Files
RLCR-5x-priority-overconf-math/trainer_state.json
ModelHub XC b0aa39424b 初始化项目,由ModelHub XC社区提供模型
Model: hector-gr/RLCR-5x-priority-overconf-math
Source: Original Platform
2026-04-22 02:36:00 +08:00

23547 lines
1.3 MiB

{
"best_metric": null,
"best_model_checkpoint": null,
"epoch": 2.5,
"eval_steps": 50,
"global_step": 1040,
"is_hyper_param_search": false,
"is_local_process_zero": true,
"is_world_process_zero": true,
"log_history": [
{
"calibration/aurc": 0.5143116826276487,
"calibration/batch_distribution_entropy": 0.24942287672258642,
"calibration/batch_entropy_100bins": 0.34826538716122146,
"calibration/batch_entropy_10bins": 0.24942287672258642,
"calibration/batch_entropy_50bins": 0.3972500110134775,
"calibration/batch_uniqueness": 0.5076220224596407,
"calibration/confidence_entropy": 0.22319858203556509,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.24942287672258642,
"calibration/distribution_entropy_100": 0.34826538716122146,
"calibration/ece": 0.4648402350505384,
"calibration/mean_confidence": 0.9214651731076637,
"calibration/unique_confidence_per_question": 0.03229166666666667,
"calibration/unique_confidences": 12.4,
"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": 0.018576388888888885,
"completions/max_length": 4027.8,
"completions/max_terminated_length": 4027.8,
"completions/mean_length": 496.7038146972656,
"completions/mean_terminated_length": 506.1095825195313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 2.4,
"epoch": 0.01201923076923077,
"grad_norm": 0.004719023127108812,
"learning_rate": 1.201923076923077e-07,
"loss": 0.0071,
"num_tokens": 8801292.0,
"reward": 0.5796843767166138,
"reward_std": 0.5260672450065613,
"rewards/accuracy_reward": 0.24774305820465087,
"rewards/brier_reward": 0.3010988771915436,
"rewards/confidence_one_or_zero": 0.3295138835906982,
"rewards/format_reward": 0.6105034708976745,
"rewards/mean_confidence_reward": 0.8376270651817321,
"sampling/batch_mean_priority_error": 0.41623627994107837,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.08493557721376419,
"sampling/error_ema_mean": 0.00029969110910315065,
"sampling/priority_kl": 0.007231711782515049,
"sampling/priority_scale": 1048575.9998779297,
"sampling/prob_entropy": 10.301719474792481,
"sampling/prob_max": 3.357546593178995e-05,
"sampling/prob_min": 0.0,
"sampling/prompt_draws_max": 1.0,
"sampling/prompt_draws_mean": 0.007200000155717134,
"sampling/prompt_draws_total": 216.0,
"sampling/seen_fraction": 0.007200000155717134,
"sampling/unseen_fraction": 0.9927999998442829,
"signal/accuracy_reward/centered_abs_mean": 0.2978624105453491,
"signal/accuracy_reward/group_std_mean": 0.3558677971363068,
"signal/accuracy_reward/group_zero_std_frac": 0.11944444626569747,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.14893120527267456,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.14893120527267456,
"signal/advantage_abs_mean": 0.4478647530078888,
"signal/advantage_pre_scale_abs_mean": 0.4478647530078888,
"signal/advantage_pre_scale_std": 0.535431969165802,
"signal/advantage_std": 0.535431969165802,
"signal/brier_reward/centered_abs_mean": 0.3062243342399597,
"signal/brier_reward/group_std_mean": 0.3604122519493103,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15311216711997985,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15311216711997985,
"signal/confidence_one_or_zero/centered_abs_mean": 0.4056097984313965,
"signal/confidence_one_or_zero/group_std_mean": 0.4532551825046539,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.002777777798473835,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.056097986904206e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.056097986904206e-06,
"signal/format_reward/centered_abs_mean": 0.4284885048866272,
"signal/format_reward/group_std_mean": 0.4680496335029602,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.2142442524433136,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.2142442524433136,
"signal/mean_confidence_reward/centered_abs_mean": 0.1593179762363434,
"signal/mean_confidence_reward/group_std_mean": 0.23351403772830964,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.5931796951917932e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.5931796951917932e-06,
"step": 5
},
{
"calibration/aurc": 0.6343878019492749,
"calibration/batch_distribution_entropy": 0.2533414958520655,
"calibration/batch_entropy_100bins": 0.3445832958692546,
"calibration/batch_entropy_10bins": 0.2533414958520655,
"calibration/batch_entropy_50bins": 0.39785691961510256,
"calibration/batch_uniqueness": 0.5116630254293695,
"calibration/confidence_entropy": 0.21199203076901546,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.2533414958520655,
"calibration/distribution_entropy_100": 0.3445832958692546,
"calibration/ece": 0.590023294162618,
"calibration/mean_confidence": 0.9236542957365893,
"calibration/unique_confidence_per_question": 0.03020833333333333,
"calibration/unique_confidences": 11.6,
"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": 0.019704861111111117,
"completions/max_length": 4034.8,
"completions/max_terminated_length": 4034.8,
"completions/mean_length": 521.3720520019531,
"completions/mean_terminated_length": 531.8583312988281,
"completions/min_length": 0.0,
"completions/min_terminated_length": 10.2,
"epoch": 0.02403846153846154,
"grad_norm": 0.00470328563824296,
"learning_rate": 2.403846153846154e-07,
"loss": 0.0038,
"num_tokens": 17889514.0,
"reward": 0.5620967268943786,
"reward_std": 0.5192024111747742,
"rewards/accuracy_reward": 0.23385416269302367,
"rewards/brier_reward": 0.28927443027496336,
"rewards/confidence_one_or_zero": 0.33159722089767457,
"rewards/format_reward": 0.6010416746139526,
"rewards/mean_confidence_reward": 0.8290581583976746,
"sampling/batch_mean_priority_error": 0.4233142476730391,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.09038958102464675,
"sampling/error_ema_mean": 0.0008069580886512995,
"sampling/priority_kl": 0.019392204470932484,
"sampling/priority_scale": 1048575.9998779297,
"sampling/prob_entropy": 10.289560317993164,
"sampling/prob_max": 3.398626940906979e-05,
"sampling/prob_min": 0.0,
"sampling/prompt_draws_max": 1.0,
"sampling/prompt_draws_mean": 0.019200000166893005,
"sampling/prompt_draws_total": 576.0,
"sampling/seen_fraction": 0.019200000166893005,
"sampling/unseen_fraction": 0.980799999833107,
"signal/accuracy_reward/centered_abs_mean": 0.2861653625965118,
"signal/accuracy_reward/group_std_mean": 0.3420232474803925,
"signal/accuracy_reward/group_zero_std_frac": 0.1500000014901161,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1430826812982559,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1430826812982559,
"signal/advantage_abs_mean": 0.4444202482700348,
"signal/advantage_pre_scale_abs_mean": 0.4444202482700348,
"signal/advantage_pre_scale_std": 0.5322044849395752,
"signal/advantage_std": 0.5322044849395752,
"signal/brier_reward/centered_abs_mean": 0.299970406293869,
"signal/brier_reward/group_std_mean": 0.3515132188796997,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.1499852031469345,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.1499852031469345,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3983506977558136,
"signal/confidence_one_or_zero/group_std_mean": 0.4477837383747101,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.00555555559694767,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.98350675823167e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.98350675823167e-06,
"signal/format_reward/centered_abs_mean": 0.43616536259651184,
"signal/format_reward/group_std_mean": 0.472662889957428,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.21808268129825592,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.21808268129825592,
"signal/mean_confidence_reward/centered_abs_mean": 0.16609691679477692,
"signal/mean_confidence_reward/group_std_mean": 0.24436136484146118,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.660969246586319e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.660969246586319e-06,
"step": 10
},
{
"calibration/aurc": 0.5529991990141164,
"calibration/batch_distribution_entropy": 0.25530890439239107,
"calibration/batch_entropy_100bins": 0.3490627169603118,
"calibration/batch_entropy_10bins": 0.25530890439239107,
"calibration/batch_entropy_50bins": 0.40684862802386473,
"calibration/batch_uniqueness": 0.507648785050892,
"calibration/confidence_entropy": 0.21876854337479537,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.25530890439239107,
"calibration/distribution_entropy_100": 0.3490627169603118,
"calibration/ece": 0.5166797503075292,
"calibration/mean_confidence": 0.9174774627907002,
"calibration/unique_confidence_per_question": 0.033854166666666664,
"calibration/unique_confidences": 13.0,
"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": 0.02074652777777779,
"completions/max_length": 4065.4,
"completions/max_terminated_length": 4065.4,
"completions/mean_length": 500.5936706542969,
"completions/mean_terminated_length": 511.26482543945315,
"completions/min_length": 0.0,
"completions/min_terminated_length": 5.8,
"epoch": 0.036057692307692304,
"grad_norm": 0.00489252433180809,
"learning_rate": 3.6057692307692306e-07,
"loss": 0.0001,
"num_tokens": 26764513.0,
"reward": 0.6016612768173217,
"reward_std": 0.5342199921607971,
"rewards/accuracy_reward": 0.25651041865348817,
"rewards/brier_reward": 0.3121532380580902,
"rewards/confidence_one_or_zero": 0.33567708134651186,
"rewards/format_reward": 0.6346354007720947,
"rewards/mean_confidence_reward": 0.8399278521537781,
"sampling/batch_mean_priority_error": 0.4096629944680966,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 1.0,
"sampling/error_ema_max": 0.09385144263505936,
"sampling/error_ema_mean": 0.0013033524388447403,
"sampling/priority_kl": 0.029195374995470046,
"sampling/priority_scale": 419434.37669820746,
"sampling/prob_entropy": 10.279756546020508,
"sampling/prob_max": 3.438788626226596e-05,
"sampling/prob_min": 5.128296152179246e-07,
"sampling/prompt_draws_max": 1.0,
"sampling/prompt_draws_mean": 0.031199999153614044,
"sampling/prompt_draws_total": 936.0,
"sampling/seen_fraction": 0.031199999153614044,
"sampling/unseen_fraction": 0.968800000846386,
"signal/accuracy_reward/centered_abs_mean": 0.31207139492034913,
"signal/accuracy_reward/group_std_mean": 0.37355145812034607,
"signal/accuracy_reward/group_zero_std_frac": 0.08333333432674409,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.15603569746017457,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.15603569746017457,
"signal/advantage_abs_mean": 0.45120006799697876,
"signal/advantage_pre_scale_abs_mean": 0.45120006799697876,
"signal/advantage_pre_scale_std": 0.541375458240509,
"signal/advantage_std": 0.541375458240509,
"signal/brier_reward/centered_abs_mean": 0.31599055528640746,
"signal/brier_reward/group_std_mean": 0.3715106964111328,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15799527764320373,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15799527764320373,
"signal/confidence_one_or_zero/centered_abs_mean": 0.401806640625,
"signal/confidence_one_or_zero/group_std_mean": 0.4505423367023468,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.002777777798473835,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.018066374555929e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.018066374555929e-06,
"signal/format_reward/centered_abs_mean": 0.42190213203430177,
"signal/format_reward/group_std_mean": 0.46364007592201234,
"signal/format_reward/group_zero_std_frac": 0.0,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.21095106601715088,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.21095106601715088,
"signal/mean_confidence_reward/centered_abs_mean": 0.15849553644657136,
"signal/mean_confidence_reward/group_std_mean": 0.23422836661338806,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.5849552255531307e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.5849552255531307e-06,
"step": 15
},
{
"calibration/aurc": 0.5617025961821958,
"calibration/batch_distribution_entropy": 0.30985545137677556,
"calibration/batch_entropy_100bins": 0.36597686438896854,
"calibration/batch_entropy_10bins": 0.30985545137677556,
"calibration/batch_entropy_50bins": 0.4230100127785527,
"calibration/batch_uniqueness": 0.5376178366974512,
"calibration/confidence_entropy": 0.23943032642759107,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.30985545137677556,
"calibration/distribution_entropy_100": 0.36597686438896854,
"calibration/ece": 0.5239923082317872,
"calibration/mean_confidence": 0.9072021034937515,
"calibration/unique_confidence_per_question": 0.03177083333333333,
"calibration/unique_confidences": 12.2,
"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": 0.018315972222222233,
"completions/max_length": 4024.0,
"completions/max_terminated_length": 4024.0,
"completions/mean_length": 480.8516540527344,
"completions/mean_terminated_length": 489.8557922363281,
"completions/min_length": 0.0,
"completions/min_terminated_length": 19.2,
"epoch": 0.04807692307692308,
"grad_norm": 0.00305964145809412,
"learning_rate": 4.807692307692308e-07,
"loss": 0.0029,
"num_tokens": 35392660.0,
"reward": 0.7069159865379333,
"reward_std": 0.5072059988975525,
"rewards/accuracy_reward": 0.30347222089767456,
"rewards/brier_reward": 0.3697111070156097,
"rewards/confidence_one_or_zero": 0.323784726858139,
"rewards/format_reward": 0.740625011920929,
"rewards/mean_confidence_reward": 0.8573902249336243,
"sampling/batch_mean_priority_error": 0.40485228530318923,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9972222222222221,
"sampling/error_ema_max": 0.09483172744512558,
"sampling/error_ema_mean": 0.0017890206538140774,
"sampling/priority_kl": 0.02999992184340954,
"sampling/priority_scale": 4.172456359863281,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 3.469034854788333e-05,
"sampling/prob_min": 2.624379021654022e-06,
"sampling/prompt_draws_max": 1.2,
"sampling/prompt_draws_mean": 0.04320000112056732,
"sampling/prompt_draws_total": 1296.0,
"sampling/seen_fraction": 0.043193334341049196,
"sampling/unseen_fraction": 0.9568066656589508,
"signal/accuracy_reward/centered_abs_mean": 0.32134330868721006,
"signal/accuracy_reward/group_std_mean": 0.376801460981369,
"signal/accuracy_reward/group_zero_std_frac": 0.09166666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16067165434360503,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.16067165434360503,
"signal/advantage_abs_mean": 0.4259904146194458,
"signal/advantage_pre_scale_abs_mean": 0.4259904146194458,
"signal/advantage_pre_scale_std": 0.5144033670425415,
"signal/advantage_std": 0.5144033670425415,
"signal/brier_reward/centered_abs_mean": 0.31932910084724425,
"signal/brier_reward/group_std_mean": 0.3702052295207977,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15966455042362213,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15966455042362213,
"signal/confidence_one_or_zero/centered_abs_mean": 0.39694010019302367,
"signal/confidence_one_or_zero/group_std_mean": 0.4463639736175537,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.00555555559694767,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.969400859205053e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.969400859205053e-06,
"signal/format_reward/centered_abs_mean": 0.34589843153953553,
"signal/format_reward/group_std_mean": 0.4145213305950165,
"signal/format_reward/group_zero_std_frac": 0.00555555559694767,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.17294921576976777,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.17294921576976777,
"signal/mean_confidence_reward/centered_abs_mean": 0.1368652194738388,
"signal/mean_confidence_reward/group_std_mean": 0.20745837390422822,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.3686520787814516e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.3686520787814516e-06,
"step": 20
},
{
"calibration/aurc": 0.6331574084168297,
"calibration/batch_distribution_entropy": 0.3056451127645691,
"calibration/batch_entropy_100bins": 0.3621909437209661,
"calibration/batch_entropy_10bins": 0.3056451127645691,
"calibration/batch_entropy_50bins": 0.4203603709425834,
"calibration/batch_uniqueness": 0.5275142501595815,
"calibration/confidence_entropy": 0.24152741604416814,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.3056451127645691,
"calibration/distribution_entropy_100": 0.3621909437209661,
"calibration/ece": 0.5781299287679891,
"calibration/mean_confidence": 0.9073145999485227,
"calibration/unique_confidence_per_question": 0.0359375,
"calibration/unique_confidences": 13.8,
"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": 0.014756944444444465,
"completions/max_length": 3893.4,
"completions/max_terminated_length": 3893.4,
"completions/mean_length": 448.126220703125,
"completions/mean_terminated_length": 454.97858276367185,
"completions/min_length": 0.0,
"completions/min_terminated_length": 35.2,
"epoch": 0.06009615384615385,
"grad_norm": 0.0026125996373593807,
"learning_rate": 6.009615384615385e-07,
"loss": -0.0093,
"num_tokens": 43616866.0,
"reward": 0.8259699463844299,
"reward_std": 0.4544887006282806,
"rewards/accuracy_reward": 0.34383679628372193,
"rewards/brier_reward": 0.4322117447853088,
"rewards/confidence_one_or_zero": 0.2889756917953491,
"rewards/format_reward": 0.8758680701255799,
"rewards/mean_confidence_reward": 0.8741405367851257,
"sampling/batch_mean_priority_error": 0.3930652402572373,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9916666666666668,
"sampling/error_ema_max": 0.11524507701396942,
"sampling/error_ema_mean": 0.002272505545988679,
"sampling/priority_kl": 0.030000098794698716,
"sampling/priority_scale": 3.1412979844026268,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 3.4941424382850526e-05,
"sampling/prob_min": 2.6300597710360306e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.0551999993622303,
"sampling/prompt_draws_total": 1656.0,
"sampling/seen_fraction": 0.05510666668415069,
"sampling/unseen_fraction": 0.9448933333158493,
"signal/accuracy_reward/centered_abs_mean": 0.3353352904319763,
"signal/accuracy_reward/group_std_mean": 0.39114522337913515,
"signal/accuracy_reward/group_zero_std_frac": 0.06944444626569748,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16766764521598815,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.16766764521598815,
"signal/advantage_abs_mean": 0.37308090925216675,
"signal/advantage_pre_scale_abs_mean": 0.37308090925216675,
"signal/advantage_pre_scale_std": 0.46204503178596495,
"signal/advantage_std": 0.46204503178596495,
"signal/brier_reward/centered_abs_mean": 0.31565704345703127,
"signal/brier_reward/group_std_mean": 0.36644824147224425,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15782852172851564,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15782852172851564,
"signal/confidence_one_or_zero/centered_abs_mean": 0.3759385883808136,
"signal/confidence_one_or_zero/group_std_mean": 0.4346859216690063,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.759385663215653e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.759385663215653e-06,
"signal/format_reward/centered_abs_mean": 0.1980034738779068,
"signal/format_reward/group_std_mean": 0.29695329666137693,
"signal/format_reward/group_zero_std_frac": 0.06388889048248529,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0990017369389534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0990017369389534,
"signal/mean_confidence_reward/centered_abs_mean": 0.11213915348052979,
"signal/mean_confidence_reward/group_std_mean": 0.1708782970905304,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.121391574088193e-06,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 1.121391574088193e-06,
"step": 25
},
{
"calibration/aurc": 0.5908165285640636,
"calibration/batch_distribution_entropy": 0.2872097804447499,
"calibration/batch_entropy_100bins": 0.3629192291377286,
"calibration/batch_entropy_10bins": 0.2872097804447499,
"calibration/batch_entropy_50bins": 0.42088789860453224,
"calibration/batch_uniqueness": 0.5167590103063137,
"calibration/confidence_entropy": 0.23885239474768563,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.0,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.2872097804447499,
"calibration/distribution_entropy_100": 0.3629192291377286,
"calibration/ece": 0.5238896392678473,
"calibration/mean_confidence": 0.9082567609618604,
"calibration/unique_confidence_per_question": 0.040625,
"calibration/unique_confidences": 15.6,
"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": 0.01119791666666663,
"completions/max_length": 3916.0,
"completions/max_terminated_length": 3916.0,
"completions/mean_length": 455.15147705078124,
"completions/mean_terminated_length": 460.3227966308594,
"completions/min_length": 0.0,
"completions/min_terminated_length": 61.6,
"epoch": 0.07211538461538461,
"grad_norm": 0.0015309693990275264,
"learning_rate": 7.211538461538461e-07,
"loss": -0.0117,
"num_tokens": 51960627.0,
"reward": 0.9048501491546631,
"reward_std": 0.40102980136871336,
"rewards/accuracy_reward": 0.38498263955116274,
"rewards/brier_reward": 0.47920804619789126,
"rewards/confidence_one_or_zero": 0.2866319477558136,
"rewards/format_reward": 0.9454861044883728,
"rewards/mean_confidence_reward": 0.8894478797912597,
"sampling/batch_mean_priority_error": 0.38290548384535295,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.986111111111111,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0027343787252902983,
"sampling/priority_kl": 0.029999807104468347,
"sampling/priority_scale": 2.616620468813926,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 3.516697615850717e-05,
"sampling/prob_min": 3.984351451435941e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.0671999990940094,
"sampling/prompt_draws_total": 2016.0,
"sampling/seen_fraction": 0.06696666553616523,
"sampling/unseen_fraction": 0.9330333344638347,
"signal/accuracy_reward/centered_abs_mean": 0.3223904073238373,
"signal/accuracy_reward/group_std_mean": 0.3792769551277161,
"signal/accuracy_reward/group_zero_std_frac": 0.08888889029622078,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16119520366191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.16119520366191864,
"signal/advantage_abs_mean": 0.33022568821907045,
"signal/advantage_pre_scale_abs_mean": 0.33022568821907045,
"signal/advantage_pre_scale_std": 0.41163812279701234,
"signal/advantage_std": 0.41163812279701234,
"signal/brier_reward/centered_abs_mean": 0.3007372856140137,
"signal/brier_reward/group_std_mean": 0.3528658151626587,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15036864280700685,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.15036864280700685,
"signal/confidence_one_or_zero/centered_abs_mean": 0.36525607109069824,
"signal/confidence_one_or_zero/group_std_mean": 0.42673117518424986,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.002777777798473835,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.652560599221033e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.652560599221033e-06,
"signal/format_reward/centered_abs_mean": 0.0925564244389534,
"signal/format_reward/group_std_mean": 0.1731490135192871,
"signal/format_reward/group_zero_std_frac": 0.3027777820825577,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0462782122194767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0462782122194767,
"signal/mean_confidence_reward/centered_abs_mean": 0.09154417216777802,
"signal/mean_confidence_reward/group_std_mean": 0.14112971723079681,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.154416829915135e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.154416829915135e-07,
"step": 30
},
{
"calibration/aurc": 0.49865686253089087,
"calibration/batch_distribution_entropy": 0.3215165241574327,
"calibration/batch_entropy_100bins": 0.36894158251549947,
"calibration/batch_entropy_10bins": 0.3215165241574327,
"calibration/batch_entropy_50bins": 0.42884229658872697,
"calibration/batch_uniqueness": 0.5198789348898746,
"calibration/confidence_entropy": 0.2638777012526705,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.0,
"calibration/coverage@30%": 0.08723404255319149,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.3215165241574327,
"calibration/distribution_entropy_100": 0.36894158251549947,
"calibration/ece": 0.4416422972833175,
"calibration/mean_confidence": 0.8992454127844949,
"calibration/unique_confidence_per_question": 0.0375,
"calibration/unique_confidences": 14.4,
"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": 0.010416666666666652,
"completions/max_length": 3919.6,
"completions/max_terminated_length": 3919.6,
"completions/mean_length": 482.71397705078124,
"completions/mean_terminated_length": 487.82008056640626,
"completions/min_length": 0.0,
"completions/min_terminated_length": 74.0,
"epoch": 0.08413461538461539,
"grad_norm": 0.0012755130883306265,
"learning_rate": 8.41346153846154e-07,
"loss": -0.0108,
"num_tokens": 60604820.0,
"reward": 0.9773583769798279,
"reward_std": 0.3674590468406677,
"rewards/accuracy_reward": 0.4388888955116272,
"rewards/brier_reward": 0.5380276739597321,
"rewards/confidence_one_or_zero": 0.23333333134651185,
"rewards/format_reward": 0.9777777671813965,
"rewards/mean_confidence_reward": 0.8878483891487121,
"sampling/batch_mean_priority_error": 0.3344713397737706,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.975,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.003155976487323642,
"sampling/priority_kl": 0.03000006154179573,
"sampling/priority_scale": 2.2848868130706252,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 3.537431184668094e-05,
"sampling/prob_min": 5.246551245363662e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.07920000106096267,
"sampling/prompt_draws_total": 2376.0,
"sampling/seen_fraction": 0.07872666716575623,
"sampling/unseen_fraction": 0.9212733328342437,
"signal/accuracy_reward/centered_abs_mean": 0.3136827230453491,
"signal/accuracy_reward/group_std_mean": 0.3738104820251465,
"signal/accuracy_reward/group_zero_std_frac": 0.08611111268401146,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.15684136152267455,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.15684136152267455,
"signal/advantage_abs_mean": 0.3014487147331238,
"signal/advantage_pre_scale_abs_mean": 0.3014487147331238,
"signal/advantage_pre_scale_std": 0.3801884174346924,
"signal/advantage_std": 0.3801884174346924,
"signal/brier_reward/centered_abs_mean": 0.2799021899700165,
"signal/brier_reward/group_std_mean": 0.3350242257118225,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.13995109498500824,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.13995109498500824,
"signal/confidence_one_or_zero/centered_abs_mean": 0.31881510019302367,
"signal/confidence_one_or_zero/group_std_mean": 0.39367093443870543,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.01111111119389534,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.1881509130471386e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.1881509130471386e-06,
"signal/format_reward/centered_abs_mean": 0.04067925326526165,
"signal/format_reward/group_std_mean": 0.08876850455999374,
"signal/format_reward/group_zero_std_frac": 0.5944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.020339626632630826,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.020339626632630826,
"signal/mean_confidence_reward/centered_abs_mean": 0.08388432711362839,
"signal/mean_confidence_reward/group_std_mean": 0.12709739059209824,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.388432092942822e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.388432092942822e-07,
"step": 35
},
{
"calibration/aurc": 0.4823138666235419,
"calibration/batch_distribution_entropy": 0.3863156724868472,
"calibration/batch_entropy_100bins": 0.38263042759035665,
"calibration/batch_entropy_10bins": 0.3863156724868472,
"calibration/batch_entropy_50bins": 0.44357007596939224,
"calibration/batch_uniqueness": 0.524572624992154,
"calibration/confidence_entropy": 0.3053943449202666,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.0,
"calibration/coverage@20%": 0.0,
"calibration/coverage@25%": 0.04656084656084656,
"calibration/coverage@30%": 0.047089947089947085,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.3863156724868472,
"calibration/distribution_entropy_100": 0.38263042759035665,
"calibration/ece": 0.4134137900803255,
"calibration/mean_confidence": 0.8813164186107763,
"calibration/unique_confidence_per_question": 0.04010416666666667,
"calibration/unique_confidences": 15.4,
"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": 0.010677083333333327,
"completions/max_length": 3680.4,
"completions/max_terminated_length": 3680.4,
"completions/mean_length": 512.7624267578125,
"completions/mean_terminated_length": 518.3632568359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 105.8,
"epoch": 0.09615384615384616,
"grad_norm": 0.0012608061078935862,
"learning_rate": 9.615384615384617e-07,
"loss": -0.0087,
"num_tokens": 69623683.0,
"reward": 1.0160284519195557,
"reward_std": 0.34829759001731875,
"rewards/accuracy_reward": 0.4693576395511627,
"rewards/brier_reward": 0.579345452785492,
"rewards/confidence_one_or_zero": 0.15078125,
"rewards/format_reward": 0.9833333253860473,
"rewards/mean_confidence_reward": 0.8721679806709289,
"sampling/batch_mean_priority_error": 0.28377850106167013,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9694444444444444,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0035211187321692707,
"sampling/priority_kl": 0.030000394210219383,
"sampling/priority_scale": 2.052108299965039,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 3.5567307349992916e-05,
"sampling/prob_min": 6.397180368367117e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.09120000004768372,
"sampling/prompt_draws_total": 2736.0,
"sampling/seen_fraction": 0.09033999890089035,
"sampling/unseen_fraction": 0.9096600010991096,
"signal/accuracy_reward/centered_abs_mean": 0.302978515625,
"signal/accuracy_reward/group_std_mean": 0.3657167613506317,
"signal/accuracy_reward/group_zero_std_frac": 0.08888889029622078,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1514892578125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1514892578125,
"signal/advantage_abs_mean": 0.2835480809211731,
"signal/advantage_pre_scale_abs_mean": 0.2835480809211731,
"signal/advantage_pre_scale_std": 0.3627303600311279,
"signal/advantage_std": 0.3627303600311279,
"signal/brier_reward/centered_abs_mean": 0.25960286259651183,
"signal/brier_reward/group_std_mean": 0.31365215182304385,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.12980143129825591,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.12980143129825591,
"signal/confidence_one_or_zero/centered_abs_mean": 0.2295952707529068,
"signal/confidence_one_or_zero/group_std_mean": 0.3169353187084198,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.08611111491918563,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.2959525495025445e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.2959525495025445e-06,
"signal/format_reward/centered_abs_mean": 0.03025173619389534,
"signal/format_reward/group_std_mean": 0.06552095487713813,
"signal/format_reward/group_zero_std_frac": 0.700000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01512586809694767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01512586809694767,
"signal/mean_confidence_reward/centered_abs_mean": 0.08143319040536881,
"signal/mean_confidence_reward/group_std_mean": 0.12138528972864152,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.143319064402022e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.143319064402022e-07,
"step": 40
},
{
"calibration/aurc": 0.3800912542992382,
"calibration/batch_distribution_entropy": 0.35937801231294425,
"calibration/batch_entropy_100bins": 0.3474659546744869,
"calibration/batch_entropy_10bins": 0.35937801231294425,
"calibration/batch_entropy_50bins": 0.4084236105276625,
"calibration/batch_uniqueness": 0.4383812566815227,
"calibration/confidence_entropy": 0.3172211438602935,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.03622047244094488,
"calibration/coverage@20%": 0.05774278215223098,
"calibration/coverage@25%": 0.19790026246719158,
"calibration/coverage@30%": 0.2,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.35937801231294425,
"calibration/distribution_entropy_100": 0.3474659546744869,
"calibration/ece": 0.2940468530345476,
"calibration/mean_confidence": 0.882873292367725,
"calibration/unique_confidence_per_question": 0.028645833333333332,
"calibration/unique_confidences": 11.0,
"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": 0.009201388888888884,
"completions/max_length": 4048.2,
"completions/max_terminated_length": 4048.2,
"completions/mean_length": 559.3792724609375,
"completions/mean_terminated_length": 564.5820068359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 117.0,
"epoch": 0.10817307692307693,
"grad_norm": 0.003110303310677409,
"learning_rate": 1.0817307692307693e-06,
"loss": -0.0034,
"num_tokens": 79186324.0,
"reward": 1.0602930545806886,
"reward_std": 0.306660133600235,
"rewards/accuracy_reward": 0.5138888955116272,
"rewards/brier_reward": 0.6227369189262391,
"rewards/confidence_one_or_zero": 0.10625,
"rewards/format_reward": 0.9839409589767456,
"rewards/mean_confidence_reward": 0.8621986746788025,
"sampling/batch_mean_priority_error": 0.23946882227667854,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9694444444444444,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0038303794339299203,
"sampling/priority_kl": 0.030000080913305284,
"sampling/priority_scale": 1.8766139150131493,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 3.5752353142015636e-05,
"sampling/prob_min": 7.404062307614367e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.10320000052452087,
"sampling/prompt_draws_total": 3096.0,
"sampling/seen_fraction": 0.10200666785240173,
"sampling/unseen_fraction": 0.8979933321475982,
"signal/accuracy_reward/centered_abs_mean": 0.2545138865709305,
"signal/accuracy_reward/group_std_mean": 0.3217469334602356,
"signal/accuracy_reward/group_zero_std_frac": 0.13888889253139497,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.12725694328546525,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.12725694328546525,
"signal/advantage_abs_mean": 0.23891697824001312,
"signal/advantage_pre_scale_abs_mean": 0.23891697824001312,
"signal/advantage_pre_scale_std": 0.3288706004619598,
"signal/advantage_std": 0.3288706004619598,
"signal/brier_reward/centered_abs_mean": 0.21723864376544952,
"signal/brier_reward/group_std_mean": 0.2735393285751343,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.10861932188272476,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.10861932188272476,
"signal/confidence_one_or_zero/centered_abs_mean": 0.1694227397441864,
"signal/confidence_one_or_zero/group_std_mean": 0.2620739072561264,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.13333333432674407,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.694227375992341e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.694227375992341e-06,
"signal/format_reward/centered_abs_mean": 0.02766384519636631,
"signal/format_reward/group_std_mean": 0.057002028077840806,
"signal/format_reward/group_zero_std_frac": 0.7444444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.013831922598183156,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.013831922598183156,
"signal/mean_confidence_reward/centered_abs_mean": 0.07951802164316177,
"signal/mean_confidence_reward/group_std_mean": 0.11687755137681961,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.951802217576187e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.951802217576187e-07,
"step": 45
},
{
"calibration/aurc": 0.30584174843599155,
"calibration/batch_distribution_entropy": 0.4698661757482432,
"calibration/batch_entropy_100bins": 0.359959773527773,
"calibration/batch_entropy_10bins": 0.4698661757482432,
"calibration/batch_entropy_50bins": 0.42373882129655405,
"calibration/batch_uniqueness": 0.4616731414535427,
"calibration/confidence_entropy": 0.37746819721208036,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.04970530029191465,
"calibration/coverage@20%": 0.182396604745731,
"calibration/coverage@25%": 0.2594019523393139,
"calibration/coverage@30%": 0.5653119338313628,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.4698661757482432,
"calibration/distribution_entropy_100": 0.359959773527773,
"calibration/ece": 0.22593068011557005,
"calibration/mean_confidence": 0.8520271057071687,
"calibration/unique_confidence_per_question": 0.03072916666666667,
"calibration/unique_confidences": 11.8,
"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": 0.012065972222222231,
"completions/max_length": 3768.4,
"completions/max_terminated_length": 3768.4,
"completions/mean_length": 568.4737854003906,
"completions/mean_terminated_length": 575.4648315429688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 120.4,
"epoch": 0.1201923076923077,
"grad_norm": 0.0008945247973315418,
"learning_rate": 1.201923076923077e-06,
"loss": -0.0054,
"num_tokens": 88842246.0,
"reward": 1.1304325819015504,
"reward_std": 0.2705526381731033,
"rewards/accuracy_reward": 0.5894097208976745,
"rewards/brier_reward": 0.6880171537399292,
"rewards/confidence_one_or_zero": 0.06059027761220932,
"rewards/format_reward": 0.9834201335906982,
"rewards/mean_confidence_reward": 0.8420838594436646,
"sampling/batch_mean_priority_error": 0.1750527001027404,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9388888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004068218637257814,
"sampling/priority_kl": 0.030000039935112,
"sampling/priority_scale": 1.7397352575790137,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.592700959416106e-05,
"sampling/prob_min": 8.291824997286312e-06,
"sampling/prompt_draws_max": 2.0,
"sampling/prompt_draws_mean": 0.11519999951124191,
"sampling/prompt_draws_total": 3456.0,
"sampling/seen_fraction": 0.11345999985933304,
"sampling/unseen_fraction": 0.8865400001406669,
"signal/accuracy_reward/centered_abs_mean": 0.21812066435813904,
"signal/accuracy_reward/group_std_mean": 0.28616309762001035,
"signal/accuracy_reward/group_zero_std_frac": 0.1916666656732559,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10906033217906952,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10906033217906952,
"signal/advantage_abs_mean": 0.2035945475101471,
"signal/advantage_pre_scale_abs_mean": 0.2035945475101471,
"signal/advantage_pre_scale_std": 0.3032866775989532,
"signal/advantage_std": 0.3032866775989532,
"signal/brier_reward/centered_abs_mean": 0.18060558736324311,
"signal/brier_reward/group_std_mean": 0.23320825695991515,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09030279368162156,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.09030279368162156,
"signal/confidence_one_or_zero/centered_abs_mean": 0.10205078125,
"signal/confidence_one_or_zero/group_std_mean": 0.17942221462726593,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.3222222238779068,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.0205078069702722e-06,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.0205078069702722e-06,
"signal/format_reward/centered_abs_mean": 0.02902018204331398,
"signal/format_reward/group_std_mean": 0.055744080990552905,
"signal/format_reward/group_zero_std_frac": 0.7666666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01451009102165699,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01451009102165699,
"signal/mean_confidence_reward/centered_abs_mean": 0.0795755386352539,
"signal/mean_confidence_reward/group_std_mean": 0.1120973750948906,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.957553975757037e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.957553975757037e-07,
"step": 50
},
{
"epoch": 0.1201923076923077,
"eval_calibration/aurc": 0.36693158510830576,
"eval_calibration/batch_distribution_entropy": 0.4879616019308547,
"eval_calibration/batch_entropy_100bins": 0.35135930466067045,
"eval_calibration/batch_entropy_10bins": 0.4879616019308547,
"eval_calibration/batch_entropy_50bins": 0.41198770433599563,
"eval_calibration/batch_uniqueness": 0.4472331427676404,
"eval_calibration/confidence_entropy": 0.40263867487886834,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.0,
"eval_calibration/coverage@20%": 0.0,
"eval_calibration/coverage@25%": 0.02920353982300885,
"eval_calibration/coverage@30%": 0.02920353982300885,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.4879616019308547,
"eval_calibration/distribution_entropy_100": 0.35135930466067045,
"eval_calibration/ece": 0.2521858407079647,
"eval_calibration/mean_confidence": 0.8436902654867258,
"eval_calibration/unique_confidence_per_question": 0.013020833333333334,
"eval_calibration/unique_confidences": 15,
"eval_completions/clipped_ratio": 0.01215277777777779,
"eval_completions/max_length": 2192.5,
"eval_completions/max_terminated_length": 2192.5,
"eval_completions/mean_length": 593.2312622070312,
"eval_completions/mean_terminated_length": 600.4430440266927,
"eval_completions/min_length": 41.833333333333336,
"eval_completions/min_terminated_length": 185.0,
"eval_loss": 0.0,
"eval_num_tokens": 88842246.0,
"eval_reward": 1.1237780451774597,
"eval_reward_std": 0.43651104470094043,
"eval_rewards/accuracy_reward": 0.5815972089767456,
"eval_rewards/brier_reward": 0.6850390136241913,
"eval_rewards/confidence_one_or_zero": 0.02951388938042025,
"eval_rewards/format_reward": 0.9809027711550394,
"eval_rewards/mean_confidence_reward": 0.8281857669353485,
"eval_runtime": 204.0598,
"eval_samples_per_second": 4.901,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4702690988779068,
"eval_signal/accuracy_reward/group_std_mean": 0.49234892924626666,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2351345494389534,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.2351345494389534,
"eval_signal/advantage_abs_mean": 0.406392698486646,
"eval_signal/advantage_pre_scale_abs_mean": 0.406392698486646,
"eval_signal/advantage_pre_scale_std": 0.4317004084587097,
"eval_signal/advantage_std": 0.4317004084587097,
"eval_signal/brier_reward/centered_abs_mean": 0.32436902821063995,
"eval_signal/brier_reward/group_std_mean": 0.35272473096847534,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.16218451410531998,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.16218451410531998,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.05544704866285125,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.1265144463007649,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.416666679084301,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 5.544704796041818e-07,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 5.544704796041818e-07,
"eval_signal/format_reward/centered_abs_mean": 0.03656684048473835,
"eval_signal/format_reward/group_std_mean": 0.09607256638507049,
"eval_signal/format_reward/group_zero_std_frac": 0.5000000074505806,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.018283420242369175,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.018283420242369175,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.08548501133918762,
"eval_signal/mean_confidence_reward/group_std_mean": 0.13385692487160364,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.54850100040494e-07,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 8.54850100040494e-07,
"eval_steps_per_second": 0.029,
"step": 50
},
{
"epoch": 0.1201923076923077,
"step": 50,
"train_probe_calibration/aurc": 0.3232102008686786,
"train_probe_calibration/batch_distribution_entropy": 0.5061747512148379,
"train_probe_calibration/batch_entropy_100bins": 0.3552038459733368,
"train_probe_calibration/batch_entropy_10bins": 0.5061747512148379,
"train_probe_calibration/batch_entropy_50bins": 0.41709690460632987,
"train_probe_calibration/batch_uniqueness": 0.45442081376408144,
"train_probe_calibration/confidence_entropy": 0.40798928804916534,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.0,
"train_probe_calibration/coverage@15%": 0.0,
"train_probe_calibration/coverage@20%": 0.0,
"train_probe_calibration/coverage@25%": 0.03353927625772286,
"train_probe_calibration/coverage@30%": 0.5189761694616064,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5061747512148379,
"train_probe_calibration/distribution_entropy_100": 0.3552038459733368,
"train_probe_calibration/ece": 0.2209002647837599,
"train_probe_calibration/mean_confidence": 0.838729037952339,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.0078125,
"train_probe_completions/max_length": 2798.0,
"train_probe_completions/max_terminated_length": 2798.0,
"train_probe_completions/mean_length": 617.9297180175781,
"train_probe_completions/mean_terminated_length": 622.6372578938802,
"train_probe_completions/min_length": 43.166666666666664,
"train_probe_completions/min_terminated_length": 179.16666666666666,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 88842246.0,
"train_probe_reward": 1.1511651674906414,
"train_probe_reward_std": 0.4263061086336772,
"train_probe_rewards/accuracy_reward": 0.6076388955116272,
"train_probe_rewards/brier_reward": 0.7111672163009644,
"train_probe_rewards/confidence_one_or_zero": 0.0329861108524104,
"train_probe_rewards/format_reward": 0.9835069477558136,
"train_probe_rewards/mean_confidence_reward": 0.824895828962326,
"train_probe_runtime": 203.2902,
"train_probe_samples_per_second": 4.919,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4641927083333333,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4889263262351354,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.23209635416666666,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.23209635416666666,
"train_probe_signal/advantage_abs_mean": 0.3949887653191884,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.3949887653191884,
"train_probe_signal/advantage_pre_scale_std": 0.4220304836829503,
"train_probe_signal/advantage_std": 0.4220304836829503,
"train_probe_signal/brier_reward/centered_abs_mean": 0.30813247958819073,
"train_probe_signal/brier_reward/group_std_mean": 0.34062348306179047,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.15406623979409537,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.15406623979409537,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.06228298662851254,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.1442195667574803,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.3333333432674408,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.228298445876135e-07,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.228298445876135e-07,
"train_probe_signal/format_reward/centered_abs_mean": 0.03162977425381541,
"train_probe_signal/format_reward/group_std_mean": 0.08433058392256498,
"train_probe_signal/format_reward/group_zero_std_frac": 0.5555555696288744,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.015814887126907706,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.015814887126907706,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.08944117774566014,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.13466068108876547,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.944117553255637e-07,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 8.944117553255637e-07,
"train_probe_steps_per_second": 0.03
},
{
"calibration/aurc": 0.3668222300226895,
"calibration/batch_distribution_entropy": 0.5336098111385617,
"calibration/batch_entropy_100bins": 0.35464743111118563,
"calibration/batch_entropy_10bins": 0.5336098111385617,
"calibration/batch_entropy_50bins": 0.41717774105665467,
"calibration/batch_uniqueness": 0.47823134437695564,
"calibration/confidence_entropy": 0.43410787120525496,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.0,
"calibration/coverage@15%": 0.02074468085106383,
"calibration/coverage@20%": 0.12925531914893618,
"calibration/coverage@25%": 0.2,
"calibration/coverage@30%": 0.21461949754632678,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.5336098111385617,
"calibration/distribution_entropy_100": 0.35464743111118563,
"calibration/ece": 0.2233237631652508,
"calibration/mean_confidence": 0.8207327630559618,
"calibration/unique_confidence_per_question": 0.02708333333333333,
"calibration/unique_confidences": 10.4,
"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": 0.01640625,
"completions/max_length": 4039.6,
"completions/max_terminated_length": 4039.6,
"completions/mean_length": 613.0697143554687,
"completions/mean_terminated_length": 623.2099731445312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 120.4,
"epoch": 0.13221153846153846,
"grad_norm": 0.0007423937786370516,
"learning_rate": 1.3221153846153848e-06,
"loss": -0.0077,
"num_tokens": 98970729.0,
"reward": 1.1381091117858886,
"reward_std": 0.2668158829212189,
"rewards/accuracy_reward": 0.5957465291023254,
"rewards/brier_reward": 0.7014619588851929,
"rewards/confidence_one_or_zero": 0.02100694477558136,
"rewards/format_reward": 0.9789930582046509,
"rewards/mean_confidence_reward": 0.810854148864746,
"sampling/batch_mean_priority_error": 0.14243097914081077,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9555555555555555,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004251902643591166,
"sampling/priority_kl": 0.030000291764736176,
"sampling/priority_scale": 1.6290871502365918,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.609485720517114e-05,
"sampling/prob_min": 8.77213133207988e-06,
"sampling/prompt_draws_max": 2.2,
"sampling/prompt_draws_mean": 0.12720000147819518,
"sampling/prompt_draws_total": 3816.0,
"sampling/seen_fraction": 0.12482666820287705,
"sampling/unseen_fraction": 0.875173331797123,
"signal/accuracy_reward/centered_abs_mean": 0.2257758229970932,
"signal/accuracy_reward/group_std_mean": 0.2867546260356903,
"signal/accuracy_reward/group_zero_std_frac": 0.225,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1128879114985466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1128879114985466,
"signal/advantage_abs_mean": 0.207180854678154,
"signal/advantage_pre_scale_abs_mean": 0.207180854678154,
"signal/advantage_pre_scale_std": 0.304352855682373,
