 0.7361111111111112, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0625, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.265625, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.640625, 'reward_components/runs_correct': 1.0, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.640625, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.640625, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 2.852860689163208, 'sampling/importance_sampling_ratio/mean': 0.7364717721939087, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.266497611999512, 'sampling/sampling_logp_difference/mean': 0.03286438807845116, 'step': 9, 'step_time': 23.690302381874062}, {'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': 0.0, 'completions/max_length': 257.0, 'completions/max_terminated_length': 257.0, 'completions/mean_length': 49.71875, 'completions/mean_terminated_length': 49.71875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10866248467937112, 'epoch': 0.013020833333333334, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03049420192837715, 'learning_rate': 1e-05, 'loss': -0.007096492685377598, 'num_tokens': 700645.0, 'reward': 0.328125, 'reward/selected_zero_std_group_fraction': 0.5, 'reward/zero_std_group_fraction': 0.7375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.34375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.015625, 'reward_components/parser_natural': 0.984375, 'reward_components/parser_unsafe': 0.609375, 'reward_components/parser_valid': 0.34375, 'reward_components/reward': 0.328125, 'reward_components/runs_correct': 0.453125, 'reward_components/runs_total': 1.125, 'reward_components/semantic_correct': 0.328125, 'reward_std': 0.4732423722743988, 'rewards/reward_func/mean': 0.328125, 'rewards/reward_func/std': 0.4732423722743988, 'sampling/importance_sampling_ratio/max': 2.1593215465545654, 'sampling/importance_sampling_ratio/mean': 0.5731668472290039, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 9.997025489807129, 'sampling/sampling_logp_difference/mean': 0.047241128981113434, 'step': 10, 'step_time': 13.953151731053367}, {'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': 0.0, 'completions/max_length': 135.0, 'completions/max_terminated_length': 135.0, 'completions/mean_length': 41.859375, 'completions/mean_terminated_length': 41.859375, 'completions/min_length': 10.0, 'completions/min_terminated_length': 10.0, 'entropy': 0.07850417867302895, 'epoch': 0.014322916666666666, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.005227868445217609, 'learning_rate': 1e-05, 'loss': -0.0016488409601151943, 'num_tokens': 762844.0, 'reward': 0.609375, 'reward/selected_zero_std_group_fraction': 0.5227272727272727, 'reward/zero_std_group_fraction': 0.75, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.78125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.21875, 'reward_components/parser_valid': 0.78125, 'reward_components/reward': 0.609375, 'reward_components/runs_correct': 0.734375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.6171875, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.609375, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 1.6967943906784058, 'sampling/importance_sampling_ratio/mean': 0.6996966600418091, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 8.862066268920898, 'sampling/sampling_logp_difference/mean': 0.06928723305463791, 'step': 11, 'step_time': 12.549372115987353}, {'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': 0.0, 'completions/max_length': 156.0, 'completions/max_terminated_length': 156.0, 'completions/mean_length': 19.046875, 'completions/mean_terminated_length': 19.046875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.0615125538315624, 'epoch': 0.015625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.010140422731637955, 'learning_rate': 1e-05, 'loss': -0.0006115362048149109, 'num_tokens': 843743.0, 'reward': 0.71875, 'reward/selected_zero_std_group_fraction': 0.5416666666666666, 'reward/zero_std_group_fraction': 0.7604166666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.84375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.84375, 'reward_components/reward': 0.71875, 'reward_components/runs_correct': 0.96875, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.78125, 'reward_std': 0.4531635046005249, 'rewards/reward_func/mean': 0.71875, 'rewards/reward_func/std': 0.4531635046005249, 'sampling/importance_sampling_ratio/max': 2.3292641639709473, 'sampling/importance_sampling_ratio/mean': 0.8091589212417603, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.2647132873535156, 'sampling/sampling_logp_difference/mean': 0.04155777022242546, 'step': 12, 'step_time': 15.754798809881322}, {'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': 0.0, 'completions/max_length': 87.0, 'completions/max_terminated_length': 87.0, 'completions/mean_length': 21.453125, 