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

rollouts/control-direct-run2/control-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": "control-direct-run2/control-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl",
+ "declared_bytes": 34150484156,
+ "streamed_bytes": 34150484156,
+ "sha256_streamed": "d9eaac752445fdb0864ebeca51fe1dbeae8ce870ff81b6d25c271c072999df03",
+ "sha256_expected": "d9eaac752445fdb0864ebeca51fe1dbeae8ce870ff81b6d25c271c072999df03",
+ "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=2299430, 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': 0.015625, 'completions/max_length': 512.0, 'completions/max_terminated_length': 403.0, 'completions/mean_length': 97.234375, 'completions/mean_terminated_length': 90.65080261230469, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.0706513561308384, 'epoch': 0.0013020833333333333, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03291924297809601, 'learning_rate': 1e-05, 'loss': 0.002471721498295665, 'num_tokens': 65807.0, 'reward': 0.03125, 'reward/selected_zero_std_group_fraction': 0.5, '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': 0.015625, 'reward_components/format_valid': 0.234375, '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': 0.640625, 'reward_components/parser_valid': 0.234375, 'reward_components/reward': 0.03125, 'reward_components/runs_correct': 0.265625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.1328125, 'reward_std': 0.17536810040473938, 'rewards/reward_func/mean': 0.03125, 'rewards/reward_func/std': 0.17536810040473938, 'sampling/importance_sampling_ratio/max': 2.4791440963745117, 'sampling/importance_sampling_ratio/mean': 0.4421990215778351, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 10.708686828613281, 'sampling/sampling_logp_difference/mean': 0.04978378862142563, 'step': 1, 'step_time': 22.99618076905608}, {'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': 415.0, 'completions/max_terminated_length': 415.0, 'completions/mean_length': 46.015625, 'completions/mean_terminated_length': 46.015625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.06747654033824801, 'epoch': 0.0026041666666666665, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.006932457443326712, 'learning_rate': 1e-05, 'loss': -0.0018002692377194762, 'num_tokens': 146128.0, 'reward': 0.21875, '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': 0.46875, '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': 0.53125, 'reward_components/parser_valid': 0.46875, 'reward_components/reward': 0.21875, 'reward_components/runs_correct': 0.34375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.28125, 'reward_std': 0.4166666865348816, 'rewards/reward_func/mean': 0.21875, 'rewards/reward_func/std': 0.4166666865348816, 'sampling/importance_sampling_ratio/max': 2.312253475189209, 'sampling/importance_sampling_ratio/mean': 0.7256616950035095, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.419851779937744, 'sampling/sampling_logp_difference/mean': 0.04783862456679344, 'step': 2, 'step_time': 21.249834468006156}, {'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': 375.0, 'completions/max_terminated_length': 375.0, 'completions/mean_length': 58.765625, 'completions/mean_terminated_length': 58.765625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.1374936979264021, 'epoch': 0.00390625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.004480303265154362, 'learning_rate': 1e-05, 'loss': 0.000855011458043009, 'num_tokens': 207041.0, 'reward': 0.265625, 'reward/selected_zero_std_group_fraction': 0.6666666666666666, 'reward/zero_std_group_fraction': 0.8333333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.46875, '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': 0.421875, 'reward_components/parser_valid': 0.46875, 'reward_components/reward': 0.265625, 'reward_components/runs_correct': 0.59375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.3046875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.265625, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 1.8186861276626587, 'sampling/importance_sampling_ratio/mean': 0.5117727518081665, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 8.39253044128418, 'sampling/sampling_logp_difference/mean': 0.05950424075126648, 'step': 3, 'step_time': 17.158962457091548}, {'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': 325.0, 'completions/mean_length': 94.78125, 'completions/mean_terminated_length': 66.96666717529297, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.15237591322511435, 'epoch': 0.005208333333333333, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.041977714747190475, 'learning_rate': 1e-05, 'loss': 0.003060427960008383, 'num_tokens': 290355.0, 'reward': 