"signal/advantage_std": 0.304352855682373,
"signal/brier_reward/centered_abs_mean": 0.17458488643169404,
"signal/brier_reward/group_std_mean": 0.22114146649837493,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08729244321584702,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.08729244321584702,
"signal/confidence_one_or_zero/centered_abs_mean": 0.03837890662252903,
"signal/confidence_one_or_zero/group_std_mean": 0.08215054646134376,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.6333333492279053,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.837890574232006e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.837890574232006e-07,
"signal/format_reward/centered_abs_mean": 0.0329210065305233,
"signal/format_reward/group_std_mean": 0.055089053511619565,
"signal/format_reward/group_zero_std_frac": 0.7916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01646050326526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01646050326526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.0847676783800125,
"signal/mean_confidence_reward/group_std_mean": 0.11370683014392853,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.47676710691303e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.47676710691303e-07,
"step": 55
},
{
"calibration/aurc": 0.3340195114348269,
"calibration/batch_distribution_entropy": 0.5726832276472693,
"calibration/batch_entropy_100bins": 0.3492907326474214,
"calibration/batch_entropy_10bins": 0.5726832276472693,
"calibration/batch_entropy_50bins": 0.410924505266504,
"calibration/batch_uniqueness": 0.4968363979020654,
"calibration/confidence_entropy": 0.47785333312889183,
"calibration/coverage@0%": 0.0026246719160104987,
"calibration/coverage@1%": 0.0026246719160104987,
"calibration/coverage@10%": 0.0026246719160104987,
"calibration/coverage@15%": 0.0026246719160104987,
"calibration/coverage@20%": 0.2347003807622639,
"calibration/coverage@25%": 0.34856220657276993,
"calibration/coverage@30%": 0.4510416666666667,
"calibration/coverage@5%": 0.0026246719160104987,
"calibration/distribution_entropy_10": 0.5726832276472693,
"calibration/distribution_entropy_100": 0.3492907326474214,
"calibration/ece": 0.17780991202081498,
"calibration/mean_confidence": 0.7972725700954527,
"calibration/unique_confidence_per_question": 0.025,
"calibration/unique_confidences": 9.6,
"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": 0.013975694444444442,
"completions/max_length": 3632.8,
"completions/max_terminated_length": 3632.8,
"completions/mean_length": 610.9026245117187,
"completions/mean_terminated_length": 619.6228271484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 156.8,
"epoch": 0.14423076923076922,
"grad_norm": 0.0008636752027086914,
"learning_rate": 1.4423076923076922e-06,
"loss": -0.008,
"num_tokens": 109068231.0,
"reward": 1.1724665880203247,
"reward_std": 0.24019188582897186,
"rewards/accuracy_reward": 0.6282986164093017,
"rewards/brier_reward": 0.7318964719772338,
"rewards/confidence_one_or_zero": 0.006597222201526165,
"rewards/format_reward": 0.9847222208976746,
"rewards/mean_confidence_reward": 0.7856779336929322,
"sampling/batch_mean_priority_error": 0.10756469318231769,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9388888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004396716970950365,
"sampling/priority_kl": 0.02999984547495842,
"sampling/priority_scale": 1.5373026251327246,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 3.625838580774143e-05,
"sampling/prob_min": 8.104404696496203e-06,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.13919999897480012,
"sampling/prompt_draws_total": 4176.0,
"sampling/seen_fraction": 0.1361933320760727,
"sampling/unseen_fraction": 0.8638066679239274,
"signal/accuracy_reward/centered_abs_mean": 0.20860460102558137,
"signal/accuracy_reward/group_std_mean": 0.2688582241535187,
"signal/accuracy_reward/group_zero_std_frac": 0.26666666865348815,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10430230051279069,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10430230051279069,
"signal/advantage_abs_mean": 0.18264085054397583,
"signal/advantage_pre_scale_abs_mean": 0.18264085054397583,
"signal/advantage_pre_scale_std": 0.2767543405294418,
"signal/advantage_std": 0.2767543405294418,
"signal/brier_reward/centered_abs_mean": 0.14837830811738967,
"signal/brier_reward/group_std_mean": 0.19285426437854766,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07418915405869483,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.07418915405869483,
"signal/confidence_one_or_zero/centered_abs_mean": 0.012413194589316844,
"signal/confidence_one_or_zero/group_std_mean": 0.03018292561173439,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.850000011920929,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.2413193815064005e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.2413193815064005e-07,
"signal/format_reward/centered_abs_mean": 0.02427300363779068,
"signal/format_reward/group_std_mean": 0.04464358240365982,
"signal/format_reward/group_zero_std_frac": 0.819444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01213650181889534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01213650181889534,
"signal/mean_confidence_reward/centered_abs_mean": 0.07837038338184357,
"signal/mean_confidence_reward/group_std_mean": 0.10648758858442306,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.837037969693483e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.837037969693483e-07,
"step": 60
},
{
"calibration/aurc": 0.36082880327797945,
"calibration/batch_distribution_entropy": 0.5609276442705863,
"calibration/batch_entropy_100bins": 0.3219518099981292,
"calibration/batch_entropy_10bins": 0.5609276442705863,
"calibration/batch_entropy_50bins": 0.37899646159315503,
"calibration/batch_uniqueness": 0.4252066320477523,
"calibration/confidence_entropy": 0.5287127000393115,
"calibration/coverage@0%": 0.0010540349123026288,
"calibration/coverage@1%": 0.0010540349123026288,
"calibration/coverage@10%": 0.034124901054034915,
"calibration/coverage@15%": 0.034124901054034915,
"calibration/coverage@20%": 0.03624130317043703,
"calibration/coverage@25%": 0.20264550264550266,
"calibration/coverage@30%": 0.24728959286407118,
"calibration/coverage@5%": 0.0010540349123026288,
"calibration/distribution_entropy_10": 0.5609276442705863,
"calibration/distribution_entropy_100": 0.3219518099981292,
"calibration/ece": 0.15328771390121856,
"calibration/mean_confidence": 0.7603495482266917,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.017013888888888884,
"completions/max_length": 4044.0,
"completions/max_terminated_length": 4044.0,
"completions/mean_length": 667.0696166992187,
"completions/mean_terminated_length": 678.5839111328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 173.0,
"epoch": 0.15625,
"grad_norm": 0.0009765251888893545,
"learning_rate": 1.5625e-06,
"loss": -0.0067,
"num_tokens": 119880969.0,
"reward": 1.1687859058380128,
"reward_std": 0.23864734768867493,
"rewards/accuracy_reward": 0.6236979126930237,
"rewards/brier_reward": 0.7352131009101868,
"rewards/confidence_one_or_zero": 0.001996527845039964,
"rewards/format_reward": 0.9786458373069763,
"rewards/mean_confidence_reward": 0.7466666579246521,
"sampling/batch_mean_priority_error": 0.08287107120565607,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.95,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0045060371980071064,
"sampling/priority_kl": 0.029999903962016107,
"sampling/priority_scale": 1.4598986505996436,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.6417489900486545e-05,
"sampling/prob_min": 8.696256554685532e-06,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.1512000024318695,
"sampling/prompt_draws_total": 4536.0,
"sampling/seen_fraction": 0.14750000238418579,
"sampling/unseen_fraction": 0.8524999976158142,
"signal/accuracy_reward/centered_abs_mean": 0.207763671875,
"signal/accuracy_reward/group_std_mean": 0.26939095854759215,
"signal/accuracy_reward/group_zero_std_frac": 0.2444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1038818359375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1038818359375,
"signal/advantage_abs_mean": 0.17969182133674622,
"signal/advantage_pre_scale_abs_mean": 0.17969182133674622,
"signal/advantage_pre_scale_std": 0.27758598923683164,
"signal/advantage_std": 0.27758598923683164,
"signal/brier_reward/centered_abs_mean": 0.13804238885641099,
"signal/brier_reward/group_std_mean": 0.1800360769033432,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06902119442820549,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.06902119442820549,
"signal/confidence_one_or_zero/centered_abs_mean": 0.003835720382630825,
"signal/confidence_one_or_zero/group_std_mean": 0.010643551312386989,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9416666865348816,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.835720150391353e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.835720150391353e-08,
"signal/format_reward/centered_abs_mean": 0.03416883647441864,
"signal/format_reward/group_std_mean": 0.0599904865026474,
"signal/format_reward/group_zero_std_frac": 0.7638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01708441823720932,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01708441823720932,
"signal/mean_confidence_reward/centered_abs_mean": 0.08167127966880798,
"signal/mean_confidence_reward/group_std_mean": 0.11141330301761627,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.167127930391871e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.167127930391871e-07,
"step": 65
},
{
"calibration/aurc": 0.35516211390845653,
"calibration/batch_distribution_entropy": 0.5715761667318768,
"calibration/batch_entropy_100bins": 0.31538070674384944,
"calibration/batch_entropy_10bins": 0.5715761667318768,
"calibration/batch_entropy_50bins": 0.37126106516177676,
"calibration/batch_uniqueness": 0.42730865779244337,
"calibration/confidence_entropy": 0.5586211599671339,
"calibration/coverage@0%": 0.000529100529100529,
"calibration/coverage@1%": 0.000529100529100529,
"calibration/coverage@10%": 0.000529100529100529,
"calibration/coverage@15%": 0.0164021164021164,
"calibration/coverage@20%": 0.028042328042328042,
"calibration/coverage@25%": 0.13589134691496896,
"calibration/coverage@30%": 0.22846310877806938,
"calibration/coverage@5%": 0.000529100529100529,
"calibration/distribution_entropy_10": 0.5715761667318768,
"calibration/distribution_entropy_100": 0.31538070674384944,
"calibration/ece": 0.13111338780650672,
"calibration/mean_confidence": 0.7339990149186095,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.018229166666666675,
"completions/max_length": 4040.4,
"completions/max_terminated_length": 4040.4,
"completions/mean_length": 675.2895874023437,
"completions/mean_terminated_length": 687.9796997070313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 204.6,
"epoch": 0.16826923076923078,
"grad_norm": 0.0007695965468883514,
"learning_rate": 1.682692307692308e-06,
"loss": -0.0098,
"num_tokens": 130770225.0,
"reward": 1.1733962297439575,
"reward_std": 0.22422350943088531,
"rewards/accuracy_reward": 0.6266493082046509,
"rewards/brier_reward": 0.7417432546615601,
"rewards/confidence_one_or_zero": 0.0006076388934161514,
"rewards/format_reward": 0.9783854126930237,
"rewards/mean_confidence_reward": 0.716662323474884,
"sampling/batch_mean_priority_error": 0.06427063559151758,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9305555555555556,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004584580659866333,
"sampling/priority_kl": 0.030000272020697594,
"sampling/priority_scale": 1.3937638641800731,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.6572145472746345e-05,
"sampling/prob_min": 9.225038775184658e-06,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.16319999992847442,
"sampling/prompt_draws_total": 4896.0,
"sampling/seen_fraction": 0.15870000123977662,
"sampling/unseen_fraction": 0.8412999987602234,
"signal/accuracy_reward/centered_abs_mean": 0.1993109792470932,
"signal/accuracy_reward/group_std_mean": 0.260605588555336,
"signal/accuracy_reward/group_zero_std_frac": 0.27777777910232543,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0996554896235466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0996554896235466,
"signal/advantage_abs_mean": 0.1672658920288086,
"signal/advantage_pre_scale_abs_mean": 0.1672658920288086,
"signal/advantage_pre_scale_std": 0.2621844053268433,
"signal/advantage_std": 0.2621844053268433,
"signal/brier_reward/centered_abs_mean": 0.12296129167079925,
"signal/brier_reward/group_std_mean": 0.1609590619802475,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.061480645835399625,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.061480645835399625,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011664496385492384,
"signal/confidence_one_or_zero/group_std_mean": 0.003138383664190769,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333253860473,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1664495502827777e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1664495502827777e-08,
"signal/format_reward/centered_abs_mean": 0.03267686627805233,
"signal/format_reward/group_std_mean": 0.0593119814991951,
"signal/format_reward/group_zero_std_frac": 0.7611111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.016338433139026165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.016338433139026165,
"signal/mean_confidence_reward/centered_abs_mean": 0.07880452424287795,
"signal/mean_confidence_reward/group_std_mean": 0.1066696047782898,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.880452130848425e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.880452130848425e-07,
"step": 70
},
{
"calibration/aurc": 0.2786799113370667,
"calibration/batch_distribution_entropy": 0.5479433779303641,
"calibration/batch_entropy_100bins": 0.3002895704040479,
"calibration/batch_entropy_10bins": 0.5479433779303641,
"calibration/batch_entropy_50bins": 0.3534960236351027,
"calibration/batch_uniqueness": 0.38395899595671973,
"calibration/confidence_entropy": 0.5804533421710255,
"calibration/coverage@0%": 0.004290037595336166,
"calibration/coverage@1%": 0.004290037595336166,
"calibration/coverage@10%": 0.019385679267933262,
"calibration/coverage@15%": 0.019385679267933262,
"calibration/coverage@20%": 0.08593125961735379,
"calibration/coverage@25%": 0.46208589649528753,
"calibration/coverage@30%": 0.6349800392679877,
"calibration/coverage@5%": 0.004290037595336166,
"calibration/distribution_entropy_10": 0.5479433779303641,
"calibration/distribution_entropy_100": 0.3002895704040479,
"calibration/ece": 0.08396643875579377,
"calibration/mean_confidence": 0.7134022072125192,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.01527777777777779,
"completions/max_length": 3922.2,
"completions/max_terminated_length": 3922.2,
"completions/mean_length": 662.4786743164062,
"completions/mean_terminated_length": 672.9049438476562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 194.4,
"epoch": 0.18028846153846154,
"grad_norm": 0.0006578608299605548,
"learning_rate": 1.8028846153846156e-06,
"loss": -0.014,
"num_tokens": 141473115.0,
"reward": 1.2087363958358766,
"reward_std": 0.2085721343755722,
"rewards/accuracy_reward": 0.6685763835906983,
"rewards/brier_reward": 0.7653754472732544,
"rewards/confidence_one_or_zero": 0.0006944444554392249,
"rewards/format_reward": 0.9835069417953491,
"rewards/mean_confidence_reward": 0.6979835271835327,
"sampling/batch_mean_priority_error": 0.050875203025745996,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9277777777777777,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004651712626218796,
"sampling/priority_kl": 0.030000195279717444,
"sampling/priority_scale": 1.336326873442158,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.672461898531765e-05,
"sampling/prob_min": 9.745861825649627e-06,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.17520000040531158,
"sampling/prompt_draws_total": 5256.0,
"sampling/seen_fraction": 0.16990666687488556,
"sampling/unseen_fraction": 0.8300933331251145,
"signal/accuracy_reward/centered_abs_mean": 0.18960503339767457,
"signal/accuracy_reward/group_std_mean": 0.2473952054977417,
"signal/accuracy_reward/group_zero_std_frac": 0.3083333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09480251669883728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09480251669883728,
"signal/advantage_abs_mean": 0.15543261170387268,
"signal/advantage_pre_scale_abs_mean": 0.15543261170387268,
"signal/advantage_pre_scale_std": 0.2481265187263489,
"signal/advantage_std": 0.2481265187263489,
"signal/brier_reward/centered_abs_mean": 0.11165534853935241,
"signal/brier_reward/group_std_mean": 0.1463429093360901,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.055827674269676206,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.055827674269676206,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0013346353895030915,
"signal/confidence_one_or_zero/group_std_mean": 0.0036294300109148024,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9805555462837219,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.3346353711085613e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.3346353711085613e-08,
"signal/format_reward/centered_abs_mean": 0.02710503488779068,
"signal/format_reward/group_std_mean": 0.05075793266296387,
"signal/format_reward/group_zero_std_frac": 0.7916666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01355251744389534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01355251744389534,
"signal/mean_confidence_reward/centered_abs_mean": 0.07513455003499984,
"signal/mean_confidence_reward/group_std_mean": 0.10091831386089326,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.513454534091579e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.513454534091579e-07,
"step": 75
},
{
"calibration/aurc": 0.3689440863507865,
"calibration/batch_distribution_entropy": 0.5184190021494378,
"calibration/batch_entropy_100bins": 0.2806227972233616,
"calibration/batch_entropy_10bins": 0.5184190021494378,
"calibration/batch_entropy_50bins": 0.33034461645252283,
"calibration/batch_uniqueness": 0.3326310367116993,
"calibration/confidence_entropy": 0.6032100330460969,
"calibration/coverage@0%": 0.001058201058201058,
"calibration/coverage@1%": 0.001058201058201058,
"calibration/coverage@10%": 0.001058201058201058,
"calibration/coverage@15%": 0.004713553538618813,
"calibration/coverage@20%": 0.004713553538618813,
"calibration/coverage@25%": 0.004713553538618813,
"calibration/coverage@30%": 0.39896942821224807,
"calibration/coverage@5%": 0.001058201058201058,
"calibration/distribution_entropy_10": 0.5184190021494378,
"calibration/distribution_entropy_100": 0.2806227972233616,
"calibration/ece": 0.1048484709501376,
"calibration/mean_confidence": 0.6910483483132289,
"calibration/unique_confidence_per_question": 0.023437499999999997,
"calibration/unique_confidences": 9.0,
"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": 0.011458333333333348,
"completions/max_length": 3935.8,
"completions/max_terminated_length": 3935.8,
"completions/mean_length": 688.7568603515625,
"completions/mean_terminated_length": 696.784912109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 183.4,
"epoch": 0.19230769230769232,
"grad_norm": 0.000616310047917068,
"learning_rate": 1.9230769230769234e-06,
"loss": -0.0075,
"num_tokens": 152522666.0,
"reward": 1.1828741550445556,
"reward_std": 0.2077293336391449,
"rewards/accuracy_reward": 0.6256944417953492,
"rewards/brier_reward": 0.7528873682022095,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9871527791023255,
"rewards/mean_confidence_reward": 0.6841449737548828,
"sampling/batch_mean_priority_error": 0.047214103962108844,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9166666666666667,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0047060281969606875,
"sampling/priority_kl": 0.030000150203704834,
"sampling/priority_scale": 1.2860350491013377,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 3.6873909994028506e-05,
"sampling/prob_min": 1.0286198084941134e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.18720000088214875,
"sampling/prompt_draws_total": 5616.0,
"sampling/seen_fraction": 0.18101999759674073,
"sampling/unseen_fraction": 0.8189800024032593,
"signal/accuracy_reward/centered_abs_mean": 0.20045573115348816,
"signal/accuracy_reward/group_std_mean": 0.2571284741163254,
"signal/accuracy_reward/group_zero_std_frac": 0.29166666269302366,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10022786557674408,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10022786557674408,
"signal/advantage_abs_mean": 0.1562511533498764,
"signal/advantage_pre_scale_abs_mean": 0.1562511533498764,
"signal/advantage_pre_scale_std": 0.24353698790073394,
"signal/advantage_std": 0.24353698790073394,
"signal/brier_reward/centered_abs_mean": 0.1068439707159996,
"signal/brier_reward/group_std_mean": 0.14024724662303925,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0534219853579998,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0534219853579998,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0003363715251907706,
"signal/confidence_one_or_zero/group_std_mean": 0.0009820926934480667,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.3637149954302003e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.3637149954302003e-09,
"signal/format_reward/centered_abs_mean": 0.02290581576526165,
"signal/format_reward/group_std_mean": 0.0464916467666626,
"signal/format_reward/group_zero_std_frac": 0.7972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011452907882630824,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011452907882630824,
"signal/mean_confidence_reward/centered_abs_mean": 0.06937960833311081,
"signal/mean_confidence_reward/group_std_mean": 0.09569962471723556,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.937960620234662e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.937960620234662e-07,
"step": 80
},
{
"calibration/aurc": 0.34587047138080185,
"calibration/batch_distribution_entropy": 0.4954573795095432,
"calibration/batch_entropy_100bins": 0.27405746184331764,
"calibration/batch_entropy_10bins": 0.4954573795095432,
"calibration/batch_entropy_50bins": 0.3226160098693715,
"calibration/batch_uniqueness": 0.3000675734530054,
"calibration/confidence_entropy": 0.6101426873701342,
"calibration/coverage@0%": 0.006401699894469639,
"calibration/coverage@1%": 0.006401699894469639,
"calibration/coverage@10%": 0.006401699894469639,
"calibration/coverage@15%": 0.006401699894469639,
"calibration/coverage@20%": 0.03345740281224152,
"calibration/coverage@25%": 0.23664042668492055,
"calibration/coverage@30%": 0.4647471549585009,
"calibration/coverage@5%": 0.006401699894469639,
"calibration/distribution_entropy_10": 0.4954573795095432,
"calibration/distribution_entropy_100": 0.27405746184331764,
"calibration/ece": 0.10014490493171968,
"calibration/mean_confidence": 0.6821050410005329,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.014843749999999978,
"completions/max_length": 3979.0,
"completions/max_terminated_length": 3979.0,
"completions/mean_length": 667.094970703125,
"completions/mean_terminated_length": 677.13818359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 175.6,
"epoch": 0.20432692307692307,
"grad_norm": 0.0007630718173459172,
"learning_rate": 2.043269230769231e-06,
"loss": -0.0113,
"num_tokens": 163328432.0,
"reward": 1.1916515588760377,
"reward_std": 0.21311030983924867,
"rewards/accuracy_reward": 0.6413194537162781,
"rewards/brier_reward": 0.7580292105674744,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9839409828186035,
"rewards/mean_confidence_reward": 0.6716553926467895,
"sampling/batch_mean_priority_error": 0.042610298929097075,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.875,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004756578896194697,
"sampling/priority_kl": 0.029999878257513046,
"sampling/priority_scale": 1.241810357524082,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.7016544956713916e-05,
"sampling/prob_min": 1.0765067963802722e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.19919999837875366,
"sampling/prompt_draws_total": 5976.0,
"sampling/seen_fraction": 0.19177333116531373,
"sampling/unseen_fraction": 0.8082266688346863,
"signal/accuracy_reward/centered_abs_mean": 0.20234375,
"signal/accuracy_reward/group_std_mean": 0.26508465111255647,
"signal/accuracy_reward/group_zero_std_frac": 0.2527777820825577,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.101171875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.101171875,
"signal/advantage_abs_mean": 0.15687242150306702,
"signal/advantage_pre_scale_abs_mean": 0.15687242150306702,
"signal/advantage_pre_scale_std": 0.2482043296098709,
"signal/advantage_std": 0.2482043296098709,
"signal/brier_reward/centered_abs_mean": 0.1052109107375145,
"signal/brier_reward/group_std_mean": 0.13991140723228454,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05260545536875725,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05260545536875725,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02864040844142437,
"signal/format_reward/group_std_mean": 0.05566352233290672,
"signal/format_reward/group_zero_std_frac": 0.7694444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.014320204220712184,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.014320204220712184,
"signal/mean_confidence_reward/centered_abs_mean": 0.0705166518688202,
"signal/mean_confidence_reward/group_std_mean": 0.09815174639225006,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.05166485204245e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.05166485204245e-07,
"step": 85
},
{
"calibration/aurc": 0.359621361820331,
"calibration/batch_distribution_entropy": 0.4875542266129493,
"calibration/batch_entropy_100bins": 0.27016203664181926,
"calibration/batch_entropy_10bins": 0.4875542266129493,
"calibration/batch_entropy_50bins": 0.31803037834961934,
"calibration/batch_uniqueness": 0.3108004230223583,
"calibration/confidence_entropy": 0.6185976385079931,
"calibration/coverage@0%": 0.0005235602094240838,
"calibration/coverage@1%": 0.0005235602094240838,
"calibration/coverage@10%": 0.027081825792079912,
"calibration/coverage@15%": 0.030875863732459312,
"calibration/coverage@20%": 0.030875863732459312,
"calibration/coverage@25%": 0.030875863732459312,
"calibration/coverage@30%": 0.0676212705566063,
"calibration/coverage@5%": 0.0005235602094240838,
"calibration/distribution_entropy_10": 0.4875542266129493,
"calibration/distribution_entropy_100": 0.27016203664181926,
"calibration/ece": 0.09774914445477971,
"calibration/mean_confidence": 0.6697879296328936,
"calibration/unique_confidence_per_question": 0.023437500000000003,
"calibration/unique_confidences": 9.0,
"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": 0.013888888888888905,
"completions/max_length": 3912.4,
"completions/max_terminated_length": 3912.4,
"completions/mean_length": 688.4250854492187,
"completions/mean_terminated_length": 698.1601928710937,
"completions/min_length": 0.0,
"completions/min_terminated_length": 199.0,
"epoch": 0.21634615384615385,
"grad_norm": 0.0006718159420415759,
"learning_rate": 2.1634615384615387e-06,
"loss": -0.0107,
"num_tokens": 174374161.0,
"reward": 1.190553903579712,
"reward_std": 0.21019268929958343,
"rewards/accuracy_reward": 0.6362847208976745,
"rewards/brier_reward": 0.7601742506027221,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9846354126930237,
"rewards/mean_confidence_reward": 0.6664713621139526,
"sampling/batch_mean_priority_error": 0.039588163981791394,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8833333333333334,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004800818022340536,
"sampling/priority_kl": 0.029999977350234984,
"sampling/priority_scale": 1.2025536180008203,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.715234051924199e-05,
"sampling/prob_min": 1.1198205720575061e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.21120000183582305,
"sampling/prompt_draws_total": 6336.0,
"sampling/seen_fraction": 0.20212666392326356,
"sampling/unseen_fraction": 0.7978733360767365,
"signal/accuracy_reward/centered_abs_mean": 0.20233290195465087,
"signal/accuracy_reward/group_std_mean": 0.2649902403354645,
"signal/accuracy_reward/group_zero_std_frac": 0.26388888657093046,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.10116645097732543,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.10116645097732543,
"signal/advantage_abs_mean": 0.15588796138763428,
"signal/advantage_pre_scale_abs_mean": 0.15588796138763428,
"signal/advantage_pre_scale_std": 0.24400971829891205,
"signal/advantage_std": 0.24400971829891205,
"signal/brier_reward/centered_abs_mean": 0.10318200141191483,
"signal/brier_reward/group_std_mean": 0.13609919250011443,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.051591000705957415,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.051591000705957415,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02562391422688961,
"signal/format_reward/group_std_mean": 0.047272267565131185,
"signal/format_reward/group_zero_std_frac": 0.8055555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012811957113444804,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012811957113444804,
"signal/mean_confidence_reward/centered_abs_mean": 0.06962320357561111,
"signal/mean_confidence_reward/group_std_mean": 0.09376237392425538,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.962320298953273e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.962320298953273e-07,
"step": 90
},
{
"calibration/aurc": 0.360027708313517,
"calibration/batch_distribution_entropy": 0.4813050950400042,
"calibration/batch_entropy_100bins": 0.2578209922325956,
"calibration/batch_entropy_10bins": 0.4813050950400042,
"calibration/batch_entropy_50bins": 0.30350270054750683,
"calibration/batch_uniqueness": 0.28367851731747284,
"calibration/confidence_entropy": 0.6131506525449577,
"calibration/coverage@0%": 0.002114180473319751,
"calibration/coverage@1%": 0.002114180473319751,
"calibration/coverage@10%": 0.002114180473319751,
"calibration/coverage@15%": 0.002114180473319751,
"calibration/coverage@20%": 0.07943111411176369,
"calibration/coverage@25%": 0.11958632788345835,
"calibration/coverage@30%": 0.305080325157146,
"calibration/coverage@5%": 0.002114180473319751,
"calibration/distribution_entropy_10": 0.4813050950400042,
"calibration/distribution_entropy_100": 0.2578209922325956,
"calibration/ece": 0.1002862750340386,
"calibration/mean_confidence": 0.6807791476508745,
"calibration/unique_confidence_per_question": 0.01875,
"calibration/unique_confidences": 7.2,
"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": 0.01059027777777779,
"completions/max_length": 3668.4,
"completions/max_terminated_length": 3668.4,
"completions/mean_length": 644.8259643554687,
"completions/mean_terminated_length": 651.8298095703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 194.4,
"epoch": 0.2283653846153846,
"grad_norm": 0.0006763488054275513,
"learning_rate": 2.283653846153846e-06,
"loss": -0.0092,
"num_tokens": 184896156.0,
"reward": 1.1970339298248291,
"reward_std": 0.1976634830236435,
"rewards/accuracy_reward": 0.6417534589767456,
"rewards/brier_reward": 0.763932728767395,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9883680462837219,
"rewards/mean_confidence_reward": 0.6744001626968383,
"sampling/batch_mean_priority_error": 0.03951753118071802,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9027777777777779,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004846246726810932,
"sampling/priority_kl": 0.02999996431171894,
"sampling/priority_scale": 1.167076528025791,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.729506061063148e-05,
"sampling/prob_min": 1.1645950689853634e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.22319999933242798,
"sampling/prompt_draws_total": 6696.0,
"sampling/seen_fraction": 0.21301333606243134,
"sampling/unseen_fraction": 0.7869866639375687,
"signal/accuracy_reward/centered_abs_mean": 0.19283311665058137,
"signal/accuracy_reward/group_std_mean": 0.2522544801235199,
"signal/accuracy_reward/group_zero_std_frac": 0.2888889014720917,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09641655832529068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09641655832529068,
"signal/advantage_abs_mean": 0.14698622226715088,
"signal/advantage_pre_scale_abs_mean": 0.14698622226715088,
"signal/advantage_pre_scale_std": 0.23368164300918579,
"signal/advantage_std": 0.23368164300918579,
"signal/brier_reward/centered_abs_mean": 0.0987538680434227,
"signal/brier_reward/group_std_mean": 0.1296857550740242,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04937693402171135,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04937693402171135,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.019194878451526166,
"signal/format_reward/group_std_mean": 0.037654457986354826,
"signal/format_reward/group_zero_std_frac": 0.8388888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009597439225763083,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009597439225763083,
"signal/mean_confidence_reward/centered_abs_mean": 0.06403515413403511,
"signal/mean_confidence_reward/group_std_mean": 0.08662768602371215,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.403514930752862e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.403514930752862e-07,
"step": 95
},
{
"calibration/aurc": 0.3596962699336317,
"calibration/batch_distribution_entropy": 0.49572307893281886,
"calibration/batch_entropy_100bins": 0.2738978163454987,
"calibration/batch_entropy_10bins": 0.49572307893281886,
"calibration/batch_entropy_50bins": 0.32242807777238164,
"calibration/batch_uniqueness": 0.3183046592648424,
"calibration/confidence_entropy": 0.6086933485955659,
"calibration/coverage@0%": 0.0021108179419525065,
"calibration/coverage@1%": 0.0021108179419525065,
"calibration/coverage@10%": 0.0021108179419525065,
"calibration/coverage@15%": 0.05593667546174143,
"calibration/coverage@20%": 0.12996273000957584,
"calibration/coverage@25%": 0.12996273000957584,
"calibration/coverage@30%": 0.38856823990575723,
"calibration/coverage@5%": 0.0021108179419525065,
"calibration/distribution_entropy_10": 0.49572307893281886,
"calibration/distribution_entropy_100": 0.2738978163454987,
"calibration/ece": 0.12810381554201036,
"calibration/mean_confidence": 0.6824774719423579,
"calibration/unique_confidence_per_question": 0.0234375,
"calibration/unique_confidences": 9.0,
"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": 0.013541666666666652,
"completions/max_length": 3520.8,
"completions/max_terminated_length": 3520.8,
"completions/mean_length": 646.881005859375,
"completions/mean_terminated_length": 655.8904174804687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 188.6,
"epoch": 0.2403846153846154,
"grad_norm": 0.0006148848915472627,
"learning_rate": 2.403846153846154e-06,
"loss": -0.0111,
"num_tokens": 195449921.0,
"reward": 1.1851803302764892,
"reward_std": 0.19183639883995057,
"rewards/accuracy_reward": 0.6269965291023254,
"rewards/brier_reward": 0.7580206274986268,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9853298664093018,
"rewards/mean_confidence_reward": 0.680933153629303,
"sampling/batch_mean_priority_error": 0.04106042591276217,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.888888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004891505371779204,
"sampling/priority_kl": 0.030000001937150956,
"sampling/priority_scale": 1.1352285265456885,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 3.743201086763293e-05,
"sampling/prob_min": 1.2065826558682602e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.23519999980926515,
"sampling/prompt_draws_total": 7056.0,
"sampling/seen_fraction": 0.22354000210762023,
"sampling/unseen_fraction": 0.7764599978923797,
"signal/accuracy_reward/centered_abs_mean": 0.17950846254825592,
"signal/accuracy_reward/group_std_mean": 0.23621359169483186,
"signal/accuracy_reward/group_zero_std_frac": 0.33333333730697634,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08975423127412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08975423127412796,
"signal/advantage_abs_mean": 0.14255423545837403,
"signal/advantage_pre_scale_abs_mean": 0.14255423545837403,
"signal/advantage_pre_scale_std": 0.2318326562643051,
"signal/advantage_std": 0.2318326562643051,
"signal/brier_reward/centered_abs_mean": 0.09948206096887588,
"signal/brier_reward/group_std_mean": 0.1306991159915924,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04974103048443794,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04974103048443794,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.023301866464316846,
"signal/format_reward/group_std_mean": 0.04023935534060001,
"signal/format_reward/group_zero_std_frac": 0.8472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011650933232158423,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011650933232158423,
"signal/mean_confidence_reward/centered_abs_mean": 0.06637315228581428,
"signal/mean_confidence_reward/group_std_mean": 0.0889115497469902,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.637314868385147e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.637314868385147e-07,
"step": 100
},
{
"epoch": 0.2403846153846154,
"eval_calibration/aurc": 0.3264960431700285,
"eval_calibration/batch_distribution_entropy": 0.5035579377534389,
"eval_calibration/batch_entropy_100bins": 0.2686276995956658,
"eval_calibration/batch_entropy_10bins": 0.5035579377534389,
"eval_calibration/batch_entropy_50bins": 0.31622418160425286,
"eval_calibration/batch_uniqueness": 0.3147685317330474,
"eval_calibration/confidence_entropy": 0.5989185390589775,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.0,
"eval_calibration/coverage@20%": 0.0,
"eval_calibration/coverage@25%": 0.0,
"eval_calibration/coverage@30%": 0.25769569041336854,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5035579377534389,
"eval_calibration/distribution_entropy_100": 0.2686276995956658,
"eval_calibration/ece": 0.06099384344766924,
"eval_calibration/mean_confidence": 0.6955584872471415,
"eval_calibration/unique_confidence_per_question": 0.011284722222222222,
"eval_calibration/unique_confidences": 13,
"eval_completions/clipped_ratio": 0.010416666666666666,
"eval_completions/max_length": 2317.3333333333335,
"eval_completions/max_terminated_length": 2317.3333333333335,
"eval_completions/mean_length": 641.1512349446615,
"eval_completions/mean_terminated_length": 647.976816813151,
"eval_completions/min_length": 49.0,
"eval_completions/min_terminated_length": 226.0,
"eval_loss": 0.0,
"eval_num_tokens": 195449921.0,
"eval_reward": 1.1953096787134807,
"eval_reward_std": 0.3530424237251282,
"eval_rewards/accuracy_reward": 0.6423611144224802,
"eval_rewards/brier_reward": 0.7612651983896891,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9869791567325592,
"eval_rewards/mean_confidence_reward": 0.6865017215410868,
"eval_runtime": 203.5004,
"eval_samples_per_second": 4.914,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4403211822112401,
"eval_signal/accuracy_reward/group_std_mean": 0.4756300499041875,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.22016059110562006,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.22016059110562006,
"eval_signal/advantage_abs_mean": 0.3166052947441737,
"eval_signal/advantage_pre_scale_abs_mean": 0.3166052947441737,
"eval_signal/advantage_pre_scale_std": 0.35058291256427765,
"eval_signal/advantage_std": 0.35058291256427765,
"eval_signal/brier_reward/centered_abs_mean": 0.17967990785837173,
"eval_signal/brier_reward/group_std_mean": 0.21290371815363565,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08983995392918587,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08983995392918587,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.025010850590964157,
"eval_signal/format_reward/group_std_mean": 0.06767813799281915,
"eval_signal/format_reward/group_zero_std_frac": 0.6388889054457346,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.012505425295482079,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.012505425295482079,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.07078721622625987,
"eval_signal/mean_confidence_reward/group_std_mean": 0.10642425095041592,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.078721561507942e-07,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 7.078721561507942e-07,
"eval_steps_per_second": 0.029,
"step": 100
},
{
"epoch": 0.2403846153846154,
"step": 100,
"train_probe_calibration/aurc": 0.2618583200446606,
"train_probe_calibration/batch_distribution_entropy": 0.5051684118856261,
"train_probe_calibration/batch_entropy_100bins": 0.2697463841725647,
"train_probe_calibration/batch_entropy_10bins": 0.5051684118856261,
"train_probe_calibration/batch_entropy_50bins": 0.3175410789880135,
"train_probe_calibration/batch_uniqueness": 0.31938374294618566,
"train_probe_calibration/confidence_entropy": 0.6020218923419944,
"train_probe_calibration/coverage@0%": 0.007029876977152899,
"train_probe_calibration/coverage@1%": 0.007029876977152899,
"train_probe_calibration/coverage@10%": 0.007029876977152899,
"train_probe_calibration/coverage@15%": 0.007029876977152899,
"train_probe_calibration/coverage@20%": 0.2398945518453427,
"train_probe_calibration/coverage@25%": 0.2398945518453427,
"train_probe_calibration/coverage@30%": 0.710896309314587,
"train_probe_calibration/coverage@5%": 0.007029876977152899,
"train_probe_calibration/distribution_entropy_10": 0.5051684118856261,
"train_probe_calibration/distribution_entropy_100": 0.2697463841725647,
"train_probe_calibration/ece": 0.039543057996485054,
"train_probe_calibration/mean_confidence": 0.6918277680140598,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.013715277777777793,
"train_probe_completions/max_length": 2729.1666666666665,
"train_probe_completions/max_terminated_length": 2729.1666666666665,
"train_probe_completions/mean_length": 639.4520975748698,
"train_probe_completions/mean_terminated_length": 648.3664245605469,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 193.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 195449921.0,
"train_probe_reward": 1.2270989616711934,
"train_probe_reward_std": 0.34227528671423596,
"train_probe_rewards/accuracy_reward": 0.683159718910853,
"train_probe_rewards/brier_reward": 0.7831770877043406,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.987847218910853,
"train_probe_rewards/mean_confidence_reward": 0.6834201316038767,
"train_probe_runtime": 200.9406,
"train_probe_samples_per_second": 4.977,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4219292501608531,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4654071430365245,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.21096462508042654,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.21096462508042654,
"train_probe_signal/advantage_abs_mean": 0.3004860281944275,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.3004860281944275,
"train_probe_signal/advantage_pre_scale_std": 0.33957551916440326,
"train_probe_signal/advantage_std": 0.33957551916440326,
"train_probe_signal/brier_reward/centered_abs_mean": 0.16428263982137045,
"train_probe_signal/brier_reward/group_std_mean": 0.1990672672788302,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08214131991068523,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08214131991068523,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.023437499689559143,
"train_probe_signal/format_reward/group_std_mean": 0.06575708525876205,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6388889104127884,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.011718749844779571,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.011718749844779571,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.06948784242073695,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.10532646502057712,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.948784611419493e-07,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 6.948784611419493e-07,
"train_probe_steps_per_second": 0.03
},
{
"calibration/aurc": 0.25365423358695904,
"calibration/batch_distribution_entropy": 0.48674473663307144,
"calibration/batch_entropy_100bins": 0.26873184440469255,
"calibration/batch_entropy_10bins": 0.48674473663307144,
"calibration/batch_entropy_50bins": 0.3163467791883904,
"calibration/batch_uniqueness": 0.29883804015135357,
"calibration/confidence_entropy": 0.603367084818194,
"calibration/coverage@0%": 0.0015706878041144342,
"calibration/coverage@1%": 0.0015706878041144342,
"calibration/coverage@10%": 0.05157068780411443,
"calibration/coverage@15%": 0.19297012061388755,
"calibration/coverage@20%": 0.32617990848146156,
"calibration/coverage@25%": 0.37274075504230814,
"calibration/coverage@30%": 0.6159436880090207,
"calibration/coverage@5%": 0.0015706878041144342,
"calibration/distribution_entropy_10": 0.48674473663307144,
"calibration/distribution_entropy_100": 0.26873184440469255,
"calibration/ece": 0.0912085220880512,
"calibration/mean_confidence": 0.6891120219009381,
"calibration/unique_confidence_per_question": 0.022395833333333334,
"calibration/unique_confidences": 8.6,
"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": 0.00902777777777779,
"completions/max_length": 3458.2,
"completions/max_terminated_length": 3458.2,
"completions/mean_length": 638.6677124023438,
"completions/mean_terminated_length": 644.5072265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 188.4,
"epoch": 0.25240384615384615,
"grad_norm": 0.0006219370989128947,
"learning_rate": 2.5240384615384618e-06,
"loss": -0.0084,
"num_tokens": 205899021.0,
"reward": 1.2294507503509522,
"reward_std": 0.19787262380123138,
"rewards/accuracy_reward": 0.6855902791023254,
"rewards/brier_reward": 0.7831931471824646,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9901041746139526,
"rewards/mean_confidence_reward": 0.6871875047683715,
"sampling/batch_mean_priority_error": 0.039179960880272646,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.9,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004937084298580885,
"sampling/priority_kl": 0.030000375211238862,
"sampling/priority_scale": 1.1062577606644481,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 3.757406011573039e-05,
"sampling/prob_min": 1.2466577754821628e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.24720000028610228,
"sampling/prompt_draws_total": 7416.0,
"sampling/seen_fraction": 0.23444666862487792,
"sampling/unseen_fraction": 0.765553331375122,
"signal/accuracy_reward/centered_abs_mean": 0.18909505307674407,
"signal/accuracy_reward/group_std_mean": 0.251471620798111,
"signal/accuracy_reward/group_zero_std_frac": 0.2777777761220932,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09454752653837203,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09454752653837203,
"signal/advantage_abs_mean": 0.14458977580070495,
"signal/advantage_pre_scale_abs_mean": 0.14458977580070495,
"signal/advantage_pre_scale_std": 0.23422183096408844,
"signal/advantage_std": 0.23422183096408844,
"signal/brier_reward/centered_abs_mean": 0.09787606298923493,
"signal/brier_reward/group_std_mean": 0.1316657304763794,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.048938031494617465,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.048938031494617465,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0003255208255723119,
"signal/confidence_one_or_zero/group_std_mean": 0.0006831518840044737,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/format_reward/centered_abs_mean": 0.017773437686264514,
"signal/format_reward/group_std_mean": 0.03596483059227466,
"signal/format_reward/group_zero_std_frac": 0.8444444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008886718843132257,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008886718843132257,
"signal/mean_confidence_reward/centered_abs_mean": 0.06357530206441879,
"signal/mean_confidence_reward/group_std_mean": 0.08628400415182114,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.357530082823359e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.357530082823359e-07,
"step": 105
},
{
"calibration/aurc": 0.2731392392789653,
"calibration/batch_distribution_entropy": 0.48010458764261943,
"calibration/batch_entropy_100bins": 0.26175482708951114,
"calibration/batch_entropy_10bins": 0.48010458764261943,
"calibration/batch_entropy_50bins": 0.30813354729215314,
"calibration/batch_uniqueness": 0.26838013652362525,
"calibration/confidence_entropy": 0.597166460926026,
"calibration/coverage@0%": 0.0026082463011314187,
"calibration/coverage@1%": 0.0026082463011314187,
"calibration/coverage@10%": 0.0026082463011314187,
"calibration/coverage@15%": 0.0694489773716275,
"calibration/coverage@20%": 0.2647040626734648,
"calibration/coverage@25%": 0.5813071148825065,
"calibration/coverage@30%": 0.6276680809399477,
"calibration/coverage@5%": 0.0026082463011314187,
"calibration/distribution_entropy_10": 0.48010458764261943,
"calibration/distribution_entropy_100": 0.26175482708951114,
"calibration/ece": 0.09699426452649675,
"calibration/mean_confidence": 0.6960098487301523,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.01128472222222221,
"completions/max_length": 3859.0,
"completions/max_terminated_length": 3859.0,
"completions/mean_length": 637.8580810546875,
"completions/mean_terminated_length": 645.131982421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 177.0,
"epoch": 0.2644230769230769,
"grad_norm": 0.000706358696334064,
"learning_rate": 2.6442307692307696e-06,
"loss": -0.0113,
"num_tokens": 216304106.0,
"reward": 1.220960521697998,
"reward_std": 0.19010037779808045,
"rewards/accuracy_reward": 0.6734375,
"rewards/brier_reward": 0.7804488062858581,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9880208253860474,
"rewards/mean_confidence_reward": 0.694079864025116,
"sampling/batch_mean_priority_error": 0.039559362733774556,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8666666666666666,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.004981566872447729,