'completions/mean_terminated_length': 21.453125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.04960785433650017, 'epoch': 0.016927083333333332, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.007238308899104595, 'learning_rate': 1e-05, 'loss': 0.0022613827604800463, 'num_tokens': 899196.0, 'reward': 0.515625, 'reward/selected_zero_std_group_fraction': 0.5576923076923077, 'reward/zero_std_group_fraction': 0.7692307692307693, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.890625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.109375, 'reward_components/parser_valid': 0.890625, 'reward_components/reward': 0.515625, 'reward_components/runs_correct': 0.78125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.578125, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.515625, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 2.127866506576538, 'sampling/importance_sampling_ratio/mean': 0.7953990697860718, 'sampling/importance_sampling_ratio/min': 0.00045816091005690396, 'sampling/sampling_logp_difference/max': 8.259763717651367, 'sampling/sampling_logp_difference/mean': 0.07180717587471008, 'step': 13, 'step_time': 10.646788284066133}, {'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': 0.0, 'completions/max_length': 72.0, 'completions/max_terminated_length': 72.0, 'completions/mean_length': 24.421875, 'completions/mean_terminated_length': 24.421875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.025012595462612808, 'epoch': 0.018229166666666668, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 979415.0, 'reward': 0.875, 'reward/selected_zero_std_group_fraction': 0.5892857142857143, 'reward/zero_std_group_fraction': 0.7857142857142857, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.078125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.875, 'reward_components/runs_correct': 1.25, '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': 2.491487979888916, 'sampling/importance_sampling_ratio/mean': 0.9066804647445679, 'sampling/importance_sampling_ratio/min': 0.17903175950050354, 'sampling/sampling_logp_difference/max': 1.1129636764526367, 'sampling/sampling_logp_difference/mean': 0.013031009584665298, 'step': 14, 'step_time': 14.33023315994069}, {'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': 0.0, 'completions/max_length': 250.0, 'completions/max_terminated_length': 250.0, 'completions/mean_length': 53.84375, 'completions/mean_terminated_length': 53.84375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.047202511166688055, 'epoch': 0.01953125, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.01271787378937006, 'learning_rate': 1e-05, 'loss': 0.016833433881402016, 'num_tokens': 1040781.0, 'reward': 0.484375, 'reward/selected_zero_std_group_fraction': 0.5833333333333334, 'reward/zero_std_group_fraction': 0.7833333333333333, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.265625, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.484375, 'reward_components/runs_correct': 0.765625, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.5703125, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.484375, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 2.6090469360351562, 'sampling/importance_sampling_ratio/mean': 0.5816086530685425, 'sampling/importance_sampling_ratio/min': 8.132651601044927e-06, 'sampling/sampling_logp_difference/max': 6.48297119140625, 'sampling/sampling_logp_difference/mean': 0.0597827322781086, 'step': 15, 'step_time': 14.568421976990066}, {'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': 0.0, 'completions/max_length': 64.0, 'completions/max_terminated_length': 64.0, 'completions/mean_length': 25.375, 'completions/mean_terminated_length': 25.375, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.044804181437939405, 'epoch': 0.020833333333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.06850574165582657, 'learning_rate': 1e-05, 'loss': -0.010874222964048386, 'num_tokens': 1111269.0, 'reward': 0.59375, 'reward/selected_zero_std_group_fraction': 0.578125, 'reward/zero_std_group_fraction': 0.78125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.59375, 'reward_components/runs_correct': 0.96875, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.6875, 'reward_std': 0.49501484632492065, 'rewards/reward_func/mean': 0.59375, 'rewards/reward_func/std': 0.49501484632492065, 'sampling/importance_sampling_ratio/max': 2.522055149078369, 'sampling/importance_sampling_ratio/mean': 0.996400773525238, 'sampling/importance_sampling_ratio/min': 0.0008604308241046965, 'sampling/sampling_logp_difference/max': 5.913875579833984, 'sampling/sampling_logp_difference/mean': 0.028771284967660904, 'step': 16, 'step_time': 13.10721344302874}, {'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': 0.0, 'completions/max_length': 77.0, 'completions/max_terminated_length': 77.0, 'completions/mean_length': 25.25, 