0.25, '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': 0.0625, 'reward_components/format_valid': 0.265625, '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': 0.609375, 'reward_components/parser_valid': 0.265625, 'reward_components/reward': 0.25, 'reward_components/runs_correct': 0.515625, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.2578125, 'reward_std': 0.4364357888698578, 'rewards/reward_func/mean': 0.25, 'rewards/reward_func/std': 0.4364357888698578, 'sampling/importance_sampling_ratio/max': 2.4476890563964844, 'sampling/importance_sampling_ratio/mean': 0.6657613515853882, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.463486194610596, 'sampling/sampling_logp_difference/mean': 0.04209759831428528, 'step': 4, 'step_time': 23.818238728214055}, {'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': 348.0, 'completions/max_terminated_length': 348.0, 'completions/mean_length': 55.578125, 'completions/mean_terminated_length': 55.578125, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.1364055722951889, 'epoch': 0.006510416666666667, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.014705697074532509, 'learning_rate': 1e-05, 'loss': -0.006099672056734562, 'num_tokens': 358424.0, 'reward': 0.515625, 'reward/selected_zero_std_group_fraction': 0.6, 'reward/zero_std_group_fraction': 0.8, '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': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.484375, 'reward_components/parser_valid': 0.515625, 'reward_components/reward': 0.515625, 'reward_components/runs_correct': 0.890625, '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': 2.822636365890503, 'sampling/importance_sampling_ratio/mean': 0.7600054740905762, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.3220744132995605, 'sampling/sampling_logp_difference/mean': 0.04339918866753578, 'step': 5, 'step_time': 17.786385164014064}, {'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': 192.0, 'completions/max_terminated_length': 192.0, 'completions/mean_length': 39.125, 'completions/mean_terminated_length': 39.125, 'completions/min_length': 13.0, 'completions/min_terminated_length': 13.0, 'entropy': 0.06149608315899968, 'epoch': 0.0078125, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.08045726269483566, 'learning_rate': 1e-05, 'loss': -5.970080383121967e-05, 'num_tokens': 427744.0, 'reward': 0.328125, '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': 0.5, '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': 0.5, 'reward_components/reward': 0.328125, 'reward_components/runs_correct': 0.65625, 'reward_components/runs_total': 1.5, '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': 2.1101651191711426, 'sampling/importance_sampling_ratio/mean': 0.7610577344894409, 'sampling/importance_sampling_ratio/min': 0.0002452212502248585, 'sampling/sampling_logp_difference/max': 5.490365982055664, 'sampling/sampling_logp_difference/mean': 0.02854968048632145, 'step': 6, 'step_time': 14.620841684169136}, {'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': 128.0, 'completions/max_terminated_length': 128.0, 'completions/mean_length': 39.265625, 'completions/mean_terminated_length': 39.265625, 'completions/min_length': 13.0, 'completions/min_terminated_length': 13.0, 'entropy': 0.09794804360717535, 'epoch': 0.009114583333333334, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 495857.0, 'reward': 0.375, 'reward/selected_zero_std_group_fraction': 0.6785714285714286, 'reward/zero_std_group_fraction': 0.8392857142857143, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, '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': 0.375, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.375, 'reward_components/runs_correct': 0.671875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.3984375, 'reward_std': 0.48795005679130554, 'rewards/reward_func/mean': 0.375, 'rewards/reward_func/std': 0.48795005679130554, 'sampling/importance_sampling_ratio/max': 2.715672731399536, 'sampling/importance_sampling_ratio/mean': 0.6931142807006836, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 10.906787872314453, 'sampling/sampling_logp_difference/mean': 0.0577496737241745, 'step': 7, 'step_time': 13.687706413911656}, {'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': 107.0, 'completions/max_terminated_length': 107.0, 'completions/mean_length': 35.609375, 'completions/mean_terminated_length': 35.609375, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.0901258836966008, 'epoch': 0.010416666666666666, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 569176.0, 'reward': 0.25, 'reward/selected_zero_std_group_fraction': 0.71875, 'reward/zero_std_group_fraction': 0.859375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.359375, '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': 