"sampling/priority_kl": 0.029999980702996254,
"sampling/priority_scale": 1.0800061345566063,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 3.771074407268316e-05,
"sampling/prob_min": 1.2843752301705536e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.2592000007629395,
"sampling/prompt_draws_total": 7776.0,
"sampling/seen_fraction": 0.24499333202838897,
"sampling/unseen_fraction": 0.755006667971611,
"signal/accuracy_reward/centered_abs_mean": 0.18021918535232545,
"signal/accuracy_reward/group_std_mean": 0.23629273772239684,
"signal/accuracy_reward/group_zero_std_frac": 0.33333333730697634,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09010959267616273,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09010959267616273,
"signal/advantage_abs_mean": 0.14149389117956163,
"signal/advantage_pre_scale_abs_mean": 0.14149389117956163,
"signal/advantage_pre_scale_std": 0.23208475708961487,
"signal/advantage_std": 0.23208475708961487,
"signal/brier_reward/centered_abs_mean": 0.09630044549703598,
"signal/brier_reward/group_std_mean": 0.12676727920770645,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04815022274851799,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04815022274851799,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02072482630610466,
"signal/format_reward/group_std_mean": 0.03876565992832184,
"signal/format_reward/group_zero_std_frac": 0.8444444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01036241315305233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01036241315305233,
"signal/mean_confidence_reward/centered_abs_mean": 0.06128065511584282,
"signal/mean_confidence_reward/group_std_mean": 0.08464662134647369,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.128065251687076e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.128065251687076e-07,
"step": 110
},
{
"calibration/aurc": 0.23900958690948393,
"calibration/batch_distribution_entropy": 0.4792917733999877,
"calibration/batch_entropy_100bins": 0.25453673066305943,
"calibration/batch_entropy_10bins": 0.4792917733999877,
"calibration/batch_entropy_50bins": 0.29963652096675586,
"calibration/batch_uniqueness": 0.23525715566464817,
"calibration/confidence_entropy": 0.5841788415392271,
"calibration/coverage@0%": 0.002645767716535433,
"calibration/coverage@1%": 0.002645767716535433,
"calibration/coverage@10%": 0.07197910104986877,
"calibration/coverage@15%": 0.15427076771653545,
"calibration/coverage@20%": 0.2620832677165354,
"calibration/coverage@25%": 0.5331317247542449,
"calibration/coverage@30%": 0.7069983914209116,
"calibration/coverage@5%": 0.002645767716535433,
"calibration/distribution_entropy_10": 0.4792917733999877,
"calibration/distribution_entropy_100": 0.25453673066305943,
"calibration/ece": 0.07616447667964926,
"calibration/mean_confidence": 0.7118249107708302,
"calibration/unique_confidence_per_question": 0.0234375,
"calibration/unique_confidences": 9.0,
"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": 0.01649305555555556,
"completions/max_length": 3940.0,
"completions/max_terminated_length": 3940.0,
"completions/mean_length": 656.9077270507812,
"completions/mean_terminated_length": 668.1058471679687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 174.6,
"epoch": 0.2764423076923077,
"grad_norm": 0.0010440150508657098,
"learning_rate": 2.7644230769230775e-06,
"loss": -0.0155,
"num_tokens": 226953891.0,
"reward": 1.2063098430633545,
"reward_std": 0.20195430517196655,
"rewards/accuracy_reward": 0.6577256917953491,
"rewards/brier_reward": 0.7725014925003052,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9823784589767456,
"rewards/mean_confidence_reward": 0.6949609279632568,
"sampling/batch_mean_priority_error": 0.04175164310245379,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8805555555555555,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0050280562601983545,
"sampling/priority_kl": 0.029999904334545135,
"sampling/priority_scale": 1.0559982656966895,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.7847199564566836e-05,
"sampling/prob_min": 1.3202533409639727e-05,
"sampling/prompt_draws_max": 3.0,
"sampling/prompt_draws_mean": 0.2712000012397766,
"sampling/prompt_draws_total": 8136.0,
"sampling/seen_fraction": 0.2555199980735779,
"sampling/unseen_fraction": 0.7444800019264222,
"signal/accuracy_reward/centered_abs_mean": 0.18530273735523223,
"signal/accuracy_reward/group_std_mean": 0.24110294580459596,
"signal/accuracy_reward/group_zero_std_frac": 0.325,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.09265136867761611,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.09265136867761611,
"signal/advantage_abs_mean": 0.14929196536540984,
"signal/advantage_pre_scale_abs_mean": 0.14929196536540984,
"signal/advantage_pre_scale_std": 0.2474428117275238,
"signal/advantage_std": 0.2474428117275238,
"signal/brier_reward/centered_abs_mean": 0.10281380563974381,
"signal/brier_reward/group_std_mean": 0.13706456124782562,
"signal/brier_reward/group_zero_std_frac": 0.002777777798473835,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.051406902819871904,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.051406902819871904,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.02937825545668602,
"signal/format_reward/group_std_mean": 0.05295464023947716,
"signal/format_reward/group_zero_std_frac": 0.7916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01468912772834301,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01468912772834301,
"signal/mean_confidence_reward/centered_abs_mean": 0.06717475205659866,
"signal/mean_confidence_reward/group_std_mean": 0.09251063913106919,
"signal/mean_confidence_reward/group_zero_std_frac": 0.002777777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.717474661854794e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.717474661854794e-07,
"step": 115
},
{
"calibration/aurc": 0.2765707780359674,
"calibration/batch_distribution_entropy": 0.46426059536033704,
"calibration/batch_entropy_100bins": 0.24633244332746132,
"calibration/batch_entropy_10bins": 0.46426059536033704,
"calibration/batch_entropy_50bins": 0.2899785666595467,
"calibration/batch_uniqueness": 0.2554472316323028,
"calibration/confidence_entropy": 0.5880420811708887,
"calibration/coverage@0%": 0.005281628929694076,
"calibration/coverage@1%": 0.005281628929694076,
"calibration/coverage@10%": 0.005281628929694076,
"calibration/coverage@15%": 0.005281628929694076,
"calibration/coverage@20%": 0.12280813811604578,
"calibration/coverage@25%": 0.3352501998796019,
"calibration/coverage@30%": 0.6500459979400643,
"calibration/coverage@5%": 0.005281628929694076,
"calibration/distribution_entropy_10": 0.46426059536033704,
"calibration/distribution_entropy_100": 0.24633244332746132,
"calibration/ece": 0.056203158003464426,
"calibration/mean_confidence": 0.7093479335837578,
"calibration/unique_confidence_per_question": 0.019270833333333334,
"calibration/unique_confidences": 7.4,
"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": 0.012673611111111116,
"completions/max_length": 3711.0,
"completions/max_terminated_length": 3711.0,
"completions/mean_length": 631.4669189453125,
"completions/mean_terminated_length": 639.7166870117187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 169.4,
"epoch": 0.28846153846153844,
"grad_norm": 0.0008303358335979283,
"learning_rate": 2.8846153846153845e-06,
"loss": -0.0093,
"num_tokens": 237324006.0,
"reward": 1.235554051399231,
"reward_std": 0.18194632232189178,
"rewards/accuracy_reward": 0.6936632037162781,
"rewards/brier_reward": 0.7907119989395142,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.986718761920929,
"rewards/mean_confidence_reward": 0.7019965291023255,
"sampling/batch_mean_priority_error": 0.03951839709516186,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8416666666666668,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0050725327804684636,
"sampling/priority_kl": 0.03000000901520252,
"sampling/priority_scale": 1.0340607046615333,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.797858953475952e-05,
"sampling/prob_min": 1.2279798283998389e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.28320000171661375,
"sampling/prompt_draws_total": 8496.0,
"sampling/seen_fraction": 0.26567999720573426,
"sampling/unseen_fraction": 0.7343200027942658,
"signal/accuracy_reward/centered_abs_mean": 0.168896484375,
"signal/accuracy_reward/group_std_mean": 0.2248100906610489,
"signal/accuracy_reward/group_zero_std_frac": 0.3555555522441864,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0844482421875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0844482421875,
"signal/advantage_abs_mean": 0.1337295189499855,
"signal/advantage_pre_scale_abs_mean": 0.1337295189499855,
"signal/advantage_pre_scale_std": 0.2254096359014511,
"signal/advantage_std": 0.2254096359014511,
"signal/brier_reward/centered_abs_mean": 0.09312592148780822,
"signal/brier_reward/group_std_mean": 0.12495446056127549,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04656296074390411,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04656296074390411,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01927625834941864,
"signal/format_reward/group_std_mean": 0.036288988590240476,
"signal/format_reward/group_zero_std_frac": 0.8527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00963812917470932,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00963812917470932,
"signal/mean_confidence_reward/centered_abs_mean": 0.05979058220982551,
"signal/mean_confidence_reward/group_std_mean": 0.08202629685401916,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.979058300908946e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.979058300908946e-07,
"step": 120
},
{
"calibration/aurc": 0.28248611122206685,
"calibration/batch_distribution_entropy": 0.4817757350877705,
"calibration/batch_entropy_100bins": 0.2525112913705817,
"calibration/batch_entropy_10bins": 0.4817757350877705,
"calibration/batch_entropy_50bins": 0.2972522066029843,
"calibration/batch_uniqueness": 0.29516464341044957,
"calibration/confidence_entropy": 0.5860149585009575,
"calibration/coverage@0%": 0.002090139843888836,
"calibration/coverage@1%": 0.002090139843888836,
"calibration/coverage@10%": 0.002090139843888836,
"calibration/coverage@15%": 0.06491736497477887,
"calibration/coverage@20%": 0.33621901321455316,
"calibration/coverage@25%": 0.4885750341569616,
"calibration/coverage@30%": 0.4938106362512024,
"calibration/coverage@5%": 0.002090139843888836,
"calibration/distribution_entropy_10": 0.4817757350877705,
"calibration/distribution_entropy_100": 0.2525112913705817,
"calibration/ece": 0.10515600893545998,
"calibration/mean_confidence": 0.707127596378001,
"calibration/unique_confidence_per_question": 0.018229166666666668,
"calibration/unique_confidences": 7.0,
"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": 0.01362847222222221,
"completions/max_length": 3608.6,
"completions/max_terminated_length": 3608.6,
"completions/mean_length": 655.3791748046875,
"completions/mean_terminated_length": 664.4575073242188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 163.0,
"epoch": 0.3004807692307692,
"grad_norm": 0.0007634198991581798,
"learning_rate": 3.0048076923076923e-06,
"loss": -0.0105,
"num_tokens": 248013366.0,
"reward": 1.201941180229187,
"reward_std": 0.20476475059986116,
"rewards/accuracy_reward": 0.6467881917953491,
"rewards/brier_reward": 0.771316409111023,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9857638835906982,
"rewards/mean_confidence_reward": 0.6920746684074401,
"sampling/batch_mean_priority_error": 0.04121937903065245,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8111111111111111,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005118600558489561,
"sampling/priority_kl": 0.029999775066971778,
"sampling/priority_scale": 1.0137596725020557,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.8107689033495265e-05,
"sampling/prob_min": 1.2597672321135178e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.29520000219345094,
"sampling/prompt_draws_total": 8856.0,
"sampling/seen_fraction": 0.2756599962711334,
"sampling/unseen_fraction": 0.7243400037288665,
"signal/accuracy_reward/centered_abs_mean": 0.2012641042470932,
"signal/accuracy_reward/group_std_mean": 0.2568613201379776,
"signal/accuracy_reward/group_zero_std_frac": 0.2972222179174423,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1006320521235466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.1006320521235466,
"signal/advantage_abs_mean": 0.1566687285900116,
"signal/advantage_pre_scale_abs_mean": 0.1566687285900116,
"signal/advantage_pre_scale_std": 0.24594251811504364,
"signal/advantage_std": 0.24594251811504364,
"signal/brier_reward/centered_abs_mean": 0.10521574020385742,
"signal/brier_reward/group_std_mean": 0.13690231144428253,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05260787010192871,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05260787010192871,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0003255208255723119,
"signal/confidence_one_or_zero/group_std_mean": 0.0006831518840044737,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/format_reward/centered_abs_mean": 0.022851562686264516,
"signal/format_reward/group_std_mean": 0.039858097583055495,
"signal/format_reward/group_zero_std_frac": 0.8472222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011425781343132258,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011425781343132258,
"signal/mean_confidence_reward/centered_abs_mean": 0.06317572891712189,
"signal/mean_confidence_reward/group_std_mean": 0.08579769432544708,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.317572797343018e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.317572797343018e-07,
"step": 125
},
{
"calibration/aurc": 0.27523049109574826,
"calibration/batch_distribution_entropy": 0.4980071222686771,
"calibration/batch_entropy_100bins": 0.2809631885026639,
"calibration/batch_entropy_10bins": 0.4980071222686771,
"calibration/batch_entropy_50bins": 0.3307453195590327,
"calibration/batch_uniqueness": 0.36682188726614934,
"calibration/confidence_entropy": 0.6059415049134447,
"calibration/coverage@0%": 0.0010610079575596816,
"calibration/coverage@1%": 0.0010610079575596816,
"calibration/coverage@10%": 0.11211363953650703,
"calibration/coverage@15%": 0.1656903532039648,
"calibration/coverage@20%": 0.2866758129509338,
"calibration/coverage@25%": 0.41138054560484977,
"calibration/coverage@30%": 0.6329594929732709,
"calibration/coverage@5%": 0.0010610079575596816,
"calibration/distribution_entropy_10": 0.4980071222686771,
"calibration/distribution_entropy_100": 0.2809631885026639,
"calibration/ece": 0.10925301852963129,
"calibration/mean_confidence": 0.6789147319741574,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.010937500000000022,
"completions/max_length": 3922.2,
"completions/max_terminated_length": 3922.2,
"completions/mean_length": 650.4537353515625,
"completions/mean_terminated_length": 657.6193115234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 188.6,
"epoch": 0.3125,
"grad_norm": 0.0013687816681340337,
"learning_rate": 3.125e-06,
"loss": -0.0093,
"num_tokens": 258602529.0,
"reward": 1.2259597539901734,
"reward_std": 0.17786626517772675,
"rewards/accuracy_reward": 0.6735243082046509,
"rewards/brier_reward": 0.7898400783538818,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9885416626930237,
"rewards/mean_confidence_reward": 0.6681119918823242,
"sampling/batch_mean_priority_error": 0.033332646520728156,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8083333333333332,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0051580403000116345,
"sampling/priority_kl": 0.029999729618430136,
"sampling/priority_scale": 0.9950135349761695,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.823342267423868e-05,
"sampling/prob_min": 1.29005729831988e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3072000026702881,
"sampling/prompt_draws_total": 9216.0,
"sampling/seen_fraction": 0.2853800058364868,
"sampling/unseen_fraction": 0.7146199941635132,
"signal/accuracy_reward/centered_abs_mean": 0.171142578125,
"signal/accuracy_reward/group_std_mean": 0.22980001866817473,
"signal/accuracy_reward/group_zero_std_frac": 0.33333333730697634,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0855712890625,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0855712890625,
"signal/advantage_abs_mean": 0.12813933938741684,
"signal/advantage_pre_scale_abs_mean": 0.12813933938741684,
"signal/advantage_pre_scale_std": 0.21815860867500306,
"signal/advantage_std": 0.21815860867500306,
"signal/brier_reward/centered_abs_mean": 0.08448321223258973,
"signal/brier_reward/group_std_mean": 0.11474492251873017,
"signal/brier_reward/group_zero_std_frac": 0.0,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.042241606116294864,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.042241606116294864,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02050781212747097,
"signal/format_reward/group_std_mean": 0.0391859769821167,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010253906063735486,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010253906063735486,
"signal/mean_confidence_reward/centered_abs_mean": 0.06431071013212204,
"signal/mean_confidence_reward/group_std_mean": 0.08673321157693863,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.431070801227179e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.431070801227179e-07,
"step": 130
},
{
"calibration/aurc": 0.2949663717931138,
"calibration/batch_distribution_entropy": 0.4976870100283005,
"calibration/batch_entropy_100bins": 0.3003926554336508,
"calibration/batch_entropy_10bins": 0.4976870100283005,
"calibration/batch_entropy_50bins": 0.35361737366404944,
"calibration/batch_uniqueness": 0.4320600368780066,
"calibration/confidence_entropy": 0.6141245140457932,
"calibration/coverage@0%": 0.0015665796344647518,
"calibration/coverage@1%": 0.0015665796344647518,
"calibration/coverage@10%": 0.0015665796344647518,
"calibration/coverage@15%": 0.0015665796344647518,
"calibration/coverage@20%": 0.19849162231964446,
"calibration/coverage@25%": 0.2945751732334826,
"calibration/coverage@30%": 0.6622418702363284,
"calibration/coverage@5%": 0.0015665796344647518,
"calibration/distribution_entropy_10": 0.4976870100283005,
"calibration/distribution_entropy_100": 0.3003926554336508,
"calibration/ece": 0.12498686483284764,
"calibration/mean_confidence": 0.6626829831819878,
"calibration/unique_confidence_per_question": 0.0171875,
"calibration/unique_confidences": 6.6,
"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": 0.009722222222222254,
"completions/max_length": 3440.0,
"completions/max_terminated_length": 3440.0,
"completions/mean_length": 674.4981811523437,
"completions/mean_terminated_length": 681.1444458007812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 188.0,
"epoch": 0.3245192307692308,
"grad_norm": 0.002276240848004818,
"learning_rate": 3.245192307692308e-06,
"loss": -0.0085,
"num_tokens": 269462412.0,
"reward": 1.2220972537994386,
"reward_std": 0.17137631475925447,
"rewards/accuracy_reward": 0.6633680582046508,
"rewards/brier_reward": 0.7907957911491394,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9900173544883728,
"rewards/mean_confidence_reward": 0.659448790550232,
"sampling/batch_mean_priority_error": 0.03146580492743215,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8055555555555556,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005192387942224741,
"sampling/priority_kl": 0.030000027641654013,
"sampling/priority_scale": 0.977628904604353,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.836079340544529e-05,
"sampling/prob_min": 1.3191566358727868e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.31919999718666076,
"sampling/prompt_draws_total": 9576.0,
"sampling/seen_fraction": 0.2951733350753784,
"sampling/unseen_fraction": 0.7048266649246215,
"signal/accuracy_reward/centered_abs_mean": 0.16806640625,
"signal/accuracy_reward/group_std_mean": 0.2234184443950653,
"signal/accuracy_reward/group_zero_std_frac": 0.3638888895511627,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.084033203125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.084033203125,
"signal/advantage_abs_mean": 0.12492303550243378,
"signal/advantage_pre_scale_abs_mean": 0.12492303550243378,
"signal/advantage_pre_scale_std": 0.2111401379108429,
"signal/advantage_std": 0.2111401379108429,
"signal/brier_reward/centered_abs_mean": 0.08217985183000565,
"signal/brier_reward/group_std_mean": 0.11212488561868668,
"signal/brier_reward/group_zero_std_frac": 0.002777777798473835,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04108992591500282,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04108992591500282,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01755099855363369,
"signal/format_reward/group_std_mean": 0.035760215297341344,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008775499276816845,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008775499276816845,
"signal/mean_confidence_reward/centered_abs_mean": 0.06346814930438996,
"signal/mean_confidence_reward/group_std_mean": 0.08517944514751434,
"signal/mean_confidence_reward/group_zero_std_frac": 0.002777777798473835,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.346814870994422e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.346814870994422e-07,
"step": 135
},
{
"calibration/aurc": 0.161821302991387,
"calibration/batch_distribution_entropy": 0.5380996745311886,
"calibration/batch_entropy_100bins": 0.3263382773335613,
"calibration/batch_entropy_10bins": 0.5380996745311886,
"calibration/batch_entropy_50bins": 0.38416013996797876,
"calibration/batch_uniqueness": 0.48281429749503796,
"calibration/confidence_entropy": 0.5990824174049726,
"calibration/coverage@0%": 0.0015693711521547933,
"calibration/coverage@1%": 0.0015693711521547933,
"calibration/coverage@10%": 0.27670903441669037,
"calibration/coverage@15%": 0.3979907857535888,
"calibration/coverage@20%": 0.7749324795283459,
"calibration/coverage@25%": 0.8180112451254938,
"calibration/coverage@30%": 0.9941868298433729,
"calibration/coverage@5%": 0.060392900563919494,
"calibration/distribution_entropy_10": 0.5380996745311886,
"calibration/distribution_entropy_100": 0.3263382773335613,
"calibration/ece": 0.08859721584863484,
"calibration/mean_confidence": 0.6633440478453676,
"calibration/unique_confidence_per_question": 0.02447916666666667,
"calibration/unique_confidences": 9.4,
"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": 0.013541666666666674,
"completions/max_length": 3597.4,
"completions/max_terminated_length": 3597.4,
"completions/mean_length": 650.0008911132812,
"completions/mean_terminated_length": 658.90849609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 180.6,
"epoch": 0.33653846153846156,
"grad_norm": 0.000727808685041964,
"learning_rate": 3.365384615384616e-06,
"loss": -0.013,
"num_tokens": 280008822.0,
"reward": 1.2597820997238158,
"reward_std": 0.1712968796491623,
"rewards/accuracy_reward": 0.7206597208976746,
"rewards/brier_reward": 0.8126063346862793,
"rewards/confidence_one_or_zero": 0.0003472222248092294,
"rewards/format_reward": 0.9862847208976746,
"rewards/mean_confidence_reward": 0.6640309572219849,
"sampling/batch_mean_priority_error": 0.031656787110785164,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8222222222222223,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005227043479681015,
"sampling/priority_kl": 0.029999838396906852,
"sampling/priority_scale": 0.961505502439104,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 3.8488073914777485e-05,
"sampling/prob_min": 1.3470378689817153e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3312000036239624,
"sampling/prompt_draws_total": 9936.0,
"sampling/seen_fraction": 0.3049199998378754,
"sampling/unseen_fraction": 0.6950800001621247,
"signal/accuracy_reward/centered_abs_mean": 0.1687391519546509,
"signal/accuracy_reward/group_std_mean": 0.2210682988166809,
"signal/accuracy_reward/group_zero_std_frac": 0.37777777314186095,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08436957597732545,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08436957597732545,
"signal/advantage_abs_mean": 0.12814967036247255,
"signal/advantage_pre_scale_abs_mean": 0.12814967036247255,
"signal/advantage_pre_scale_std": 0.21529237926006317,
"signal/advantage_std": 0.21529237926006317,
"signal/brier_reward/centered_abs_mean": 0.08760491609573365,
"signal/brier_reward/group_std_mean": 0.1167238175868988,
"signal/brier_reward/group_zero_std_frac": 0.002777777798473835,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043802458047866824,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043802458047866824,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0006727430503815412,
"signal/confidence_one_or_zero/group_std_mean": 0.0019641853868961334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 6.7274299908604006e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 6.7274299908604006e-09,
"signal/format_reward/centered_abs_mean": 0.020887586660683154,
"signal/format_reward/group_std_mean": 0.03523364253342152,
"signal/format_reward/group_zero_std_frac": 0.8638888835906983,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010443793330341577,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010443793330341577,
"signal/mean_confidence_reward/centered_abs_mean": 0.07050719410181046,
"signal/mean_confidence_reward/group_std_mean": 0.09147275984287262,
"signal/mean_confidence_reward/group_zero_std_frac": 0.00555555559694767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.050719091239443e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.050719091239443e-07,
"step": 140
},
{
"calibration/aurc": 0.2002019075525349,
"calibration/batch_distribution_entropy": 0.5122078027614163,
"calibration/batch_entropy_100bins": 0.29976493766985196,
"calibration/batch_entropy_10bins": 0.5122078027614163,
"calibration/batch_entropy_50bins": 0.35287843446889366,
"calibration/batch_uniqueness": 0.4097952308753282,
"calibration/confidence_entropy": 0.590766473840011,
"calibration/coverage@0%": 0.002086053089643168,
"calibration/coverage@1%": 0.002086053089643168,
"calibration/coverage@10%": 0.1858980635335074,
"calibration/coverage@15%": 0.5017274400600803,
"calibration/coverage@20%": 0.7080111738117296,
"calibration/coverage@25%": 0.8,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.002086053089643168,
"calibration/distribution_entropy_10": 0.5122078027614163,
"calibration/distribution_entropy_100": 0.29976493766985196,
"calibration/ece": 0.10117033702167381,
"calibration/mean_confidence": 0.6790366267827563,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.011805555555555559,
"completions/max_length": 3492.0,
"completions/max_terminated_length": 3492.0,
"completions/mean_length": 636.8101684570313,
"completions/mean_terminated_length": 644.3724243164063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 178.2,
"epoch": 0.3485576923076923,
"grad_norm": 0.000746583566069603,
"learning_rate": 3.4855769230769233e-06,
"loss": -0.0119,
"num_tokens": 290450027.0,
"reward": 1.236671781539917,
"reward_std": 0.1658506065607071,
"rewards/accuracy_reward": 0.682031261920929,
"rewards/brier_reward": 0.8031913042068481,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9881076335906982,
"rewards/mean_confidence_reward": 0.6686701536178589,
"sampling/batch_mean_priority_error": 0.03593021038001605,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8111111111111111,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0052638577297329904,
"sampling/priority_kl": 0.030000052228569985,
"sampling/priority_scale": 0.9464996993308887,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 3.861593140754849e-05,
"sampling/prob_min": 1.3738372217630967e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3431999981403351,
"sampling/prompt_draws_total": 10296.0,
"sampling/seen_fraction": 0.31465333700180054,
"sampling/unseen_fraction": 0.6853466629981995,
"signal/accuracy_reward/centered_abs_mean": 0.16121419370174409,
"signal/accuracy_reward/group_std_mean": 0.2107767105102539,
"signal/accuracy_reward/group_zero_std_frac": 0.4055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08060709685087204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08060709685087204,
"signal/advantage_abs_mean": 0.12273389399051667,
"signal/advantage_pre_scale_abs_mean": 0.12273389399051667,
"signal/advantage_pre_scale_std": 0.211455237865448,
"signal/advantage_std": 0.211455237865448,
"signal/brier_reward/centered_abs_mean": 0.08627397716045379,
"signal/brier_reward/group_std_mean": 0.11665287166833878,
"signal/brier_reward/group_zero_std_frac": 0.008333333395421505,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043136988580226895,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043136988580226895,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.02012261264026165,
"signal/format_reward/group_std_mean": 0.03786379247903824,
"signal/format_reward/group_zero_std_frac": 0.8444444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.010061306320130825,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.010061306320130825,
"signal/mean_confidence_reward/centered_abs_mean": 0.06955652832984924,
"signal/mean_confidence_reward/group_std_mean": 0.09108842760324479,
"signal/mean_confidence_reward/group_zero_std_frac": 0.008333333395421505,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.955652793294576e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.955652793294576e-07,
"step": 145
},
{
"calibration/aurc": 0.22469899536459623,
"calibration/batch_distribution_entropy": 0.49352183964938645,
"calibration/batch_entropy_100bins": 0.2907041538671515,
"calibration/batch_entropy_10bins": 0.49352183964938645,
"calibration/batch_entropy_50bins": 0.3422122263786084,
"calibration/batch_uniqueness": 0.37005133120849526,
"calibration/confidence_entropy": 0.58310095568642,
"calibration/coverage@0%": 0.001045767716535433,
"calibration/coverage@1%": 0.001045767716535433,
"calibration/coverage@10%": 0.1145874343832021,
"calibration/coverage@15%": 0.28893460881481636,
"calibration/coverage@20%": 0.4735757592549591,
"calibration/coverage@25%": 0.6433276057980674,
"calibration/coverage@30%": 0.7105192068479361,
"calibration/coverage@5%": 0.001045767716535433,
"calibration/distribution_entropy_10": 0.49352183964938645,
"calibration/distribution_entropy_100": 0.2907041538671515,
"calibration/ece": 0.0853479206156124,
"calibration/mean_confidence": 0.6878991824983854,
"calibration/unique_confidence_per_question": 0.019270833333333334,
"calibration/unique_confidences": 7.4,
"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": 0.015885416666666652,
"completions/max_length": 3841.4,
"completions/max_terminated_length": 3841.4,
"completions/mean_length": 671.563720703125,
"completions/mean_terminated_length": 682.4291015625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 201.6,
"epoch": 0.3605769230769231,
"grad_norm": 0.0005632630200125277,
"learning_rate": 3.605769230769231e-06,
"loss": -0.0151,
"num_tokens": 301300905.0,
"reward": 1.2439425945281983,
"reward_std": 0.17063007950782777,
"rewards/accuracy_reward": 0.6962673544883728,
"rewards/brier_reward": 0.8076630473136902,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9839409828186035,
"rewards/mean_confidence_reward": 0.685508668422699,
"sampling/batch_mean_priority_error": 0.03787541302960014,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7638888888888888,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.00530406478792429,
"sampling/priority_kl": 0.03000020757317543,
"sampling/priority_scale": 0.9324293434852734,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 3.873952809954062e-05,
"sampling/prob_min": 1.3995698463986627e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.35520000457763673,
"sampling/prompt_draws_total": 10656.0,
"sampling/seen_fraction": 0.32404000163078306,
"sampling/unseen_fraction": 0.6759599983692169,
"signal/accuracy_reward/centered_abs_mean": 0.14857313334941863,
"signal/accuracy_reward/group_std_mean": 0.20339624285697938,
"signal/accuracy_reward/group_zero_std_frac": 0.3944444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07428656667470931,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07428656667470931,
"signal/advantage_abs_mean": 0.12221142053604125,
"signal/advantage_pre_scale_abs_mean": 0.12221142053604125,
"signal/advantage_pre_scale_std": 0.21580576300621032,
"signal/advantage_std": 0.21580576300621032,
"signal/brier_reward/centered_abs_mean": 0.09183170050382614,
"signal/brier_reward/group_std_mean": 0.1252654179930687,
"signal/brier_reward/group_zero_std_frac": 0.03611111212521791,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04591585025191307,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04591585025191307,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02568901889026165,
"signal/format_reward/group_std_mean": 0.04536781832575798,
"signal/format_reward/group_zero_std_frac": 0.819444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012844509445130825,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012844509445130825,
"signal/mean_confidence_reward/centered_abs_mean": 0.07101247161626816,
"signal/mean_confidence_reward/group_std_mean": 0.09455030411481857,
"signal/mean_confidence_reward/group_zero_std_frac": 0.04722222331911326,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.101247319951654e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.101247319951654e-07,
"step": 150
},
{
"epoch": 0.3605769230769231,
"eval_calibration/aurc": 0.21648289229710788,
"eval_calibration/batch_distribution_entropy": 0.48338692999776756,
"eval_calibration/batch_entropy_100bins": 0.27455615561797053,
"eval_calibration/batch_entropy_10bins": 0.48338692999776756,
"eval_calibration/batch_entropy_50bins": 0.32320306411209526,
"eval_calibration/batch_uniqueness": 0.2413918376060083,
"eval_calibration/confidence_entropy": 0.5636802490436134,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.0,
"eval_calibration/coverage@20%": 0.5665198237885463,
"eval_calibration/coverage@25%": 0.7541850220264317,
"eval_calibration/coverage@30%": 0.8176211453744493,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.48338692999776756,
"eval_calibration/distribution_entropy_100": 0.27455615561797053,
"eval_calibration/ece": 0.05352422907488988,
"eval_calibration/mean_confidence": 0.7074449339207047,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.013020833333333334,
"eval_completions/max_length": 2320.5,
"eval_completions/max_terminated_length": 2320.5,
"eval_completions/mean_length": 684.08056640625,
"eval_completions/mean_terminated_length": 693.1365661621094,
"eval_completions/min_length": 54.666666666666664,
"eval_completions/min_terminated_length": 242.16666666666666,
"eval_loss": 0.0,
"eval_num_tokens": 301300905.0,
"eval_reward": 1.227612296740214,
"eval_reward_std": 0.34397614002227783,
"eval_rewards/accuracy_reward": 0.6710069378217062,
"eval_rewards/brier_reward": 0.7989605267842611,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9852430522441864,
"eval_rewards/mean_confidence_reward": 0.6970052123069763,
"eval_runtime": 210.4163,
"eval_samples_per_second": 4.752,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4160698751608531,
"eval_signal/accuracy_reward/group_std_mean": 0.4616414209206899,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20803493758042654,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20803493758042654,
"eval_signal/advantage_abs_mean": 0.2960173139969508,
"eval_signal/advantage_pre_scale_abs_mean": 0.2960173139969508,
"eval_signal/advantage_pre_scale_std": 0.34291206300258636,
"eval_signal/advantage_std": 0.34291206300258636,
"eval_signal/brier_reward/centered_abs_mean": 0.17483383417129517,
"eval_signal/brier_reward/group_std_mean": 0.22024944176276526,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08741691708564758,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08741691708564758,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.02794053762530287,
"eval_signal/format_reward/group_std_mean": 0.06800450353572766,
"eval_signal/format_reward/group_zero_std_frac": 0.6666666865348816,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.013970268812651435,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.013970268812651435,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.11518825963139534,
"eval_signal/mean_confidence_reward/group_std_mean": 0.15239649017651877,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.1518825620745095e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.1518825620745095e-06,
"eval_steps_per_second": 0.029,
"step": 150
},
{
"epoch": 0.3605769230769231,
"step": 150,
"train_probe_calibration/aurc": 0.20118106887974196,
"train_probe_calibration/batch_distribution_entropy": 0.48299539334492897,
"train_probe_calibration/batch_entropy_100bins": 0.2740406292224008,
"train_probe_calibration/batch_entropy_10bins": 0.48299539334492897,
"train_probe_calibration/batch_entropy_50bins": 0.32259619478037777,
"train_probe_calibration/batch_uniqueness": 0.25400156219004166,
"train_probe_calibration/confidence_entropy": 0.5653509032070053,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.0,
"train_probe_calibration/coverage@15%": 0.006161971830985915,
"train_probe_calibration/coverage@20%": 0.5607394366197183,
"train_probe_calibration/coverage@25%": 0.8116197183098591,
"train_probe_calibration/coverage@30%": 0.9964788732394366,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.48299539334492897,
"train_probe_calibration/distribution_entropy_100": 0.2740406292224008,
"train_probe_calibration/ece": 0.04286971830985919,
"train_probe_calibration/mean_confidence": 0.7078345070422535,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.014583333333333356,
"train_probe_completions/max_length": 2331.5,
"train_probe_completions/max_terminated_length": 2331.5,
"train_probe_completions/mean_length": 686.4166870117188,
"train_probe_completions/mean_terminated_length": 696.6106872558594,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 215.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 301300905.0,
"train_probe_reward": 1.2423616449038188,
"train_probe_reward_std": 0.3410983880360921,
"train_probe_rewards/accuracy_reward": 0.6901041666666666,
"train_probe_rewards/brier_reward": 0.8084939221541086,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9861111243565878,
"train_probe_rewards/mean_confidence_reward": 0.6980034708976746,
"train_probe_runtime": 212.6957,
"train_probe_samples_per_second": 4.702,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.4138997445503871,
"train_probe_signal/accuracy_reward/group_std_mean": 0.46029503146807355,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20694987227519354,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.20694987227519354,
"train_probe_signal/advantage_abs_mean": 0.2931293447812398,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2931293447812398,
"train_probe_signal/advantage_pre_scale_std": 0.33921169737974805,
"train_probe_signal/advantage_std": 0.33921169737974805,
"train_probe_signal/brier_reward/centered_abs_mean": 0.16598876068989435,
"train_probe_signal/brier_reward/group_std_mean": 0.2144825185338656,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08299438034494717,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08299438034494717,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.026258680193374555,
"train_probe_signal/format_reward/group_std_mean": 0.06309404006848733,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6944444676240286,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.013129340096687278,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.013129340096687278,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1145426481962204,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.14749648794531822,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.1454264381427492e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.1454264381427492e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.22965670297129837,
"calibration/batch_distribution_entropy": 0.4561284084250222,
"calibration/batch_entropy_100bins": 0.25662249926372777,
"calibration/batch_entropy_10bins": 0.4561284084250222,
"calibration/batch_entropy_50bins": 0.3020918540160094,
"calibration/batch_uniqueness": 0.16051952690981258,
"calibration/confidence_entropy": 0.558045726461945,
"calibration/coverage@0%": 0.002655612171382654,
"calibration/coverage@1%": 0.002655612171382654,
"calibration/coverage@10%": 0.002655612171382654,
"calibration/coverage@15%": 0.002655612171382654,
"calibration/coverage@20%": 0.43610356000332223,
"calibration/coverage@25%": 0.7411703971289201,
"calibration/coverage@30%": 0.9039506908389366,
"calibration/coverage@5%": 0.002655612171382654,
"calibration/distribution_entropy_10": 0.4561284084250222,
"calibration/distribution_entropy_100": 0.25662249926372777,
"calibration/ece": 0.07278548683083944,
"calibration/mean_confidence": 0.7137022027408687,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.013975694444444464,
"completions/max_length": 3620.6,
"completions/max_terminated_length": 3620.6,
"completions/mean_length": 692.3322021484375,
"completions/mean_terminated_length": 702.1637329101562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 187.2,
"epoch": 0.37259615384615385,
"grad_norm": 0.0006764737772755325,
"learning_rate": 3.725961538461539e-06,
"loss": -0.0135,
"num_tokens": 312347452.0,
"reward": 1.2194071769714356,
"reward_std": 0.17755380570888518,
"rewards/accuracy_reward": 0.6553819417953491,
"rewards/brier_reward": 0.7974809050559998,
"rewards/confidence_one_or_zero": 0.00026041667442768814,
"rewards/format_reward": 0.985937488079071,
"rewards/mean_confidence_reward": 0.6956162929534913,
"sampling/batch_mean_priority_error": 0.044587768701128454,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7972222222222223,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005351596139371395,
"sampling/priority_kl": 0.02999967634677887,
"sampling/priority_scale": 0.9192980109946802,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 3.886347622028552e-05,
"sampling/prob_min": 1.4243225450627506e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.3671999990940094,
"sampling/prompt_draws_total": 11016.0,
"sampling/seen_fraction": 0.33339332938194277,
"sampling/unseen_fraction": 0.6666066706180572,
"signal/accuracy_reward/centered_abs_mean": 0.16404079794883727,
"signal/accuracy_reward/group_std_mean": 0.2167521059513092,
"signal/accuracy_reward/group_zero_std_frac": 0.38611111640930174,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.08202039897441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.08202039897441864,
"signal/advantage_abs_mean": 0.12833702713251113,
"signal/advantage_pre_scale_abs_mean": 0.12833702713251113,
"signal/advantage_pre_scale_std": 0.221464666724205,
"signal/advantage_std": 0.221464666724205,
"signal/brier_reward/centered_abs_mean": 0.09387256652116775,
"signal/brier_reward/group_std_mean": 0.12882587313652039,
"signal/brier_reward/group_zero_std_frac": 0.04722222331911326,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.046936283260583876,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.046936283260583876,
"signal/confidence_one_or_zero/centered_abs_mean": 0.00047200522385537624,
"signal/confidence_one_or_zero/group_std_mean": 0.0008226238191127777,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 4.7200515496115255e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 4.7200515496115255e-09,
"signal/format_reward/centered_abs_mean": 0.023242188058793543,
"signal/format_reward/group_std_mean": 0.04568904340267181,
"signal/format_reward/group_zero_std_frac": 0.8055555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.011621094029396772,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.011621094029396772,
"signal/mean_confidence_reward/centered_abs_mean": 0.06486843228340149,
"signal/mean_confidence_reward/group_std_mean": 0.09012808799743652,
"signal/mean_confidence_reward/group_zero_std_frac": 0.06111111231148243,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.486842949016137e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.486842949016137e-07,
"step": 155
},
{
"calibration/aurc": 0.23205115891678957,
"calibration/batch_distribution_entropy": 0.43420742788405137,
"calibration/batch_entropy_100bins": 0.2390598214395338,
"calibration/batch_entropy_10bins": 0.43420742788405137,
"calibration/batch_entropy_50bins": 0.2814173538431148,
"calibration/batch_uniqueness": 0.10195991736118337,
"calibration/confidence_entropy": 0.552720533534938,
"calibration/coverage@0%": 0.004207814359698442,
"calibration/coverage@1%": 0.004207814359698442,
"calibration/coverage@10%": 0.13457430650629532,
"calibration/coverage@15%": 0.16651147928116442,
"calibration/coverage@20%": 0.407872735825667,
"calibration/coverage@25%": 0.5434927885961156,
"calibration/coverage@30%": 0.7420298591455085,
"calibration/coverage@5%": 0.004207814359698442,
"calibration/distribution_entropy_10": 0.43420742788405137,
"calibration/distribution_entropy_100": 0.2390598214395338,
"calibration/ece": 0.10113857306348302,
"calibration/mean_confidence": 0.7223825960740257,
"calibration/unique_confidence_per_question": 0.018229166666666668,
"calibration/unique_confidences": 7.0,
"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": 0.008767361111111116,
"completions/max_length": 2959.4,
"completions/max_terminated_length": 2959.4,
"completions/mean_length": 677.7904663085938,
"completions/mean_terminated_length": 683.8215942382812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 197.6,
"epoch": 0.38461538461538464,
"grad_norm": 0.0006002213340252638,
"learning_rate": 3.846153846153847e-06,
"loss": -0.0078,
"num_tokens": 323238542.0,
"reward": 1.2424018144607545,
"reward_std": 0.1606945961713791,
"rewards/accuracy_reward": 0.6875867962837219,
"rewards/brier_reward": 0.8059702754020691,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9912326455116272,
"rewards/mean_confidence_reward": 0.6932335138320923,
"sampling/batch_mean_priority_error": 0.04362556407230752,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.8138888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005402308888733387,
"sampling/priority_kl": 0.030000124871730805,
"sampling/priority_scale": 0.9071862280601636,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 3.899485018337146e-05,
"sampling/prob_min": 1.4481612379313446e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.37919999957084655,
"sampling/prompt_draws_total": 11376.0,
"sampling/seen_fraction": 0.34318000078201294,
"sampling/unseen_fraction": 0.6568199992179871,
"signal/accuracy_reward/centered_abs_mean": 0.15255534052848815,
"signal/accuracy_reward/group_std_mean": 0.20133095383644103,
"signal/accuracy_reward/group_zero_std_frac": 0.4305555522441864,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07627767026424408,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07627767026424408,
"signal/advantage_abs_mean": 0.11574586480855942,
"signal/advantage_pre_scale_abs_mean": 0.11574586480855942,
"signal/advantage_pre_scale_std": 0.2071862369775772,
"signal/advantage_std": 0.2071862369775772,
"signal/brier_reward/centered_abs_mean": 0.08608684539794922,
"signal/brier_reward/group_std_mean": 0.11757852882146835,
"signal/brier_reward/group_zero_std_frac": 0.06944444626569748,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04304342269897461,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04304342269897461,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.016097005270421506,
"signal/format_reward/group_std_mean": 0.0349950447678566,
"signal/format_reward/group_zero_std_frac": 0.8416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008048502635210753,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008048502635210753,
"signal/mean_confidence_reward/centered_abs_mean": 0.06561442390084267,
"signal/mean_confidence_reward/group_std_mean": 0.08899795264005661,
"signal/mean_confidence_reward/group_zero_std_frac": 0.08055555745959282,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.561442205565981e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.561442205565981e-07,
"step": 160
},
{
"calibration/aurc": 0.22540844687267875,
"calibration/batch_distribution_entropy": 0.50386111626777,
"calibration/batch_entropy_100bins": 0.27361577246112806,
"calibration/batch_entropy_10bins": 0.50386111626777,
"calibration/batch_entropy_50bins": 0.3220960602751323,
"calibration/batch_uniqueness": 0.2634407387572316,
"calibration/confidence_entropy": 0.5741133615322693,
"calibration/coverage@0%": 0.0005221932114882506,
"calibration/coverage@1%": 0.0005221932114882506,
"calibration/coverage@10%": 0.1364801984608321,
"calibration/coverage@15%": 0.22525304441383467,
"calibration/coverage@20%": 0.4553799642709908,
"calibration/coverage@25%": 0.5910975643984225,
"calibration/coverage@30%": 0.7239594937576296,
"calibration/coverage@5%": 0.10708387300151449,
"calibration/distribution_entropy_10": 0.50386111626777,
"calibration/distribution_entropy_100": 0.27361577246112806,
"calibration/ece": 0.10513270199830167,
"calibration/mean_confidence": 0.6773881662806778,
"calibration/unique_confidence_per_question": 0.018750000000000003,
"calibration/unique_confidences": 7.2,
"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": 0.00789930555555558,
"completions/max_length": 3310.8,
"completions/max_terminated_length": 3310.8,
"completions/mean_length": 687.6014038085938,
"completions/mean_terminated_length": 693.0419189453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 190.0,
"epoch": 0.39663461538461536,
"grad_norm": 0.0007785888155922294,
"learning_rate": 3.966346153846154e-06,
"loss": -0.0083,
"num_tokens": 334271806.0,
"reward": 1.232865309715271,
"reward_std": 0.1643844425678253,
"rewards/accuracy_reward": 0.6660590291023254,