'completions/mean_terminated_length': 25.25, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.12381407269276679, 'epoch': 0.022135416666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.014247248880565166, 'learning_rate': 1e-05, 'loss': 0.0011658023577183485, 'num_tokens': 1169205.0, 'reward': 0.53125, 'reward/selected_zero_std_group_fraction': 0.75, 'reward/zero_std_group_fraction': 0.875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.890625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.109375, 'reward_components/parser_valid': 0.890625, 'reward_components/reward': 0.53125, 'reward_components/runs_correct': 0.765625, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.6484375, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.53125, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.2833750247955322, 'sampling/importance_sampling_ratio/mean': 0.7679867744445801, 'sampling/importance_sampling_ratio/min': 0.0006078102742321789, 'sampling/sampling_logp_difference/max': 3.555668592453003, 'sampling/sampling_logp_difference/mean': 0.06167206913232803, 'step': 17, 'step_time': 12.730754989082925}, {'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': 0.0, 'completions/max_length': 138.0, 'completions/max_terminated_length': 138.0, 'completions/mean_length': 33.328125, 'completions/mean_terminated_length': 33.328125, 'completions/min_length': 12.0, 'completions/min_terminated_length': 12.0, 'entropy': 0.11625635204836726, 'epoch': 0.0234375, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.10339552164077759, 'learning_rate': 1e-05, 'loss': -0.0010639484971761703, 'num_tokens': 1234186.0, 'reward': 0.6875, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.953125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.953125, 'reward_components/reward': 0.6875, 'reward_components/runs_correct': 1.1875, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.7578125, 'reward_std': 0.467176616191864, 'rewards/reward_func/mean': 0.6875, 'rewards/reward_func/std': 0.467176616191864, 'sampling/importance_sampling_ratio/max': 2.7782413959503174, 'sampling/importance_sampling_ratio/mean': 0.880646824836731, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.119138240814209, 'sampling/sampling_logp_difference/mean': 0.03643210977315903, 'step': 18, 'step_time': 12.969250632915646}, {'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': 0.0, 'completions/max_length': 59.0, 'completions/max_terminated_length': 59.0, 'completions/mean_length': 22.484375, 'completions/mean_terminated_length': 22.484375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.14818466920405626, 'epoch': 0.024739583333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.030023228377103806, 'learning_rate': 1e-05, 'loss': 0.0038805645890533924, 'num_tokens': 1314665.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.5833333333333334, 'reward/zero_std_group_fraction': 0.7916666666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.265625, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.75, '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': 2.6159238815307617, 'sampling/importance_sampling_ratio/mean': 0.7614646553993225, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 8.795028686523438, 'sampling/sampling_logp_difference/mean': 0.062246691435575485, 'step': 19, 'step_time': 14.398780534858815}, {'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': 0.0, 'completions/max_length': 413.0, 'completions/max_terminated_length': 413.0, 'completions/mean_length': 36.21875, 'completions/mean_terminated_length': 36.21875, 'completions/min_length': 10.0, 'completions/min_terminated_length': 10.0, 'entropy': 0.06865742988884449, 'epoch': 0.026041666666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.024288076907396317, 'learning_rate': 1e-05, 'loss': 0.0015844210283830762, 'num_tokens': 1382775.0, 'reward': 0.828125, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.828125, 'reward_components/runs_correct': 1.328125, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.8515625, 'reward_std': 0.38025420904159546, 'rewards/reward_func/mean': 0.828125, 'rewards/reward_func/std': 0.38025420904159546, 'sampling/importance_sampling_ratio/max': 2.713305711746216, 'sampling/importance_sampling_ratio/mean': 0.7808318138122559, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.5530402660369873, 'sampling/sampling_logp_difference/mean': 0.038360346108675, 'step': 20, 'step_time': 17.963437022059225}, {'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': 0.0, 'completions/max_length': 120.0, 'completions/max_terminated_length': 120.0, 'completions/mean_length': 25.59375, 'completions/mean_terminated_length': 25.59375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.160244002006948, 'epoch': 0.02734375, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.009228933602571487, 'learning_rate': 