0.640625, 'reward_components/parser_valid': 0.359375, 'reward_components/reward': 0.25, 'reward_components/runs_correct': 0.609375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.3046875, 'reward_std': 0.4364357888698578, 'rewards/reward_func/mean': 0.25, 'rewards/reward_func/std': 0.4364357888698578, 'sampling/importance_sampling_ratio/max': 2.1649227142333984, 'sampling/importance_sampling_ratio/mean': 0.5307821035385132, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 15.446427345275879, 'sampling/sampling_logp_difference/mean': 0.07964935153722763, 'step': 8, 'step_time': 14.664501897990704}, {'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': 0.03125, 'completions/max_length': 512.0, 'completions/max_terminated_length': 510.0, 'completions/mean_length': 107.078125, 'completions/mean_terminated_length': 94.01612854003906, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.14985968172550201, 'epoch': 0.01171875, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.06829945743083954, 'learning_rate': 1e-05, 'loss': -0.0023049316368997097, 'num_tokens': 656285.0, 'reward': 0.21875, 'reward/selected_zero_std_group_fraction': 0.6666666666666666, 'reward/zero_std_group_fraction': 0.8333333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.03125, 'reward_components/format_valid': 0.21875, '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': 0.78125, 'reward_components/parser_valid': 0.21875, 'reward_components/reward': 0.21875, 'reward_components/runs_correct': 0.359375, 'reward_components/runs_total': 1.625, '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': 2.930047035217285, 'sampling/importance_sampling_ratio/mean': 0.684585452079773, 'sampling/importance_sampling_ratio/min': 5.981505182717228e-06, 'sampling/sampling_logp_difference/max': 5.064081192016602, 'sampling/sampling_logp_difference/mean': 0.04212861880660057, 'step': 9, 'step_time': 24.4348031738773}, {'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': 171.0, 'completions/max_terminated_length': 171.0, 'completions/mean_length': 38.03125, 'completions/mean_terminated_length': 38.03125, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.06273522460833192, 'epoch': 0.013020833333333334, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.014614474028348923, 'learning_rate': 1e-05, 'loss': -0.0004707158077508211, 'num_tokens': 712671.0, 'reward': 0.296875, 'reward/selected_zero_std_group_fraction': 0.675, 'reward/zero_std_group_fraction': 0.8375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.296875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.03125, 'reward_components/parser_natural': 0.96875, 'reward_components/parser_unsafe': 0.640625, 'reward_components/parser_valid': 0.296875, 'reward_components/reward': 0.296875, 'reward_components/runs_correct': 0.421875, 'reward_components/runs_total': 1.125, 'reward_components/semantic_correct': 0.296875, 'reward_std': 0.4604927599430084, 'rewards/reward_func/mean': 0.296875, 'rewards/reward_func/std': 0.4604927599430084, 'sampling/importance_sampling_ratio/max': 1.7523019313812256, 'sampling/importance_sampling_ratio/mean': 0.8029743432998657, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.1399953365325928, 'sampling/sampling_logp_difference/mean': 0.02783369831740856, 'step': 10, 'step_time': 12.12212030601222}, {'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': 0.09375, 'completions/max_length': 512.0, 'completions/max_terminated_length': 246.0, 'completions/mean_length': 83.609375, 'completions/mean_terminated_length': 39.2931022644043, 'completions/min_length': 10.0, 'completions/min_terminated_length': 10.0, 'entropy': 0.03614974149968475, 'epoch': 0.014322916666666666, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 777542.0, 'reward': 0.375, 'reward/selected_zero_std_group_fraction': 0.7045454545454546, 'reward/zero_std_group_fraction': 0.8522727272727273, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.09375, 'reward_components/format_valid': 0.375, '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': 0.53125, 'reward_components/parser_valid': 0.375, 'reward_components/reward': 0.375, 'reward_components/runs_correct': 0.375, 'reward_components/runs_total': 1.375, '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': 2.5379526615142822, 'sampling/importance_sampling_ratio/mean': 0.8068971633911133, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.363430023193359, 'sampling/sampling_logp_difference/mean': 0.04846201464533806, 'step': 11, 'step_time': 19.209286472178064}, {'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': 0.046875, 'completions/max_length': 512.0, 'completions/max_terminated_length': 339.0, 'completions/mean_length': 75.578125, 'completions/mean_terminated_length': 54.114749908447266, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10354180412832648, 'epoch': 0.015625, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 862059.