"rewards/brier_reward": 0.8076440930366516,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9920138955116272,
"rewards/mean_confidence_reward": 0.6794357776641846,
"sampling/batch_mean_priority_error": 0.04254050978087628,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7833333333333333,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005447699595242739,
"sampling/priority_kl": 0.029999880865216256,
"sampling/priority_scale": 0.8958297192817554,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 3.91241439501755e-05,
"sampling/prob_min": 1.470900224376237e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.39120000004768374,
"sampling/prompt_draws_total": 11736.0,
"sampling/seen_fraction": 0.3527533352375031,
"sampling/unseen_fraction": 0.6472466647624969,
"signal/accuracy_reward/centered_abs_mean": 0.166943359375,
"signal/accuracy_reward/group_std_mean": 0.217330726981163,
"signal/accuracy_reward/group_zero_std_frac": 0.3861111104488373,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0834716796875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0834716796875,
"signal/advantage_abs_mean": 0.11999084949493408,
"signal/advantage_pre_scale_abs_mean": 0.11999084949493408,
"signal/advantage_pre_scale_std": 0.2072890281677246,
"signal/advantage_std": 0.2072890281677246,
"signal/brier_reward/centered_abs_mean": 0.08629768192768097,
"signal/brier_reward/group_std_mean": 0.11852459907531739,
"signal/brier_reward/group_zero_std_frac": 0.09444444701075554,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04314884096384049,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04314884096384049,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01467013880610466,
"signal/format_reward/group_std_mean": 0.03149766884744167,
"signal/format_reward/group_zero_std_frac": 0.8583333373069764,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00733506940305233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00733506940305233,
"signal/mean_confidence_reward/centered_abs_mean": 0.06508653238415718,
"signal/mean_confidence_reward/group_std_mean": 0.08856085240840912,
"signal/mean_confidence_reward/group_zero_std_frac": 0.11666666865348815,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.508653086712002e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.508653086712002e-07,
"step": 165
},
{
"calibration/aurc": 0.17985976022582273,
"calibration/batch_distribution_entropy": 0.413745285715431,
"calibration/batch_entropy_100bins": 0.22452526520252097,
"calibration/batch_entropy_10bins": 0.413745285715431,
"calibration/batch_entropy_50bins": 0.26430750940805287,
"calibration/batch_uniqueness": 0.11678018501184825,
"calibration/confidence_entropy": 0.573921612271716,
"calibration/coverage@0%": 0.0010526315789473684,
"calibration/coverage@1%": 0.0010526315789473684,
"calibration/coverage@10%": 0.0010526315789473684,
"calibration/coverage@15%": 0.4791675433804964,
"calibration/coverage@20%": 0.7099612139451155,
"calibration/coverage@25%": 0.7926837270341207,
"calibration/coverage@30%": 0.9301837270341207,
"calibration/coverage@5%": 0.0010526315789473684,
"calibration/distribution_entropy_10": 0.413745285715431,
"calibration/distribution_entropy_100": 0.22452526520252097,
"calibration/ece": 0.06974335424247227,
"calibration/mean_confidence": 0.688039419033681,
"calibration/unique_confidence_per_question": 0.014583333333333334,
"calibration/unique_confidences": 5.6,
"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": 0.011458333333333326,
"completions/max_length": 3253.0,
"completions/max_terminated_length": 3253.0,
"completions/mean_length": 712.0092041015625,
"completions/mean_terminated_length": 720.3050659179687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 187.4,
"epoch": 0.40865384615384615,
"grad_norm": 0.0007355434936471283,
"learning_rate": 4.086538461538462e-06,
"loss": -0.0097,
"num_tokens": 345598280.0,
"reward": 1.2475274801254272,
"reward_std": 0.1580900192260742,
"rewards/accuracy_reward": 0.6924479246139527,
"rewards/brier_reward": 0.8140516638755798,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9885416626930237,
"rewards/mean_confidence_reward": 0.6819965600967407,
"sampling/batch_mean_priority_error": 0.04085833234276857,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7277777777777777,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005493382271379232,
"sampling/priority_kl": 0.02999972701072693,
"sampling/priority_scale": 0.8849816142814234,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.9243818901013586e-05,
"sampling/prob_min": 1.4929330063750967e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.4032000005245209,
"sampling/prompt_draws_total": 12096.0,
"sampling/seen_fraction": 0.3616133391857147,
"sampling/unseen_fraction": 0.6383866608142853,
"signal/accuracy_reward/centered_abs_mean": 0.15483398288488387,
"signal/accuracy_reward/group_std_mean": 0.2024546504020691,
"signal/accuracy_reward/group_zero_std_frac": 0.4361111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07741699144244193,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07741699144244193,
"signal/advantage_abs_mean": 0.1167249247431755,
"signal/advantage_pre_scale_abs_mean": 0.1167249247431755,
"signal/advantage_pre_scale_std": 0.2062102288007736,
"signal/advantage_std": 0.2062102288007736,
"signal/brier_reward/centered_abs_mean": 0.08543108999729157,
"signal/brier_reward/group_std_mean": 0.11673992872238159,
"signal/brier_reward/group_zero_std_frac": 0.1250000014901161,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.042715544998645785,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.042715544998645785,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.018695746455341576,
"signal/format_reward/group_std_mean": 0.033667823672294615,
"signal/format_reward/group_zero_std_frac": 0.8666666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009347873227670788,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009347873227670788,
"signal/mean_confidence_reward/centered_abs_mean": 0.06826469749212265,
"signal/mean_confidence_reward/group_std_mean": 0.09210723340511322,
"signal/mean_confidence_reward/group_zero_std_frac": 0.14166666716337203,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.826469984844152e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.826469984844152e-07,
"step": 170
},
{
"calibration/aurc": 0.21927099575947104,
"calibration/batch_distribution_entropy": 0.3604094691063196,
"calibration/batch_entropy_100bins": 0.1965044571986865,
"calibration/batch_entropy_10bins": 0.3604094691063196,
"calibration/batch_entropy_50bins": 0.23132186759881396,
"calibration/batch_uniqueness": -0.05702418518762982,
"calibration/confidence_entropy": 0.5533325414996917,
"calibration/coverage@0%": 0.0010416666666666667,
"calibration/coverage@1%": 0.0010416666666666667,
"calibration/coverage@10%": 0.3256489965095986,
"calibration/coverage@15%": 0.340832242582897,
"calibration/coverage@20%": 0.488612921434072,
"calibration/coverage@25%": 0.49070169428002497,
"calibration/coverage@30%": 0.6922723749082973,
"calibration/coverage@5%": 0.0010416666666666667,
"calibration/distribution_entropy_10": 0.3604094691063196,
"calibration/distribution_entropy_100": 0.1965044571986865,
"calibration/ece": 0.1368860904827325,
"calibration/mean_confidence": 0.7124982051032311,
"calibration/unique_confidence_per_question": 0.016145833333333335,
"calibration/unique_confidences": 6.2,
"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": 0.0052951388888889065,
"completions/max_length": 2967.6,
"completions/max_terminated_length": 2967.6,
"completions/mean_length": 720.8404663085937,
"completions/mean_terminated_length": 724.6657348632813,
"completions/min_length": 0.0,
"completions/min_terminated_length": 195.2,
"epoch": 0.4206730769230769,
"grad_norm": 0.0006969543755985796,
"learning_rate": 4.20673076923077e-06,
"loss": -0.0049,
"num_tokens": 356992282.0,
"reward": 1.2345721244812011,
"reward_std": 0.1517510175704956,
"rewards/accuracy_reward": 0.6669270753860473,
"rewards/brier_reward": 0.8076720833778381,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99453125,
"rewards/mean_confidence_reward": 0.689978277683258,
"sampling/batch_mean_priority_error": 0.04690258188712983,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7444444444444444,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.00554170049726963,
"sampling/priority_kl": 0.029999922215938567,
"sampling/priority_scale": 0.8748339593177661,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 3.9364156691590324e-05,
"sampling/prob_min": 1.5141977200983092e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.415200001001358,
"sampling/prompt_draws_total": 12456.0,
"sampling/seen_fraction": 0.3704466700553894,
"sampling/unseen_fraction": 0.6295533299446106,
"signal/accuracy_reward/centered_abs_mean": 0.15778537094593048,
"signal/accuracy_reward/group_std_mean": 0.20124109387397765,
"signal/accuracy_reward/group_zero_std_frac": 0.45277778506278993,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07889268547296524,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07889268547296524,
"signal/advantage_abs_mean": 0.1146531030535698,
"signal/advantage_pre_scale_abs_mean": 0.1146531030535698,
"signal/advantage_pre_scale_std": 0.19675582647323608,
"signal/advantage_std": 0.19675582647323608,
"signal/brier_reward/centered_abs_mean": 0.08641349822282791,
"signal/brier_reward/group_std_mean": 0.1162311390042305,
"signal/brier_reward/group_zero_std_frac": 0.16111111342906953,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04320674911141396,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04320674911141396,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009901258815079927,
"signal/format_reward/group_std_mean": 0.022229401022195817,
"signal/format_reward/group_zero_std_frac": 0.8944444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004950629407539963,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004950629407539963,
"signal/mean_confidence_reward/centered_abs_mean": 0.06902343928813934,
"signal/mean_confidence_reward/group_std_mean": 0.09159857630729676,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18888889253139496,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.902343670844857e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.902343670844857e-07,
"step": 175
},
{
"calibration/aurc": 0.167481550360431,
"calibration/batch_distribution_entropy": 0.3882773633167407,
"calibration/batch_entropy_100bins": 0.21081035817577246,
"calibration/batch_entropy_10bins": 0.3882773633167407,
"calibration/batch_entropy_50bins": 0.24816254276150107,
"calibration/batch_uniqueness": 0.025158058255912157,
"calibration/confidence_entropy": 0.5634601281146097,
"calibration/coverage@0%": 0.002096989185252367,
"calibration/coverage@1%": 0.002096989185252367,
"calibration/coverage@10%": 0.22303939756221572,
"calibration/coverage@15%": 0.6428123840540875,
"calibration/coverage@20%": 0.6590496214151241,
"calibration/coverage@25%": 0.6957950282392711,
"calibration/coverage@30%": 0.8132053519488075,
"calibration/coverage@5%": 0.12146871693394345,
"calibration/distribution_entropy_10": 0.3882773633167407,
"calibration/distribution_entropy_100": 0.21081035817577246,
"calibration/ece": 0.11586185334078787,
"calibration/mean_confidence": 0.6786814728079384,
"calibration/unique_confidence_per_question": 0.01875,
"calibration/unique_confidences": 7.2,
"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": 0.006510416666666652,
"completions/max_length": 2608.0,
"completions/max_terminated_length": 2608.0,
"completions/mean_length": 692.8581665039062,
"completions/mean_terminated_length": 697.3784790039062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 228.6,
"epoch": 0.4326923076923077,
"grad_norm": 0.0006377384415827692,
"learning_rate": 4.326923076923077e-06,
"loss": -0.0057,
"num_tokens": 368071608.0,
"reward": 1.2389385461807252,
"reward_std": 0.14253281354904174,
"rewards/accuracy_reward": 0.6736111164093017,
"rewards/brier_reward": 0.8110234379768372,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9932291746139527,
"rewards/mean_confidence_reward": 0.6653298616409302,
"sampling/batch_mean_priority_error": 0.046377329209952564,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7166666666666668,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005593094788491726,
"sampling/priority_kl": 0.029999877139925957,
"sampling/priority_scale": 0.8652638375526294,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 3.948584635509178e-05,
"sampling/prob_min": 1.5348353190347553e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.4272000014781952,
"sampling/prompt_draws_total": 12816.0,
"sampling/seen_fraction": 0.37928666472434996,
"sampling/unseen_fraction": 0.62071333527565,
"signal/accuracy_reward/centered_abs_mean": 0.14998914897441865,
"signal/accuracy_reward/group_std_mean": 0.19401742815971373,
"signal/accuracy_reward/group_zero_std_frac": 0.45833333730697634,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07499457448720932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07499457448720932,
"signal/advantage_abs_mean": 0.10474234223365783,
"signal/advantage_pre_scale_abs_mean": 0.10474234223365783,
"signal/advantage_pre_scale_std": 0.1851011574268341,
"signal/advantage_std": 0.1851011574268341,
"signal/brier_reward/centered_abs_mean": 0.08834660351276398,
"signal/brier_reward/group_std_mean": 0.11847157329320908,
"signal/brier_reward/group_zero_std_frac": 0.14166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04417330175638199,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04417330175638199,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0115234375,
"signal/format_reward/group_std_mean": 0.025313834473490716,
"signal/format_reward/group_zero_std_frac": 0.8805555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00576171875,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00576171875,
"signal/mean_confidence_reward/centered_abs_mean": 0.08030165135860443,
"signal/mean_confidence_reward/group_std_mean": 0.10517692267894745,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1638888895511627,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.030164622141456e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.030164622141456e-07,
"step": 180
},
{
"calibration/aurc": 0.19044998736844715,
"calibration/batch_distribution_entropy": 0.4409406822397132,
"calibration/batch_entropy_100bins": 0.24965460938374448,
"calibration/batch_entropy_10bins": 0.4409406822397132,
"calibration/batch_entropy_50bins": 0.2938893667888068,
"calibration/batch_uniqueness": 0.14487271940651691,
"calibration/confidence_entropy": 0.558334050340173,
"calibration/coverage@0%": 0.006266318537859009,
"calibration/coverage@1%": 0.006266318537859009,
"calibration/coverage@10%": 0.006266318537859009,
"calibration/coverage@15%": 0.5150388925152307,
"calibration/coverage@20%": 0.7192069190600522,
"calibration/coverage@25%": 0.7990208877284595,
"calibration/coverage@30%": 0.9112271540469974,
"calibration/coverage@5%": 0.006266318537859009,
"calibration/distribution_entropy_10": 0.4409406822397132,
"calibration/distribution_entropy_100": 0.24965460938374448,
"calibration/ece": 0.07587018603133158,
"calibration/mean_confidence": 0.6548105689730199,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.004166666666666652,
"completions/max_length": 2245.6,
"completions/max_terminated_length": 2245.6,
"completions/mean_length": 736.2003540039062,
"completions/mean_terminated_length": 739.3214111328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 231.4,
"epoch": 0.44471153846153844,
"grad_norm": 0.0005959332338534296,
"learning_rate": 4.447115384615385e-06,
"loss": -0.0035,
"num_tokens": 379675452.0,
"reward": 1.248991894721985,
"reward_std": 0.13299419730901718,
"rewards/accuracy_reward": 0.6863715410232544,
"rewards/brier_reward": 0.8158528804779053,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9957465291023254,
"rewards/mean_confidence_reward": 0.6423176884651184,
"sampling/batch_mean_priority_error": 0.044696025078938384,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7138888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005641408730298281,
"sampling/priority_kl": 0.030000098049640656,
"sampling/priority_scale": 0.856269246363081,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 3.960601825383492e-05,
"sampling/prob_min": 1.554702212160919e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.43920000195503234,
"sampling/prompt_draws_total": 13176.0,
"sampling/seen_fraction": 0.3879533350467682,
"sampling/unseen_fraction": 0.6120466649532318,
"signal/accuracy_reward/centered_abs_mean": 0.13031141310930253,
"signal/accuracy_reward/group_std_mean": 0.1774735778570175,
"signal/accuracy_reward/group_zero_std_frac": 0.4777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06515570655465126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06515570655465126,
"signal/advantage_abs_mean": 0.09407005459070206,
"signal/advantage_pre_scale_abs_mean": 0.09407005459070206,
"signal/advantage_pre_scale_std": 0.169356307387352,
"signal/advantage_std": 0.169356307387352,
"signal/brier_reward/centered_abs_mean": 0.08961217999458312,
"signal/brier_reward/group_std_mean": 0.12095813304185868,
"signal/brier_reward/group_zero_std_frac": 0.0722222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04480608999729156,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04480608999729156,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.00793728269636631,
"signal/format_reward/group_std_mean": 0.01877547800540924,
"signal/format_reward/group_zero_std_frac": 0.9083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003968641348183155,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003968641348183155,
"signal/mean_confidence_reward/centered_abs_mean": 0.08483316749334335,
"signal/mean_confidence_reward/group_std_mean": 0.111886428296566,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0833333358168602,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.483316150886821e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.483316150886821e-07,
"step": 185
},
{
"calibration/aurc": 0.16341015205736542,
"calibration/batch_distribution_entropy": 0.4751064047048981,
"calibration/batch_entropy_100bins": 0.28294152401087824,
"calibration/batch_entropy_10bins": 0.4751064047048981,
"calibration/batch_entropy_50bins": 0.3330741841101023,
"calibration/batch_uniqueness": 0.23920660908986213,
"calibration/confidence_entropy": 0.5537682233129289,
"calibration/coverage@0%": 0.0047257908550904825,
"calibration/coverage@1%": 0.0047257908550904825,
"calibration/coverage@10%": 0.12252683797550934,
"calibration/coverage@15%": 0.5624311863974538,
"calibration/coverage@20%": 0.6029575021869276,
"calibration/coverage@25%": 0.9241356941957267,
"calibration/coverage@30%": 0.9821989528795811,
"calibration/coverage@5%": 0.12252683797550934,
"calibration/distribution_entropy_10": 0.4751064047048981,
"calibration/distribution_entropy_100": 0.28294152401087824,
"calibration/ece": 0.13411462163037816,
"calibration/mean_confidence": 0.636389035043661,
"calibration/unique_confidence_per_question": 0.023437499999999997,
"calibration/unique_confidences": 9.0,
"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": 0.005381944444444442,
"completions/max_length": 3022.0,
"completions/max_terminated_length": 3022.0,
"completions/mean_length": 770.71875,
"completions/mean_terminated_length": 774.9037719726563,
"completions/min_length": 0.0,
"completions/min_terminated_length": 265.8,
"epoch": 0.4567307692307692,
"grad_norm": 0.0006286805146373808,
"learning_rate": 4.567307692307692e-06,
"loss": -0.0059,
"num_tokens": 391665524.0,
"reward": 1.246759819984436,
"reward_std": 0.1368389680981636,
"rewards/accuracy_reward": 0.6818576335906983,
"rewards/brier_reward": 0.8174648761749268,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9941840291023254,
"rewards/mean_confidence_reward": 0.6571961641311646,
"sampling/batch_mean_priority_error": 0.047586199898986284,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6944444444444444,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005692024063318968,
"sampling/priority_kl": 0.03000013455748558,
"sampling/priority_scale": 0.8477106154197827,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 3.972284248447977e-05,
"sampling/prob_min": 1.5739300579298288e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.451199996471405,
"sampling/prompt_draws_total": 13536.0,
"sampling/seen_fraction": 0.39632666707038877,
"sampling/unseen_fraction": 0.6036733329296112,
"signal/accuracy_reward/centered_abs_mean": 0.14089084267616273,
"signal/accuracy_reward/group_std_mean": 0.1887511819601059,
"signal/accuracy_reward/group_zero_std_frac": 0.4499999940395355,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07044542133808136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07044542133808136,
"signal/advantage_abs_mean": 0.0984200268983841,
"signal/advantage_pre_scale_abs_mean": 0.0984200268983841,
"signal/advantage_pre_scale_std": 0.17806504666805267,
"signal/advantage_std": 0.17806504666805267,
"signal/brier_reward/centered_abs_mean": 0.09338046461343766,
"signal/brier_reward/group_std_mean": 0.12290722280740737,
"signal/brier_reward/group_zero_std_frac": 0.15000000447034836,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04669023230671883,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04669023230671883,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0003363715251907706,
"signal/confidence_one_or_zero/group_std_mean": 0.0009820926934480667,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.3637149954302003e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.3637149954302003e-09,
"signal/format_reward/centered_abs_mean": 0.010660807369276881,
"signal/format_reward/group_std_mean": 0.024061377346515655,
"signal/format_reward/group_zero_std_frac": 0.8861111164093017,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005330403684638441,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005330403684638441,
"signal/mean_confidence_reward/centered_abs_mean": 0.07968723326921463,
"signal/mean_confidence_reward/group_std_mean": 0.10576378107070923,
"signal/mean_confidence_reward/group_zero_std_frac": 0.17777778208255768,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.968723252815834e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.968723252815834e-07,
"step": 190
},
{
"calibration/aurc": 0.18329079440989587,
"calibration/batch_distribution_entropy": 0.41771489644191373,
"calibration/batch_entropy_100bins": 0.24362862638440336,
"calibration/batch_entropy_10bins": 0.41771489644191373,
"calibration/batch_entropy_50bins": 0.2867956771015705,
"calibration/batch_uniqueness": 0.044625812665381105,
"calibration/confidence_entropy": 0.5434150954937691,
"calibration/coverage@0%": 0.012588820332088645,
"calibration/coverage@1%": 0.012588820332088645,
"calibration/coverage@10%": 0.1366529914550833,
"calibration/coverage@15%": 0.3187713645579442,
"calibration/coverage@20%": 0.3588475991818673,
"calibration/coverage@25%": 0.8739051324242475,
"calibration/coverage@30%": 0.9413612565445026,
"calibration/coverage@5%": 0.10563694867433464,
"calibration/distribution_entropy_10": 0.41771489644191373,
"calibration/distribution_entropy_100": 0.24362862638440336,
"calibration/ece": 0.09409223753500193,
"calibration/mean_confidence": 0.6793051824351858,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.0050347222222222095,
"completions/max_length": 3301.4,
"completions/max_terminated_length": 3301.4,
"completions/mean_length": 730.9253540039062,
"completions/mean_terminated_length": 734.6098022460938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 236.8,
"epoch": 0.46875,
"grad_norm": 0.0006991397240199149,
"learning_rate": 4.6875000000000004e-06,
"loss": -0.0045,
"num_tokens": 403195064.0,
"reward": 1.2842619657516479,
"reward_std": 0.13217405527830123,
"rewards/accuracy_reward": 0.7235243201255799,
"rewards/brier_reward": 0.8509754776954651,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9940104246139526,
"rewards/mean_confidence_reward": 0.6825746297836304,
"sampling/batch_mean_priority_error": 0.036192118732255085,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.711111111111111,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005736053269356489,
"sampling/priority_kl": 0.029999953135848044,
"sampling/priority_scale": 0.8396184146171436,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 3.984217764809728e-05,
"sampling/prob_min": 1.5926705600577407e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.46320000290870667,
"sampling/prompt_draws_total": 13896.0,
"sampling/seen_fraction": 0.40477333664894105,
"sampling/unseen_fraction": 0.595226663351059,
"signal/accuracy_reward/centered_abs_mean": 0.13550889641046523,
"signal/accuracy_reward/group_std_mean": 0.17976902425289154,
"signal/accuracy_reward/group_zero_std_frac": 0.4888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06775444820523262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06775444820523262,
"signal/advantage_abs_mean": 0.0955138310790062,
"signal/advantage_pre_scale_abs_mean": 0.0955138310790062,
"signal/advantage_pre_scale_std": 0.18250684440135956,
"signal/advantage_std": 0.18250684440135956,
"signal/brier_reward/centered_abs_mean": 0.08265027552843093,
"signal/brier_reward/group_std_mean": 0.11094668358564377,
"signal/brier_reward/group_zero_std_frac": 0.16944444477558135,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04132513776421547,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04132513776421547,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008083767141215503,
"signal/confidence_one_or_zero/group_std_mean": 0.0018047165125608445,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.083766545041726e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.083766545041726e-09,
"signal/format_reward/centered_abs_mean": 0.010519748367369176,
"signal/format_reward/group_std_mean": 0.020848751068115234,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005259874183684588,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005259874183684588,
"signal/mean_confidence_reward/centered_abs_mean": 0.0648464672267437,
"signal/mean_confidence_reward/group_std_mean": 0.08931698352098465,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1972222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.48464674668503e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.48464674668503e-07,
"step": 195
},
{
"calibration/aurc": 0.2065555207393436,
"calibration/batch_distribution_entropy": 0.5505419111167639,
"calibration/batch_entropy_100bins": 0.35560265388757684,
"calibration/batch_entropy_10bins": 0.5505419111167639,
"calibration/batch_entropy_50bins": 0.41860969055372027,
"calibration/batch_uniqueness": 0.5245940876165731,
"calibration/confidence_entropy": 0.5985963082906771,
"calibration/coverage@0%": 0.0005208333333333333,
"calibration/coverage@1%": 0.0005208333333333333,
"calibration/coverage@10%": 0.3602311242766602,
"calibration/coverage@15%": 0.5076925461559658,
"calibration/coverage@20%": 0.5092550461559658,
"calibration/coverage@25%": 0.5629008794892991,
"calibration/coverage@30%": 0.5629008794892991,
"calibration/coverage@5%": 0.2745240596583081,
"calibration/distribution_entropy_10": 0.5505419111167639,
"calibration/distribution_entropy_100": 0.35560265388757684,
"calibration/ece": 0.14544184003857816,
"calibration/mean_confidence": 0.5095282963626342,
"calibration/unique_confidence_per_question": 0.022395833333333334,
"calibration/unique_confidences": 8.6,
"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": 0.005381944444444442,
"completions/max_length": 3359.2,
"completions/max_terminated_length": 3359.2,
"completions/mean_length": 743.1461059570313,
"completions/mean_terminated_length": 747.1810913085938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 216.6,
"epoch": 0.4807692307692308,
"grad_norm": 0.0005398523062467575,
"learning_rate": 4.807692307692308e-06,
"loss": -0.0049,
"num_tokens": 414880587.0,
"reward": 1.2484180927276611,
"reward_std": 0.11528781056404114,
"rewards/accuracy_reward": 0.6973090410232544,
"rewards/brier_reward": 0.8050713539123535,
"rewards/confidence_one_or_zero": 0.0006944444612599909,
"rewards/format_reward": 0.9944444417953491,
"rewards/mean_confidence_reward": 0.5597823977470398,
"sampling/batch_mean_priority_error": 0.040991808575262516,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.675,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005779322423040867,
"sampling/priority_kl": 0.029999599605798722,
"sampling/priority_scale": 0.8319383798865602,
"sampling/prob_entropy": 10.278953552246094,
"sampling/prob_max": 3.995946171926334e-05,
"sampling/prob_min": 1.610812141734641e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.47519999742507935,
"sampling/prompt_draws_total": 14256.0,
"sampling/seen_fraction": 0.41300666332244873,
"sampling/unseen_fraction": 0.5869933366775513,
"signal/accuracy_reward/centered_abs_mean": 0.13489040732383728,
"signal/accuracy_reward/group_std_mean": 0.18231622874736786,
"signal/accuracy_reward/group_zero_std_frac": 0.45833333730697634,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06744520366191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06744520366191864,
"signal/advantage_abs_mean": 0.08110309839248657,
"signal/advantage_pre_scale_abs_mean": 0.08110309839248657,
"signal/advantage_pre_scale_std": 0.1474313974380493,
"signal/advantage_std": 0.1474313974380493,
"signal/brier_reward/centered_abs_mean": 0.08825185149908066,
"signal/brier_reward/group_std_mean": 0.11570080667734146,
"signal/brier_reward/group_zero_std_frac": 0.055555556528270246,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04412592574954033,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04412592574954033,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0012695312267169355,
"signal/confidence_one_or_zero/group_std_mean": 0.00259860772639513,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.2695311824018063e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.2695311824018063e-08,
"signal/format_reward/centered_abs_mean": 0.010199652798473836,
"signal/format_reward/group_std_mean": 0.021548817306756972,
"signal/format_reward/group_zero_std_frac": 0.9055555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005099826399236918,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005099826399236918,
"signal/mean_confidence_reward/centered_abs_mean": 0.07937933951616287,
"signal/mean_confidence_reward/group_std_mean": 0.10239440947771072,
"signal/mean_confidence_reward/group_zero_std_frac": 0.058333334140479566,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.937933673929365e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.937933673929365e-07,
"step": 200
},
{
"epoch": 0.4807692307692308,
"eval_calibration/aurc": 0.16129420631495728,
"eval_calibration/batch_distribution_entropy": 0.5326029169461522,
"eval_calibration/batch_entropy_100bins": 0.316344106044224,
"eval_calibration/batch_entropy_10bins": 0.5326029169461522,
"eval_calibration/batch_entropy_50bins": 0.3723951632305077,
"eval_calibration/batch_uniqueness": 0.37980282603306575,
"eval_calibration/confidence_entropy": 0.5775948003888051,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.4847161572052402,
"eval_calibration/coverage@15%": 0.6349344978165938,
"eval_calibration/coverage@20%": 0.6349344978165938,
"eval_calibration/coverage@25%": 0.8724890829694323,
"eval_calibration/coverage@30%": 0.9973799126637555,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5326029169461522,
"eval_calibration/distribution_entropy_100": 0.316344106044224,
"eval_calibration/ece": 0.05973799126637561,
"eval_calibration/mean_confidence": 0.6391266375545851,
"eval_calibration/unique_confidence_per_question": 0.008680555555555556,
"eval_calibration/unique_confidences": 10,
"eval_completions/clipped_ratio": 0.0026041666666666665,
"eval_completions/max_length": 2023.0,
"eval_completions/max_terminated_length": 2023.0,
"eval_completions/mean_length": 741.9574788411459,
"eval_completions/mean_terminated_length": 743.9277547200521,
"eval_completions/min_length": 184.5,
"eval_completions/min_terminated_length": 290.5,
"eval_loss": 0.0,
"eval_num_tokens": 414880587.0,
"eval_reward": 1.2596179842948914,
"eval_reward_std": 0.2900431454181671,
"eval_rewards/accuracy_reward": 0.6944444477558136,
"eval_rewards/brier_reward": 0.8308550417423248,
"eval_rewards/confidence_one_or_zero": 0.0017361111628512542,
"eval_rewards/format_reward": 0.9939236144224802,
"eval_rewards/mean_confidence_reward": 0.6352430582046509,
"eval_runtime": 172.7696,
"eval_samples_per_second": 5.788,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4055989633003871,
"eval_signal/accuracy_reward/group_std_mean": 0.45562995473543805,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20279948165019354,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20279948165019354,
"eval_signal/advantage_abs_mean": 0.24742527306079865,
"eval_signal/advantage_pre_scale_abs_mean": 0.24742527306079865,
"eval_signal/advantage_pre_scale_std": 0.2890806297461192,
"eval_signal/advantage_std": 0.2890806297461192,
"eval_signal/brier_reward/centered_abs_mean": 0.13593003898859024,
"eval_signal/brier_reward/group_std_mean": 0.1747422789533933,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06796501949429512,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.06796501949429512,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0033637151742974916,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.009820927555362383,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.944444457689921,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.3637149954302004e-08,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.3637149954302004e-08,
"eval_signal/format_reward/centered_abs_mean": 0.011447482276707888,
"eval_signal/format_reward/group_std_mean": 0.02786809237052997,
"eval_signal/format_reward/group_zero_std_frac": 0.8611111243565878,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.005723741138353944,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.005723741138353944,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.16506347556908926,
"eval_signal/mean_confidence_reward/group_std_mean": 0.19205797463655472,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6506346772378795e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6506346772378795e-06,
"eval_steps_per_second": 0.035,
"step": 200
},
{
"epoch": 0.4807692307692308,
"step": 200,
"train_probe_calibration/aurc": 0.14485941592010415,
"train_probe_calibration/batch_distribution_entropy": 0.5516900974776684,
"train_probe_calibration/batch_entropy_100bins": 0.329665641102721,
"train_probe_calibration/batch_entropy_10bins": 0.5516900974776684,
"train_probe_calibration/batch_entropy_50bins": 0.3880770587607389,
"train_probe_calibration/batch_uniqueness": 0.41108156757398834,
"train_probe_calibration/confidence_entropy": 0.5772922826691077,
"train_probe_calibration/coverage@0%": 0.002617801047120419,
"train_probe_calibration/coverage@1%": 0.002617801047120419,
"train_probe_calibration/coverage@10%": 0.4755671902268761,
"train_probe_calibration/coverage@15%": 0.6492146596858639,
"train_probe_calibration/coverage@20%": 0.6492146596858639,
"train_probe_calibration/coverage@25%": 0.8673647469458988,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.002617801047120419,
"train_probe_calibration/distribution_entropy_10": 0.5516900974776684,
"train_probe_calibration/distribution_entropy_100": 0.329665641102721,
"train_probe_calibration/ece": 0.08909249563699832,
"train_probe_calibration/mean_confidence": 0.6388307155322861,
"train_probe_calibration/unique_confidence_per_question": 0.008680555555555556,
"train_probe_calibration/unique_confidences": 10,
"train_probe_completions/clipped_ratio": 0.0034722222222222285,
"train_probe_completions/max_length": 2349.0,
"train_probe_completions/max_terminated_length": 2349.0,
"train_probe_completions/mean_length": 758.0345764160156,
"train_probe_completions/mean_terminated_length": 760.6690470377604,
"train_probe_completions/min_length": 107.83333333333333,
"train_probe_completions/min_terminated_length": 233.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 414880587.0,
"train_probe_reward": 1.2768206199010212,
"train_probe_reward_std": 0.282529279589653,
"train_probe_rewards/accuracy_reward": 0.7222222288449606,
"train_probe_rewards/brier_reward": 0.836614578962326,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9947916666666666,
"train_probe_rewards/mean_confidence_reward": 0.6355034708976746,
"train_probe_runtime": 185.1737,
"train_probe_samples_per_second": 5.4,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3904079894224803,
"train_probe_signal/accuracy_reward/group_std_mean": 0.44654062887032825,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19520399471124014,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.19520399471124014,
"train_probe_signal/advantage_abs_mean": 0.23706388225158057,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23706388225158057,
"train_probe_signal/advantage_pre_scale_std": 0.2810659855604172,
"train_probe_signal/advantage_std": 0.2810659855604172,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13185519352555275,
"train_probe_signal/brier_reward/group_std_mean": 0.17271685103575388,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06592759676277637,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06592759676277637,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.009982638681928316,
"train_probe_signal/format_reward/group_std_mean": 0.026473373795549076,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8611111342906952,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.004991319340964158,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.004991319340964158,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.16340060532093048,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.19176515191793442,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6340060445448519e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6340060445448519e-06,
"train_probe_steps_per_second": 0.032
},
{
"calibration/aurc": 0.19018132797279527,
"calibration/batch_distribution_entropy": 0.44957798572890495,
"calibration/batch_entropy_100bins": 0.26184080956041245,
"calibration/batch_entropy_10bins": 0.44957798572890495,
"calibration/batch_entropy_50bins": 0.3082347644657134,
"calibration/batch_uniqueness": 0.12632964276617803,
"calibration/confidence_entropy": 0.5583439692211252,
"calibration/coverage@0%": 0.0015984030434751163,
"calibration/coverage@1%": 0.0015984030434751163,
"calibration/coverage@10%": 0.2676351484502993,
"calibration/coverage@15%": 0.2676351484502993,
"calibration/coverage@20%": 0.2949317363768085,
"calibration/coverage@25%": 0.8166204747775374,
"calibration/coverage@30%": 0.9447403590741317,
"calibration/coverage@5%": 0.2088970907075171,
"calibration/distribution_entropy_10": 0.44957798572890495,
"calibration/distribution_entropy_100": 0.26184080956041245,
"calibration/ece": 0.09613425195341234,
"calibration/mean_confidence": 0.6744759805451221,
"calibration/unique_confidence_per_question": 0.019270833333333334,
"calibration/unique_confidences": 7.4,
"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": 0.007031249999999978,
"completions/max_length": 2448.8,
"completions/max_terminated_length": 2448.8,
"completions/mean_length": 721.4182250976562,
"completions/mean_terminated_length": 726.6839233398438,
"completions/min_length": 38.6,
"completions/min_terminated_length": 213.6,
"epoch": 0.49278846153846156,
"grad_norm": 0.000698321033269167,
"learning_rate": 4.927884615384616e-06,
"loss": -0.0069,
"num_tokens": 426280477.0,
"reward": 1.2819555044174193,
"reward_std": 0.13396106362342836,
"rewards/accuracy_reward": 0.7280381798744202,
"rewards/brier_reward": 0.8430637955665589,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9927951335906983,
"rewards/mean_confidence_reward": 0.6937847256660461,
"sampling/batch_mean_priority_error": 0.03715462012595625,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6888888888888889,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005817731469869613,
"sampling/priority_kl": 0.02999994046986103,
"sampling/priority_scale": 0.8246608435874805,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 4.007791067124344e-05,
"sampling/prob_min": 1.6206031432375312e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.487200003862381,
"sampling/prompt_draws_total": 14616.0,
"sampling/seen_fraction": 0.4212199985980988,
"sampling/unseen_fraction": 0.5787800014019012,
"signal/accuracy_reward/centered_abs_mean": 0.13053928017616273,
"signal/accuracy_reward/group_std_mean": 0.17609834372997285,
"signal/accuracy_reward/group_zero_std_frac": 0.4833333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06526964008808137,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06526964008808137,
"signal/advantage_abs_mean": 0.09769376963377,
"signal/advantage_pre_scale_abs_mean": 0.09769376963377,
"signal/advantage_pre_scale_std": 0.18367029428482057,
"signal/advantage_std": 0.18367029428482057,
"signal/brier_reward/centered_abs_mean": 0.07794546037912368,
"signal/brier_reward/group_std_mean": 0.10496081113815307,
"signal/brier_reward/group_zero_std_frac": 0.23055556267499924,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03897273018956184,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03897273018956184,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.01070421002805233,
"signal/format_reward/group_std_mean": 0.02027086652815342,
"signal/format_reward/group_zero_std_frac": 0.9166666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005352105014026165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005352105014026165,
"signal/mean_confidence_reward/centered_abs_mean": 0.054693204909563066,
"signal/mean_confidence_reward/group_std_mean": 0.07309356331825256,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2972222313284874,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.469320342399442e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.469320342399442e-07,
"step": 205
},
{
"calibration/aurc": 0.20358953124556783,
"calibration/batch_distribution_entropy": 0.30459638634612196,
"calibration/batch_entropy_100bins": 0.16357802323120227,
"calibration/batch_entropy_10bins": 0.30459638634612196,
"calibration/batch_entropy_50bins": 0.19256140227752971,
"calibration/batch_uniqueness": -0.3185311723570564,
"calibration/confidence_entropy": 0.5280267888744263,
"calibration/coverage@0%": 0.008444148936170213,
"calibration/coverage@1%": 0.008444148936170213,
"calibration/coverage@10%": 0.17302748226950354,
"calibration/coverage@15%": 0.19698581560283687,
"calibration/coverage@20%": 0.7025530522446252,
"calibration/coverage@25%": 0.754122340425532,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.008444148936170213,
"calibration/distribution_entropy_10": 0.30459638634612196,
"calibration/distribution_entropy_100": 0.16357802323120227,
"calibration/ece": 0.08418190912281812,
"calibration/mean_confidence": 0.7395930984643754,
"calibration/unique_confidence_per_question": 0.01875,
"calibration/unique_confidences": 7.2,
"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": 0.004513888888888906,
"completions/max_length": 3212.2,
"completions/max_terminated_length": 3212.2,
"completions/mean_length": 779.1319580078125,
"completions/mean_terminated_length": 782.6882568359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 250.2,
"epoch": 0.5048076923076923,
"grad_norm": 0.0009137173765338957,
"learning_rate": 4.987980769230769e-06,
"loss": -0.0034,
"num_tokens": 438366413.0,
"reward": 1.2391991853713988,
"reward_std": 0.1426423504948616,
"rewards/accuracy_reward": 0.6729166626930236,
"rewards/brier_reward": 0.8111093878746033,
"rewards/confidence_one_or_zero": 0.0004340277868323028,
"rewards/format_reward": 0.9943576455116272,
"rewards/mean_confidence_reward": 0.732387113571167,
"sampling/batch_mean_priority_error": 0.05414736051964284,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6833333333333333,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005871547013521194,
"sampling/priority_kl": 0.030000075325369836,
"sampling/priority_scale": 0.8178263127570972,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 4.019822226837277e-05,
"sampling/prob_min": 1.6377108840970324e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.49920000433921813,
"sampling/prompt_draws_total": 14976.0,
"sampling/seen_fraction": 0.42946000695228576,
"sampling/unseen_fraction": 0.5705399930477142,
"signal/accuracy_reward/centered_abs_mean": 0.12706163227558137,
"signal/accuracy_reward/group_std_mean": 0.17422983646392823,
"signal/accuracy_reward/group_zero_std_frac": 0.48055556416511536,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06353081613779069,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06353081613779069,
"signal/advantage_abs_mean": 0.10214050263166427,
"signal/advantage_pre_scale_abs_mean": 0.10214050263166427,
"signal/advantage_pre_scale_std": 0.19253622591495514,
"signal/advantage_std": 0.19253622591495514,
"signal/brier_reward/centered_abs_mean": 0.08288372755050659,
"signal/brier_reward/group_std_mean": 0.11273103803396226,
"signal/brier_reward/group_zero_std_frac": 0.29444445073604586,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.041441863775253295,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.041441863775253295,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008192274020984769,
"signal/confidence_one_or_zero/group_std_mean": 0.0018573501612991095,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9916666626930237,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.192274236762388e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.192274236762388e-09,
"signal/format_reward/centered_abs_mean": 0.01031358502805233,
"signal/format_reward/group_std_mean": 0.02181966695934534,
"signal/format_reward/group_zero_std_frac": 0.9027777791023255,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005156792514026165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005156792514026165,
"signal/mean_confidence_reward/centered_abs_mean": 0.050381416082382204,
"signal/mean_confidence_reward/group_std_mean": 0.06970751732587814,
"signal/mean_confidence_reward/group_zero_std_frac": 0.39166667461395266,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.038141409841046e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.038141409841046e-07,
"step": 210
},
{
"calibration/aurc": 0.19383833760635832,
"calibration/batch_distribution_entropy": 0.319087947530728,
"calibration/batch_entropy_100bins": 0.17407298528855475,
"calibration/batch_entropy_10bins": 0.319087947530728,
"calibration/batch_entropy_50bins": 0.20491590180438157,
"calibration/batch_uniqueness": -0.2964349564207656,
"calibration/confidence_entropy": 0.5218764549310007,
"calibration/coverage@0%": 0.012703504052971526,
"calibration/coverage@1%": 0.012703504052971526,
"calibration/coverage@10%": 0.21165638363412337,
"calibration/coverage@15%": 0.3772813836341234,
"calibration/coverage@20%": 0.5916749338624339,
"calibration/coverage@25%": 0.7656332671957673,
"calibration/coverage@30%": 0.8005291005291006,
"calibration/coverage@5%": 0.18966685483831186,
"calibration/distribution_entropy_10": 0.319087947530728,
"calibration/distribution_entropy_100": 0.17407298528855475,
"calibration/ece": 0.1256729673216957,
"calibration/mean_confidence": 0.7253777652148183,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.007204861111111094,
"completions/max_length": 2856.6,
"completions/max_terminated_length": 2856.6,
"completions/mean_length": 772.7975830078125,
"completions/mean_terminated_length": 778.4013061523438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 241.8,
"epoch": 0.5168269230769231,
"grad_norm": 0.0008226765785366297,
"learning_rate": 4.957932692307692e-06,
"loss": -0.0055,
"num_tokens": 450353393.0,
"reward": 1.269078254699707,
"reward_std": 0.1451116070151329,
"rewards/accuracy_reward": 0.7244791626930237,
"rewards/brier_reward": 0.8210409998893737,
"rewards/confidence_one_or_zero": 0.0004340277810115367,
"rewards/format_reward": 0.9926215291023255,
"rewards/mean_confidence_reward": 0.7352039813995361,
"sampling/batch_mean_priority_error": 0.052838198647012245,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.7,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0059312603436410425,
"sampling/priority_kl": 0.029999976605176927,
"sampling/priority_scale": 0.8115076482063159,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 4.0320714469999074e-05,
"sampling/prob_min": 1.6541368677280843e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5111999988555909,
"sampling/prompt_draws_total": 15336.0,
"sampling/seen_fraction": 0.4377600014209747,
"sampling/unseen_fraction": 0.5622399985790253,
"signal/accuracy_reward/centered_abs_mean": 0.13215061128139496,
"signal/accuracy_reward/group_std_mean": 0.180630099773407,
"signal/accuracy_reward/group_zero_std_frac": 0.4611111104488373,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06607530564069748,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06607530564069748,
"signal/advantage_abs_mean": 0.10373011976480484,
"signal/advantage_pre_scale_abs_mean": 0.10373011976480484,
"signal/advantage_pre_scale_std": 0.19957785606384276,
"signal/advantage_std": 0.19957785606384276,
"signal/brier_reward/centered_abs_mean": 0.08286137729883195,
"signal/brier_reward/group_std_mean": 0.11241399496793747,
"signal/brier_reward/group_zero_std_frac": 0.3277777791023254,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04143068864941597,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04143068864941597,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0008300781133584678,
"signal/confidence_one_or_zero/group_std_mean": 0.0021562909241765738,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 8.300780862668944e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 8.300780862668944e-09,