1e-05, 'loss': -0.0014966953312978148, 'num_tokens': 1435165.0, 'reward': 0.703125, 'reward/selected_zero_std_group_fraction': 0.55, 'reward/zero_std_group_fraction': 0.775, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.703125, 'reward_components/runs_correct': 0.953125, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.7109375, 'reward_std': 0.4604927599430084, 'rewards/reward_func/mean': 0.703125, 'rewards/reward_func/std': 0.4604927599430084, 'sampling/importance_sampling_ratio/max': 2.143580913543701, 'sampling/importance_sampling_ratio/mean': 0.5983307361602783, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.6071319580078125, 'sampling/sampling_logp_difference/mean': 0.09631633013486862, 'step': 21, 'step_time': 10.796623947215267}, {'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': 0.0, 'completions/max_length': 71.0, 'completions/max_terminated_length': 71.0, 'completions/mean_length': 30.421875, 'completions/mean_terminated_length': 30.421875, 'completions/min_length': 22.0, 'completions/min_terminated_length': 22.0, 'entropy': 0.03828439279459417, 'epoch': 0.028645833333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.034914661198854446, 'learning_rate': 1e-05, 'loss': -0.0012782011181116104, 'num_tokens': 1504952.0, 'reward': 0.71875, 'reward/selected_zero_std_group_fraction': 0.5416666666666666, 'reward/zero_std_group_fraction': 0.7708333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.9375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.0625, 'reward_components/parser_valid': 0.9375, 'reward_components/reward': 0.71875, 'reward_components/runs_correct': 1.65625, 'reward_components/runs_total': 2.0, 'reward_components/semantic_correct': 0.828125, 'reward_std': 0.4531635046005249, 'rewards/reward_func/mean': 0.71875, 'rewards/reward_func/std': 0.4531635046005249, 'sampling/importance_sampling_ratio/max': 2.414232015609741, 'sampling/importance_sampling_ratio/mean': 0.7812410593032837, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.401658058166504, 'sampling/sampling_logp_difference/mean': 0.0367971807718277, 'step': 22, 'step_time': 12.855918646790087}, {'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': 0.0, 'completions/max_length': 136.0, 'completions/max_terminated_length': 136.0, 'completions/mean_length': 27.90625, 'completions/mean_terminated_length': 27.90625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.06023486936464906, 'epoch': 0.029947916666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.005541967228055, 'learning_rate': 1e-05, 'loss': -0.0017026570858433843, 'num_tokens': 1575986.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.5714285714285714, 'reward/zero_std_group_fraction': 0.7857142857142857, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.484375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.515625, 'reward_components/parser_valid': 0.484375, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.484375, 'reward_components/runs_total': 1.5, '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': 2.9945693016052246, 'sampling/importance_sampling_ratio/mean': 0.8866766691207886, 'sampling/importance_sampling_ratio/min': 0.01300936471670866, 'sampling/sampling_logp_difference/max': 3.6674928665161133, 'sampling/sampling_logp_difference/mean': 0.026879629120230675, 'step': 23, 'step_time': 13.970327518763952}, {'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': 0.0, 'completions/max_length': 186.0, 'completions/max_terminated_length': 186.0, 'completions/mean_length': 27.890625, 'completions/mean_terminated_length': 27.890625, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.12846213718876243, 'epoch': 0.03125, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03954000398516655, 'learning_rate': 1e-05, 'loss': -0.00484201405197382, 'num_tokens': 1645547.0, 'reward': 0.796875, 'reward/selected_zero_std_group_fraction': 0.5625, 'reward/zero_std_group_fraction': 0.78125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.984375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.984375, 'reward_components/reward': 0.796875, 'reward_components/runs_correct': 1.671875, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.890625, 'reward_std': 0.40550529956817627, 'rewards/reward_func/mean': 0.796875, 'rewards/reward_func/std': 0.40550529956817627, 'sampling/importance_sampling_ratio/max': 2.801384449005127, 'sampling/importance_sampling_ratio/mean': 0.9004610180854797, 'sampling/importance_sampling_ratio/min': 0.0429828017950058, 'sampling/sampling_logp_difference/max': 3.417893648147583, 'sampling/sampling_logp_difference/mean': 0.039159439504146576, 'step': 24, 'step_time': 14.240607869927771}, {'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': 0.0, 'completions/max_length': 90.0, 'completions/max_terminated_length': 90.0, 'completions/mean_length': 30.1875, 