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.7291666666666666, 'reward/zero_std_group_fraction': 0.8645833333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.046875, 'reward_components/format_valid': 0.5, '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': 0.5, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.5, 'reward_components/runs_total': 1.375, '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': 2.332786798477173, 'sampling/importance_sampling_ratio/mean': 0.7766380310058594, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.60979700088501, 'sampling/sampling_logp_difference/mean': 0.04004963859915733, 'step': 12, 'step_time': 23.399780742125586}, {'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': 108.0, 'completions/max_terminated_length': 108.0, 'completions/mean_length': 25.046875, 'completions/mean_terminated_length': 25.046875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.07552982866764069, 'epoch': 0.016927083333333332, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.00924766156822443, 'learning_rate': 1e-05, 'loss': -0.0003540041798260063, 'num_tokens': 917742.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.7307692307692307, 'reward/zero_std_group_fraction': 0.8653846153846154, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.5, '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': 0.46875, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.484375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.421875, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.359375, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 1.95522141456604, 'sampling/importance_sampling_ratio/mean': 0.7101733684539795, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.5716552734375, 'sampling/sampling_logp_difference/mean': 0.04394473880529404, 'step': 13, 'step_time': 10.89887742116116}, {'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': 27.765625, 'completions/mean_terminated_length': 27.765625, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.03464210531092249, 'epoch': 0.018229166666666668, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 998175.0, 'reward': 0.625, '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': 0.625, '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': 0.625, 'reward_components/reward': 0.625, 'reward_components/runs_correct': 0.875, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.625, 'reward_std': 0.48795005679130554, 'rewards/reward_func/mean': 0.625, 'rewards/reward_func/std': 0.48795005679130554, 'sampling/importance_sampling_ratio/max': 2.136075496673584, 'sampling/importance_sampling_ratio/mean': 0.7437152862548828, 'sampling/importance_sampling_ratio/min': 0.08363161236047745, 'sampling/sampling_logp_difference/max': 4.497424125671387, 'sampling/sampling_logp_difference/mean': 0.027775714173913002, 'step': 14, 'step_time': 14.282457209075801}, {'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': 0.015625, 'completions/max_length': 512.0, 'completions/max_terminated_length': 281.0, 'completions/mean_length': 69.3125, 'completions/mean_terminated_length': 62.28571701049805, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.1304083769209683, 'epoch': 0.01953125, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.005323031917214394, 'learning_rate': 1e-05, 'loss': -0.00019478613103274256, 'num_tokens': 1060531.0, 'reward': 0.359375, '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': 0.015625, 'reward_components/format_valid': 0.40625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.015625, 'reward_components/parser_natural': 0.859375, 'reward_components/parser_unsafe': 0.578125, 'reward_components/parser_valid': 0.40625, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.359375, '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': 2.680710792541504, 'sampling/importance_sampling_ratio/mean': 0.66759192943573, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.064492702484131, 'sampling/sampling_logp_difference/mean': 0.043533965945243835, 'step': 15, 'step_time': 18.64824915898498}, {'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': 56.0, 'completions/max_terminated_length': 56.0, 'completions/mean_length': 24.0, 'completions/mean_terminated_length': 24.0, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.05367545783519745, 'epoch': 0.020833333333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03530122712254524, 'learning_rate': 1e-05, 'loss': 0.0016329901991412044, 'num_tokens': 1130931.