"signal/format_reward/centered_abs_mean": 0.01175672747194767,
"signal/format_reward/group_std_mean": 0.024286934360861778,
"signal/format_reward/group_zero_std_frac": 0.8888888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005878363735973835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005878363735973835,
"signal/mean_confidence_reward/centered_abs_mean": 0.04709501713514328,
"signal/mean_confidence_reward/group_std_mean": 0.06495100557804108,
"signal/mean_confidence_reward/group_zero_std_frac": 0.45277778506278993,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.709501695288054e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.709501695288054e-07,
"step": 215
},
{
"calibration/aurc": 0.23251423646772898,
"calibration/batch_distribution_entropy": 0.3889181188463192,
"calibration/batch_entropy_100bins": 0.2171264230417509,
"calibration/batch_entropy_10bins": 0.3889181188463192,
"calibration/batch_entropy_50bins": 0.2555977121286575,
"calibration/batch_uniqueness": -0.07294295671342735,
"calibration/confidence_entropy": 0.539465534455553,
"calibration/coverage@0%": 0.004742396889320927,
"calibration/coverage@1%": 0.004742396889320927,
"calibration/coverage@10%": 0.12825514476467503,
"calibration/coverage@15%": 0.44024514615337107,
"calibration/coverage@20%": 0.49824820128460284,
"calibration/coverage@25%": 0.5766150672337371,
"calibration/coverage@30%": 0.5766150672337371,
"calibration/coverage@5%": 0.004742396889320927,
"calibration/distribution_entropy_10": 0.3889181188463192,
"calibration/distribution_entropy_100": 0.2171264230417509,
"calibration/ece": 0.11262302634465074,
"calibration/mean_confidence": 0.6882683367362527,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.00659722222222221,
"completions/max_length": 3048.4,
"completions/max_terminated_length": 3048.4,
"completions/mean_length": 868.336474609375,
"completions/mean_terminated_length": 874.126708984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 208.6,
"epoch": 0.5288461538461539,
"grad_norm": 0.0008023114060051739,
"learning_rate": 4.927884615384616e-06,
"loss": -0.0069,
"num_tokens": 463435221.0,
"reward": 1.2563817501068115,
"reward_std": 0.14041633605957032,
"rewards/accuracy_reward": 0.6907118082046508,
"rewards/brier_reward": 0.8314127564430237,
"rewards/confidence_one_or_zero": 0.0013020833546761423,
"rewards/format_reward": 0.9906249880790711,
"rewards/mean_confidence_reward": 0.6941753387451172,
"sampling/batch_mean_priority_error": 0.04295309574473831,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6777777777777778,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.005983918812125922,
"sampling/priority_kl": 0.02999996654689312,
"sampling/priority_scale": 0.8054787575965747,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 4.044408997287974e-05,
"sampling/prob_min": 1.670217352511827e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.523199999332428,
"sampling/prompt_draws_total": 15696.0,
"sampling/seen_fraction": 0.44600666165351865,
"sampling/unseen_fraction": 0.5539933383464813,
"signal/accuracy_reward/centered_abs_mean": 0.1377658426761627,
"signal/accuracy_reward/group_std_mean": 0.180549818277359,
"signal/accuracy_reward/group_zero_std_frac": 0.4916666805744171,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06888292133808135,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06888292133808135,
"signal/advantage_abs_mean": 0.10294992327690125,
"signal/advantage_pre_scale_abs_mean": 0.10294992327690125,
"signal/advantage_pre_scale_std": 0.19110060930252076,
"signal/advantage_std": 0.19110060930252076,
"signal/brier_reward/centered_abs_mean": 0.08256558924913407,
"signal/brier_reward/group_std_mean": 0.11115524917840958,
"signal/brier_reward/group_zero_std_frac": 0.30833334028720855,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.041282794624567035,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.041282794624567035,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0024034288129769267,
"signal/confidence_one_or_zero/group_std_mean": 0.005086476355791092,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9777777791023254,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.4034287449126168e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.4034287449126168e-08,
"signal/format_reward/centered_abs_mean": 0.01320529505610466,
"signal/format_reward/group_std_mean": 0.02615750953555107,
"signal/format_reward/group_zero_std_frac": 0.8916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00660264752805233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00660264752805233,
"signal/mean_confidence_reward/centered_abs_mean": 0.06290501430630684,
"signal/mean_confidence_reward/group_std_mean": 0.08351254016160965,
"signal/mean_confidence_reward/group_zero_std_frac": 0.36666667461395264,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.290501346484234e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.290501346484234e-07,
"step": 220
},
{
"calibration/aurc": 0.12652826645345755,
"calibration/batch_distribution_entropy": 0.46529389381214703,
"calibration/batch_entropy_100bins": 0.26020387344041745,
"calibration/batch_entropy_10bins": 0.46529389381214703,
"calibration/batch_entropy_50bins": 0.30630778975065986,
"calibration/batch_uniqueness": 0.23280162375100857,
"calibration/confidence_entropy": 0.5795074087022364,
"calibration/coverage@0%": 0.09057318935427575,
"calibration/coverage@1%": 0.09580879144851659,
"calibration/coverage@10%": 0.6117064790575916,
"calibration/coverage@15%": 0.8069298038016214,
"calibration/coverage@20%": 0.8168774477806788,
"calibration/coverage@25%": 0.8611482811140121,
"calibration/coverage@30%": 0.8611482811140121,
"calibration/coverage@5%": 0.23067133980606855,
"calibration/distribution_entropy_10": 0.46529389381214703,
"calibration/distribution_entropy_100": 0.26020387344041745,
"calibration/ece": 0.19566735272875574,
"calibration/mean_confidence": 0.6187747341189015,
"calibration/unique_confidence_per_question": 0.01979166666666667,
"calibration/unique_confidences": 7.6,
"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": 0.003993055555555558,
"completions/max_length": 3369.0,
"completions/max_terminated_length": 3369.0,
"completions/mean_length": 886.9040893554687,
"completions/mean_terminated_length": 890.4674682617188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 221.8,
"epoch": 0.5408653846153846,
"grad_norm": 0.0006064812187105417,
"learning_rate": 4.897836538461539e-06,
"loss": -0.0029,
"num_tokens": 476719620.0,
"reward": 1.2701764583587647,
"reward_std": 0.11144869178533554,
"rewards/accuracy_reward": 0.720225703716278,
"rewards/brier_reward": 0.8250629425048828,
"rewards/confidence_one_or_zero": 0.0017361111240461469,
"rewards/format_reward": 0.9950520753860473,
"rewards/mean_confidence_reward": 0.604869794845581,
"sampling/batch_mean_priority_error": 0.036432831587038625,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6083333333333334,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0060226328670978544,
"sampling/priority_kl": 0.029999875649809837,
"sampling/priority_scale": 0.799505728459917,
"sampling/prob_entropy": 10.278953742980956,
"sampling/prob_max": 4.055845565744676e-05,
"sampling/prob_min": 1.685902716417331e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5351999998092651,
"sampling/prompt_draws_total": 16056.0,
"sampling/seen_fraction": 0.45362000465393065,
"sampling/unseen_fraction": 0.5463799953460693,
"signal/accuracy_reward/centered_abs_mean": 0.12177191823720931,
"signal/accuracy_reward/group_std_mean": 0.16531633734703063,
"signal/accuracy_reward/group_zero_std_frac": 0.5138888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06088595911860466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06088595911860466,
"signal/advantage_abs_mean": 0.07888294160366058,
"signal/advantage_pre_scale_abs_mean": 0.07888294160366058,
"signal/advantage_pre_scale_std": 0.1528717279434204,
"signal/advantage_std": 0.1528717279434204,
"signal/brier_reward/centered_abs_mean": 0.07004087418317795,
"signal/brier_reward/group_std_mean": 0.09422325640916825,
"signal/brier_reward/group_zero_std_frac": 0.1722222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035020437091588974,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035020437091588974,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002734375,
"signal/confidence_one_or_zero/group_std_mean": 0.004100846825167537,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.7343747888153302e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.7343747888153302e-08,
"signal/format_reward/centered_abs_mean": 0.008696831669658423,
"signal/format_reward/group_std_mean": 0.018851579166948795,
"signal/format_reward/group_zero_std_frac": 0.9111111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004348415834829211,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004348415834829211,
"signal/mean_confidence_reward/centered_abs_mean": 0.06936849504709244,
"signal/mean_confidence_reward/group_std_mean": 0.09045136123895645,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18055555522441863,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.936849104022258e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.936849104022258e-07,
"step": 225
},
{
"calibration/aurc": 0.25964683947564937,
"calibration/batch_distribution_entropy": 0.4298122178001658,
"calibration/batch_entropy_100bins": 0.23594515591118478,
"calibration/batch_entropy_10bins": 0.4298122178001658,
"calibration/batch_entropy_50bins": 0.27775081997800827,
"calibration/batch_uniqueness": 0.05135679139027812,
"calibration/confidence_entropy": 0.5489465221626701,
"calibration/coverage@0%": 0.15678039531122578,
"calibration/coverage@1%": 0.15678039531122578,
"calibration/coverage@10%": 0.18750956197789245,
"calibration/coverage@15%": 0.2015720619778924,
"calibration/coverage@20%": 0.4533445721373601,
"calibration/coverage@25%": 0.6474226561268613,
"calibration/coverage@30%": 0.6944200451608039,
"calibration/coverage@5%": 0.15678039531122578,
"calibration/distribution_entropy_10": 0.4298122178001658,
"calibration/distribution_entropy_100": 0.23594515591118478,
"calibration/ece": 0.13530504305269925,
"calibration/mean_confidence": 0.6636817097652362,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.005468749999999978,
"completions/max_length": 2951.6,
"completions/max_terminated_length": 2951.6,
"completions/mean_length": 809.384814453125,
"completions/mean_terminated_length": 813.767333984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 226.0,
"epoch": 0.5528846153846154,
"grad_norm": 0.0007383719203062356,
"learning_rate": 4.867788461538462e-06,
"loss": -0.0054,
"num_tokens": 489174581.0,
"reward": 1.2462389707565307,
"reward_std": 0.13540728837251664,
"rewards/accuracy_reward": 0.6755208253860474,
"rewards/brier_reward": 0.8228465676307678,
"rewards/confidence_one_or_zero": 0.0008680555794853718,
"rewards/format_reward": 0.9940972208976746,
"rewards/mean_confidence_reward": 0.6592578172683716,
"sampling/batch_mean_priority_error": 0.045255449784256524,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6194444444444444,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.006066837068647146,
"sampling/priority_kl": 0.03000005558133125,
"sampling/priority_scale": 0.793832951807417,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 4.067181362188421e-05,
"sampling/prob_min": 1.701090259302873e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5472000002861023,
"sampling/prompt_draws_total": 16416.0,
"sampling/seen_fraction": 0.46108666658401487,
"sampling/unseen_fraction": 0.5389133334159851,
"signal/accuracy_reward/centered_abs_mean": 0.1392578139901161,
"signal/accuracy_reward/group_std_mean": 0.1852429300546646,
"signal/accuracy_reward/group_zero_std_frac": 0.4666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06962890699505805,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06962890699505805,
"signal/advantage_abs_mean": 0.09845461696386337,
"signal/advantage_pre_scale_abs_mean": 0.09845461696386337,
"signal/advantage_pre_scale_std": 0.18213818967342377,
"signal/advantage_std": 0.18213818967342377,
"signal/brier_reward/centered_abs_mean": 0.08291573077440262,
"signal/brier_reward/group_std_mean": 0.11135628968477249,
"signal/brier_reward/group_zero_std_frac": 0.23888889253139495,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04145786538720131,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04145786538720131,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0016276041395030915,
"signal/confidence_one_or_zero/group_std_mean": 0.003662066720426083,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9833333253860473,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6276040426532744e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6276040426532744e-08,
"signal/format_reward/centered_abs_mean": 0.010481770988553763,
"signal/format_reward/group_std_mean": 0.022597668692469597,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005240885494276881,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005240885494276881,
"signal/mean_confidence_reward/centered_abs_mean": 0.06652046516537666,
"signal/mean_confidence_reward/group_std_mean": 0.08856887370347977,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2944444417953491,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.652046408817114e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.652046408817114e-07,
"step": 230
},
{
"calibration/aurc": 0.21916975459389473,
"calibration/batch_distribution_entropy": 0.36100771717627633,
"calibration/batch_entropy_100bins": 0.19316641280299807,
"calibration/batch_entropy_10bins": 0.36100771717627633,
"calibration/batch_entropy_50bins": 0.2273923757453154,
"calibration/batch_uniqueness": -0.18844872173518384,
"calibration/confidence_entropy": 0.508086913449876,
"calibration/coverage@0%": 0.013162141012056872,
"calibration/coverage@1%": 0.013162141012056872,
"calibration/coverage@10%": 0.04981135567174273,
"calibration/coverage@15%": 0.2021673766141511,
"calibration/coverage@20%": 0.2131621410120569,
"calibration/coverage@25%": 0.7295691211841004,
"calibration/coverage@30%": 0.9237121188512386,
"calibration/coverage@5%": 0.04981135567174273,
"calibration/distribution_entropy_10": 0.36100771717627633,
"calibration/distribution_entropy_100": 0.19316641280299807,
"calibration/ece": 0.06566668041868481,
"calibration/mean_confidence": 0.7343678993951881,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.004427083333333348,
"completions/max_length": 3289.6,
"completions/max_terminated_length": 3289.6,
"completions/mean_length": 761.0531372070312,
"completions/mean_terminated_length": 764.4302368164062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 180.8,
"epoch": 0.5649038461538461,
"grad_norm": 0.0006777613307349384,
"learning_rate": 4.837740384615385e-06,
"loss": -0.0046,
"num_tokens": 501045081.0,
"reward": 1.2459373474121094,
"reward_std": 0.15442993640899658,
"rewards/accuracy_reward": 0.6864583373069764,
"rewards/brier_reward": 0.8100889563560486,
"rewards/confidence_one_or_zero": 0.0006076389050576836,
"rewards/format_reward": 0.9953125,
"rewards/mean_confidence_reward": 0.7428732514381409,
"sampling/batch_mean_priority_error": 0.06387436928556019,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6555555555555556,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.0061315404251217846,
"sampling/priority_kl": 0.029999935254454612,
"sampling/priority_scale": 0.7885626494651661,
"sampling/prob_entropy": 10.278954887390137,
"sampling/prob_max": 4.079045102116652e-05,
"sampling/prob_min": 1.7144251614809036e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5592000126838684,
"sampling/prompt_draws_total": 16776.0,
"sampling/seen_fraction": 0.4687666654586792,
"sampling/unseen_fraction": 0.5312333345413208,
"signal/accuracy_reward/centered_abs_mean": 0.1421875014901161,
"signal/accuracy_reward/group_std_mean": 0.1907571256160736,
"signal/accuracy_reward/group_zero_std_frac": 0.4444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07109375074505805,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07109375074505805,
"signal/advantage_abs_mean": 0.1121572047472,
"signal/advantage_pre_scale_abs_mean": 0.1121572047472,
"signal/advantage_pre_scale_std": 0.2046825647354126,
"signal/advantage_std": 0.2046825647354126,
"signal/brier_reward/centered_abs_mean": 0.09700408428907395,
"signal/brier_reward/group_std_mean": 0.1282535523176193,
"signal/brier_reward/group_zero_std_frac": 0.11111111119389534,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.048502042144536975,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.048502042144536975,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0010145399603061378,
"signal/confidence_one_or_zero/group_std_mean": 0.0015925956889986993,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9944444417953491,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.01453991874223e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.01453991874223e-08,
"signal/format_reward/centered_abs_mean": 0.008778211940079928,
"signal/format_reward/group_std_mean": 0.021230710670351982,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004389105970039964,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004389105970039964,
"signal/mean_confidence_reward/centered_abs_mean": 0.062408319860696795,
"signal/mean_confidence_reward/group_std_mean": 0.08640324026346206,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18055555671453477,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.240831908144173e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.240831908144173e-07,
"step": 235
},
{
"calibration/aurc": 0.1801804727272827,
"calibration/batch_distribution_entropy": 0.44021960039486946,
"calibration/batch_entropy_100bins": 0.23149426160615266,
"calibration/batch_entropy_10bins": 0.44021960039486946,
"calibration/batch_entropy_50bins": 0.27251129921699097,
"calibration/batch_uniqueness": 0.07269222939658837,
"calibration/confidence_entropy": 0.4649029441679257,
"calibration/coverage@0%": 0.01994750656167979,
"calibration/coverage@1%": 0.01994750656167979,
"calibration/coverage@10%": 0.31052382105128745,
"calibration/coverage@15%": 0.6252309032967183,
"calibration/coverage@20%": 0.6445466126091638,
"calibration/coverage@25%": 0.7958834692047656,
"calibration/coverage@30%": 0.8183982329842932,
"calibration/coverage@5%": 0.01994750656167979,
"calibration/distribution_entropy_10": 0.44021960039486946,
"calibration/distribution_entropy_100": 0.23149426160615266,
"calibration/ece": 0.11495716548071525,
"calibration/mean_confidence": 0.7526511497869869,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.00477430555555558,
"completions/max_length": 3072.4,
"completions/max_terminated_length": 3072.4,
"completions/mean_length": 743.3960205078125,
"completions/mean_terminated_length": 746.9897094726563,
"completions/min_length": 0.0,
"completions/min_terminated_length": 147.6,
"epoch": 0.5769230769230769,
"grad_norm": 0.0007029211265034974,
"learning_rate": 4.807692307692308e-06,
"loss": -0.004,
"num_tokens": 512686891.0,
"reward": 1.2595340251922607,
"reward_std": 0.14976598024368287,
"rewards/accuracy_reward": 0.6979166626930237,
"rewards/brier_reward": 0.8260842323303222,
"rewards/confidence_one_or_zero": 0.0007812500058207661,
"rewards/format_reward": 0.9950520873069764,
"rewards/mean_confidence_reward": 0.7551562547683716,
"sampling/batch_mean_priority_error": 0.06334802356864544,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6611111111111111,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.006200295034795999,
"sampling/priority_kl": 0.02999994494020939,
"sampling/priority_scale": 0.7836340844398364,
"sampling/prob_entropy": 10.278955841064453,
"sampling/prob_max": 4.0912131953518835e-05,
"sampling/prob_min": 1.7265845599467866e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5711999893188476,
"sampling/prompt_draws_total": 17136.0,
"sampling/seen_fraction": 0.476526665687561,
"sampling/unseen_fraction": 0.5234733343124389,
"signal/accuracy_reward/centered_abs_mean": 0.12988281548023223,
"signal/accuracy_reward/group_std_mean": 0.17232809364795684,
"signal/accuracy_reward/group_zero_std_frac": 0.5055555522441864,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06494140774011611,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06494140774011611,
"signal/advantage_abs_mean": 0.11209896206855774,
"signal/advantage_pre_scale_abs_mean": 0.11209896206855774,
"signal/advantage_pre_scale_std": 0.20345027148723602,
"signal/advantage_std": 0.20345027148723602,
"signal/brier_reward/centered_abs_mean": 0.10559439659118652,
"signal/brier_reward/group_std_mean": 0.13607816100120546,
"signal/brier_reward/group_zero_std_frac": 0.05000000149011612,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05279719829559326,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05279719829559326,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0014160156133584678,
"signal/confidence_one_or_zero/group_std_mean": 0.0027380796149373055,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9888888835906983,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.4160154648834578e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.4160154648834578e-08,
"signal/format_reward/centered_abs_mean": 0.00910915806889534,
"signal/format_reward/group_std_mean": 0.020607628300786017,
"signal/format_reward/group_zero_std_frac": 0.9027777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00455457903444767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00455457903444767,
"signal/mean_confidence_reward/centered_abs_mean": 0.07177354246377946,
"signal/mean_confidence_reward/group_std_mean": 0.09573450088500976,
"signal/mean_confidence_reward/group_zero_std_frac": 0.07500000149011612,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.177354063969688e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.177354063969688e-07,
"step": 240
},
{
"calibration/aurc": 0.11709602782683955,
"calibration/batch_distribution_entropy": 0.5672294931403771,
"calibration/batch_entropy_100bins": 0.3116243412218339,
"calibration/batch_entropy_10bins": 0.5672294931403771,
"calibration/batch_entropy_50bins": 0.36683913244674493,
"calibration/batch_uniqueness": 0.28723508706075185,
"calibration/confidence_entropy": 0.42654062776644575,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.6111260214833772,
"calibration/coverage@15%": 0.7574644152004357,
"calibration/coverage@20%": 0.8191817958759573,
"calibration/coverage@25%": 0.8933654725177265,
"calibration/coverage@30%": 0.9848563968668408,
"calibration/coverage@5%": 0.08952879581151832,
"calibration/distribution_entropy_10": 0.5672294931403771,
"calibration/distribution_entropy_100": 0.3116243412218339,
"calibration/ece": 0.058913449990573484,
"calibration/mean_confidence": 0.7108805560567061,
"calibration/unique_confidence_per_question": 0.025520833333333333,
"calibration/unique_confidences": 9.8,
"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": 0.004513888888888884,
"completions/max_length": 2893.0,
"completions/max_terminated_length": 2893.0,
"completions/mean_length": 759.0476684570312,
"completions/mean_terminated_length": 762.4732666015625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 154.4,
"epoch": 0.5889423076923077,
"grad_norm": 0.0005396166816353798,
"learning_rate": 4.777644230769231e-06,
"loss": -0.0049,
"num_tokens": 524523792.0,
"reward": 1.2858156204223632,
"reward_std": 0.13980449438095094,
"rewards/accuracy_reward": 0.7180555462837219,
"rewards/brier_reward": 0.8593769550323487,
"rewards/confidence_one_or_zero": 0.0010416666802484543,
"rewards/format_reward": 0.9941840291023254,
"rewards/mean_confidence_reward": 0.7379383563995361,
"sampling/batch_mean_priority_error": 0.0541321167838522,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6694444444444445,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.006265456974506378,
"sampling/priority_kl": 0.03000013791024685,
"sampling/priority_scale": 0.7791448295814917,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 4.104044201085344e-05,
"sampling/prob_min": 1.7405801918357612e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5832000017166138,
"sampling/prompt_draws_total": 17496.0,
"sampling/seen_fraction": 0.48456000089645385,
"sampling/unseen_fraction": 0.5154399991035461,
"signal/accuracy_reward/centered_abs_mean": 0.12662760615348817,
"signal/accuracy_reward/group_std_mean": 0.1733509123325348,
"signal/accuracy_reward/group_zero_std_frac": 0.47777777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06331380307674409,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06331380307674409,
"signal/advantage_abs_mean": 0.09974031746387482,
"signal/advantage_pre_scale_abs_mean": 0.09974031746387482,
"signal/advantage_pre_scale_std": 0.18898828327655792,
"signal/advantage_std": 0.18898828327655792,
"signal/brier_reward/centered_abs_mean": 0.0990310087800026,
"signal/brier_reward/group_std_mean": 0.13221175074577332,
"signal/brier_reward/group_zero_std_frac": 0.20277777761220933,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0495155043900013,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0495155043900013,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0019531249883584679,
"signal/confidence_one_or_zero/group_std_mean": 0.004861733969300985,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9749999880790711,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.953124808551365e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.953124808551365e-08,
"signal/format_reward/centered_abs_mean": 0.010606553871184587,
"signal/format_reward/group_std_mean": 0.02274163216352463,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005303276935592294,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005303276935592294,
"signal/mean_confidence_reward/centered_abs_mean": 0.08483045399188996,
"signal/mean_confidence_reward/group_std_mean": 0.11241712123155594,
"signal/mean_confidence_reward/group_zero_std_frac": 0.21388888657093047,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.48304478040518e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.48304478040518e-07,
"step": 245
},
{
"calibration/aurc": 0.1636056684986535,
"calibration/batch_distribution_entropy": 0.6186128158819945,
"calibration/batch_entropy_100bins": 0.3487107609376868,
"calibration/batch_entropy_10bins": 0.6186128158819945,
"calibration/batch_entropy_50bins": 0.4104966656830034,
"calibration/batch_uniqueness": 0.429661900750044,
"calibration/confidence_entropy": 0.45239855037732674,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.175,
"calibration/coverage@10%": 0.40646542321116935,
"calibration/coverage@15%": 0.47216677574171023,
"calibration/coverage@20%": 0.5295729712041884,
"calibration/coverage@25%": 0.73539212478185,
"calibration/coverage@30%": 0.8873245757180157,
"calibration/coverage@5%": 0.19947916666666668,
"calibration/distribution_entropy_10": 0.6186128158819945,
"calibration/distribution_entropy_100": 0.3487107609376868,
"calibration/ece": 0.12899922009919845,
"calibration/mean_confidence": 0.6672979726566239,
"calibration/unique_confidence_per_question": 0.025,
"calibration/unique_confidences": 9.6,
"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": 0.004513888888888862,
"completions/max_length": 3556.0,
"completions/max_terminated_length": 3556.0,
"completions/mean_length": 787.1635498046875,
"completions/mean_terminated_length": 790.782568359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 153.4,
"epoch": 0.6009615384615384,
"grad_norm": 0.0005077553796581924,
"learning_rate": 4.747596153846154e-06,
"loss": -0.0035,
"num_tokens": 536723724.0,
"reward": 1.258297610282898,
"reward_std": 0.13198113888502122,
"rewards/accuracy_reward": 0.6806423544883728,
"rewards/brier_reward": 0.8412350416183472,
"rewards/confidence_one_or_zero": 0.0015625000465661286,
"rewards/format_reward": 0.9947048664093018,
"rewards/mean_confidence_reward": 0.6485720157623291,
"sampling/batch_mean_priority_error": 0.05104182940510933,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6361111111111112,
"sampling/error_ema_max": 0.12885397672653198,
"sampling/error_ema_mean": 0.006326197646558285,
"sampling/priority_kl": 0.02999974898993969,
"sampling/priority_scale": 0.774927383591421,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 4.1167728340951726e-05,
"sampling/prob_min": 1.7496926739113404e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.5952000021934509,
"sampling/prompt_draws_total": 17856.0,
"sampling/seen_fraction": 0.492440003156662,
"sampling/unseen_fraction": 0.507559996843338,
"signal/accuracy_reward/centered_abs_mean": 0.1427029088139534,
"signal/accuracy_reward/group_std_mean": 0.19304186701774598,
"signal/accuracy_reward/group_zero_std_frac": 0.4277777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0713514544069767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0713514544069767,
"signal/advantage_abs_mean": 0.09508290439844132,
"signal/advantage_pre_scale_abs_mean": 0.09508290439844132,
"signal/advantage_pre_scale_std": 0.17090495228767394,
"signal/advantage_std": 0.17090495228767394,
"signal/brier_reward/centered_abs_mean": 0.10632044672966004,
"signal/brier_reward/group_std_mean": 0.1394522801041603,
"signal/brier_reward/group_zero_std_frac": 0.11388889029622078,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.05316022336483002,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.05316022336483002,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002875434048473835,
"signal/confidence_one_or_zero/group_std_mean": 0.006179308518767357,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.87543400645518e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.87543400645518e-08,
"signal/format_reward/centered_abs_mean": 0.008837890625,
"signal/format_reward/group_std_mean": 0.016633689403533936,
"signal/format_reward/group_zero_std_frac": 0.9305555820465088,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0044189453125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0044189453125,
"signal/mean_confidence_reward/centered_abs_mean": 0.09797850847244263,
"signal/mean_confidence_reward/group_std_mean": 0.12640962302684783,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1305555559694767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.797850339055004e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.797850339055004e-07,
"step": 250
},
{
"epoch": 0.6009615384615384,
"eval_calibration/aurc": 0.14130535061583444,
"eval_calibration/batch_distribution_entropy": 0.6530374121770919,
"eval_calibration/batch_entropy_100bins": 0.36960797941874046,
"eval_calibration/batch_entropy_10bins": 0.6530374121770919,
"eval_calibration/batch_entropy_50bins": 0.4350965331647375,
"eval_calibration/batch_uniqueness": 0.47262016049727446,
"eval_calibration/confidence_entropy": 0.45554960791222726,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.4520905923344948,
"eval_calibration/coverage@15%": 0.6890243902439024,
"eval_calibration/coverage@20%": 0.7813588850174216,
"eval_calibration/coverage@25%": 0.9059233449477352,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6530374121770919,
"eval_calibration/distribution_entropy_100": 0.36960797941874046,
"eval_calibration/ece": 0.03135888501742151,
"eval_calibration/mean_confidence": 0.6818815331010453,
"eval_calibration/unique_confidence_per_question": 0.009548611111111112,
"eval_calibration/unique_confidences": 11,
"eval_completions/clipped_ratio": 0.0034722222222222467,
"eval_completions/max_length": 2078.3333333333335,
"eval_completions/max_terminated_length": 2078.3333333333335,
"eval_completions/mean_length": 772.7741495768229,
"eval_completions/mean_terminated_length": 775.481201171875,
"eval_completions/min_length": 95.66666666666667,
"eval_completions/min_terminated_length": 241.0,
"eval_loss": 0.0,
"eval_num_tokens": 536723724.0,
"eval_reward": 1.2739282846450806,
"eval_reward_std": 0.29537801444530487,
"eval_rewards/accuracy_reward": 0.7022569477558136,
"eval_rewards/brier_reward": 0.8490581413110098,
"eval_rewards/confidence_one_or_zero": 0.0008680555814256271,
"eval_rewards/format_reward": 0.9965277910232544,
"eval_rewards/mean_confidence_reward": 0.6795138816038767,
"eval_runtime": 173.6277,
"eval_samples_per_second": 5.759,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4012044270833333,
"eval_signal/accuracy_reward/group_std_mean": 0.453156977891922,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20060221354166666,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20060221354166666,
"eval_signal/advantage_abs_mean": 0.25084766497214633,
"eval_signal/advantage_pre_scale_abs_mean": 0.25084766497214633,
"eval_signal/advantage_pre_scale_std": 0.29384053746859234,
"eval_signal/advantage_std": 0.29384053746859234,
"eval_signal/brier_reward/centered_abs_mean": 0.17003757506608963,
"eval_signal/brier_reward/group_std_mean": 0.22511051098505655,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08501878753304482,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08501878753304482,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0016818575871487458,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0049104637776811915,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 0.9722222288449606,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151002e-08,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151002e-08,
"eval_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"eval_signal/format_reward/group_std_mean": 0.019641855110724766,
"eval_signal/format_reward/group_zero_std_frac": 0.8888889153798422,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.23637964576482773,
"eval_signal/mean_confidence_reward/group_std_mean": 0.27153053879737854,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.363796359835154e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.363796359835154e-06,
"eval_steps_per_second": 0.035,
"step": 250
},
{
"epoch": 0.6009615384615384,
"step": 250,
"train_probe_calibration/aurc": 0.12413227849884392,
"train_probe_calibration/batch_distribution_entropy": 0.6596172202182647,
"train_probe_calibration/batch_entropy_100bins": 0.3680004020914199,
"train_probe_calibration/batch_entropy_10bins": 0.6596172202182647,
"train_probe_calibration/batch_entropy_50bins": 0.43320411914539897,
"train_probe_calibration/batch_uniqueness": 0.45417553734211713,
"train_probe_calibration/confidence_entropy": 0.4511331460653976,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.46596858638743455,
"train_probe_calibration/coverage@15%": 0.7390924956369983,
"train_probe_calibration/coverage@20%": 0.7914485165794066,
"train_probe_calibration/coverage@25%": 0.9685863874345549,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.6596172202182647,
"train_probe_calibration/distribution_entropy_100": 0.3680004020914199,
"train_probe_calibration/ece": 0.04528795811518314,
"train_probe_calibration/mean_confidence": 0.6881326352530541,
"train_probe_calibration/unique_confidence_per_question": 0.010416666666666666,
"train_probe_calibration/unique_confidences": 12,
"train_probe_completions/clipped_ratio": 0.008506944444444461,
"train_probe_completions/max_length": 2231.0,
"train_probe_completions/max_terminated_length": 2231.0,
"train_probe_completions/mean_length": 773.5767618815104,
"train_probe_completions/mean_terminated_length": 780.2432556152344,
"train_probe_completions/min_length": 30.166666666666668,
"train_probe_completions/min_terminated_length": 195.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 536723724.0,
"train_probe_reward": 1.287823756535848,
"train_probe_reward_std": 0.293773889541626,
"train_probe_rewards/accuracy_reward": 0.7291666467984518,
"train_probe_rewards/brier_reward": 0.8516753911972046,
"train_probe_rewards/confidence_one_or_zero": 0.0017361111628512542,
"train_probe_rewards/format_reward": 0.9947916766007742,
"train_probe_rewards/mean_confidence_reward": 0.6845485866069794,
"train_probe_runtime": 186.6636,
"train_probe_samples_per_second": 5.357,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3862847139437993,
"train_probe_signal/accuracy_reward/group_std_mean": 0.44479583700497943,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19314235697189966,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.19314235697189966,
"train_probe_signal/advantage_abs_mean": 0.24279569586118063,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.24279569586118063,
"train_probe_signal/advantage_pre_scale_std": 0.292436346411705,
"train_probe_signal/advantage_std": 0.292436346411705,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1708850140372912,
"train_probe_signal/brier_reward/group_std_mean": 0.2263486310839653,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0854425070186456,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.0854425070186456,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.009820927555362383,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 0.944444457689921,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.3637149954302004e-08,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.3637149954302004e-08,
"train_probe_signal/format_reward/centered_abs_mean": 0.010091145522892475,
"train_probe_signal/format_reward/group_std_mean": 0.02946278266608715,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8333333631356558,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0050455727614462376,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0050455727614462376,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.23532441755135855,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.27096671362717945,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.3532442128271214e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.3532442128271214e-06,
"train_probe_steps_per_second": 0.032
},
{
"calibration/aurc": 0.10192401164575704,
"calibration/batch_distribution_entropy": 0.5630897990918653,
"calibration/batch_entropy_100bins": 0.3115759678912134,
"calibration/batch_entropy_10bins": 0.5630897990918653,
"calibration/batch_entropy_50bins": 0.3667821881446125,
"calibration/batch_uniqueness": 0.2728590050713087,
"calibration/confidence_entropy": 0.43422457263377695,
"calibration/coverage@0%": 0.1171875,
"calibration/coverage@1%": 0.3534325787401575,
"calibration/coverage@10%": 0.6791543635170604,
"calibration/coverage@15%": 0.7370529855643044,
"calibration/coverage@20%": 0.7505946522309712,
"calibration/coverage@25%": 0.8765625,
"calibration/coverage@30%": 0.921875,
"calibration/coverage@5%": 0.4003075787401575,
"calibration/distribution_entropy_10": 0.5630897990918653,
"calibration/distribution_entropy_100": 0.3115759678912134,
"calibration/ece": 0.12341678969816257,
"calibration/mean_confidence": 0.7173136072834646,
"calibration/unique_confidence_per_question": 0.025520833333333333,
"calibration/unique_confidences": 9.8,
"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": 0.002517361111111116,
"completions/max_length": 3401.0,
"completions/max_terminated_length": 3401.0,
"completions/mean_length": 781.3870727539063,
"completions/mean_terminated_length": 783.37958984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 177.2,
"epoch": 0.6129807692307693,
"grad_norm": 0.0004924332606606185,
"learning_rate": 4.7175480769230775e-06,
"loss": -0.0004,
"num_tokens": 548826199.0,
"reward": 1.2913060426712035,
"reward_std": 0.12112262547016144,
"rewards/accuracy_reward": 0.7307291626930237,
"rewards/brier_reward": 0.8547330856323242,
"rewards/confidence_one_or_zero": 0.0016493055794853718,
"rewards/format_reward": 0.9971354126930236,
"rewards/mean_confidence_reward": 0.7186979174613952,
"sampling/batch_mean_priority_error": 0.0507233970853445,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6305555555555555,
"sampling/error_ema_max": 0.12311078608036041,
"sampling/error_ema_mean": 0.006377698108553887,
"sampling/priority_kl": 0.029999877139925957,
"sampling/priority_scale": 0.7706882416969165,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 4.12878165661823e-05,
"sampling/prob_min": 1.7625923283048907e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6072000026702881,
"sampling/prompt_draws_total": 18216.0,
"sampling/seen_fraction": 0.4998066782951355,
"sampling/unseen_fraction": 0.5001933217048645,
"signal/accuracy_reward/centered_abs_mean": 0.12176649272441864,
"signal/accuracy_reward/group_std_mean": 0.1668856054544449,
"signal/accuracy_reward/group_zero_std_frac": 0.5055555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06088324636220932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06088324636220932,
"signal/advantage_abs_mean": 0.08548608571290969,
"signal/advantage_pre_scale_abs_mean": 0.08548608571290969,
"signal/advantage_pre_scale_std": 0.1691525399684906,
"signal/advantage_std": 0.1691525399684906,
"signal/brier_reward/centered_abs_mean": 0.0856711134314537,
"signal/brier_reward/group_std_mean": 0.11683738827705384,
"signal/brier_reward/group_zero_std_frac": 0.18611111640930175,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04283555671572685,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04283555671572685,
"signal/confidence_one_or_zero/centered_abs_mean": 0.002620442712213844,
"signal/confidence_one_or_zero/group_std_mean": 0.004098849650472402,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.6204425296327827e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.6204425296327827e-08,
"signal/format_reward/centered_abs_mean": 0.00539822056889534,
"signal/format_reward/group_std_mean": 0.013028488121926784,
"signal/format_reward/group_zero_std_frac": 0.9361111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00269911028444767,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00269911028444767,
"signal/mean_confidence_reward/centered_abs_mean": 0.07201225459575653,
"signal/mean_confidence_reward/group_std_mean": 0.09611042886972428,
"signal/mean_confidence_reward/group_zero_std_frac": 0.21111111342906952,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 7.201225457720284e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 7.201225457720284e-07,
"step": 255
},
{
"calibration/aurc": 0.14828991525526353,
"calibration/batch_distribution_entropy": 0.44975362442630307,
"calibration/batch_entropy_100bins": 0.24745196947920672,
"calibration/batch_entropy_10bins": 0.44975362442630307,
"calibration/batch_entropy_50bins": 0.29129645473160004,
"calibration/batch_uniqueness": 0.03987591217030008,
"calibration/confidence_entropy": 0.4098157739993427,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.42926183660909684,
"calibration/coverage@15%": 0.6075963331944033,
"calibration/coverage@20%": 0.6421161978926266,
"calibration/coverage@25%": 0.9016763637819739,
"calibration/coverage@30%": 0.9607803963983823,
"calibration/coverage@5%": 0.1234375,
"calibration/distribution_entropy_10": 0.44975362442630307,
"calibration/distribution_entropy_100": 0.24745196947920672,
"calibration/ece": 0.0922259926249394,
"calibration/mean_confidence": 0.767020542663863,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.002690972222222232,
"completions/max_length": 2928.2,
"completions/max_terminated_length": 2928.2,
"completions/mean_length": 778.8683227539062,
"completions/mean_terminated_length": 780.974951171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 180.4,
"epoch": 0.625,
"grad_norm": 0.0004822597838938236,
"learning_rate": 4.6875000000000004e-06,
"loss": -0.0008,
"num_tokens": 560880266.0,
"reward": 1.300320601463318,
"reward_std": 0.1322122886776924,
"rewards/accuracy_reward": 0.7479166626930237,
"rewards/brier_reward": 0.8555737972259522,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9971354246139527,
"rewards/mean_confidence_reward": 0.7657639026641846,
"sampling/batch_mean_priority_error": 0.05647566379535647,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6083333333333334,
"sampling/error_ema_max": 0.11928199231624603,
"sampling/error_ema_mean": 0.0064342057332396506,
"sampling/priority_kl": 0.0299997441470623,
"sampling/priority_scale": 0.7668565093772486,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 4.1415702435187994e-05,
"sampling/prob_min": 1.775550190359354e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6192000031471252,
"sampling/prompt_draws_total": 18576.0,
"sampling/seen_fraction": 0.5074999928474426,
"sampling/unseen_fraction": 0.4925000071525574,
"signal/accuracy_reward/centered_abs_mean": 0.13364800214767455,
"signal/accuracy_reward/group_std_mean": 0.17856489717960358,
"signal/accuracy_reward/group_zero_std_frac": 0.4833333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06682400107383728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06682400107383728,
"signal/advantage_abs_mean": 0.09538109004497528,
"signal/advantage_pre_scale_abs_mean": 0.09538109004497528,
"signal/advantage_pre_scale_std": 0.18343801498413087,
"signal/advantage_std": 0.18343801498413087,
"signal/brier_reward/centered_abs_mean": 0.08802536427974701,
"signal/brier_reward/group_std_mean": 0.12009987086057664,
"signal/brier_reward/group_zero_std_frac": 0.2805555522441864,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.044012682139873506,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.044012682139873506,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005289713619276881,
"signal/format_reward/group_std_mean": 0.011597984097898006,
"signal/format_reward/group_zero_std_frac": 0.9472222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0026448568096384406,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0026448568096384406,
"signal/mean_confidence_reward/centered_abs_mean": 0.06126871481537819,
"signal/mean_confidence_reward/group_std_mean": 0.0842192992568016,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3111111164093018,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.126871085143648e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.126871085143648e-07,
"step": 260
},
{
"calibration/aurc": 0.17156296594583784,
"calibration/batch_distribution_entropy": 0.5713246895783481,
"calibration/batch_entropy_100bins": 0.3150922720135943,
"calibration/batch_entropy_10bins": 0.5713246895783481,
"calibration/batch_entropy_50bins": 0.3709215244641552,
"calibration/batch_uniqueness": 0.3472916017679485,
"calibration/confidence_entropy": 0.4582822522468408,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.39228697203717355,
"calibration/coverage@15%": 0.6017746302279601,
"calibration/coverage@20%": 0.6673629242819843,
"calibration/coverage@25%": 0.7258485639686685,
"calibration/coverage@30%": 0.7550913838120105,
"calibration/coverage@5%": 0.12189973614775726,
"calibration/distribution_entropy_10": 0.5713246895783481,
"calibration/distribution_entropy_100": 0.3150922720135943,
"calibration/ece": 0.10203320370357609,
"calibration/mean_confidence": 0.7285038118209939,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.003993055555555536,
"completions/max_length": 2905.8,
"completions/max_terminated_length": 2905.8,
"completions/mean_length": 828.4642456054687,
"completions/mean_terminated_length": 831.7786865234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 219.4,
"epoch": 0.6370192307692307,
"grad_norm": 0.0005070755141787231,
"learning_rate": 4.657451923076923e-06,
"loss": -0.0038,
"num_tokens": 573521518.0,
"reward": 1.2752347469329834,
"reward_std": 0.12059136927127838,
"rewards/accuracy_reward": 0.7099826455116272,
"rewards/brier_reward": 0.8448992967605591,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9955729126930237,
"rewards/mean_confidence_reward": 0.73536456823349,
"sampling/batch_mean_priority_error": 0.053314529740651026,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6416666666666666,
"sampling/error_ema_max": 0.11928199231624603,
"sampling/error_ema_mean": 0.006500357948243618,
"sampling/priority_kl": 0.030000258609652518,
"sampling/priority_scale": 0.7631947935326024,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 4.154189373366535e-05,
"sampling/prob_min": 1.7881768508232197e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6311999917030334,
"sampling/prompt_draws_total": 18936.0,
"sampling/seen_fraction": 0.5149933338165283,
"sampling/unseen_fraction": 0.48500666618347166,
"signal/accuracy_reward/centered_abs_mean": 0.1154568150639534,
"signal/accuracy_reward/group_std_mean": 0.1554502695798874,
"signal/accuracy_reward/group_zero_std_frac": 0.55,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0577284075319767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0577284075319767,
"signal/advantage_abs_mean": 0.08649851530790328,
"signal/advantage_pre_scale_abs_mean": 0.08649851530790328,
"signal/advantage_pre_scale_std": 0.1742924690246582,
"signal/advantage_std": 0.1742924690246582,
"signal/brier_reward/centered_abs_mean": 0.08069736361503602,
"signal/brier_reward/group_std_mean": 0.10713728815317154,
"signal/brier_reward/group_zero_std_frac": 0.21944445073604585,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04034868180751801,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04034868180751801,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.008132595662027597,
"signal/format_reward/group_std_mean": 0.017994208075106145,
"signal/format_reward/group_zero_std_frac": 0.9166666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0040662978310137985,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0040662978310137985,
"signal/mean_confidence_reward/centered_abs_mean": 0.06501844376325608,
"signal/mean_confidence_reward/group_std_mean": 0.08517636507749557,
"signal/mean_confidence_reward/group_zero_std_frac": 0.24166666567325593,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.501843927253504e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.501843927253504e-07,
"step": 265
},
{
"calibration/aurc": 0.1724353102680505,
"calibration/batch_distribution_entropy": 0.6705310899852044,
"calibration/batch_entropy_100bins": 0.36594505639391034,
"calibration/batch_entropy_10bins": 0.6705310899852044,