'completions/mean_terminated_length': 30.1875, 'completions/min_length': 15.0, 'completions/min_terminated_length': 15.0, 'entropy': 0.047863063984550536, 'epoch': 0.032552083333333336, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.029943237081170082, 'learning_rate': 1e-05, 'loss': -0.002707455540075898, 'num_tokens': 1726263.0, 'reward': 0.828125, 'reward/selected_zero_std_group_fraction': 0.5833333333333334, 'reward/zero_std_group_fraction': 0.7916666666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 1.0, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.0, 'reward_components/parser_valid': 1.0, 'reward_components/reward': 0.828125, 'reward_components/runs_correct': 1.703125, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.9140625, 'reward_std': 0.38025420904159546, 'rewards/reward_func/mean': 0.828125, 'rewards/reward_func/std': 0.38025420904159546, 'sampling/importance_sampling_ratio/max': 1.4446991682052612, 'sampling/importance_sampling_ratio/mean': 0.8583819270133972, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 1.8968439102172852, 'sampling/sampling_logp_difference/mean': 0.01647956669330597, 'step': 25, 'step_time': 14.741396218771115}, {'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': 0.0, 'completions/max_length': 117.0, 'completions/max_terminated_length': 117.0, 'completions/mean_length': 27.046875, 'completions/mean_terminated_length': 27.046875, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.0646327305585146, 'epoch': 0.033854166666666664, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1784442.0, 'reward': 0.75, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.75, 'reward_components/runs_correct': 1.125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.75, 'reward_std': 0.4364357888698578, 'rewards/reward_func/mean': 0.75, 'rewards/reward_func/std': 0.4364357888698578, 'sampling/importance_sampling_ratio/max': 2.685920238494873, 'sampling/importance_sampling_ratio/mean': 0.8782329559326172, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.3586173057556152, 'sampling/sampling_logp_difference/mean': 0.030394528061151505, 'step': 26, 'step_time': 11.5534611629555}, {'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': 512.0, 'completions/max_terminated_length': 241.0, 'completions/mean_length': 88.015625, 'completions/mean_terminated_length': 35.94736862182617, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.020254440434655407, 'epoch': 0.03515625, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1865723.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.6590909090909091, 'reward/zero_std_group_fraction': 0.8295454545454546, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.109375, 'reward_components/format_valid': 0.640625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.25, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.75, 'reward_components/parser_unsafe': 0.359375, 'reward_components/parser_valid': 0.640625, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.765625, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.5078125, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 1.2473052740097046, 'sampling/importance_sampling_ratio/mean': 0.740852952003479, 'sampling/importance_sampling_ratio/min': 2.575777635627219e-08, 'sampling/sampling_logp_difference/max': 9.576803207397461, 'sampling/sampling_logp_difference/mean': 0.09739453345537186, 'step': 27, 'step_time': 22.384662643889897}, {'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': 0.0, 'completions/max_length': 152.0, 'completions/max_terminated_length': 152.0, 'completions/mean_length': 28.265625, 'completions/mean_terminated_length': 28.265625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.07239008927717805, 'epoch': 0.036458333333333336, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.019313916563987732, 'learning_rate': 1e-05, 'loss': -0.0024633954744786024, 'num_tokens': 1935052.0, 'reward': 0.84375, 'reward/selected_zero_std_group_fraction': 0.6458333333333334, 'reward/zero_std_group_fraction': 0.8229166666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.90625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.09375, 'reward_components/parser_valid': 0.90625, 'reward_components/reward': 0.84375, 'reward_components/runs_correct': 1.375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.84375, 'reward_std': 0.36596253514289856, 'rewards/reward_func/mean': 0.84375, 'rewards/reward_func/std': 0.36596253514289856, 'sampling/importance_sampling_ratio/max': 2.7767739295959473, 'sampling/importance_sampling_ratio/mean': 0.8580807447433472, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.267211675643921, 'sampling/sampling_logp_difference/mean': 0.033327288925647736, 'step': 28, 'step_time': 13.90948970313184}, {'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': 0.0, 'completions/max_length': 96.0, 