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.734375, 'reward/zero_std_group_fraction': 0.8671875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, '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': 0.375, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.859375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.5625, 'reward_std': 0.5039526224136353, 'rewards/reward_func/mean': 0.5, 'rewards/reward_func/std': 0.5039526224136353, 'sampling/importance_sampling_ratio/max': 1.9526467323303223, 'sampling/importance_sampling_ratio/mean': 0.7544420957565308, 'sampling/importance_sampling_ratio/min': 0.0006818603724241257, 'sampling/sampling_logp_difference/max': 7.2836198806762695, 'sampling/sampling_logp_difference/mean': 0.06341870129108429, 'step': 16, 'step_time': 12.47579638089519}, {'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': 92.0, 'completions/max_terminated_length': 92.0, 'completions/mean_length': 22.78125, 'completions/mean_terminated_length': 22.78125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.09992378298193216, 'epoch': 0.022135416666666668, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.060755230486392975, 'learning_rate': 1e-05, 'loss': -0.0005033347988501191, 'num_tokens': 1188709.0, 'reward': 0.34375, 'reward/selected_zero_std_group_fraction': 0.5, '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': 0.65625, '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': 0.34375, 'reward_components/parser_valid': 0.65625, 'reward_components/reward': 0.34375, 'reward_components/runs_correct': 0.453125, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.359375, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.34375, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 1.7463427782058716, 'sampling/importance_sampling_ratio/mean': 0.6594699621200562, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.982542991638184, 'sampling/sampling_logp_difference/mean': 0.06757418811321259, 'step': 17, 'step_time': 13.05647733504884}, {'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': 377.0, 'completions/max_terminated_length': 377.0, 'completions/mean_length': 49.1875, 'completions/mean_terminated_length': 49.1875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.12593155424110591, 'epoch': 0.0234375, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.001671101781539619, 'learning_rate': 1e-05, 'loss': 0.0008836997440084815, 'num_tokens': 1254705.0, 'reward': 0.390625, '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': 0.71875, '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': 0.28125, 'reward_components/parser_valid': 0.71875, 'reward_components/reward': 0.390625, 'reward_components/runs_correct': 0.640625, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.4453125, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.390625, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 2.7278246879577637, 'sampling/importance_sampling_ratio/mean': 0.6458708047866821, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.0829596519470215, 'sampling/sampling_logp_difference/mean': 0.05826945602893829, 'step': 18, 'step_time': 17.30448768415954}, {'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': 56.0, 'completions/max_terminated_length': 56.0, 'completions/mean_length': 22.890625, 'completions/mean_terminated_length': 22.890625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.06935105426236987, 'epoch': 0.024739583333333332, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.0046625761315226555, 'learning_rate': 1e-05, 'loss': -0.00046711473260074854, 'num_tokens': 1335210.0, 'reward': 0.34375, 'reward/selected_zero_std_group_fraction': 0.6666666666666666, 'reward/zero_std_group_fraction': 0.8333333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.609375, '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': 0.390625, 'reward_components/parser_valid': 0.609375, 'reward_components/reward': 0.34375, 'reward_components/runs_correct': 0.59375, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.34375, 'reward_std': 0.4787135720252991, 'rewards/reward_func/mean': 0.34375, 'rewards/reward_func/std': 0.4787135720252991, 'sampling/importance_sampling_ratio/max': 1.9923979043960571, 'sampling/importance_sampling_ratio/mean': 0.7147588729858398, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 9.998953819274902, 'sampling/sampling_logp_difference/mean': 0.11976424604654312, 'step': 19, 'step_time': 14.53097786323633}, {'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': 42.0, 'completions/max_terminated_length': 42.0, 'completions/mean_length': 24.46875, 'completions/mean_terminated_length': 24.46875, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.030989954946562648, 'epoch': 0.026041666666666668, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1402568.0, 'reward': 0.625, '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': 