"calibration/batch_entropy_50bins": 0.43078459944550557,
"calibration/batch_uniqueness": 0.5443845630429488,
"calibration/confidence_entropy": 0.5165166687743069,
"calibration/coverage@0%": 0.023097112860892388,
"calibration/coverage@1%": 0.023097112860892388,
"calibration/coverage@10%": 0.4130767230293049,
"calibration/coverage@15%": 0.5197839987121903,
"calibration/coverage@20%": 0.6050889263380436,
"calibration/coverage@25%": 0.6672000011857372,
"calibration/coverage@30%": 0.8000747076369044,
"calibration/coverage@5%": 0.14690370734908137,
"calibration/distribution_entropy_10": 0.6705310899852044,
"calibration/distribution_entropy_100": 0.36594505639391034,
"calibration/ece": 0.11203265838968064,
"calibration/mean_confidence": 0.6719259889611031,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.005555555555555558,
"completions/max_length": 3264.8,
"completions/max_terminated_length": 3264.8,
"completions/mean_length": 870.48125,
"completions/mean_terminated_length": 875.4301147460938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 274.8,
"epoch": 0.6490384615384616,
"grad_norm": 0.0004944394459016621,
"learning_rate": 4.627403846153847e-06,
"loss": -0.0045,
"num_tokens": 586663670.0,
"reward": 1.2752594470977783,
"reward_std": 0.1263850063085556,
"rewards/accuracy_reward": 0.7150173544883728,
"rewards/brier_reward": 0.8412169933319091,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9942708253860474,
"rewards/mean_confidence_reward": 0.6813281416893006,
"sampling/batch_mean_priority_error": 0.04256142612508279,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5972222222222222,
"sampling/error_ema_max": 0.11928199231624603,
"sampling/error_ema_mean": 0.006547117419540882,
"sampling/priority_kl": 0.029999774321913718,
"sampling/priority_scale": 0.7595782100455836,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.1664379386929797e-05,
"sampling/prob_min": 1.800627251213882e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6432000041007996,
"sampling/prompt_draws_total": 19296.0,
"sampling/seen_fraction": 0.5221800208091736,
"sampling/unseen_fraction": 0.4778199791908264,
"signal/accuracy_reward/centered_abs_mean": 0.12918294370174407,
"signal/accuracy_reward/group_std_mean": 0.1769692748785019,
"signal/accuracy_reward/group_zero_std_frac": 0.46666666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06459147185087204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06459147185087204,
"signal/advantage_abs_mean": 0.08929220288991928,
"signal/advantage_pre_scale_abs_mean": 0.08929220288991928,
"signal/advantage_pre_scale_std": 0.17134988605976104,
"signal/advantage_std": 0.17134988605976104,
"signal/brier_reward/centered_abs_mean": 0.07582532465457917,
"signal/brier_reward/group_std_mean": 0.10502927154302596,
"signal/brier_reward/group_zero_std_frac": 0.05833333525806665,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037912662327289584,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037912662327289584,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01001519076526165,
"signal/format_reward/group_std_mean": 0.020045911055058242,
"signal/format_reward/group_zero_std_frac": 0.9138888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005007595382630825,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005007595382630825,
"signal/mean_confidence_reward/centered_abs_mean": 0.0677403338253498,
"signal/mean_confidence_reward/group_std_mean": 0.08821502774953842,
"signal/mean_confidence_reward/group_zero_std_frac": 0.06944444570690393,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.774032954126597e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.774032954126597e-07,
"step": 270
},
{
"calibration/aurc": 0.20918962012232867,
"calibration/batch_distribution_entropy": 0.6260484810131134,
"calibration/batch_entropy_100bins": 0.3459900673899717,
"calibration/batch_entropy_10bins": 0.6260484810131134,
"calibration/batch_entropy_50bins": 0.40729390925908593,
"calibration/batch_uniqueness": 0.49000524103582255,
"calibration/confidence_entropy": 0.4940136629397375,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.13594836422061554,
"calibration/coverage@15%": 0.39209262202554085,
"calibration/coverage@20%": 0.6095946243339592,
"calibration/coverage@25%": 0.6824626807529294,
"calibration/coverage@30%": 0.8821657109615225,
"calibration/coverage@5%": 0.13594836422061554,
"calibration/distribution_entropy_10": 0.6260484810131134,
"calibration/distribution_entropy_100": 0.3459900673899717,
"calibration/ece": 0.11715939758055464,
"calibration/mean_confidence": 0.7095371226439267,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.005642361111111116,
"completions/max_length": 2745.8,
"completions/max_terminated_length": 2745.8,
"completions/mean_length": 861.8537353515625,
"completions/mean_terminated_length": 866.7883422851562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 203.4,
"epoch": 0.6610576923076923,
"grad_norm": 0.0007629328756593168,
"learning_rate": 4.597355769230769e-06,
"loss": -0.0056,
"num_tokens": 599688769.0,
"reward": 1.2769941091537476,
"reward_std": 0.12798523157835007,
"rewards/accuracy_reward": 0.7092013955116272,
"rewards/brier_reward": 0.8505017399787903,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9942708373069763,
"rewards/mean_confidence_reward": 0.716163182258606,
"sampling/batch_mean_priority_error": 0.03905228317416913,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6000000000000001,
"sampling/error_ema_max": 0.119870525598526,
"sampling/error_ema_mean": 0.006589326169341802,
"sampling/priority_kl": 0.030000057816505433,
"sampling/priority_scale": 0.7562616884941236,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 4.179128518444486e-05,
"sampling/prob_min": 1.8127395378542133e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6552000045776367,
"sampling/prompt_draws_total": 19656.0,
"sampling/seen_fraction": 0.5294933319091797,
"sampling/unseen_fraction": 0.4705066680908203,
"signal/accuracy_reward/centered_abs_mean": 0.11976996660232545,
"signal/accuracy_reward/group_std_mean": 0.16761533915996552,
"signal/accuracy_reward/group_zero_std_frac": 0.4888889014720917,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05988498330116272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05988498330116272,
"signal/advantage_abs_mean": 0.08943271785974502,
"signal/advantage_pre_scale_abs_mean": 0.08943271785974502,
"signal/advantage_pre_scale_std": 0.1768662929534912,
"signal/advantage_std": 0.1768662929534912,
"signal/brier_reward/centered_abs_mean": 0.07352758347988128,
"signal/brier_reward/group_std_mean": 0.10236799418926239,
"signal/brier_reward/group_zero_std_frac": 0.08611111268401146,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03676379173994064,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03676379173994064,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01022135429084301,
"signal/format_reward/group_std_mean": 0.020866769552230834,
"signal/format_reward/group_zero_std_frac": 0.9083333492279053,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005110677145421505,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005110677145421505,
"signal/mean_confidence_reward/centered_abs_mean": 0.06516871526837349,
"signal/mean_confidence_reward/group_std_mean": 0.08634473383426666,
"signal/mean_confidence_reward/group_zero_std_frac": 0.0888888917863369,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.516871280837222e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.516871280837222e-07,
"step": 275
},
{
"calibration/aurc": 0.14039470835629952,
"calibration/batch_distribution_entropy": 0.6472011937223521,
"calibration/batch_entropy_100bins": 0.36074211480402196,
"calibration/batch_entropy_10bins": 0.6472011937223521,
"calibration/batch_entropy_50bins": 0.42465978078877853,
"calibration/batch_uniqueness": 0.5245013563165913,
"calibration/confidence_entropy": 0.49849425976719336,
"calibration/coverage@0%": 0.06307277628032346,
"calibration/coverage@1%": 0.14932614555256066,
"calibration/coverage@10%": 0.42254102650494163,
"calibration/coverage@15%": 0.6666587476947085,
"calibration/coverage@20%": 0.7297859312195583,
"calibration/coverage@25%": 0.7877999473920652,
"calibration/coverage@30%": 0.8885936267555682,
"calibration/coverage@5%": 0.2825934973045822,
"calibration/distribution_entropy_10": 0.6472011937223521,
"calibration/distribution_entropy_100": 0.36074211480402196,
"calibration/ece": 0.10694779631862672,
"calibration/mean_confidence": 0.6907570665460823,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.010069444444444442,
"completions/max_length": 3203.6,
"completions/max_terminated_length": 3203.6,
"completions/mean_length": 870.253662109375,
"completions/mean_terminated_length": 879.0881958007812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 217.2,
"epoch": 0.6730769230769231,
"grad_norm": 0.0005508947651833296,
"learning_rate": 4.567307692307692e-06,
"loss": -0.0091,
"num_tokens": 612789739.0,
"reward": 1.26469144821167,
"reward_std": 0.14498854875564576,
"rewards/accuracy_reward": 0.6958333373069763,
"rewards/brier_reward": 0.8436918377876281,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.98984375,
"rewards/mean_confidence_reward": 0.6993142485618591,
"sampling/batch_mean_priority_error": 0.03846872324937353,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.6222222222222221,
"sampling/error_ema_max": 0.12075332552194595,
"sampling/error_ema_mean": 0.006628950219601392,
"sampling/priority_kl": 0.029999958351254463,
"sampling/priority_scale": 0.7531239926582203,
"sampling/prob_entropy": 10.27894992828369,
"sampling/prob_max": 4.1917922499123963e-05,
"sampling/prob_min": 1.824544197006617e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6672000050544739,
"sampling/prompt_draws_total": 20016.0,
"sampling/seen_fraction": 0.5366933345794678,
"sampling/unseen_fraction": 0.4633066654205322,
"signal/accuracy_reward/centered_abs_mean": 0.1374348968267441,
"signal/accuracy_reward/group_std_mean": 0.1876349091529846,
"signal/accuracy_reward/group_zero_std_frac": 0.4444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06871744841337205,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06871744841337205,
"signal/advantage_abs_mean": 0.10304587036371231,
"signal/advantage_pre_scale_abs_mean": 0.10304587036371231,
"signal/advantage_pre_scale_std": 0.19462815523147584,
"signal/advantage_std": 0.19462815523147584,
"signal/brier_reward/centered_abs_mean": 0.08086232841014862,
"signal/brier_reward/group_std_mean": 0.11166905611753464,
"signal/brier_reward/group_zero_std_frac": 0.08333333507180214,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04043116420507431,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04043116420507431,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01641167551279068,
"signal/format_reward/group_std_mean": 0.029581551998853685,
"signal/format_reward/group_zero_std_frac": 0.8805555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00820583775639534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00820583775639534,
"signal/mean_confidence_reward/centered_abs_mean": 0.06651312336325646,
"signal/mean_confidence_reward/group_std_mean": 0.08824529945850372,
"signal/mean_confidence_reward/group_zero_std_frac": 0.08888889104127884,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.65131210553227e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.65131210553227e-07,
"step": 280
},
{
"calibration/aurc": 0.14288680384347863,
"calibration/batch_distribution_entropy": 0.6650315258167521,
"calibration/batch_entropy_100bins": 0.37373616932584974,
"calibration/batch_entropy_10bins": 0.6650315258167521,
"calibration/batch_entropy_50bins": 0.4399561715298336,
"calibration/batch_uniqueness": 0.5287880229643276,
"calibration/confidence_entropy": 0.5082635866477374,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.10548302872062662,
"calibration/coverage@10%": 0.4817853906989272,
"calibration/coverage@15%": 0.7174201159144425,
"calibration/coverage@20%": 0.7431254323607392,
"calibration/coverage@25%": 0.8367832665460192,
"calibration/coverage@30%": 0.8789591695119057,
"calibration/coverage@5%": 0.20065376945375304,
"calibration/distribution_entropy_10": 0.6650315258167521,
"calibration/distribution_entropy_100": 0.37373616932584974,
"calibration/ece": 0.09754575100242655,
"calibration/mean_confidence": 0.6825601722922887,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.02152777777777779,
"completions/max_length": 3658.2,
"completions/max_terminated_length": 3658.2,
"completions/mean_length": 883.3541870117188,
"completions/mean_terminated_length": 902.7584106445313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 198.8,
"epoch": 0.6850961538461539,
"grad_norm": 0.034632738679647446,
"learning_rate": 4.537259615384616e-06,
"loss": -0.0225,
"num_tokens": 626066267.0,
"reward": 1.2527711153030396,
"reward_std": 0.1646108791232109,
"rewards/accuracy_reward": 0.6983506917953491,
"rewards/brier_reward": 0.8310499072074891,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9761284708976745,
"rewards/mean_confidence_reward": 0.6590017795562744,
"sampling/batch_mean_priority_error": 0.03579175608843308,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5833333333333334,
"sampling/error_ema_max": 0.12075332552194595,
"sampling/error_ema_mean": 0.006667227391153574,
"sampling/priority_kl": 0.02999993450939655,
"sampling/priority_scale": 0.7503358900779858,
"sampling/prob_entropy": 10.278953552246094,
"sampling/prob_max": 4.205153236398473e-05,
"sampling/prob_min": 1.8355593783780933e-05,
"sampling/prompt_draws_max": 4.0,
"sampling/prompt_draws_mean": 0.6791999936103821,
"sampling/prompt_draws_total": 20376.0,
"sampling/seen_fraction": 0.5441400051116944,
"sampling/unseen_fraction": 0.45585999488830564,
"signal/accuracy_reward/centered_abs_mean": 0.1542697474360466,
"signal/accuracy_reward/group_std_mean": 0.20294737815856934,
"signal/accuracy_reward/group_zero_std_frac": 0.4277777761220932,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0771348737180233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0771348737180233,
"signal/advantage_abs_mean": 0.11804689913988113,
"signal/advantage_pre_scale_abs_mean": 0.11804689913988113,
"signal/advantage_pre_scale_std": 0.2127869665622711,
"signal/advantage_std": 0.2127869665622711,
"signal/brier_reward/centered_abs_mean": 0.09634966850280761,
"signal/brier_reward/group_std_mean": 0.12952123284339906,
"signal/brier_reward/group_zero_std_frac": 0.05277777910232544,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04817483425140381,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04817483425140381,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0365831159055233,
"signal/format_reward/group_std_mean": 0.061961127817630766,
"signal/format_reward/group_zero_std_frac": 0.7694444477558136,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.01829155795276165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.01829155795276165,
"signal/mean_confidence_reward/centered_abs_mean": 0.08290661871433258,
"signal/mean_confidence_reward/group_std_mean": 0.10661832988262177,
"signal/mean_confidence_reward/group_zero_std_frac": 0.06388889029622077,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 8.290661867249582e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 8.290661867249582e-07,
"step": 285
},
{
"calibration/aurc": 0.28724837900990297,
"calibration/batch_distribution_entropy": 0.5682252479718652,
"calibration/batch_entropy_100bins": 0.3655933130909036,
"calibration/batch_entropy_10bins": 0.5682252479718652,
"calibration/batch_entropy_50bins": 0.4205142224408179,
"calibration/batch_uniqueness": 0.2985340044223322,
"calibration/confidence_entropy": 0.4648818296102114,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4457751396648044,
"calibration/coverage@15%": 0.5015770484171321,
"calibration/coverage@20%": 0.6608335870759612,
"calibration/coverage@25%": 0.7395583237306184,
"calibration/coverage@30%": 0.7770234986945169,
"calibration/coverage@5%": 0.13125,
"calibration/distribution_entropy_10": 0.5682252479718652,
"calibration/distribution_entropy_100": 0.3655933130909036,
"calibration/ece": 0.1118047358055387,
"calibration/mean_confidence": 0.6590319810025146,
"calibration/unique_confidence_per_question": 0.051041666666666666,
"calibration/unique_confidences": 19.6,
"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": 0.04357638888888891,
"completions/max_length": 3553.8,
"completions/max_terminated_length": 3553.8,
"completions/mean_length": 739.6080718994141,
"completions/mean_terminated_length": 754.0138488769531,
"completions/min_length": 0.0,
"completions/min_terminated_length": 159.4,
"epoch": 0.6971153846153846,
"grad_norm": 0.0005248843808658421,
"learning_rate": 4.507211538461539e-06,
"loss": -0.036,
"num_tokens": 637696888.0,
"reward": 1.13420969247818,
"reward_std": 0.1965183973312378,
"rewards/accuracy_reward": 0.5455729156732559,
"rewards/brier_reward": 0.7833370208740235,
"rewards/confidence_one_or_zero": 0.01180555522441864,
"rewards/format_reward": 0.9394965291023254,
"rewards/mean_confidence_reward": 0.6383805871009827,
"sampling/batch_mean_priority_error": 0.06071788752333244,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5555555555555556,
"sampling/error_ema_max": 0.12075332552194595,
"sampling/error_ema_mean": 0.006719445157796144,
"sampling/priority_kl": 0.029999922960996628,
"sampling/priority_scale": 0.7473562896018848,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 4.217153327772394e-05,
"sampling/prob_min": 1.759280130499974e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.6911999940872192,
"sampling/prompt_draws_total": 20736.0,
"sampling/seen_fraction": 0.5507933259010315,
"sampling/unseen_fraction": 0.44920667409896853,
"signal/accuracy_reward/centered_abs_mean": 0.10755750983953476,
"signal/accuracy_reward/group_std_mean": 0.15215527266263962,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05377875491976738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05377875491976738,
"signal/advantage_abs_mean": 0.14942664355039598,
"signal/advantage_pre_scale_abs_mean": 0.14942664355039598,
"signal/advantage_pre_scale_std": 0.23570600152015686,
"signal/advantage_std": 0.23570600152015686,
"signal/brier_reward/centered_abs_mean": 0.1448768064379692,
"signal/brier_reward/group_std_mean": 0.18426230549812317,
"signal/brier_reward/group_zero_std_frac": 0.19444445222616197,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0724384032189846,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0724384032189846,
"signal/confidence_one_or_zero/centered_abs_mean": 0.02027994841337204,
"signal/confidence_one_or_zero/group_std_mean": 0.03520326614379883,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.8694444477558136,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 2.027994696618407e-07,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 2.027994696618407e-07,
"signal/format_reward/centered_abs_mean": 0.0818847662769258,
"signal/format_reward/group_std_mean": 0.10462786592543125,
"signal/format_reward/group_zero_std_frac": 0.7166666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0409423831384629,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0409423831384629,
"signal/mean_confidence_reward/centered_abs_mean": 0.09347325339913368,
"signal/mean_confidence_reward/group_std_mean": 0.12237994074821472,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23611111491918563,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 9.347324635200494e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 9.347324635200494e-07,
"step": 290
},
{
"calibration/aurc": 0.15274101278407712,
"calibration/batch_distribution_entropy": 0.3099765802059592,
"calibration/batch_entropy_100bins": 0.17101656435086784,
"calibration/batch_entropy_10bins": 0.3099765802059592,
"calibration/batch_entropy_50bins": 0.2013179325290131,
"calibration/batch_uniqueness": -0.3301233201476169,
"calibration/confidence_entropy": 0.38117362988310044,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.37017620603392887,
"calibration/coverage@15%": 0.5733911431064572,
"calibration/coverage@20%": 0.6,
"calibration/coverage@25%": 0.759375,
"calibration/coverage@30%": 0.9276041666666666,
"calibration/coverage@5%": 0.3330622120760598,
"calibration/distribution_entropy_10": 0.3099765802059592,
"calibration/distribution_entropy_100": 0.17101656435086784,
"calibration/ece": 0.10969831779512347,
"calibration/mean_confidence": 0.8102238274005625,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.0015625,
"completions/max_length": 3030.8,
"completions/max_terminated_length": 3030.8,
"completions/mean_length": 902.4171875,
"completions/mean_terminated_length": 903.819921875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 208.6,
"epoch": 0.7091346153846154,
"grad_norm": 0.0005306955426931381,
"learning_rate": 4.477163461538462e-06,
"loss": 0.0003,
"num_tokens": 651179902.0,
"reward": 1.2625298500061035,
"reward_std": 0.13462880551815032,
"rewards/accuracy_reward": 0.69296875,
"rewards/brier_reward": 0.833638048171997,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9984375,
"rewards/mean_confidence_reward": 0.7773871302604676,
"sampling/batch_mean_priority_error": 0.07865727422187206,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5194444444444445,
"sampling/error_ema_max": 0.12075332552194595,
"sampling/error_ema_mean": 0.006801041215658188,
"sampling/priority_kl": 0.030000052228569985,
"sampling/priority_scale": 0.7444859921699389,
"sampling/prob_entropy": 10.278954315185548,
"sampling/prob_max": 4.228979887557216e-05,
"sampling/prob_min": 1.7701475371723063e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7032000064849854,
"sampling/prompt_draws_total": 21096.0,
"sampling/seen_fraction": 0.557260000705719,
"sampling/unseen_fraction": 0.442739999294281,
"signal/accuracy_reward/centered_abs_mean": 0.12332356721162796,
"signal/accuracy_reward/group_std_mean": 0.16784208714962007,
"signal/accuracy_reward/group_zero_std_frac": 0.5027777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06166178360581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06166178360581398,
"signal/advantage_abs_mean": 0.0977450504899025,
"signal/advantage_pre_scale_abs_mean": 0.0977450504899025,
"signal/advantage_pre_scale_std": 0.18939689993858339,
"signal/advantage_std": 0.18939689993858339,
"signal/brier_reward/centered_abs_mean": 0.09247624278068542,
"signal/brier_reward/group_std_mean": 0.12620881497859954,
"signal/brier_reward/group_zero_std_frac": 0.32222222089767455,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04623812139034271,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04623812139034271,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.002951388875953853,
"signal/format_reward/group_std_mean": 0.006992555782198906,
"signal/format_reward/group_zero_std_frac": 0.9666666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0014756944379769265,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0014756944379769265,
"signal/mean_confidence_reward/centered_abs_mean": 0.0591384619474411,
"signal/mean_confidence_reward/group_std_mean": 0.07959082275629044,
"signal/mean_confidence_reward/group_zero_std_frac": 0.41666666865348817,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.913845825489261e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.913845825489261e-07,
"step": 295
},
{
"calibration/aurc": 0.17787271048902797,
"calibration/batch_distribution_entropy": 0.5103355610682992,
"calibration/batch_entropy_100bins": 0.2808516359903569,
"calibration/batch_entropy_10bins": 0.5103355610682992,
"calibration/batch_entropy_50bins": 0.33061400174644945,
"calibration/batch_uniqueness": 0.17855177361487,
"calibration/confidence_entropy": 0.4409625069876807,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4376278285465623,
"calibration/coverage@15%": 0.4856396866840731,
"calibration/coverage@20%": 0.5389033942558747,
"calibration/coverage@25%": 0.7477806788511749,
"calibration/coverage@30%": 0.9177083333333332,
"calibration/coverage@5%": 0.10809399477806789,
"calibration/distribution_entropy_10": 0.5103355610682992,
"calibration/distribution_entropy_100": 0.2808516359903569,
"calibration/ece": 0.1495082680591819,
"calibration/mean_confidence": 0.7491250543951262,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.001388888888888884,
"completions/max_length": 3413.8,
"completions/max_terminated_length": 3413.8,
"completions/mean_length": 868.06884765625,
"completions/mean_terminated_length": 869.28193359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 211.6,
"epoch": 0.7211538461538461,
"grad_norm": 0.0003403428418096155,
"learning_rate": 4.447115384615385e-06,
"loss": 0.0,
"num_tokens": 664253655.0,
"reward": 1.2810194969177247,
"reward_std": 0.12045833468437195,
"rewards/accuracy_reward": 0.7264756917953491,
"rewards/brier_reward": 0.8370242953300476,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.998524296283722,
"rewards/mean_confidence_reward": 0.7348958492279053,
"sampling/batch_mean_priority_error": 0.06383506851105139,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5805555555555555,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.006883678119629621,
"sampling/priority_kl": 0.029999998211860657,
"sampling/priority_scale": 0.7418554007774218,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 4.2413204209879043e-05,
"sampling/prob_min": 1.7807840413297528e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7152000069618225,
"sampling/prompt_draws_total": 21456.0,
"sampling/seen_fraction": 0.5638733267784118,
"sampling/unseen_fraction": 0.43612667322158816,
"signal/accuracy_reward/centered_abs_mean": 0.12248806357383728,
"signal/accuracy_reward/group_std_mean": 0.16501044929027558,
"signal/accuracy_reward/group_zero_std_frac": 0.5138888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06124403178691864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06124403178691864,
"signal/advantage_abs_mean": 0.08717213496565819,
"signal/advantage_pre_scale_abs_mean": 0.08717213496565819,
"signal/advantage_pre_scale_std": 0.16690560579299926,
"signal/advantage_std": 0.16690560579299926,
"signal/brier_reward/centered_abs_mean": 0.08108467981219292,
"signal/brier_reward/group_std_mean": 0.11095312237739563,
"signal/brier_reward/group_zero_std_frac": 0.2527777820825577,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04054233990609646,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04054233990609646,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0028266058885492384,
"signal/format_reward/group_std_mean": 0.007450965978205204,
"signal/format_reward/group_zero_std_frac": 0.9611111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0014133029442746192,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0014133029442746192,
"signal/mean_confidence_reward/centered_abs_mean": 0.06620009019970893,
"signal/mean_confidence_reward/group_std_mean": 0.08651411533355713,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3138888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.620008434765623e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.620008434765623e-07,
"step": 300
},
{
"epoch": 0.7211538461538461,
"eval_calibration/aurc": 0.14878378298575334,
"eval_calibration/batch_distribution_entropy": 0.6329108288230492,
"eval_calibration/batch_entropy_100bins": 0.3565150525929009,
"eval_calibration/batch_entropy_10bins": 0.6329108288230492,
"eval_calibration/batch_entropy_50bins": 0.4196837515471404,
"eval_calibration/batch_uniqueness": 0.44409782503858014,
"eval_calibration/confidence_entropy": 0.46713917160786056,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.5894097222222222,
"eval_calibration/coverage@20%": 0.7630208333333334,
"eval_calibration/coverage@25%": 0.9088541666666666,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6329108288230492,
"eval_calibration/distribution_entropy_100": 0.3565150525929009,
"eval_calibration/ece": 0.031770833333333456,
"eval_calibration/mean_confidence": 0.6878472222222223,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.0,
"eval_completions/max_length": 2147.0,
"eval_completions/max_terminated_length": 2147.0,
"eval_completions/mean_length": 819.9404703776041,
"eval_completions/mean_terminated_length": 819.9404703776041,
"eval_completions/min_length": 268.6666666666667,
"eval_completions/min_terminated_length": 268.6666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 664253655.0,
"eval_reward": 1.278730869293213,
"eval_reward_std": 0.2882691373427709,
"eval_rewards/accuracy_reward": 0.7100694378217062,
"eval_rewards/brier_reward": 0.8473784724871317,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 1.0,
"eval_rewards/mean_confidence_reward": 0.6878472069899241,
"eval_runtime": 124.8589,
"eval_samples_per_second": 8.009,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3951822916666667,
"eval_signal/accuracy_reward/group_std_mean": 0.4499455789724986,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19759114583333334,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19759114583333334,
"eval_signal/advantage_abs_mean": 0.245485690732797,
"eval_signal/advantage_pre_scale_abs_mean": 0.245485690732797,
"eval_signal/advantage_pre_scale_std": 0.28670944770177204,
"eval_signal/advantage_std": 0.28670944770177204,
"eval_signal/brier_reward/centered_abs_mean": 0.16264214863379797,
"eval_signal/brier_reward/group_std_mean": 0.2163154954711596,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08132107431689899,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08132107431689899,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0,
"eval_signal/format_reward/group_std_mean": 0.0,
"eval_signal/format_reward/group_zero_std_frac": 1.0,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.21807722747325897,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2531433776021004,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.180772374534475e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.180772374534475e-06,
"eval_steps_per_second": 0.048,
"step": 300
},
{
"epoch": 0.7211538461538461,
"step": 300,
"train_probe_calibration/aurc": 0.11106416293807442,
"train_probe_calibration/batch_distribution_entropy": 0.6166980952291539,
"train_probe_calibration/batch_entropy_100bins": 0.346543914125164,
"train_probe_calibration/batch_entropy_10bins": 0.6166980952291539,
"train_probe_calibration/batch_entropy_50bins": 0.40794588867458925,
"train_probe_calibration/batch_uniqueness": 0.41111171392746915,
"train_probe_calibration/confidence_entropy": 0.4680949122400047,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.5954861111111112,
"train_probe_calibration/coverage@15%": 0.7708333333333334,
"train_probe_calibration/coverage@20%": 0.8585069444444444,
"train_probe_calibration/coverage@25%": 0.9340277777777778,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.6166980952291539,
"train_probe_calibration/distribution_entropy_100": 0.346543914125164,
"train_probe_calibration/ece": 0.04765624999999997,
"train_probe_calibration/mean_confidence": 0.6978298611111112,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.0,
"train_probe_completions/max_length": 2437.5,
"train_probe_completions/max_terminated_length": 2437.5,
"train_probe_completions/mean_length": 839.3019307454427,
"train_probe_completions/mean_terminated_length": 839.3019307454427,
"train_probe_completions/min_length": 197.66666666666666,
"train_probe_completions/min_terminated_length": 197.66666666666666,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 664253655.0,
"train_probe_reward": 1.3032057682673137,
"train_probe_reward_std": 0.2715952495733897,
"train_probe_rewards/accuracy_reward": 0.7447916666666666,
"train_probe_rewards/brier_reward": 0.8616058925787607,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 1.0,
"train_probe_rewards/mean_confidence_reward": 0.6978298425674438,
"train_probe_runtime": 143.155,
"train_probe_samples_per_second": 6.985,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3674045105775197,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4328818420569102,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18370225528875986,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18370225528875986,
"train_probe_signal/advantage_abs_mean": 0.22503923873106638,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.22503923873106638,
"train_probe_signal/advantage_pre_scale_std": 0.2698550472656886,
"train_probe_signal/advantage_std": 0.2698550472656886,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14530708640813828,
"train_probe_signal/brier_reward/group_std_mean": 0.1937358702222506,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07265354320406914,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07265354320406914,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.0,
"train_probe_signal/format_reward/group_std_mean": 0.0,
"train_probe_signal/format_reward/group_zero_std_frac": 1.0,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.21216904123624167,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.24356854458649954,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.1216902723608655e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.1216902723608655e-06,
"train_probe_steps_per_second": 0.042
},
{
"calibration/aurc": 0.13489746861381913,
"calibration/batch_distribution_entropy": 0.5663912873579798,
"calibration/batch_entropy_100bins": 0.31968976387481074,
"calibration/batch_entropy_10bins": 0.5663912873579798,
"calibration/batch_entropy_50bins": 0.37633361749638417,
"calibration/batch_uniqueness": 0.2982378782233609,
"calibration/confidence_entropy": 0.4514197585485686,
"calibration/coverage@0%": 0.1,
"calibration/coverage@1%": 0.11197916666666667,
"calibration/coverage@10%": 0.49733327893820717,
"calibration/coverage@15%": 0.6512483681462141,
"calibration/coverage@20%": 0.70863386640557,
"calibration/coverage@25%": 0.8930292645778938,
"calibration/coverage@30%": 0.9065709312445606,
"calibration/coverage@5%": 0.3082381418624891,
"calibration/distribution_entropy_10": 0.5663912873579798,
"calibration/distribution_entropy_100": 0.31968976387481074,
"calibration/ece": 0.11733763054830285,
"calibration/mean_confidence": 0.7128647193211489,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0007812500000000222,
"completions/max_length": 3041.8,
"completions/max_terminated_length": 3041.8,
"completions/mean_length": 869.4628540039063,
"completions/mean_terminated_length": 870.12236328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 180.4,
"epoch": 0.7331730769230769,
"grad_norm": 0.0006279823137447238,
"learning_rate": 4.4170673076923085e-06,
"loss": 0.0003,
"num_tokens": 677379851.0,
"reward": 1.2907153844833374,
"reward_std": 0.11878290176391601,
"rewards/accuracy_reward": 0.7276041626930236,
"rewards/brier_reward": 0.8545937538146973,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.999218738079071,
"rewards/mean_confidence_reward": 0.7068228960037232,
"sampling/batch_mean_priority_error": 0.045256717291975654,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5555555555555556,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.006935930624604225,
"sampling/priority_kl": 0.03000014051795006,
"sampling/priority_scale": 0.7394883930915966,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 4.2540891445241866e-05,
"sampling/prob_min": 1.7829657736001536e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7271999955177307,
"sampling/prompt_draws_total": 21816.0,
"sampling/seen_fraction": 0.5706066608428955,
"sampling/unseen_fraction": 0.4293933391571045,
"signal/accuracy_reward/centered_abs_mean": 0.133094622194767,
"signal/accuracy_reward/group_std_mean": 0.17181747853755952,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0665473110973835,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0665473110973835,
"signal/advantage_abs_mean": 0.09010841697454453,
"signal/advantage_pre_scale_abs_mean": 0.09010841697454453,
"signal/advantage_pre_scale_std": 0.16782027184963227,
"signal/advantage_std": 0.16782027184963227,
"signal/brier_reward/centered_abs_mean": 0.07871444821357727,
"signal/brier_reward/group_std_mean": 0.10398375689983368,
"signal/brier_reward/group_zero_std_frac": 0.18333333134651184,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03935722410678864,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03935722410678864,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0014702690648846328,
"signal/format_reward/group_std_mean": 0.003469961090013385,
"signal/format_reward/group_zero_std_frac": 0.9833333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0007351345324423164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0007351345324423164,
"signal/mean_confidence_reward/centered_abs_mean": 0.06634602397680282,
"signal/mean_confidence_reward/group_std_mean": 0.08630629777908325,
"signal/mean_confidence_reward/group_zero_std_frac": 0.21388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.634602186750271e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.634602186750271e-07,
"step": 305
},
{
"calibration/aurc": 0.1590332219273491,
"calibration/batch_distribution_entropy": 0.5941489175456391,
"calibration/batch_entropy_100bins": 0.32546430178528685,
"calibration/batch_entropy_10bins": 0.5941489175456391,
"calibration/batch_entropy_50bins": 0.38313131009335605,
"calibration/batch_uniqueness": 0.3948182590011388,
"calibration/confidence_entropy": 0.4614033471335378,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.35505738685813754,
"calibration/coverage@15%": 0.4348020017406441,
"calibration/coverage@20%": 0.6619329308093995,
"calibration/coverage@25%": 0.8572644691035685,
"calibration/coverage@30%": 0.9447916666666666,
"calibration/coverage@5%": 0.14203655352480418,
"calibration/distribution_entropy_10": 0.5941489175456391,
"calibration/distribution_entropy_100": 0.32546430178528685,
"calibration/ece": 0.10625652741514358,
"calibration/mean_confidence": 0.7430390557006092,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0009548611111111161,
"completions/max_length": 3017.6,
"completions/max_terminated_length": 3017.6,
"completions/mean_length": 802.1393188476562,
"completions/mean_terminated_length": 802.9186279296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 226.4,
"epoch": 0.7451923076923077,
"grad_norm": 0.0005608039791695774,
"learning_rate": 4.3870192307692315e-06,
"loss": -0.0001,
"num_tokens": 689739984.0,
"reward": 1.2944949626922608,
"reward_std": 0.11501410156488419,
"rewards/accuracy_reward": 0.731249988079071,
"rewards/brier_reward": 0.8586796998977662,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451216697693,
"rewards/mean_confidence_reward": 0.7596788287162781,
"sampling/batch_mean_priority_error": 0.0495483870967742,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5083333333333333,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.0069914465770125386,
"sampling/priority_kl": 0.030000025779008864,
"sampling/priority_scale": 0.7370796263450756,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.266466785338707e-05,
"sampling/prob_min": 1.7870402007247322e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7391999959945679,
"sampling/prompt_draws_total": 22176.0,
"sampling/seen_fraction": 0.5770333290100098,
"sampling/unseen_fraction": 0.42296667098999025,
"signal/accuracy_reward/centered_abs_mean": 0.10629340261220932,
"signal/accuracy_reward/group_std_mean": 0.15068531036376953,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777731418609,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05314670130610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05314670130610466,
"signal/advantage_abs_mean": 0.08016361147165299,
"signal/advantage_pre_scale_abs_mean": 0.08016361147165299,
"signal/advantage_pre_scale_std": 0.16271564960479737,
"signal/advantage_std": 0.16271564960479737,
"signal/brier_reward/centered_abs_mean": 0.07374851629137993,
"signal/brier_reward/group_std_mean": 0.10301677286624908,
"signal/brier_reward/group_zero_std_frac": 0.12222222536802292,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03687425814568997,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03687425814568997,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0018283419660292566,
"signal/format_reward/group_std_mean": 0.004803628288209438,
"signal/format_reward/group_zero_std_frac": 0.975,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0009141709830146283,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0009141709830146283,
"signal/mean_confidence_reward/centered_abs_mean": 0.06200139597058296,
"signal/mean_confidence_reward/group_std_mean": 0.08194329738616943,
"signal/mean_confidence_reward/group_zero_std_frac": 0.14722222685813904,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.200139523571125e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.200139523571125e-07,
"step": 310
},
{
"calibration/aurc": 0.13479263979375694,
"calibration/batch_distribution_entropy": 0.5943594151592482,
"calibration/batch_entropy_100bins": 0.320886936344589,
"calibration/batch_entropy_10bins": 0.5943594151592482,
"calibration/batch_entropy_50bins": 0.37774290955771883,
"calibration/batch_uniqueness": 0.3900287770099487,
"calibration/confidence_entropy": 0.46822462705007134,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.08177083333333333,
"calibration/coverage@10%": 0.516490835492006,
"calibration/coverage@15%": 0.6732616688253394,
"calibration/coverage@20%": 0.7740828103691673,
"calibration/coverage@25%": 0.9017757545931758,
"calibration/coverage@30%": 0.9546136811023622,
"calibration/coverage@5%": 0.153125,
"calibration/distribution_entropy_10": 0.5943594151592482,
"calibration/distribution_entropy_100": 0.320886936344589,
"calibration/ece": 0.09571776724710983,
"calibration/mean_confidence": 0.7316965047490801,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0009548611111111383,
"completions/max_length": 3089.4,
"completions/max_terminated_length": 3089.4,
"completions/mean_length": 820.1656372070313,
"completions/mean_terminated_length": 820.9537719726562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 233.6,
"epoch": 0.7572115384615384,
"grad_norm": 0.000576628081034869,
"learning_rate": 4.356971153846154e-06,
"loss": 0.0013,
"num_tokens": 702283716.0,
"reward": 1.3008342266082764,
"reward_std": 0.12076780349016189,
"rewards/accuracy_reward": 0.7386284828186035,
"rewards/brier_reward": 0.8643272757530213,
"rewards/confidence_one_or_zero": 0.0007812500174622983,
"rewards/format_reward": 0.9986979126930237,
"rewards/mean_confidence_reward": 0.7404079675674439,
"sampling/batch_mean_priority_error": 0.04232408260335557,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5194444444444444,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.0070373106747865675,
"sampling/priority_kl": 0.029999876394867897,
"sampling/priority_scale": 0.7346825778251513,
"sampling/prob_entropy": 10.278953742980956,
"sampling/prob_max": 4.278513515600935e-05,
"sampling/prob_min": 1.7971647685044446e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.751199996471405,
"sampling/prompt_draws_total": 22536.0,
"sampling/seen_fraction": 0.5832133173942566,
"sampling/unseen_fraction": 0.4167866826057434,
"signal/accuracy_reward/centered_abs_mean": 0.12458224892616272,
"signal/accuracy_reward/group_std_mean": 0.1669006824493408,
"signal/accuracy_reward/group_zero_std_frac": 0.5138888895511627,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06229112446308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06229112446308136,
"signal/advantage_abs_mean": 0.08788303434848785,
"signal/advantage_pre_scale_abs_mean": 0.08788303434848785,
"signal/advantage_pre_scale_std": 0.17004837095737457,
"signal/advantage_std": 0.17004837095737457,
"signal/brier_reward/centered_abs_mean": 0.07569783031940461,
"signal/brier_reward/group_std_mean": 0.10299409478902817,
"signal/brier_reward/group_zero_std_frac": 0.08611111342906952,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037848915159702304,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037848915159702304,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0014485677354969084,
"signal/confidence_one_or_zero/group_std_mean": 0.0031183866783976557,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9861111044883728,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.4485675592368353e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.4485675592368353e-08,
"signal/format_reward/centered_abs_mean": 0.0024793836753815414,
"signal/format_reward/group_std_mean": 0.006416239216923714,
"signal/format_reward/group_zero_std_frac": 0.9666666507720947,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0012396918376907707,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0012396918376907707,
"signal/mean_confidence_reward/centered_abs_mean": 0.06360838785767556,
"signal/mean_confidence_reward/group_std_mean": 0.08352760523557663,
"signal/mean_confidence_reward/group_zero_std_frac": 0.11388889104127883,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.360838597174734e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.360838597174734e-07,
"step": 315
},
{
"calibration/aurc": 0.13073502348042956,
"calibration/batch_distribution_entropy": 0.5765002360538408,
"calibration/batch_entropy_100bins": 0.31331468009648444,
"calibration/batch_entropy_10bins": 0.5765002360538408,
"calibration/batch_entropy_50bins": 0.3688289720205298,
"calibration/batch_uniqueness": 0.3522939916000804,
"calibration/confidence_entropy": 0.43345649050027235,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5114032689477306,
"calibration/coverage@15%": 0.7500894030889598,
"calibration/coverage@20%": 0.7834609126875635,
"calibration/coverage@25%": 0.8748218559617058,
"calibration/coverage@30%": 0.9118513381201044,
"calibration/coverage@5%": 0.16605744125326372,
"calibration/distribution_entropy_10": 0.5765002360538408,
"calibration/distribution_entropy_100": 0.31331468009648444,
"calibration/ece": 0.0996640623932032,
"calibration/mean_confidence": 0.7197991089736581,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.0021701388888888838,
"completions/max_length": 3174.0,
"completions/max_terminated_length": 3174.0,
"completions/mean_length": 847.2627563476562,
"completions/mean_terminated_length": 849.2025634765625,
"completions/min_length": 36.4,
"completions/min_terminated_length": 219.8,
"epoch": 0.7692307692307693,
"grad_norm": 0.0004925581743009388,
"learning_rate": 4.326923076923077e-06,
"loss": -0.0007,
"num_tokens": 715118423.0,
"reward": 1.2951796770095825,
"reward_std": 0.11276318728923798,
"rewards/accuracy_reward": 0.7394965291023254,
"rewards/brier_reward": 0.8530182361602783,
"rewards/confidence_one_or_zero": 0.00086805559694767,
"rewards/format_reward": 0.9978298544883728,
"rewards/mean_confidence_reward": 0.7319531440734863,
"sampling/batch_mean_priority_error": 0.05047228719001652,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.49444444444444446,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.007085711508989334,
"sampling/priority_kl": 0.02999986745417118,
"sampling/priority_scale": 0.7324336468940601,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 4.290657088859007e-05,
"sampling/prob_min": 1.807057451514993e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7632000088691712,
"sampling/prompt_draws_total": 22896.0,
"sampling/seen_fraction": 0.5893399953842163,
"sampling/unseen_fraction": 0.41066000461578367,
"signal/accuracy_reward/centered_abs_mean": 0.11623806059360504,
"signal/accuracy_reward/group_std_mean": 0.1584540769457817,
"signal/accuracy_reward/group_zero_std_frac": 0.5249999940395356,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05811903029680252,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05811903029680252,
"signal/advantage_abs_mean": 0.08010389134287835,
"signal/advantage_pre_scale_abs_mean": 0.08010389134287835,
"signal/advantage_pre_scale_std": 0.1603192687034607,
"signal/advantage_std": 0.1603192687034607,
"signal/brier_reward/centered_abs_mean": 0.07845103666186333,
"signal/brier_reward/group_std_mean": 0.10729333162307739,
"signal/brier_reward/group_zero_std_frac": 0.10277777835726738,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.039225518330931665,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.039225518330931665,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0011935763992369174,
"signal/confidence_one_or_zero/group_std_mean": 0.0013081363402307033,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.1935763666315324e-08,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.1935763666315324e-08,
"signal/format_reward/centered_abs_mean": 0.0036295573459938168,
"signal/format_reward/group_std_mean": 0.007045127963647246,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0018147786729969084,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0018147786729969084,
"signal/mean_confidence_reward/centered_abs_mean": 0.068427724391222,
"signal/mean_confidence_reward/group_std_mean": 0.09000243991613388,
"signal/mean_confidence_reward/group_zero_std_frac": 0.11111111417412758,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.842772108939243e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.842772108939243e-07,
"step": 320
},
{
"calibration/aurc": 0.13901186236414578,
"calibration/batch_distribution_entropy": 0.49057468410709976,
"calibration/batch_entropy_100bins": 0.27327159981394644,
"calibration/batch_entropy_10bins": 0.49057468410709976,
"calibration/batch_entropy_50bins": 0.32169090580353693,
"calibration/batch_uniqueness": 0.12938756523973152,
"calibration/confidence_entropy": 0.4157152736968742,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4143706369982548,
"calibration/coverage@15%": 0.6253739664926022,
"calibration/coverage@20%": 0.8666517080069627,
"calibration/coverage@25%": 0.9114583333333334,
"calibration/coverage@30%": 0.971875,
"calibration/coverage@5%": 0.07291666666666666,
"calibration/distribution_entropy_10": 0.49057468410709976,
"calibration/distribution_entropy_100": 0.27327159981394644,