'completions/max_terminated_length': 96.0, 'completions/mean_length': 40.5625, 'completions/mean_terminated_length': 40.5625, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.15820372756570578, 'epoch': 0.037760416666666664, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.043206073343753815, 'learning_rate': 1e-05, 'loss': 0.0037743595894426107, 'num_tokens': 2011696.0, 'reward': 0.84375, 'reward/selected_zero_std_group_fraction': 0.6346153846153846, 'reward/zero_std_group_fraction': 0.8173076923076923, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.84375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.84375, 'reward_components/reward': 0.84375, 'reward_components/runs_correct': 1.125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.84375, 'reward_std': 0.36596253514289856, 'rewards/reward_func/mean': 0.84375, 'rewards/reward_func/std': 0.36596253514289856, 'sampling/importance_sampling_ratio/max': 2.5543203353881836, 'sampling/importance_sampling_ratio/mean': 0.8347257375717163, 'sampling/importance_sampling_ratio/min': 0.0035655552055686712, 'sampling/sampling_logp_difference/max': 4.608963966369629, 'sampling/sampling_logp_difference/mean': 0.0468142069876194, 'step': 29, 'step_time': 14.320101961144246}, {'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': 0.0, 'completions/max_length': 139.0, 'completions/max_terminated_length': 139.0, 'completions/mean_length': 38.078125, 'completions/mean_terminated_length': 38.078125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.08051013760268688, 'epoch': 0.0390625, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.021993232890963554, 'learning_rate': 1e-05, 'loss': -0.00483764847740531, 'num_tokens': 2091765.0, 'reward': 0.46875, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.84375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.84375, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 1.0625, 'reward_components/runs_total': 1.75, 'reward_components/semantic_correct': 0.6015625, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.718433380126953, 'sampling/importance_sampling_ratio/mean': 0.5965124368667603, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.111157417297363, 'sampling/sampling_logp_difference/mean': 0.06402274966239929, 'step': 30, 'step_time': 15.57376961002592}, {'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': 0.0, 'completions/max_length': 174.0, 'completions/max_terminated_length': 174.0, 'completions/mean_length': 27.328125, 'completions/mean_terminated_length': 27.328125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.12859956454485655, 'epoch': 0.040364583333333336, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.047997020184993744, 'learning_rate': 1e-05, 'loss': 0.0012259188806638122, 'num_tokens': 2156746.0, 'reward': 0.9375, 'reward/selected_zero_std_group_fraction': 0.6166666666666667, 'reward/zero_std_group_fraction': 0.8083333333333333, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.9375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.9375, 'reward_components/reward': 0.9375, 'reward_components/runs_correct': 1.3125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.9375, 'reward_std': 0.24397502839565277, 'rewards/reward_func/mean': 0.9375, 'rewards/reward_func/std': 0.24397502839565277, 'sampling/importance_sampling_ratio/max': 2.3956685066223145, 'sampling/importance_sampling_ratio/mean': 0.7724407315254211, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.942480564117432, 'sampling/sampling_logp_difference/mean': 0.057173993438482285, 'step': 31, 'step_time': 13.688914143131115}, {'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': 0.0, 'completions/max_length': 166.0, 'completions/max_terminated_length': 166.0, 'completions/mean_length': 35.8125, 'completions/mean_terminated_length': 35.8125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10049664042890072, 'epoch': 0.041666666666666664, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.044129617512226105, 'learning_rate': 1e-05, 'loss': 0.0008574683452025056, 'num_tokens': 2215230.0, 'reward': 0.390625, 'reward/selected_zero_std_group_fraction': 0.609375, 'reward/zero_std_group_fraction': 0.8046875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.84375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.84375, 'reward_components/reward': 0.390625, 'reward_components/runs_correct': 0.640625, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.4296875, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.390625, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 2.309192180633545, 'sampling/importance_sampling_ratio/mean': 0.5349934101104736, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.517840385437012, 'sampling/sampling_logp_difference/mean': 0.07096005231142044, 'step': 32, 'step_time': 12.296852588071488}], 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."
+}