0.75, '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': 0.625, 'reward_components/runs_correct': 1.0, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.625, 'reward_std': 0.48795005679130554, 'rewards/reward_func/mean': 0.625, 'rewards/reward_func/std': 0.48795005679130554, 'sampling/importance_sampling_ratio/max': 2.1830294132232666, 'sampling/importance_sampling_ratio/mean': 0.8137635588645935, 'sampling/importance_sampling_ratio/min': 9.443721751267731e-07, 'sampling/sampling_logp_difference/max': 5.674090385437012, 'sampling/sampling_logp_difference/mean': 0.06479974836111069, 'step': 20, 'step_time': 12.179355298983864}, {'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': 83.0, 'completions/max_terminated_length': 83.0, 'completions/mean_length': 26.78125, 'completions/mean_terminated_length': 26.78125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.15589527692645788, 'epoch': 0.02734375, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.0604284293949604, 'learning_rate': 1e-05, 'loss': 0.001882106065750122, 'num_tokens': 1455034.0, 'reward': 0.46875, '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': 0.515625, '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': 0.484375, 'reward_components/parser_valid': 0.515625, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 0.609375, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.4765625, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.7926242351531982, 'sampling/importance_sampling_ratio/mean': 0.6839655637741089, 'sampling/importance_sampling_ratio/min': 3.7149558920646086e-05, 'sampling/sampling_logp_difference/max': 8.571677207946777, 'sampling/sampling_logp_difference/mean': 0.10058866441249847, 'step': 21, 'step_time': 10.443959277123213}, {'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': 175.0, 'completions/max_terminated_length': 175.0, 'completions/mean_length': 45.1875, 'completions/mean_terminated_length': 45.1875, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.0769885906483978, 'epoch': 0.028645833333333332, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1525766.0, 'reward': 0.25, 'reward/selected_zero_std_group_fraction': 0.7916666666666666, 'reward/zero_std_group_fraction': 0.8958333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.5, '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': 0.5, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.25, 'reward_components/runs_correct': 0.625, 'reward_components/runs_total': 2.0, 'reward_components/semantic_correct': 0.3125, 'reward_std': 0.4364357888698578, 'rewards/reward_func/mean': 0.25, 'rewards/reward_func/std': 0.4364357888698578, 'sampling/importance_sampling_ratio/max': 1.2990950345993042, 'sampling/importance_sampling_ratio/mean': 0.4253673553466797, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.219188213348389, 'sampling/sampling_logp_difference/mean': 0.08914432674646378, 'step': 22, 'step_time': 14.650309790973552}, {'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': 130.0, 'completions/max_terminated_length': 130.0, 'completions/mean_length': 29.4375, 'completions/mean_terminated_length': 29.4375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.05914927739650011, 'epoch': 0.029947916666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.004939020611345768, 'learning_rate': 1e-05, 'loss': -0.0006702466635033488, 'num_tokens': 1596898.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.7857142857142857, 'reward/zero_std_group_fraction': 0.8928571428571429, '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': 2.5562663078308105, 'sampling/importance_sampling_ratio/mean': 0.6484360694885254, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.890910625457764, 'sampling/sampling_logp_difference/mean': 0.06656914204359055, 'step': 23, 'step_time': 14.211359249195084}, {'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': 62.0, 'completions/max_terminated_length': 62.0, 'completions/mean_length': 29.796875, 'completions/mean_terminated_length': 29.796875, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.043122350703924894, 'epoch': 0.03125, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.843564510345459, 'learning_rate': 1e-05, 'loss': -0.0012085638009011745, 'num_tokens': 1666581.0, 'reward': 0.578125, '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': 0.609375, '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': 0.390625, 'reward_components/parser_valid': 0.609375, 'reward_components/reward': 0.578125, 'reward_components/runs_correct': 1.046875, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.5859375, 'reward_std': 0.49776285886764526, 'rewards/reward_func/mean': 0.578125, 'rewards/reward_func/std': 0.49776285886764526, 'sampling/importance_sampling_ratio/max': 