"calibration/ece": 0.10245163746417334,
"calibration/mean_confidence": 0.753003616991101,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.0012152777777777902,
"completions/max_length": 3229.4,
"completions/max_terminated_length": 3229.4,
"completions/mean_length": 874.606689453125,
"completions/mean_terminated_length": 875.6975708007812,
"completions/min_length": 43.0,
"completions/min_terminated_length": 218.4,
"epoch": 0.78125,
"grad_norm": 0.0005906697479076684,
"learning_rate": 4.296875e-06,
"loss": 0.0005,
"num_tokens": 728261700.0,
"reward": 1.2714888095855712,
"reward_std": 0.13772770762443542,
"rewards/accuracy_reward": 0.709375,
"rewards/brier_reward": 0.8348029732704163,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9987847208976746,
"rewards/mean_confidence_reward": 0.7502343893051148,
"sampling/batch_mean_priority_error": 0.0678190865203879,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5083333333333333,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.007151188887655735,
"sampling/priority_kl": 0.030000296607613563,
"sampling/priority_scale": 0.7303502858849242,
"sampling/prob_entropy": 10.278953742980956,
"sampling/prob_max": 4.302758679841645e-05,
"sampling/prob_min": 1.81500847247662e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7751999974250794,
"sampling/prompt_draws_total": 23256.0,
"sampling/seen_fraction": 0.5953666567802429,
"sampling/unseen_fraction": 0.4046333432197571,
"signal/accuracy_reward/centered_abs_mean": 0.1295681416988373,
"signal/accuracy_reward/group_std_mean": 0.18199051022529603,
"signal/accuracy_reward/group_zero_std_frac": 0.4388888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06478407084941865,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06478407084941865,
"signal/advantage_abs_mean": 0.09816538095474243,
"signal/advantage_pre_scale_abs_mean": 0.09816538095474243,
"signal/advantage_pre_scale_std": 0.18509195148944854,
"signal/advantage_std": 0.18509195148944854,
"signal/brier_reward/centered_abs_mean": 0.09397627711296082,
"signal/brier_reward/group_std_mean": 0.12922771871089936,
"signal/brier_reward/group_zero_std_frac": 0.1583333358168602,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04698813855648041,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04698813855648041,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.002289496478624642,
"signal/format_reward/group_std_mean": 0.005327311297878623,
"signal/format_reward/group_zero_std_frac": 0.975,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001144748239312321,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001144748239312321,
"signal/mean_confidence_reward/centered_abs_mean": 0.06724716573953629,
"signal/mean_confidence_reward/group_std_mean": 0.09011977612972259,
"signal/mean_confidence_reward/group_zero_std_frac": 0.19722222536802292,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.724716740791337e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.724716740791337e-07,
"step": 325
},
{
"calibration/aurc": 0.15108210556594276,
"calibration/batch_distribution_entropy": 0.4955656129923048,
"calibration/batch_entropy_100bins": 0.27147566325949657,
"calibration/batch_entropy_10bins": 0.4955656129923048,
"calibration/batch_entropy_50bins": 0.3195767583496484,
"calibration/batch_uniqueness": 0.0848271456646099,
"calibration/confidence_entropy": 0.40205932981181736,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.29648954930078053,
"calibration/coverage@15%": 0.6091366724536245,
"calibration/coverage@20%": 0.7798693748034939,
"calibration/coverage@25%": 0.8649695387293299,
"calibration/coverage@30%": 0.9140244234116622,
"calibration/coverage@5%": 0.09973890339425587,
"calibration/distribution_entropy_10": 0.4955656129923048,
"calibration/distribution_entropy_100": 0.27147566325949657,
"calibration/ece": 0.10098273481607048,
"calibration/mean_confidence": 0.7487735536592258,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0004340277777777901,
"completions/max_length": 3020.2,
"completions/max_terminated_length": 3020.2,
"completions/mean_length": 851.8645874023438,
"completions/mean_terminated_length": 852.2350708007813,
"completions/min_length": 49.0,
"completions/min_terminated_length": 218.0,
"epoch": 0.7932692307692307,
"grad_norm": 0.0005713775753974915,
"learning_rate": 4.266826923076923e-06,
"loss": 0.0,
"num_tokens": 741202636.0,
"reward": 1.263979959487915,
"reward_std": 0.12538683861494065,
"rewards/accuracy_reward": 0.6946180701255799,
"rewards/brier_reward": 0.8337604522705078,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9995659589767456,
"rewards/mean_confidence_reward": 0.7758159756660461,
"sampling/batch_mean_priority_error": 0.07355097726254482,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.49444444444444435,
"sampling/error_ema_max": 0.14215169847011566,
"sampling/error_ema_mean": 0.007231994159519672,
"sampling/priority_kl": 0.0299999613314867,
"sampling/priority_scale": 0.7283593594795092,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 4.314823090680875e-05,
"sampling/prob_min": 1.8233119772048666e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7871999979019165,
"sampling/prompt_draws_total": 23616.0,
"sampling/seen_fraction": 0.6012866616249084,
"sampling/unseen_fraction": 0.39871333837509154,
"signal/accuracy_reward/centered_abs_mean": 0.11459418386220932,
"signal/accuracy_reward/group_std_mean": 0.15884294211864472,
"signal/accuracy_reward/group_zero_std_frac": 0.522222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05729709193110466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05729709193110466,
"signal/advantage_abs_mean": 0.09014993458986283,
"signal/advantage_pre_scale_abs_mean": 0.09014993458986283,
"signal/advantage_pre_scale_std": 0.17779311537742615,
"signal/advantage_std": 0.17779311537742615,
"signal/brier_reward/centered_abs_mean": 0.08623187392950057,
"signal/brier_reward/group_std_mean": 0.11791259646415711,
"signal/brier_reward/group_zero_std_frac": 0.24722222089767457,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04311593696475029,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04311593696475029,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0008409288129769266,
"signal/format_reward/group_std_mean": 0.0024552317336201668,
"signal/format_reward/group_zero_std_frac": 0.9861111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0004204644064884633,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0004204644064884633,
"signal/mean_confidence_reward/centered_abs_mean": 0.05817870870232582,
"signal/mean_confidence_reward/group_std_mean": 0.07863243967294693,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3000000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.817870373903134e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.817870373903134e-07,
"step": 330
},
{
"calibration/aurc": 0.11525137357263186,
"calibration/batch_distribution_entropy": 0.5396938674173277,
"calibration/batch_entropy_100bins": 0.2985571907929471,
"calibration/batch_entropy_10bins": 0.5396938674173277,
"calibration/batch_entropy_50bins": 0.3514566943865821,
"calibration/batch_uniqueness": 0.2981337965761003,
"calibration/confidence_entropy": 0.45188166379727307,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.07864583333333333,
"calibration/coverage@10%": 0.5964172479255806,
"calibration/coverage@15%": 0.7365567714242751,
"calibration/coverage@20%": 0.7610576961414205,
"calibration/coverage@25%": 0.794422306672317,
"calibration/coverage@30%": 0.8849982229026834,
"calibration/coverage@5%": 0.442828002610966,
"calibration/distribution_entropy_10": 0.5396938674173277,
"calibration/distribution_entropy_100": 0.2985571907929471,
"calibration/ece": 0.10572534690420521,
"calibration/mean_confidence": 0.7414600043459145,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0005208333333333482,
"completions/max_length": 3505.2,
"completions/max_terminated_length": 3505.2,
"completions/mean_length": 867.2281372070313,
"completions/mean_terminated_length": 867.6787475585937,
"completions/min_length": 60.2,
"completions/min_terminated_length": 218.4,
"epoch": 0.8052884615384616,
"grad_norm": 0.0005073934444226325,
"learning_rate": 4.236778846153847e-06,
"loss": 0.0017,
"num_tokens": 754332176.0,
"reward": 1.2985543012619019,
"reward_std": 0.11553947627544403,
"rewards/accuracy_reward": 0.7366319417953491,
"rewards/brier_reward": 0.8609826326370239,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9994791507720947,
"rewards/mean_confidence_reward": 0.7424305438995361,
"sampling/batch_mean_priority_error": 0.04359199988799283,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.49444444444444446,
"sampling/error_ema_max": 0.15842704474925995,
"sampling/error_ema_mean": 0.0072890852577984335,
"sampling/priority_kl": 0.029999826103448868,
"sampling/priority_scale": 0.7265909253852442,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 4.327336573624052e-05,
"sampling/prob_min": 1.8324282063986174e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.7991999983787537,
"sampling/prompt_draws_total": 23976.0,
"sampling/seen_fraction": 0.607313334941864,
"sampling/unseen_fraction": 0.392686665058136,
"signal/accuracy_reward/centered_abs_mean": 0.11971571147441865,
"signal/accuracy_reward/group_std_mean": 0.1570712596178055,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05985785573720932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05985785573720932,
"signal/advantage_abs_mean": 0.0869685485959053,
"signal/advantage_pre_scale_abs_mean": 0.0869685485959053,
"signal/advantage_pre_scale_std": 0.16786086559295654,
"signal/advantage_std": 0.16786086559295654,
"signal/brier_reward/centered_abs_mean": 0.07740350812673569,
"signal/brier_reward/group_std_mean": 0.10209325551986695,
"signal/brier_reward/group_zero_std_frac": 0.1388888880610466,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.038701754063367844,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.038701754063367844,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0010091145755723118,
"signal/format_reward/group_std_mean": 0.0029462780803442,
"signal/format_reward/group_zero_std_frac": 0.9833333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0005045572877861559,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0005045572877861559,
"signal/mean_confidence_reward/centered_abs_mean": 0.0635205015540123,
"signal/mean_confidence_reward/group_std_mean": 0.08177418410778045,
"signal/mean_confidence_reward/group_zero_std_frac": 0.14444444328546524,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.352049808810988e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.352049808810988e-07,
"step": 335
},
{
"calibration/aurc": 0.12451635772769279,
"calibration/batch_distribution_entropy": 0.6427127041171415,
"calibration/batch_entropy_100bins": 0.3597033637976214,
"calibration/batch_entropy_10bins": 0.6427127041171415,
"calibration/batch_entropy_50bins": 0.42343697991089424,
"calibration/batch_uniqueness": 0.46505167498570865,
"calibration/confidence_entropy": 0.4800151184802828,
"calibration/coverage@0%": 0.1671875,
"calibration/coverage@1%": 0.1671875,
"calibration/coverage@10%": 0.5347398816521537,
"calibration/coverage@15%": 0.6217899470971798,
"calibration/coverage@20%": 0.7536695706829066,
"calibration/coverage@25%": 0.8006331193981563,
"calibration/coverage@30%": 0.8701068237689045,
"calibration/coverage@5%": 0.43518436245494607,
"calibration/distribution_entropy_10": 0.6427127041171415,
"calibration/distribution_entropy_100": 0.3597033637976214,
"calibration/ece": 0.14596912037100324,
"calibration/mean_confidence": 0.7001734506787146,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.0009548611111111161,
"completions/max_length": 3395.2,
"completions/max_terminated_length": 3395.2,
"completions/mean_length": 935.81875,
"completions/mean_terminated_length": 936.7029907226563,
"completions/min_length": 64.6,
"completions/min_terminated_length": 292.2,
"epoch": 0.8173076923076923,
"grad_norm": 0.0005600933800451458,
"learning_rate": 4.20673076923077e-06,
"loss": 0.002,
"num_tokens": 768201608.0,
"reward": 1.278798723220825,
"reward_std": 0.10951859503984451,
"rewards/accuracy_reward": 0.7026041626930237,
"rewards/brier_reward": 0.8559340238571167,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451335906982,
"rewards/mean_confidence_reward": 0.7040451526641845,
"sampling/batch_mean_priority_error": 0.038162505133661886,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.48611111111111105,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007329526357352734,
"sampling/priority_kl": 0.02999986633658409,
"sampling/priority_scale": 0.7247206031577662,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 4.33936613262631e-05,
"sampling/prob_min": 1.8415910381008872e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8111999988555908,
"sampling/prompt_draws_total": 24336.0,
"sampling/seen_fraction": 0.6130200028419495,
"sampling/unseen_fraction": 0.38697999715805054,
"signal/accuracy_reward/centered_abs_mean": 0.11417100727558135,
"signal/accuracy_reward/group_std_mean": 0.1579512119293213,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444596767426,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05708550363779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05708550363779068,
"signal/advantage_abs_mean": 0.0791636049747467,
"signal/advantage_pre_scale_abs_mean": 0.0791636049747467,
"signal/advantage_pre_scale_std": 0.15533441305160522,
"signal/advantage_std": 0.15533441305160522,
"signal/brier_reward/centered_abs_mean": 0.0714503824710846,
"signal/brier_reward/group_std_mean": 0.09577226787805557,
"signal/brier_reward/group_zero_std_frac": 0.12222222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0357251912355423,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0357251912355423,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0018391926889307797,
"signal/format_reward/group_std_mean": 0.0051025690510869024,
"signal/format_reward/group_zero_std_frac": 0.9722222208976745,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0009195963444653899,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0009195963444653899,
"signal/mean_confidence_reward/centered_abs_mean": 0.06465005502104759,
"signal/mean_confidence_reward/group_std_mean": 0.0848843514919281,
"signal/mean_confidence_reward/group_zero_std_frac": 0.13055555820465087,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.465005753852893e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.465005753852893e-07,
"step": 340
},
{
"calibration/aurc": 0.09457681193417841,
"calibration/batch_distribution_entropy": 0.7022329895287385,
"calibration/batch_entropy_100bins": 0.40857958618301726,
"calibration/batch_entropy_10bins": 0.7022329895287385,
"calibration/batch_entropy_50bins": 0.4809732780923296,
"calibration/batch_uniqueness": 0.6365140520714129,
"calibration/confidence_entropy": 0.5093651656548829,
"calibration/coverage@0%": 0.33503605820746946,
"calibration/coverage@1%": 0.36760606769325693,
"calibration/coverage@10%": 0.5745122571916504,
"calibration/coverage@15%": 0.6721027239758789,
"calibration/coverage@20%": 0.9057291666666668,
"calibration/coverage@25%": 0.9255208333333332,
"calibration/coverage@30%": 0.9442708333333332,
"calibration/coverage@5%": 0.5033105497188031,
"calibration/distribution_entropy_10": 0.7022329895287385,
"calibration/distribution_entropy_100": 0.40857958618301726,
"calibration/ece": 0.1802956842027404,
"calibration/mean_confidence": 0.6166906055436112,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002430555555555558,
"completions/max_length": 3666.2,
"completions/max_terminated_length": 3666.2,
"completions/mean_length": 917.8632080078125,
"completions/mean_terminated_length": 920.127734375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 253.2,
"epoch": 0.8293269230769231,
"grad_norm": 0.000359359139110893,
"learning_rate": 4.176682692307693e-06,
"loss": -0.0005,
"num_tokens": 781840704.0,
"reward": 1.2887306213378906,
"reward_std": 0.09829263091087341,
"rewards/accuracy_reward": 0.7401041626930237,
"rewards/brier_reward": 0.8397751927375794,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.997569465637207,
"rewards/mean_confidence_reward": 0.6232031345367431,
"sampling/batch_mean_priority_error": 0.037444839205341335,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4916666666666667,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007368226815015077,
"sampling/priority_kl": 0.029999981075525282,
"sampling/priority_scale": 0.7232257783180103,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 4.352282339823432e-05,
"sampling/prob_min": 1.8503407045500352e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.823200011253357,
"sampling/prompt_draws_total": 24696.0,
"sampling/seen_fraction": 0.6190266728401184,
"sampling/unseen_fraction": 0.3809733271598816,
"signal/accuracy_reward/centered_abs_mean": 0.12099609225988388,
"signal/accuracy_reward/group_std_mean": 0.16088730096817017,
"signal/accuracy_reward/group_zero_std_frac": 0.5361111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06049804612994194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06049804612994194,
"signal/advantage_abs_mean": 0.07144818976521491,
"signal/advantage_pre_scale_abs_mean": 0.07144818976521491,
"signal/advantage_pre_scale_std": 0.1398210719227791,
"signal/advantage_std": 0.1398210719227791,
"signal/brier_reward/centered_abs_mean": 0.06847175061702729,
"signal/brier_reward/group_std_mean": 0.09024935513734818,
"signal/brier_reward/group_zero_std_frac": 0.09444444626569748,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034235875308513644,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034235875308513644,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004589843703433872,
"signal/format_reward/group_std_mean": 0.010953563451766967,
"signal/format_reward/group_zero_std_frac": 0.9472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002294921851716936,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002294921851716936,
"signal/mean_confidence_reward/centered_abs_mean": 0.06193251013755798,
"signal/mean_confidence_reward/group_std_mean": 0.08066735863685608,
"signal/mean_confidence_reward/group_zero_std_frac": 0.10555555671453476,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.193250669639383e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.193250669639383e-07,
"step": 345
},
{
"calibration/aurc": 0.15021197683923637,
"calibration/batch_distribution_entropy": 0.6270485825789845,
"calibration/batch_entropy_100bins": 0.3587935523162376,
"calibration/batch_entropy_10bins": 0.6270485825789845,
"calibration/batch_entropy_50bins": 0.42236596455563563,
"calibration/batch_uniqueness": 0.46232869790464354,
"calibration/confidence_entropy": 0.4789339090468678,
"calibration/coverage@0%": 0.05759162303664922,
"calibration/coverage@1%": 0.10575916230366492,
"calibration/coverage@10%": 0.34087601999905665,
"calibration/coverage@15%": 0.49549831081081075,
"calibration/coverage@20%": 0.7307291666666667,
"calibration/coverage@25%": 0.8890625,
"calibration/coverage@30%": 0.9,
"calibration/coverage@5%": 0.22252440922597988,
"calibration/distribution_entropy_10": 0.6270485825789845,
"calibration/distribution_entropy_100": 0.3587935523162376,
"calibration/ece": 0.11578408889910854,
"calibration/mean_confidence": 0.6800286469388237,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.00321180555555558,
"completions/max_length": 3813.4,
"completions/max_terminated_length": 3813.4,
"completions/mean_length": 950.54541015625,
"completions/mean_terminated_length": 953.6885009765625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 270.0,
"epoch": 0.8413461538461539,
"grad_norm": 0.0005441403482109308,
"learning_rate": 4.146634615384616e-06,
"loss": -0.0012,
"num_tokens": 795912747.0,
"reward": 1.2941713333129883,
"reward_std": 0.11140554696321488,
"rewards/accuracy_reward": 0.7353298664093018,
"rewards/brier_reward": 0.8562109470367432,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967881917953492,
"rewards/mean_confidence_reward": 0.6828558921813965,
"sampling/batch_mean_priority_error": 0.04163396657615974,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4722222222222222,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007407169230282307,
"sampling/priority_kl": 0.03000009208917618,
"sampling/priority_scale": 0.7217172205680982,
"sampling/prob_entropy": 10.278950309753418,
"sampling/prob_max": 4.364752385299653e-05,
"sampling/prob_min": 1.8589558385428974e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8351999878883362,
"sampling/prompt_draws_total": 25056.0,
"sampling/seen_fraction": 0.6247599959373474,
"sampling/unseen_fraction": 0.37524000406265257,
"signal/accuracy_reward/centered_abs_mean": 0.12787000387907027,
"signal/accuracy_reward/group_std_mean": 0.1749707505106926,
"signal/accuracy_reward/group_zero_std_frac": 0.48611111044883726,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06393500193953514,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06393500193953514,
"signal/advantage_abs_mean": 0.0786064624786377,
"signal/advantage_pre_scale_abs_mean": 0.0786064624786377,
"signal/advantage_pre_scale_std": 0.15402153134346008,
"signal/advantage_std": 0.15402153134346008,
"signal/brier_reward/centered_abs_mean": 0.06964456140995026,
"signal/brier_reward/group_std_mean": 0.09630894064903259,
"signal/brier_reward/group_zero_std_frac": 0.16111110914498566,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03482228070497513,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03482228070497513,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005126953125,
"signal/format_reward/group_std_mean": 0.010197317413985729,
"signal/format_reward/group_zero_std_frac": 0.9555555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025634765625,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0025634765625,
"signal/mean_confidence_reward/centered_abs_mean": 0.0615044504404068,
"signal/mean_confidence_reward/group_std_mean": 0.0810565173625946,
"signal/mean_confidence_reward/group_zero_std_frac": 0.18333333767950535,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.150444505692576e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.150444505692576e-07,
"step": 350
},
{
"epoch": 0.8413461538461539,
"eval_calibration/aurc": 0.14071806497163786,
"eval_calibration/batch_distribution_entropy": 0.5927960453163179,
"eval_calibration/batch_entropy_100bins": 0.3351960947265209,
"eval_calibration/batch_entropy_10bins": 0.5927960453163179,
"eval_calibration/batch_entropy_50bins": 0.3945874192846862,
"eval_calibration/batch_uniqueness": 0.32732551984877123,
"eval_calibration/confidence_entropy": 0.44637156937290573,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.551304347826087,
"eval_calibration/coverage@15%": 0.697391304347826,
"eval_calibration/coverage@20%": 0.7521739130434782,
"eval_calibration/coverage@25%": 0.8704347826086957,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5927960453163179,
"eval_calibration/distribution_entropy_100": 0.3351960947265209,
"eval_calibration/ece": 0.044521739130434675,
"eval_calibration/mean_confidence": 0.7302608695652174,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.0017361111111111234,
"eval_completions/max_length": 2879.6666666666665,
"eval_completions/max_terminated_length": 2879.6666666666665,
"eval_completions/mean_length": 956.6055704752604,
"eval_completions/mean_terminated_length": 958.2885335286459,
"eval_completions/min_length": 192.0,
"eval_completions/min_terminated_length": 286.3333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 795912747.0,
"eval_reward": 1.2779152592023213,
"eval_reward_std": 0.3102490504582723,
"eval_rewards/accuracy_reward": 0.7022569477558136,
"eval_rewards/brier_reward": 0.8552951316038767,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9982638955116272,
"eval_rewards/mean_confidence_reward": 0.7289930184682211,
"eval_runtime": 167.9151,
"eval_samples_per_second": 5.955,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4070638070503871,
"eval_signal/accuracy_reward/group_std_mean": 0.45701082050800323,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20353190352519354,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20353190352519354,
"eval_signal/advantage_abs_mean": 0.2656337320804596,
"eval_signal/advantage_pre_scale_abs_mean": 0.2656337320804596,
"eval_signal/advantage_pre_scale_std": 0.3075862228870392,
"eval_signal/advantage_std": 0.3075862228870392,
"eval_signal/brier_reward/centered_abs_mean": 0.16687446584304175,
"eval_signal/brier_reward/group_std_mean": 0.22345413515965143,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08343723292152087,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08343723292152087,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/group_std_mean": 0.009820927555362383,
"eval_signal/format_reward/group_zero_std_frac": 0.944444457689921,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.19890405982732773,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2373119369149208,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9890404511594775e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9890404511594775e-06,
"eval_steps_per_second": 0.036,
"step": 350
},
{
"epoch": 0.8413461538461539,
"step": 350,
"train_probe_calibration/aurc": 0.12366766369114877,
"train_probe_calibration/batch_distribution_entropy": 0.5780326992172357,
"train_probe_calibration/batch_entropy_100bins": 0.32548332021851784,
"train_probe_calibration/batch_entropy_10bins": 0.5780326992172357,
"train_probe_calibration/batch_entropy_50bins": 0.3831536982852399,
"train_probe_calibration/batch_uniqueness": 0.29336063220676245,
"train_probe_calibration/confidence_entropy": 0.44255715769472015,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.5683202785030461,
"train_probe_calibration/coverage@15%": 0.7745865970409052,
"train_probe_calibration/coverage@20%": 0.8772845953002611,
"train_probe_calibration/coverage@25%": 0.9530026109660574,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5780326992172357,
"train_probe_calibration/distribution_entropy_100": 0.32548332021851784,
"train_probe_calibration/ece": 0.0268929503916449,
"train_probe_calibration/mean_confidence": 0.7407310704960837,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.002604166666666685,
"train_probe_completions/max_length": 2846.6666666666665,
"train_probe_completions/max_terminated_length": 2846.6666666666665,
"train_probe_completions/mean_length": 952.1272786458334,
"train_probe_completions/mean_terminated_length": 954.6122334798177,
"train_probe_completions/min_length": 120.83333333333333,
"train_probe_completions/min_terminated_length": 213.16666666666666,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 795912747.0,
"train_probe_reward": 1.2993086377779643,
"train_probe_reward_std": 0.3012443433205287,
"train_probe_rewards/accuracy_reward": 0.7413194378217062,
"train_probe_rewards/brier_reward": 0.8598871727784475,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9973958432674408,
"train_probe_rewards/mean_confidence_reward": 0.7388020753860474,
"train_probe_runtime": 184.2927,
"train_probe_samples_per_second": 5.426,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3741319427887599,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4373552103837331,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18706597139437994,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18706597139437994,
"train_probe_signal/advantage_abs_mean": 0.2482676605383555,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2482676605383555,
"train_probe_signal/advantage_pre_scale_std": 0.29956333835919696,
"train_probe_signal/advantage_std": 0.29956333835919696,
"train_probe_signal/brier_reward/centered_abs_mean": 0.16151477148135504,
"train_probe_signal/brier_reward/group_std_mean": 0.22173232833544412,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08075738574067752,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08075738574067752,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.0050455727614462376,
"train_probe_signal/format_reward/group_std_mean": 0.014731391333043575,
"train_probe_signal/format_reward/group_zero_std_frac": 0.9166666865348816,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025227863807231188,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0025227863807231188,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1901095782717069,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.22979472080866495,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9010956862075545e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9010956862075545e-06,
"train_probe_steps_per_second": 0.033
},
{
"calibration/aurc": 0.17479328506484096,
"calibration/batch_distribution_entropy": 0.49078581596270715,
"calibration/batch_entropy_100bins": 0.2738594645868757,
"calibration/batch_entropy_10bins": 0.49078581596270715,
"calibration/batch_entropy_50bins": 0.32238293070265683,
"calibration/batch_uniqueness": 0.10383284877673402,
"calibration/confidence_entropy": 0.42451245543918326,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.22788842425787253,
"calibration/coverage@15%": 0.4377930601013447,
"calibration/coverage@20%": 0.6457713541939623,
"calibration/coverage@25%": 0.6966080957945986,
"calibration/coverage@30%": 0.924549551637301,
"calibration/coverage@5%": 0.10757180156657964,
"calibration/distribution_entropy_10": 0.49078581596270715,
"calibration/distribution_entropy_100": 0.2738594645868757,
"calibration/ece": 0.1168297092438673,
"calibration/mean_confidence": 0.7653914820591261,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.0038194444444444643,
"completions/max_length": 3793.2,
"completions/max_terminated_length": 3793.2,
"completions/mean_length": 995.2534790039062,
"completions/mean_terminated_length": 999.286328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 243.8,
"epoch": 0.8533653846153846,
"grad_norm": 0.0005921661504544318,
"learning_rate": 4.116586538461539e-06,
"loss": -0.0041,
"num_tokens": 810478099.0,
"reward": 1.2731634616851806,
"reward_std": 0.13975580632686616,
"rewards/accuracy_reward": 0.7113715291023255,
"rewards/brier_reward": 0.8388463497161865,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99609375,
"rewards/mean_confidence_reward": 0.7715624809265137,
"sampling/batch_mean_priority_error": 0.06671788019592899,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5166666666666666,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007469820789992809,
"sampling/priority_kl": 0.030000051110982896,
"sampling/priority_scale": 0.7204637587303295,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 4.3779354746220635e-05,
"sampling/prob_min": 1.8673363229027018e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8472000002861023,
"sampling/prompt_draws_total": 25416.0,
"sampling/seen_fraction": 0.6306733250617981,
"sampling/unseen_fraction": 0.3693266749382019,
"signal/accuracy_reward/centered_abs_mean": 0.13954535275697708,
"signal/accuracy_reward/group_std_mean": 0.18298389613628388,
"signal/accuracy_reward/group_zero_std_frac": 0.4777777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06977267637848854,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06977267637848854,
"signal/advantage_abs_mean": 0.10368146002292633,
"signal/advantage_pre_scale_abs_mean": 0.10368146002292633,
"signal/advantage_pre_scale_std": 0.1949189454317093,
"signal/advantage_std": 0.1949189454317093,
"signal/brier_reward/centered_abs_mean": 0.08842111974954606,
"signal/brier_reward/group_std_mean": 0.11987781524658203,
"signal/brier_reward/group_zero_std_frac": 0.32500000298023224,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04421055987477303,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04421055987477303,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007123480970039964,
"signal/format_reward/group_std_mean": 0.014730273373425007,
"signal/format_reward/group_zero_std_frac": 0.9361111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003561740485019982,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003561740485019982,
"signal/mean_confidence_reward/centered_abs_mean": 0.05848633646965027,
"signal/mean_confidence_reward/group_std_mean": 0.07816922888159752,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3833333432674408,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.848633236382739e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.848633236382739e-07,
"step": 355
},
{
"calibration/aurc": 0.14227408809205797,
"calibration/batch_distribution_entropy": 0.3833968902727182,
"calibration/batch_entropy_100bins": 0.20720136062897182,
"calibration/batch_entropy_10bins": 0.3833968902727182,
"calibration/batch_entropy_50bins": 0.24391408924249838,
"calibration/batch_uniqueness": -0.17578774698721988,
"calibration/confidence_entropy": 0.39237565439323796,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5404726936466492,
"calibration/coverage@15%": 0.5415143603133159,
"calibration/coverage@20%": 0.7298369688123187,
"calibration/coverage@25%": 0.9333810116871298,
"calibration/coverage@30%": 0.9884514435695537,
"calibration/coverage@5%": 0.184375,
"calibration/distribution_entropy_10": 0.3833968902727182,
"calibration/distribution_entropy_100": 0.20720136062897182,
"calibration/ece": 0.09318675761811303,
"calibration/mean_confidence": 0.822795640520307,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.005642361111111116,
"completions/max_length": 3793.2,
"completions/max_terminated_length": 3793.2,
"completions/mean_length": 953.820751953125,
"completions/mean_terminated_length": 959.3001342773438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 260.6,
"epoch": 0.8653846153846154,
"grad_norm": 0.0006347987800836563,
"learning_rate": 4.086538461538462e-06,
"loss": -0.0045,
"num_tokens": 824565474.0,
"reward": 1.2802110195159913,
"reward_std": 0.12795323729515076,
"rewards/accuracy_reward": 0.7278645873069763,
"rewards/brier_reward": 0.8384444713592529,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9940972208976746,
"rewards/mean_confidence_reward": 0.7921961784362793,
"sampling/batch_mean_priority_error": 0.07182012144971406,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.47777777777777775,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007543521653860808,
"sampling/priority_kl": 0.029999841749668122,
"sampling/priority_scale": 0.7193483173148707,
"sampling/prob_entropy": 10.278953170776367,
"sampling/prob_max": 4.391378824948333e-05,
"sampling/prob_min": 1.8755430210148917e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8592000007629395,
"sampling/prompt_draws_total": 25776.0,
"sampling/seen_fraction": 0.6365933299064637,
"sampling/unseen_fraction": 0.3634066700935364,
"signal/accuracy_reward/centered_abs_mean": 0.11312391310930252,
"signal/accuracy_reward/group_std_mean": 0.15549919307231902,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05656195655465126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05656195655465126,
"signal/advantage_abs_mean": 0.091581329703331,
"signal/advantage_pre_scale_abs_mean": 0.091581329703331,
"signal/advantage_pre_scale_std": 0.1872180163860321,
"signal/advantage_std": 0.1872180163860321,
"signal/brier_reward/centered_abs_mean": 0.08145019859075546,
"signal/brier_reward/group_std_mean": 0.11254193782806396,
"signal/brier_reward/group_zero_std_frac": 0.3555555582046509,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04072509929537773,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04072509929537773,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009537760308012367,
"signal/format_reward/group_std_mean": 0.01679510176181793,
"signal/format_reward/group_zero_std_frac": 0.9333333611488343,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004768880154006183,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004768880154006183,
"signal/mean_confidence_reward/centered_abs_mean": 0.053370780497789386,
"signal/mean_confidence_reward/group_std_mean": 0.0711805447936058,
"signal/mean_confidence_reward/group_zero_std_frac": 0.44722222685813906,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.337078050615673e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.337078050615673e-07,
"step": 360
},
{
"calibration/aurc": 0.140028992425538,
"calibration/batch_distribution_entropy": 0.3930293886864735,
"calibration/batch_entropy_100bins": 0.2175984067067936,
"calibration/batch_entropy_10bins": 0.3930293886864735,
"calibration/batch_entropy_50bins": 0.2561533236625141,
"calibration/batch_uniqueness": -0.08713060293914288,
"calibration/confidence_entropy": 0.40117444762509863,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4989583333333333,
"calibration/coverage@15%": 0.6760416666666667,
"calibration/coverage@20%": 0.7057291666666666,
"calibration/coverage@25%": 0.7270833333333333,
"calibration/coverage@30%": 0.9098169604003481,
"calibration/coverage@5%": 0.2635416666666667,
"calibration/distribution_entropy_10": 0.3930293886864735,
"calibration/distribution_entropy_100": 0.2175984067067936,
"calibration/ece": 0.11969225957354215,
"calibration/mean_confidence": 0.7895940763707573,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.004166666666666674,
"completions/max_length": 3381.4,
"completions/max_terminated_length": 3381.4,
"completions/mean_length": 917.1393310546875,
"completions/mean_terminated_length": 921.1011962890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 236.0,
"epoch": 0.8774038461538461,
"grad_norm": 0.000542930094525218,
"learning_rate": 4.0564903846153846e-06,
"loss": -0.0015,
"num_tokens": 838224583.0,
"reward": 1.2964860677719117,
"reward_std": 0.10884315818548203,
"rewards/accuracy_reward": 0.7424479007720948,
"rewards/brier_reward": 0.8546753764152527,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9958333373069763,
"rewards/mean_confidence_reward": 0.7798090219497681,
"sampling/batch_mean_priority_error": 0.060249924041115656,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.48888888888888893,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007615257333964109,
"sampling/priority_kl": 0.030000029504299162,
"sampling/priority_scale": 0.7183378756279126,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 4.404875508043915e-05,
"sampling/prob_min": 1.8835568334907292e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8712000012397766,
"sampling/prompt_draws_total": 26136.0,
"sampling/seen_fraction": 0.6424266695976257,
"sampling/unseen_fraction": 0.35757333040237427,
"signal/accuracy_reward/centered_abs_mean": 0.09465603232383728,
"signal/accuracy_reward/group_std_mean": 0.13468586504459382,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333253860473,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04732801616191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04732801616191864,
"signal/advantage_abs_mean": 0.07618805170059204,
"signal/advantage_pre_scale_abs_mean": 0.07618805170059204,
"signal/advantage_pre_scale_std": 0.16468151807785034,
"signal/advantage_std": 0.16468151807785034,
"signal/brier_reward/centered_abs_mean": 0.07008838132023812,
"signal/brier_reward/group_std_mean": 0.0985906183719635,
"signal/brier_reward/group_zero_std_frac": 0.35,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03504419066011906,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03504419066011906,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006065538234543056,
"signal/format_reward/group_std_mean": 0.010566732659935952,
"signal/format_reward/group_zero_std_frac": 0.9583333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003032769117271528,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003032769117271528,
"signal/mean_confidence_reward/centered_abs_mean": 0.05134060978889465,
"signal/mean_confidence_reward/group_std_mean": 0.06885762214660644,
"signal/mean_confidence_reward/group_zero_std_frac": 0.41666666865348817,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.134061041189852e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.134061041189852e-07,
"step": 365
},
{
"calibration/aurc": 0.1258378555398733,
"calibration/batch_distribution_entropy": 0.46751652925453246,
"calibration/batch_entropy_100bins": 0.2515257216364021,
"calibration/batch_entropy_10bins": 0.46751652925453246,
"calibration/batch_entropy_50bins": 0.29609200985829276,
"calibration/batch_uniqueness": 0.07823023946937402,
"calibration/confidence_entropy": 0.4175166052930705,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.10628272251308901,
"calibration/coverage@10%": 0.285970222513089,
"calibration/coverage@15%": 0.675792700721104,
"calibration/coverage@20%": 0.8834558556894002,
"calibration/coverage@25%": 0.971641380826737,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.27190772251308903,
"calibration/distribution_entropy_10": 0.46751652925453246,
"calibration/distribution_entropy_100": 0.2515257216364021,
"calibration/ece": 0.09827271825945014,
"calibration/mean_confidence": 0.7923587312701746,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.003385416666666652,
"completions/max_length": 3863.6,
"completions/max_terminated_length": 3863.6,
"completions/mean_length": 915.6849975585938,
"completions/mean_terminated_length": 918.7907958984375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 253.0,
"epoch": 0.8894230769230769,
"grad_norm": 0.0006251902668736875,
"learning_rate": 4.026442307692308e-06,
"loss": -0.0026,
"num_tokens": 851860378.0,
"reward": 1.2964604139328002,
"reward_std": 0.12138742804527283,
"rewards/accuracy_reward": 0.7434895753860473,
"rewards/brier_reward": 0.852888023853302,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9965277791023255,
"rewards/mean_confidence_reward": 0.7711111187934876,
"sampling/batch_mean_priority_error": 0.056980975083554763,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.5111111111111111,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007679272443056107,
"sampling/priority_kl": 0.029999953880906104,
"sampling/priority_scale": 0.7177827418083325,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 4.419601973495446e-05,
"sampling/prob_min": 1.891095052997116e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.8832000017166137,
"sampling/prompt_draws_total": 26496.0,
"sampling/seen_fraction": 0.6486466765403748,
"sampling/unseen_fraction": 0.3513533234596252,
"signal/accuracy_reward/centered_abs_mean": 0.118115234375,
"signal/accuracy_reward/group_std_mean": 0.15658425986766816,
"signal/accuracy_reward/group_zero_std_frac": 0.55277778506279,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0590576171875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0590576171875,
"signal/advantage_abs_mean": 0.08963771909475327,
"signal/advantage_pre_scale_abs_mean": 0.08963771909475327,
"signal/advantage_pre_scale_std": 0.18096743822097777,
"signal/advantage_std": 0.18096743822097777,
"signal/brier_reward/centered_abs_mean": 0.07543071657419205,
"signal/brier_reward/group_std_mean": 0.10182640105485916,
"signal/brier_reward/group_zero_std_frac": 0.33888888657093047,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037715358287096025,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037715358287096025,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006477864552289247,
"signal/format_reward/group_std_mean": 0.013997751846909523,
"signal/format_reward/group_zero_std_frac": 0.9388888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0032389322761446236,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0032389322761446236,
"signal/mean_confidence_reward/centered_abs_mean": 0.05826389938592911,
"signal/mean_confidence_reward/group_std_mean": 0.07576072439551354,
"signal/mean_confidence_reward/group_zero_std_frac": 0.38333333730697633,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.826389838148316e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.826389838148316e-07,
"step": 370
},
{
"calibration/aurc": 0.09348026206368412,
"calibration/batch_distribution_entropy": 0.588975726328109,
"calibration/batch_entropy_100bins": 0.3265009176487136,
"calibration/batch_entropy_10bins": 0.588975726328109,
"calibration/batch_entropy_50bins": 0.3843515975154779,
"calibration/batch_uniqueness": 0.40700246648161914,
"calibration/confidence_entropy": 0.4716141812235014,
"calibration/coverage@0%": 0.09712793733681462,
"calibration/coverage@1%": 0.2689295039164491,
"calibration/coverage@10%": 0.6583557606432182,
"calibration/coverage@15%": 0.7130799773534009,
"calibration/coverage@20%": 0.803554878132134,
"calibration/coverage@25%": 0.8672546039351314,
"calibration/coverage@30%": 0.9305590323819029,
"calibration/coverage@5%": 0.6009145073795106,
"calibration/distribution_entropy_10": 0.588975726328109,
"calibration/distribution_entropy_100": 0.3265009176487136,
"calibration/ece": 0.12552451354808858,
"calibration/mean_confidence": 0.7361619589251296,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.0028645833333333258,
"completions/max_length": 3791.2,
"completions/max_terminated_length": 3791.2,
"completions/mean_length": 902.0409790039063,
"completions/mean_terminated_length": 904.7165649414062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 196.6,
"epoch": 0.9014423076923077,
"grad_norm": 0.0006674884352833033,
"learning_rate": 3.996394230769231e-06,
"loss": -0.0005,
"num_tokens": 865337586.0,
"reward": 1.297727370262146,
"reward_std": 0.11251561045646667,
"rewards/accuracy_reward": 0.7408854126930237,
"rewards/brier_reward": 0.8575060844421387,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9970486044883728,
"rewards/mean_confidence_reward": 0.7293142437934875,
"sampling/batch_mean_priority_error": 0.04492750631761033,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4222222222222222,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007732342928647995,
"sampling/priority_kl": 0.030000034347176553,
"sampling/priority_scale": 0.7166644394630566,
"sampling/prob_entropy": 10.278953170776367,
"sampling/prob_max": 4.432367641129531e-05,
"sampling/prob_min": 1.8990676107932813e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.895199990272522,
"sampling/prompt_draws_total": 26856.0,
"sampling/seen_fraction": 0.6539800047874451,
"sampling/unseen_fraction": 0.34601999521255494,
"signal/accuracy_reward/centered_abs_mean": 0.12051323503255844,
"signal/accuracy_reward/group_std_mean": 0.16058450937271118,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000178813934,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06025661751627922,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06025661751627922,
"signal/advantage_abs_mean": 0.0811893805861473,
"signal/advantage_pre_scale_abs_mean": 0.0811893805861473,
"signal/advantage_pre_scale_std": 0.16239385306835175,
"signal/advantage_std": 0.16239385306835175,
"signal/brier_reward/centered_abs_mean": 0.06375756785273552,
"signal/brier_reward/group_std_mean": 0.08792267739772797,
"signal/brier_reward/group_zero_std_frac": 0.1777777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03187878392636776,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03187878392636776,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005447048670612275,
"signal/format_reward/group_std_mean": 0.012354053743183612,
"signal/format_reward/group_zero_std_frac": 0.9416666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0027235243353061376,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0027235243353061376,
"signal/mean_confidence_reward/centered_abs_mean": 0.05573513433337211,
"signal/mean_confidence_reward/group_std_mean": 0.07394854128360748,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1972222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.573513362833183e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.573513362833183e-07,
"step": 375
},
{
"calibration/aurc": 0.08495235291893864,
"calibration/batch_distribution_entropy": 0.6155774549758437,
"calibration/batch_entropy_100bins": 0.3351735776080944,
"calibration/batch_entropy_10bins": 0.6155774549758437,
"calibration/batch_entropy_50bins": 0.39456091249719843,
"calibration/batch_uniqueness": 0.44102586574209024,
"calibration/confidence_entropy": 0.5091412586682026,
"calibration/coverage@0%": 0.0203125,
"calibration/coverage@1%": 0.35401708006962573,
"calibration/coverage@10%": 0.6745866402116402,
"calibration/coverage@15%": 0.8016699735449734,
"calibration/coverage@20%": 0.8672949735449735,
"calibration/coverage@25%": 0.8896908068783069,
"calibration/coverage@30%": 0.9447172619047619,
"calibration/coverage@5%": 0.5850586085899402,
"calibration/distribution_entropy_10": 0.6155774549758437,
"calibration/distribution_entropy_100": 0.3351735776080944,
"calibration/ece": 0.128370848702115,
"calibration/mean_confidence": 0.7126485893530606,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.005034722222222255,
"completions/max_length": 3856.2,
"completions/max_terminated_length": 3856.2,
"completions/mean_length": 935.269091796875,
"completions/mean_terminated_length": 940.030419921875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 203.8,
"epoch": 0.9134615384615384,
"grad_norm": 0.0005975854583084583,
"learning_rate": 3.966346153846154e-06,
"loss": -0.004,
"num_tokens": 879227438.0,
"reward": 1.2894989728927613,
"reward_std": 0.10981154143810272,
"rewards/accuracy_reward": 0.7271701335906983,
"rewards/brier_reward": 0.8570225596427917,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9947916626930237,
"rewards/mean_confidence_reward": 0.6774826407432556,
"sampling/batch_mean_priority_error": 0.03030392225748877,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.43888888888888894,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007766260858625173,
"sampling/priority_kl": 0.030000124126672745,
"sampling/priority_scale": 0.7155722082359717,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 4.445084559847601e-05,
"sampling/prob_min": 1.906977740873117e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9072000026702881,
"sampling/prompt_draws_total": 27216.0,
"sampling/seen_fraction": 0.6592066645622253,