2.04976487159729, 'sampling/importance_sampling_ratio/mean': 0.7125258445739746, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.695780277252197, 'sampling/sampling_logp_difference/mean': 0.04758775234222412, 'step': 24, 'step_time': 12.776823744759895}, {'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': 24.875, 'completions/mean_terminated_length': 24.875, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.055804528295993805, 'epoch': 0.032552083333333336, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.004768621642142534, 'learning_rate': 1e-05, 'loss': -0.0014160224236547947, 'num_tokens': 1746957.0, 'reward': 0.484375, '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': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0625, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.9375, 'reward_components/parser_unsafe': 0.25, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.484375, 'reward_components/runs_correct': 1.21875, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.609375, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.484375, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 1.8020468950271606, 'sampling/importance_sampling_ratio/mean': 0.834571361541748, 'sampling/importance_sampling_ratio/min': 0.0012101101456210017, 'sampling/sampling_logp_difference/max': 5.868072032928467, 'sampling/sampling_logp_difference/mean': 0.054860781878232956, 'step': 25, 'step_time': 14.583849129849114}, {'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': 28.25, 'completions/mean_terminated_length': 28.25, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.11564640048891306, 'epoch': 0.033854166666666664, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.02289400063455105, 'learning_rate': 1e-05, 'loss': -0.0008288491517305374, 'num_tokens': 1805213.0, 'reward': 0.59375, 'reward/selected_zero_std_group_fraction': 0.675, 'reward/zero_std_group_fraction': 0.8375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, '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': 0.375, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.59375, 'reward_components/runs_correct': 0.828125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.59375, 'reward_std': 0.49501484632492065, 'rewards/reward_func/mean': 0.59375, 'rewards/reward_func/std': 0.49501484632492065, 'sampling/importance_sampling_ratio/max': 2.668942451477051, 'sampling/importance_sampling_ratio/mean': 0.4509754776954651, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 18.108680725097656, 'sampling/sampling_logp_difference/mean': 0.14276234805583954, 'step': 26, 'step_time': 10.87025969196111}, {'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': 341.0, 'completions/max_terminated_length': 341.0, 'completions/mean_length': 35.984375, 'completions/mean_terminated_length': 35.984375, 'completions/min_length': 15.0, 'completions/min_terminated_length': 15.0, 'entropy': 0.028091311891330406, 'epoch': 0.03515625, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1883164.0, 'reward': 0.375, 'reward/selected_zero_std_group_fraction': 0.7045454545454546, 'reward/zero_std_group_fraction': 0.8522727272727273, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.375, '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': 0.625, 'reward_components/parser_valid': 0.375, 'reward_components/reward': 0.375, 'reward_components/runs_correct': 0.5, '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': 1.5105257034301758, 'sampling/importance_sampling_ratio/mean': 0.7471547722816467, 'sampling/importance_sampling_ratio/min': 0.0006328212330117822, 'sampling/sampling_logp_difference/max': 6.985882759094238, 'sampling/sampling_logp_difference/mean': 0.045362215489149094, 'step': 27, 'step_time': 18.352597938035615}, {'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': 131.0, 'completions/max_terminated_length': 131.0, 'completions/mean_length': 25.96875, 'completions/mean_terminated_length': 25.96875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.04772661323659122, 'epoch': 0.036458333333333336, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1952346.0, 'reward': 0.5, 'reward/selected_zero_std_group_fraction': 0.7291666666666666, 'reward/zero_std_group_fraction': 0.8645833333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, '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': 0.375, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.5, 'reward_components/runs_correct': 0.890625, 'reward_components/runs_total': 1.625, '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': 2.0489203929901123, 'sampling/importance_sampling_ratio/mean': 0.7075769305229187, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.585416793823242, 'sampling/sampling_logp_difference/mean': 0.07491427659988403, 'step': 28, 'step_time': 13.946502693113871}, {'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': 105.0, 'completions/max_terminated_length': 105.0, 'completions/mean_length': 44.75, 'completions/mean_terminated_length': 44.75, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.07954265782609582, 'epoch': 0.037760416666666664, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.020363550633192062, 'learning_rate': 1e-05, 'loss': -0.000875548692420125, 'num_tokens': 2029258.0, 'reward': 0.3125, 'reward/selected_zero_std_group_fraction': 0.7307692307692307, 'reward/zero_std_group_fraction': 0.8653846153846154, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.4375, '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': 0.5, 'reward_components/parser_valid': 0.4375, 'reward_components/reward': 0.3125, 'reward_components/runs_correct': 0.3125, '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': 2.9170873165130615, 'sampling/importance_sampling_ratio/mean': 0.7177724838256836, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.941206932067871, 'sampling/sampling_logp_difference/mean': 0.055506911128759384, 'step': 29, 'step_time': 14.832541061914526}, {'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': 263.0, 'completions/max_terminated_length': 263.0, 'completions/mean_length': 46.3125, 'completions/mean_terminated_length': 46.3125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.11673849081853405, 'epoch': 0.0390625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.02981884777545929, 'learning_rate': 1e-05, 'loss': 0.0018443928565829992, 'num_tokens': 2109854.0, 'reward': 0.203125, 'reward/selected_zero_std_group_fraction': 0.7321428571428571, 'reward/zero_std_group_fraction': 0.8660714285714286, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, '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': 0.375, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.203125, 'reward_components/runs_correct': 0.671875, 'reward_components/runs_total': 1.75, 'reward_components/semantic_correct': 0.3359375, 'reward_std': 0.40550529956817627, 'rewards/reward_func/mean': 0.203125, 'rewards/reward_func/std': 0.40550529956817627, 'sampling/importance_sampling_ratio/max': 2.2310125827789307, 'sampling/importance_sampling_ratio/mean': 0.44867584109306335, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.255682945251465, 'sampling/sampling_logp_difference/mean': 0.0773380920290947, 'step': 30, 'step_time': 18.071602616924793}, {'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': 61.0, 'completions/max_terminated_length': 61.0, 'completions/mean_length': 23.46875, 'completions/mean_terminated_length': 23.46875, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.11529165925458074, 'epoch': 0.040364583333333336, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.022707628086209297, 'learning_rate': 1e-05, 'loss': -0.00022746599279344082, 'num_tokens': 2174588.0, 'reward': 0.6875, 'reward/selected_zero_std_group_fraction': 0.7333333333333333, 'reward/zero_std_group_fraction': 0.8666666666666667, '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': 0.6875, 'reward_components/runs_correct': 0.9375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.71875, 'reward_std': 0.467176616191864, 'rewards/reward_func/mean': 0.6875, 'rewards/reward_func/std': 0.467176616191864, 'sampling/importance_sampling_ratio/max': 2.4983232021331787, 'sampling/importance_sampling_ratio/mean': 0.5413348078727722, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.9086360931396484, 'sampling/sampling_logp_difference/mean': 0.11122269928455353, 'step': 31, 'step_time': 12.021246707183309}, {'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': 30.171875, 'completions/mean_terminated_length': 30.171875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.08716653956798837, 'epoch': 0.041666666666666664, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 2232711.0, 'reward': 0.25, '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': 0.5, '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': 0.5, 'reward_components/parser_valid': 0.5, 'reward_components/reward': 0.25, 'reward_components/runs_correct': 0.5, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.3125, 'reward_std': 0.4364357888698578, 'rewards/reward_func/mean': 0.25, 'rewards/reward_func/std': 0.4364357888698578, 'sampling/importance_sampling_ratio/max': 1.9323734045028687, 'sampling/importance_sampling_ratio/mean': 0.7391115427017212, 'sampling/importance_sampling_ratio/min': 8.425581654591952e-07, 'sampling/sampling_logp_difference/max': 6.946174144744873, 'sampling/sampling_logp_difference/mean': 0.08737648278474808, 'step': 32, 'step_time': 11.962580899824388}], 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."
+}