"sampling/unseen_fraction": 0.3407933354377747,
"signal/accuracy_reward/centered_abs_mean": 0.1240505650639534,
"signal/accuracy_reward/group_std_mean": 0.16211120784282684,
"signal/accuracy_reward/group_zero_std_frac": 0.544444465637207,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0620252825319767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0620252825319767,
"signal/advantage_abs_mean": 0.07988931685686111,
"signal/advantage_pre_scale_abs_mean": 0.07988931685686111,
"signal/advantage_pre_scale_std": 0.16046163737773894,
"signal/advantage_std": 0.16046163737773894,
"signal/brier_reward/centered_abs_mean": 0.06219109818339348,
"signal/brier_reward/group_std_mean": 0.08544109016656876,
"signal/brier_reward/group_zero_std_frac": 0.14166666567325592,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03109554909169674,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03109554909169674,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009016927040647715,
"signal/format_reward/group_std_mean": 0.016939852014184,
"signal/format_reward/group_zero_std_frac": 0.9305555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004508463520323858,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004508463520323858,
"signal/mean_confidence_reward/centered_abs_mean": 0.05640299767255783,
"signal/mean_confidence_reward/group_std_mean": 0.07368175387382507,
"signal/mean_confidence_reward/group_zero_std_frac": 0.15555555671453475,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.64029949146061e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.64029949146061e-07,
"step": 380
},
{
"calibration/aurc": 0.1405584194396033,
"calibration/batch_distribution_entropy": 0.6175718779005612,
"calibration/batch_entropy_100bins": 0.33658476146835625,
"calibration/batch_entropy_10bins": 0.6175718779005612,
"calibration/batch_entropy_50bins": 0.3962221353047351,
"calibration/batch_uniqueness": 0.4358823734176201,
"calibration/confidence_entropy": 0.49378913317443135,
"calibration/coverage@0%": 0.05416666666666666,
"calibration/coverage@1%": 0.19930191972076788,
"calibration/coverage@10%": 0.3779477530541012,
"calibration/coverage@15%": 0.5394169938917976,
"calibration/coverage@20%": 0.6689811403237553,
"calibration/coverage@25%": 0.7433357939632546,
"calibration/coverage@30%": 0.9380577427821523,
"calibration/coverage@5%": 0.2701352530541012,
"calibration/distribution_entropy_10": 0.6175718779005612,
"calibration/distribution_entropy_100": 0.33658476146835625,
"calibration/ece": 0.09195633176333981,
"calibration/mean_confidence": 0.7355146774642372,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.0032986111111111384,
"completions/max_length": 3793.8,
"completions/max_terminated_length": 3793.8,
"completions/mean_length": 903.2882080078125,
"completions/mean_terminated_length": 906.3167114257812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 214.4,
"epoch": 0.9254807692307693,
"grad_norm": 0.0006492780521512032,
"learning_rate": 3.936298076923077e-06,
"loss": -0.002,
"num_tokens": 892689414.0,
"reward": 1.3000523090362548,
"reward_std": 0.11470878720283509,
"rewards/accuracy_reward": 0.7401041626930237,
"rewards/brier_reward": 0.8632847309112549,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967013955116272,
"rewards/mean_confidence_reward": 0.716371500492096,
"sampling/batch_mean_priority_error": 0.033226227266539954,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666667,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007798460777848959,
"sampling/priority_kl": 0.029999995604157448,
"sampling/priority_scale": 0.7147068679099903,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 4.458282419363968e-05,
"sampling/prob_min": 1.9145980331813917e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9192000031471252,
"sampling/prompt_draws_total": 27576.0,
"sampling/seen_fraction": 0.6645333409309387,
"sampling/unseen_fraction": 0.33546665906906126,
"signal/accuracy_reward/centered_abs_mean": 0.1258029520511627,
"signal/accuracy_reward/group_std_mean": 0.16285853683948517,
"signal/accuracy_reward/group_zero_std_frac": 0.55,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06290147602558135,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06290147602558135,
"signal/advantage_abs_mean": 0.08486108630895614,
"signal/advantage_pre_scale_abs_mean": 0.08486108630895614,
"signal/advantage_pre_scale_std": 0.16529499590396882,
"signal/advantage_std": 0.16529499590396882,
"signal/brier_reward/centered_abs_mean": 0.0634469136595726,
"signal/brier_reward/group_std_mean": 0.08727488517761231,
"signal/brier_reward/group_zero_std_frac": 0.15555555671453475,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0317234568297863,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0317234568297863,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00593532978091389,
"signal/format_reward/group_std_mean": 0.01148662269115448,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002967664890456945,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002967664890456945,
"signal/mean_confidence_reward/centered_abs_mean": 0.057197265326976776,
"signal/mean_confidence_reward/group_std_mean": 0.07460127323865891,
"signal/mean_confidence_reward/group_zero_std_frac": 0.17222222536802292,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.719726232200629e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.719726232200629e-07,
"step": 385
},
{
"calibration/aurc": 0.15444463437730716,
"calibration/batch_distribution_entropy": 0.6379871401540195,
"calibration/batch_entropy_100bins": 0.35082089900099217,
"calibration/batch_entropy_10bins": 0.6379871401540195,
"calibration/batch_entropy_50bins": 0.41298068606937866,
"calibration/batch_uniqueness": 0.45707237434195686,
"calibration/confidence_entropy": 0.47556850046240673,
"calibration/coverage@0%": 0.23986092788038701,
"calibration/coverage@1%": 0.23986092788038701,
"calibration/coverage@10%": 0.34862027264731754,
"calibration/coverage@15%": 0.4109842238346526,
"calibration/coverage@20%": 0.6173785180299032,
"calibration/coverage@25%": 0.735137254589043,
"calibration/coverage@30%": 0.8669522849462366,
"calibration/coverage@5%": 0.34862027264731754,
"calibration/distribution_entropy_10": 0.6379871401540195,
"calibration/distribution_entropy_100": 0.35082089900099217,
"calibration/ece": 0.14569204028002386,
"calibration/mean_confidence": 0.7182665767653985,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.009114583333333348,
"completions/max_length": 3975.0,
"completions/max_terminated_length": 3975.0,
"completions/mean_length": 965.8712768554688,
"completions/mean_terminated_length": 974.8541625976562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 224.4,
"epoch": 0.9375,
"grad_norm": 0.0006160195334814489,
"learning_rate": 3.90625e-06,
"loss": -0.0115,
"num_tokens": 906936155.0,
"reward": 1.2957585334777832,
"reward_std": 0.12280676662921905,
"rewards/accuracy_reward": 0.7367187380790711,
"rewards/brier_reward": 0.8644192695617676,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9903645753860474,
"rewards/mean_confidence_reward": 0.72361980676651,
"sampling/batch_mean_priority_error": 0.036464205832859656,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.47777777777777786,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007832664065063,
"sampling/priority_kl": 0.03000004217028618,
"sampling/priority_scale": 0.714122933219187,
"sampling/prob_entropy": 10.278953552246094,
"sampling/prob_max": 4.4722390157403424e-05,
"sampling/prob_min": 1.9219180467189288e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9312000036239624,
"sampling/prompt_draws_total": 27936.0,
"sampling/seen_fraction": 0.6700466632843017,
"sampling/unseen_fraction": 0.32995333671569826,
"signal/accuracy_reward/centered_abs_mean": 0.11191948801279068,
"signal/accuracy_reward/group_std_mean": 0.15588756203651427,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05595974400639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05595974400639534,
"signal/advantage_abs_mean": 0.08563981354236602,
"signal/advantage_pre_scale_abs_mean": 0.08563981354236602,
"signal/advantage_pre_scale_std": 0.17761370241642,
"signal/advantage_std": 0.17761370241642,
"signal/brier_reward/centered_abs_mean": 0.07111435532569885,
"signal/brier_reward/group_std_mean": 0.098927903175354,
"signal/brier_reward/group_zero_std_frac": 0.23888888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.035557177662849423,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.035557177662849423,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.016693793423473834,
"signal/format_reward/group_std_mean": 0.030034196749329566,
"signal/format_reward/group_zero_std_frac": 0.8833333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008346896711736917,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008346896711736917,
"signal/mean_confidence_reward/centered_abs_mean": 0.06112522184848786,
"signal/mean_confidence_reward/group_std_mean": 0.0802394762635231,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.112521873546938e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.112521873546938e-07,
"step": 390
},
{
"calibration/aurc": 0.14231824250564923,
"calibration/batch_distribution_entropy": 0.5188558679111168,
"calibration/batch_entropy_100bins": 0.27990488901705934,
"calibration/batch_entropy_10bins": 0.5188558679111168,
"calibration/batch_entropy_50bins": 0.3294995065277213,
"calibration/batch_uniqueness": 0.2114929455399162,
"calibration/confidence_entropy": 0.4412506440080336,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4840907854656223,
"calibration/coverage@15%": 0.6611199057036931,
"calibration/coverage@20%": 0.7449531602283396,
"calibration/coverage@25%": 0.7973753280839895,
"calibration/coverage@30%": 0.9729166666666667,
"calibration/coverage@5%": 0.2,
"calibration/distribution_entropy_10": 0.5188558679111168,
"calibration/distribution_entropy_100": 0.27990488901705934,
"calibration/ece": 0.10563489391665462,
"calibration/mean_confidence": 0.7899494168157181,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.012673611111111094,
"completions/max_length": 3903.8,
"completions/max_terminated_length": 3903.8,
"completions/mean_length": 939.616748046875,
"completions/mean_terminated_length": 951.5771362304688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 222.2,
"epoch": 0.9495192307692307,
"grad_norm": 0.000712635344825685,
"learning_rate": 3.876201923076923e-06,
"loss": -0.0117,
"num_tokens": 920847516.0,
"reward": 1.3004497051239015,
"reward_std": 0.13428637385368347,
"rewards/accuracy_reward": 0.7579861164093018,
"rewards/brier_reward": 0.856179702281952,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.98671875,
"rewards/mean_confidence_reward": 0.7419097185134887,
"sampling/batch_mean_priority_error": 0.04323727973120382,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4416666666666666,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007874753326177597,
"sampling/priority_kl": 0.030000117421150208,
"sampling/priority_scale": 0.7136170566780493,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 4.4862589857075366e-05,
"sampling/prob_min": 1.9290969066787512e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9431999921798706,
"sampling/prompt_draws_total": 28296.0,
"sampling/seen_fraction": 0.6754800081253052,
"sampling/unseen_fraction": 0.3245199918746948,
"signal/accuracy_reward/centered_abs_mean": 0.12306857705116273,
"signal/accuracy_reward/group_std_mean": 0.17122169733047485,
"signal/accuracy_reward/group_zero_std_frac": 0.47222223281860354,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06153428852558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06153428852558136,
"signal/advantage_abs_mean": 0.0938779279589653,
"signal/advantage_pre_scale_abs_mean": 0.0938779279589653,
"signal/advantage_pre_scale_std": 0.18882446885108947,
"signal/advantage_std": 0.18882446885108947,
"signal/brier_reward/centered_abs_mean": 0.07441985458135605,
"signal/brier_reward/group_std_mean": 0.10555129051208496,
"signal/brier_reward/group_zero_std_frac": 0.24166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03720992729067803,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03720992729067803,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.016726345475763082,
"signal/format_reward/group_std_mean": 0.028362412378191947,
"signal/format_reward/group_zero_std_frac": 0.8916666746139527,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008363172737881541,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008363172737881541,
"signal/mean_confidence_reward/centered_abs_mean": 0.05717773661017418,
"signal/mean_confidence_reward/group_std_mean": 0.0777185320854187,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444417953491,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.717773547075921e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.717773547075921e-07,
"step": 395
},
{
"calibration/aurc": 0.2103259096874605,
"calibration/batch_distribution_entropy": 0.5207171743631474,
"calibration/batch_entropy_100bins": 0.3006230264744981,
"calibration/batch_entropy_10bins": 0.5207171743631474,
"calibration/batch_entropy_50bins": 0.35388856272596264,
"calibration/batch_uniqueness": 0.18626633909579246,
"calibration/confidence_entropy": 0.45076468811045334,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.1404699738903394,
"calibration/coverage@15%": 0.18580330722367275,
"calibration/coverage@20%": 0.2562019147084421,
"calibration/coverage@25%": 0.7477652478378727,
"calibration/coverage@30%": 0.8831667349726777,
"calibration/coverage@5%": 0.13577023498694515,
"calibration/distribution_entropy_10": 0.5207171743631474,
"calibration/distribution_entropy_100": 0.3006230264744981,
"calibration/ece": 0.1485258476552534,
"calibration/mean_confidence": 0.7193914784120652,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.01822916666666665,
"completions/max_length": 3858.2,
"completions/max_terminated_length": 3858.2,
"completions/mean_length": 969.39208984375,
"completions/mean_terminated_length": 987.3859252929688,
"completions/min_length": 0.0,
"completions/min_terminated_length": 268.2,
"epoch": 0.9615384615384616,
"grad_norm": 0.0005787439295090735,
"learning_rate": 3.846153846153847e-06,
"loss": -0.0206,
"num_tokens": 935088481.0,
"reward": 1.2711968421936035,
"reward_std": 0.13464723378419877,
"rewards/accuracy_reward": 0.7215277791023255,
"rewards/brier_reward": 0.8390807509422302,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9817708373069763,
"rewards/mean_confidence_reward": 0.7178385496139527,
"sampling/batch_mean_priority_error": 0.051486737806012664,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.44722222222222224,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.007927501015365124,
"sampling/priority_kl": 0.029999881982803345,
"sampling/priority_scale": 0.7131928384071216,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 4.5004417916061354e-05,
"sampling/prob_min": 1.9361668091733008e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9551999926567077,
"sampling/prompt_draws_total": 28656.0,
"sampling/seen_fraction": 0.6808599948883056,
"sampling/unseen_fraction": 0.3191400051116943,
"signal/accuracy_reward/centered_abs_mean": 0.12125650942325591,
"signal/accuracy_reward/group_std_mean": 0.1649042248725891,
"signal/accuracy_reward/group_zero_std_frac": 0.5138888895511627,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06062825471162796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06062825471162796,
"signal/advantage_abs_mean": 0.09647919833660126,
"signal/advantage_pre_scale_abs_mean": 0.09647919833660126,
"signal/advantage_pre_scale_std": 0.19579894542694093,
"signal/advantage_std": 0.19579894542694093,
"signal/brier_reward/centered_abs_mean": 0.08560052514076233,
"signal/brier_reward/group_std_mean": 0.11464979350566865,
"signal/brier_reward/group_zero_std_frac": 0.2750000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.042800262570381165,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.042800262570381165,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.02554253526031971,
"signal/format_reward/group_std_mean": 0.039585836976766584,
"signal/format_reward/group_zero_std_frac": 0.8611111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.012771267630159854,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.012771267630159854,
"signal/mean_confidence_reward/centered_abs_mean": 0.06342068463563919,
"signal/mean_confidence_reward/group_std_mean": 0.08286617994308472,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3166666775941849,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.342068104459031e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.342068104459031e-07,
"step": 400
},
{
"epoch": 0.9615384615384616,
"eval_calibration/aurc": 0.1372654401238463,
"eval_calibration/batch_distribution_entropy": 0.6162748041187558,
"eval_calibration/batch_entropy_100bins": 0.35525026458055475,
"eval_calibration/batch_entropy_10bins": 0.6162748041187558,
"eval_calibration/batch_entropy_50bins": 0.4181948635631051,
"eval_calibration/batch_uniqueness": 0.3992356427296172,
"eval_calibration/confidence_entropy": 0.4599076584380884,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5528169014084507,
"eval_calibration/coverage@15%": 0.6575704225352113,
"eval_calibration/coverage@20%": 0.8107394366197183,
"eval_calibration/coverage@25%": 0.8802816901408451,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6162748041187558,
"eval_calibration/distribution_entropy_100": 0.35525026458055475,
"eval_calibration/ece": 0.036355633802816784,
"eval_calibration/mean_confidence": 0.6766725352112677,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.013888888888888914,
"eval_completions/max_length": 3091.8333333333335,
"eval_completions/max_terminated_length": 3091.8333333333335,
"eval_completions/mean_length": 965.8703206380209,
"eval_completions/mean_terminated_length": 979.3125712076823,
"eval_completions/min_length": 72.16666666666667,
"eval_completions/min_terminated_length": 301.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 935088481.0,
"eval_reward": 1.2657662431399028,
"eval_reward_std": 0.3163274774948756,
"eval_rewards/accuracy_reward": 0.703125,
"eval_rewards/brier_reward": 0.8422830005486807,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9861111243565878,
"eval_rewards/mean_confidence_reward": 0.667274276415507,
"eval_runtime": 214.8399,
"eval_samples_per_second": 4.655,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4060329844554265,
"eval_signal/accuracy_reward/group_std_mean": 0.456397940715154,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.20301649222771326,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.20301649222771326,
"eval_signal/advantage_abs_mean": 0.26114028692245483,
"eval_signal/advantage_pre_scale_abs_mean": 0.26114028692245483,
"eval_signal/advantage_pre_scale_std": 0.31408119946718216,
"eval_signal/advantage_std": 0.31408119946718216,
"eval_signal/brier_reward/centered_abs_mean": 0.17281847819685936,
"eval_signal/brier_reward/group_std_mean": 0.23123475164175034,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08640923909842968,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08640923909842968,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.026692708022892475,
"eval_signal/format_reward/group_std_mean": 0.07258860456446807,
"eval_signal/format_reward/group_zero_std_frac": 0.6111111268401146,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.013346354011446238,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.013346354011446238,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.24405381083488464,
"eval_signal/mean_confidence_reward/group_std_mean": 0.278071865439415,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.4405380069462503e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.4405380069462503e-06,
"eval_steps_per_second": 0.028,
"step": 400
},
{
"epoch": 0.9615384615384616,
"step": 400,
"train_probe_calibration/aurc": 0.09306832002660875,
"train_probe_calibration/batch_distribution_entropy": 0.6328398216611376,
"train_probe_calibration/batch_entropy_100bins": 0.3528405852438377,
"train_probe_calibration/batch_entropy_10bins": 0.6328398216611376,
"train_probe_calibration/batch_entropy_50bins": 0.4153582280362068,
"train_probe_calibration/batch_uniqueness": 0.39234035658169547,
"train_probe_calibration/confidence_entropy": 0.4567299333781084,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6769638128861429,
"train_probe_calibration/coverage@15%": 0.8420123565754634,
"train_probe_calibration/coverage@20%": 0.8861429832303619,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.6328398216611376,
"train_probe_calibration/distribution_entropy_100": 0.3528405852438377,
"train_probe_calibration/ece": 0.07458075904677838,
"train_probe_calibration/mean_confidence": 0.6906443071491616,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.02309027777777778,
"train_probe_completions/max_length": 3173.0,
"train_probe_completions/max_terminated_length": 3173.0,
"train_probe_completions/mean_length": 945.7347412109375,
"train_probe_completions/mean_terminated_length": 967.8654988606771,
"train_probe_completions/min_length": 37.5,
"train_probe_completions/min_terminated_length": 231.16666666666666,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 935088481.0,
"train_probe_reward": 1.297945221265157,
"train_probe_reward_std": 0.2997801552216212,
"train_probe_rewards/accuracy_reward": 0.7526041567325592,
"train_probe_rewards/brier_reward": 0.8597656190395355,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9835069378217062,
"train_probe_rewards/mean_confidence_reward": 0.6792534788449606,
"train_probe_runtime": 214.6557,
"train_probe_samples_per_second": 4.659,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3627929737170537,
"train_probe_signal/accuracy_reward/group_std_mean": 0.43074525396029156,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18139648685852686,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18139648685852686,
"train_probe_signal/advantage_abs_mean": 0.23724218209584555,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23724218209584555,
"train_probe_signal/advantage_pre_scale_std": 0.3022302836179733,
"train_probe_signal/advantage_std": 0.3022302836179733,
"train_probe_signal/brier_reward/centered_abs_mean": 0.15757650882005692,
"train_probe_signal/brier_reward/group_std_mean": 0.2167892505725225,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07878825441002846,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07878825441002846,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.031087239272892475,
"train_probe_signal/format_reward/group_std_mean": 0.07184661490221818,
"train_probe_signal/format_reward/group_zero_std_frac": 0.6666666840513548,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.015543619636446238,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.015543619636446238,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.2371364881594976,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.27242950598398846,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.371364795787182e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.371364795787182e-06,
"train_probe_steps_per_second": 0.028
},
{
"calibration/aurc": 0.09160201940281179,
"calibration/batch_distribution_entropy": 0.5302120579669701,
"calibration/batch_entropy_100bins": 0.29462174083381854,
"calibration/batch_entropy_10bins": 0.5302120579669701,
"calibration/batch_entropy_50bins": 0.3468239463697428,
"calibration/batch_uniqueness": 0.19472494727986614,
"calibration/confidence_entropy": 0.4459124291933553,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5100920743816573,
"calibration/coverage@15%": 0.877302995137047,
"calibration/coverage@20%": 0.9473958333333334,
"calibration/coverage@25%": 0.9713541666666666,
"calibration/coverage@30%": 0.9833333333333334,
"calibration/coverage@5%": 0.4398577063679068,
"calibration/distribution_entropy_10": 0.5302120579669701,
"calibration/distribution_entropy_100": 0.29462174083381854,
"calibration/ece": 0.14295423778611177,
"calibration/mean_confidence": 0.7306168172362433,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.012586805555555558,
"completions/max_length": 3763.4,
"completions/max_terminated_length": 3763.4,
"completions/mean_length": 998.8343017578125,
"completions/mean_terminated_length": 1011.7352783203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 254.2,
"epoch": 0.9735576923076923,
"grad_norm": 0.0004764749901369214,
"learning_rate": 3.81610576923077e-06,
"loss": -0.0115,
"num_tokens": 949714508.0,
"reward": 1.2655861377716064,
"reward_std": 0.12250902205705642,
"rewards/accuracy_reward": 0.7224826335906982,
"rewards/brier_reward": 0.8213498592376709,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9873263835906982,
"rewards/mean_confidence_reward": 0.668897545337677,
"sampling/batch_mean_priority_error": 0.05678419767544698,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4083333333333333,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.00798400416970253,
"sampling/priority_kl": 0.02999998740851879,
"sampling/priority_scale": 0.7126380980247632,
"sampling/prob_entropy": 10.278954887390137,
"sampling/prob_max": 4.5139902067603546e-05,
"sampling/prob_min": 1.943270181072876e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9672000050544739,
"sampling/prompt_draws_total": 29016.0,
"sampling/seen_fraction": 0.6859200119972229,
"sampling/unseen_fraction": 0.3140799880027771,
"signal/accuracy_reward/centered_abs_mean": 0.12692600041627883,
"signal/accuracy_reward/group_std_mean": 0.17087701857089996,
"signal/accuracy_reward/group_zero_std_frac": 0.4972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06346300020813941,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06346300020813941,
"signal/advantage_abs_mean": 0.08730639815330506,
"signal/advantage_pre_scale_abs_mean": 0.08730639815330506,
"signal/advantage_pre_scale_std": 0.1753060758113861,
"signal/advantage_std": 0.1753060758113861,
"signal/brier_reward/centered_abs_mean": 0.08770604431629181,
"signal/brier_reward/group_std_mean": 0.11753087639808654,
"signal/brier_reward/group_zero_std_frac": 0.24166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.043853022158145905,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.043853022158145905,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01763237863779068,
"signal/format_reward/group_std_mean": 0.028114334866404533,
"signal/format_reward/group_zero_std_frac": 0.9000000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00881618931889534,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00881618931889534,
"signal/mean_confidence_reward/centered_abs_mean": 0.06661078855395317,
"signal/mean_confidence_reward/group_std_mean": 0.08901985883712768,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2722222298383713,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.661078714387258e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.661078714387258e-07,
"step": 405
},
{
"calibration/aurc": 0.146472110841029,
"calibration/batch_distribution_entropy": 0.5287236575059514,
"calibration/batch_entropy_100bins": 0.2992488069162606,
"calibration/batch_entropy_10bins": 0.5287236575059514,
"calibration/batch_entropy_50bins": 0.35227085369669153,
"calibration/batch_uniqueness": 0.20660984918907693,
"calibration/confidence_entropy": 0.4404203378198561,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.43997464813960063,
"calibration/coverage@15%": 0.47216462175437623,
"calibration/coverage@20%": 0.6504767488936355,
"calibration/coverage@25%": 0.8245111648494333,
"calibration/coverage@30%": 0.9152815013404826,
"calibration/coverage@5%": 0.3947077991416191,
"calibration/distribution_entropy_10": 0.5287236575059514,
"calibration/distribution_entropy_100": 0.2992488069162606,
"calibration/ece": 0.11603427993313538,
"calibration/mean_confidence": 0.7137951490105662,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.011284722222222232,
"completions/max_length": 3923.0,
"completions/max_terminated_length": 3923.0,
"completions/mean_length": 979.064501953125,
"completions/mean_terminated_length": 990.3482177734375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 216.6,
"epoch": 0.9855769230769231,
"grad_norm": 0.0004954978358000517,
"learning_rate": 3.7860576923076927e-06,
"loss": -0.0151,
"num_tokens": 964046835.0,
"reward": 1.274612021446228,
"reward_std": 0.1268946871161461,
"rewards/accuracy_reward": 0.7220486164093017,
"rewards/brier_reward": 0.8384461641311646,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9887152791023255,
"rewards/mean_confidence_reward": 0.7021701455116272,
"sampling/batch_mean_priority_error": 0.052499148894268544,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.43888888888888894,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.008041384257376194,
"sampling/priority_kl": 0.02999994605779648,
"sampling/priority_scale": 0.7123283087974415,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 4.5282146311365067e-05,
"sampling/prob_min": 1.950107980519533e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.979200005531311,
"sampling/prompt_draws_total": 29376.0,
"sampling/seen_fraction": 0.6911133289337158,
"sampling/unseen_fraction": 0.3088866710662842,
"signal/accuracy_reward/centered_abs_mean": 0.12586805671453477,
"signal/accuracy_reward/group_std_mean": 0.16451444029808043,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888835906982,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06293402835726739,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06293402835726739,
"signal/advantage_abs_mean": 0.09020028412342071,
"signal/advantage_pre_scale_abs_mean": 0.09020028412342071,
"signal/advantage_pre_scale_std": 0.18385487496852876,
"signal/advantage_std": 0.18385487496852876,
"signal/brier_reward/centered_abs_mean": 0.08396928757429123,
"signal/brier_reward/group_std_mean": 0.11364553421735764,
"signal/brier_reward/group_zero_std_frac": 0.347222226858139,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.041984643787145615,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.041984643787145615,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.019259982742369175,
"signal/format_reward/group_std_mean": 0.03404767997562885,
"signal/format_reward/group_zero_std_frac": 0.8694444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.009629991371184587,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.009629991371184587,
"signal/mean_confidence_reward/centered_abs_mean": 0.06446019485592842,
"signal/mean_confidence_reward/group_std_mean": 0.08440773785114289,
"signal/mean_confidence_reward/group_zero_std_frac": 0.39166666865348815,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.446019142458681e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.446019142458681e-07,
"step": 410
},
{
"calibration/aurc": 0.20424447506264923,
"calibration/batch_distribution_entropy": 0.5956056427737836,
"calibration/batch_entropy_100bins": 0.34593084307848915,
"calibration/batch_entropy_10bins": 0.5956056427737836,
"calibration/batch_entropy_50bins": 0.40722419135785,
"calibration/batch_uniqueness": 0.4067028193692369,
"calibration/confidence_entropy": 0.4794688579045291,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.08421052631578947,
"calibration/coverage@10%": 0.3074001933968781,
"calibration/coverage@15%": 0.45175714877745543,
"calibration/coverage@20%": 0.5347810471059539,
"calibration/coverage@25%": 0.5909490261085786,
"calibration/coverage@30%": 0.8287976153474237,
"calibration/coverage@5%": 0.0968421052631579,
"calibration/distribution_entropy_10": 0.5956056427737836,
"calibration/distribution_entropy_100": 0.34593084307848915,
"calibration/ece": 0.12277091535433071,
"calibration/mean_confidence": 0.6573581252590137,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.00859375,
"completions/max_length": 3644.8,
"completions/max_terminated_length": 3644.8,
"completions/mean_length": 1015.5580810546875,
"completions/mean_terminated_length": 1024.4833251953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 256.4,
"epoch": 0.9975961538461539,
"grad_norm": 0.0004494949243962765,
"learning_rate": 3.756009615384616e-06,
"loss": -0.0094,
"num_tokens": 978866064.0,
"reward": 1.2763642072677612,
"reward_std": 0.11199232935905457,
"rewards/accuracy_reward": 0.70625,
"rewards/brier_reward": 0.8552326440811158,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9912326335906982,
"rewards/mean_confidence_reward": 0.6571180343627929,
"sampling/batch_mean_priority_error": 0.04054459026418521,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.43611111111111106,
"sampling/error_ema_max": 0.16249588131904602,
"sampling/error_ema_mean": 0.008088457770645619,
"sampling/priority_kl": 0.029999810457229614,
"sampling/priority_scale": 0.7122232496039942,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 4.542987808235921e-05,
"sampling/prob_min": 1.956713203981053e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 0.9911999940872193,
"sampling/prompt_draws_total": 29736.0,
"sampling/seen_fraction": 0.6963999986648559,
"sampling/unseen_fraction": 0.30360000133514403,
"signal/accuracy_reward/centered_abs_mean": 0.12038845419883729,
"signal/accuracy_reward/group_std_mean": 0.1608419895172119,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06019422709941864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06019422709941864,
"signal/advantage_abs_mean": 0.07947031259536744,
"signal/advantage_pre_scale_abs_mean": 0.07947031259536744,
"signal/advantage_pre_scale_std": 0.1640724152326584,
"signal/advantage_std": 0.1640724152326584,
"signal/brier_reward/centered_abs_mean": 0.07228801995515824,
"signal/brier_reward/group_std_mean": 0.10005511939525605,
"signal/brier_reward/group_zero_std_frac": 0.26111111640930174,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03614400997757912,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03614400997757912,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.014339192397892475,
"signal/format_reward/group_std_mean": 0.0262321874499321,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007169596198946237,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007169596198946237,
"signal/mean_confidence_reward/centered_abs_mean": 0.061816409975290296,
"signal/mean_confidence_reward/group_std_mean": 0.08097582757472992,
"signal/mean_confidence_reward/group_zero_std_frac": 0.28333333432674407,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.181640742397576e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.181640742397576e-07,
"step": 415
},
{
"calibration/aurc": 0.15625822134062678,
"calibration/batch_distribution_entropy": 0.5187429287304861,
"calibration/batch_entropy_100bins": 0.29191831446718414,
"calibration/batch_entropy_10bins": 0.5187429287304861,
"calibration/batch_entropy_50bins": 0.3436415165920129,
"calibration/batch_uniqueness": 0.17664039554168148,
"calibration/confidence_entropy": 0.454614391014447,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.05511811023622047,
"calibration/coverage@10%": 0.3866551837270341,
"calibration/coverage@15%": 0.5913426837270341,
"calibration/coverage@20%": 0.7276694381129991,
"calibration/coverage@25%": 0.7998906962287609,
"calibration/coverage@30%": 0.8527614311368975,
"calibration/coverage@5%": 0.05931758530183727,
"calibration/distribution_entropy_10": 0.5187429287304861,
"calibration/distribution_entropy_100": 0.29191831446718414,
"calibration/ece": 0.07153017077635034,
"calibration/mean_confidence": 0.7071844047865727,
"calibration/unique_confidence_per_question": 0.021354166666666664,
"calibration/unique_confidences": 8.2,
"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": 0.002430555555555558,
"completions/max_length": 3184.6,
"completions/max_terminated_length": 3184.6,
"completions/mean_length": 974.892626953125,
"completions/mean_terminated_length": 977.2836303710938,
"completions/min_length": 86.6,
"completions/min_terminated_length": 263.4,
"epoch": 1.0096153846153846,
"grad_norm": 0.00075355643639341,
"learning_rate": 3.725961538461539e-06,
"loss": -0.0027,
"num_tokens": 993413498.0,
"reward": 1.267124342918396,
"reward_std": 0.12146431356668472,
"rewards/accuracy_reward": 0.6927083373069763,
"rewards/brier_reward": 0.8456927061080932,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9958333253860474,
"rewards/mean_confidence_reward": 0.7187413096427917,
"sampling/batch_mean_priority_error": 0.05120892290939198,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4083333333333334,
"sampling/error_ema_max": 0.19449620842933654,
"sampling/error_ema_mean": 0.008134695142507553,
"sampling/priority_kl": 0.030000004172325134,
"sampling/priority_scale": 0.7118521988624706,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 4.5568116183858366e-05,
"sampling/prob_min": 1.963528666237835e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0032000064849853,
"sampling/prompt_draws_total": 30096.0,
"sampling/seen_fraction": 0.701253342628479,
"sampling/unseen_fraction": 0.298746657371521,
"signal/accuracy_reward/centered_abs_mean": 0.12330729216337204,
"signal/accuracy_reward/group_std_mean": 0.16879068613052367,
"signal/accuracy_reward/group_zero_std_frac": 0.4972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06165364608168602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06165364608168602,
"signal/advantage_abs_mean": 0.08633148670196533,
"signal/advantage_pre_scale_abs_mean": 0.08633148670196533,
"signal/advantage_pre_scale_std": 0.17180612981319426,
"signal/advantage_std": 0.17180612981319426,
"signal/brier_reward/centered_abs_mean": 0.07244243398308754,
"signal/brier_reward/group_std_mean": 0.10006694793701172,
"signal/brier_reward/group_zero_std_frac": 0.3,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03622121699154377,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03622121699154377,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00702039924217388,
"signal/format_reward/group_std_mean": 0.013451122678816319,
"signal/format_reward/group_zero_std_frac": 0.9416666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00351019962108694,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00351019962108694,
"signal/mean_confidence_reward/centered_abs_mean": 0.05719347670674324,
"signal/mean_confidence_reward/group_std_mean": 0.07559116035699845,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3555555582046509,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.71934742765734e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.71934742765734e-07,
"step": 420
},
{
"calibration/aurc": 0.06865467727784927,
"calibration/batch_distribution_entropy": 0.4543795898254017,
"calibration/batch_entropy_100bins": 0.24974455856950017,
"calibration/batch_entropy_10bins": 0.4543795898254017,
"calibration/batch_entropy_50bins": 0.29399525351491274,
"calibration/batch_uniqueness": 0.013456829224856321,
"calibration/confidence_entropy": 0.4199663795097931,
"calibration/coverage@0%": 0.15,
"calibration/coverage@1%": 0.15,
"calibration/coverage@10%": 0.6710055774278215,
"calibration/coverage@15%": 0.9165231299212598,
"calibration/coverage@20%": 0.9682291666666666,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5911991469816273,
"calibration/distribution_entropy_10": 0.4543795898254017,
"calibration/distribution_entropy_100": 0.24974455856950017,
"calibration/ece": 0.086389435695538,
"calibration/mean_confidence": 0.7852444225721785,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.001736111111111116,
"completions/max_length": 3634.6,
"completions/max_terminated_length": 3634.6,
"completions/mean_length": 960.628564453125,
"completions/mean_terminated_length": 962.2773803710937,
"completions/min_length": 0.0,
"completions/min_terminated_length": 240.6,
"epoch": 1.0216346153846154,
"grad_norm": 0.0008929867180995643,
"learning_rate": 3.695913461538462e-06,
"loss": -0.0011,
"num_tokens": 1007578723.0,
"reward": 1.2825724124908446,
"reward_std": 0.12003176510334015,
"rewards/accuracy_reward": 0.7144097208976745,
"rewards/brier_reward": 0.8527161598205566,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9980034828186035,
"rewards/mean_confidence_reward": 0.7765104055404664,
"sampling/batch_mean_priority_error": 0.05508253028193806,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4333333333333333,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008189113438129425,
"sampling/priority_kl": 0.030000052973628046,
"sampling/priority_scale": 0.7119282067054883,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 4.572009711409919e-05,
"sampling/prob_min": 1.9699002950801515e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0152000188827515,
"sampling/prompt_draws_total": 30456.0,
"sampling/seen_fraction": 0.7064599990844727,
"sampling/unseen_fraction": 0.29354000091552734,
"signal/accuracy_reward/centered_abs_mean": 0.11453992873430252,
"signal/accuracy_reward/group_std_mean": 0.1561146765947342,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333432674408,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05726996436715126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05726996436715126,
"signal/advantage_abs_mean": 0.08596721366047859,
"signal/advantage_pre_scale_abs_mean": 0.08596721366047859,
"signal/advantage_pre_scale_std": 0.17472383975982667,
"signal/advantage_std": 0.17472383975982667,
"signal/brier_reward/centered_abs_mean": 0.06893061101436615,
"signal/brier_reward/group_std_mean": 0.0971571832895279,
"signal/brier_reward/group_zero_std_frac": 0.3444444537162781,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034465305507183075,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034465305507183075,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0037272135145030917,
"signal/format_reward/group_std_mean": 0.008416965417563916,
"signal/format_reward/group_zero_std_frac": 0.9611111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0018636067572515458,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0018636067572515458,
"signal/mean_confidence_reward/centered_abs_mean": 0.04610461071133613,
"signal/mean_confidence_reward/group_std_mean": 0.06151590570807457,
"signal/mean_confidence_reward/group_zero_std_frac": 0.39444445371627807,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.6104609623398575e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.6104609623398575e-07,
"step": 425
},
{
"calibration/aurc": 0.16252197159696252,
"calibration/batch_distribution_entropy": 0.551132185909566,
"calibration/batch_entropy_100bins": 0.3044312413497231,
"calibration/batch_entropy_10bins": 0.551132185909566,
"calibration/batch_entropy_50bins": 0.358371531660677,
"calibration/batch_uniqueness": 0.25293511284722225,
"calibration/confidence_entropy": 0.45229610428378975,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.12291666666666667,
"calibration/coverage@10%": 0.2604166666666667,
"calibration/coverage@15%": 0.4239583333333334,
"calibration/coverage@20%": 0.6843750000000001,
"calibration/coverage@25%": 0.8161458333333333,
"calibration/coverage@30%": 0.8630208333333332,
"calibration/coverage@5%": 0.2453125,
"calibration/distribution_entropy_10": 0.551132185909566,
"calibration/distribution_entropy_100": 0.3044312413497231,
"calibration/ece": 0.15208333333333335,
"calibration/mean_confidence": 0.7494791666666668,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.001041666666666674,
"completions/max_length": 3636.6,
"completions/max_terminated_length": 3636.6,
"completions/mean_length": 972.2225708007812,
"completions/mean_terminated_length": 973.2240600585938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 238.0,
"epoch": 1.0336538461538463,
"grad_norm": 0.0007758528809063137,
"learning_rate": 3.665865384615385e-06,
"loss": -0.0007,
"num_tokens": 1021881479.0,
"reward": 1.2820632934570313,
"reward_std": 0.1231762483716011,
"rewards/accuracy_reward": 0.7171875,
"rewards/brier_reward": 0.8480520844459534,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9988715171813964,
"rewards/mean_confidence_reward": 0.7767274141311645,
"sampling/batch_mean_priority_error": 0.052858080641298974,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4138888888888889,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008248506300151347,
"sampling/priority_kl": 0.029999766498804092,
"sampling/priority_scale": 0.7118548452155664,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 4.586404174915515e-05,
"sampling/prob_min": 1.9762736701522954e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.0271999835968018,
"sampling/prompt_draws_total": 30816.0,
"sampling/seen_fraction": 0.7113333344459534,
"sampling/unseen_fraction": 0.28866666555404663,
"signal/accuracy_reward/centered_abs_mean": 0.12096354067325592,
"signal/accuracy_reward/group_std_mean": 0.1624488949775696,
"signal/accuracy_reward/group_zero_std_frac": 0.5250000178813934,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06048177033662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06048177033662796,
"signal/advantage_abs_mean": 0.08952582180500031,
"signal/advantage_pre_scale_abs_mean": 0.08952582180500031,
"signal/advantage_pre_scale_std": 0.1779074639081955,
"signal/advantage_std": 0.1779074639081955,
"signal/brier_reward/centered_abs_mean": 0.0699217140674591,
"signal/brier_reward/group_std_mean": 0.09810963273048401,
"signal/brier_reward/group_zero_std_frac": 0.32777778506278993,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03496085703372955,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03496085703372955,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0021538628148846327,
"signal/format_reward/group_std_mean": 0.005733087472617626,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0010769314074423164,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0010769314074423164,
"signal/mean_confidence_reward/centered_abs_mean": 0.0436925008893013,
"signal/mean_confidence_reward/group_std_mean": 0.05805739238858223,
"signal/mean_confidence_reward/group_zero_std_frac": 0.40277777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.369250177660433e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.369250177660433e-07,
"step": 430
},
{
"calibration/aurc": 0.1326655917417161,
"calibration/batch_distribution_entropy": 0.4668999632133756,
"calibration/batch_entropy_100bins": 0.25158488405771856,
"calibration/batch_entropy_10bins": 0.4668999632133756,
"calibration/batch_entropy_50bins": 0.29616165490342655,
"calibration/batch_uniqueness": 0.03741925538473879,
"calibration/confidence_entropy": 0.4208925983815625,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.15287958115183248,
"calibration/coverage@10%": 0.25787185761349496,
"calibration/coverage@15%": 0.7178876734948214,
"calibration/coverage@20%": 0.8331021540469974,
"calibration/coverage@25%": 0.9232851936466492,
"calibration/coverage@30%": 0.9566579634464751,
"calibration/coverage@5%": 0.1649214659685864,
"calibration/distribution_entropy_10": 0.4668999632133756,
"calibration/distribution_entropy_100": 0.25158488405771856,
"calibration/ece": 0.11240595837377372,
"calibration/mean_confidence": 0.7994348054477146,
"calibration/unique_confidence_per_question": 0.01927083333333333,
"calibration/unique_confidences": 7.4,
"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": 0.00199652777777779,
"completions/max_length": 3659.0,
"completions/max_terminated_length": 3659.0,
"completions/mean_length": 985.06005859375,
"completions/mean_terminated_length": 987.0644653320312,
"completions/min_length": 56.6,
"completions/min_terminated_length": 256.8,
"epoch": 1.0456730769230769,
"grad_norm": 0.0006942002801224589,
"learning_rate": 3.635817307692308e-06,
"loss": -0.0008,
"num_tokens": 1036339675.0,
"reward": 1.2937331438064574,
"reward_std": 0.1319456309080124,
"rewards/accuracy_reward": 0.7370659828186035,
"rewards/brier_reward": 0.8524679064750671,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166746139526,
"rewards/mean_confidence_reward": 0.7897829651832581,
"sampling/batch_mean_priority_error": 0.0520680581107638,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.4,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008302754536271096,
"sampling/priority_kl": 0.030000053718686102,
"sampling/priority_scale": 0.7120462120277807,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 4.6015283442102375e-05,
"sampling/prob_min": 1.9823418188025244e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.039199995994568,
"sampling/prompt_draws_total": 31176.0,
"sampling/seen_fraction": 0.7163466691970826,
"sampling/unseen_fraction": 0.2836533308029175,
"signal/accuracy_reward/centered_abs_mean": 0.13167860358953476,
"signal/accuracy_reward/group_std_mean": 0.1694795623421669,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06583930179476738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06583930179476738,
"signal/advantage_abs_mean": 0.10044280737638474,
"signal/advantage_pre_scale_abs_mean": 0.10044280737638474,
"signal/advantage_pre_scale_std": 0.18933759331703187,
"signal/advantage_std": 0.18933759331703187,
"signal/brier_reward/centered_abs_mean": 0.0774193748831749,
"signal/brier_reward/group_std_mean": 0.10275954306125641,
"signal/brier_reward/group_zero_std_frac": 0.31666666865348814,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03870968744158745,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03870968744158745,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0038411458488553764,
"signal/format_reward/group_std_mean": 0.008209389261901378,
"signal/format_reward/group_zero_std_frac": 0.9638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019205729244276882,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019205729244276882,
"signal/mean_confidence_reward/centered_abs_mean": 0.04758844077587128,
"signal/mean_confidence_reward/group_std_mean": 0.06242263168096542,
"signal/mean_confidence_reward/group_zero_std_frac": 0.36111111640930177,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.758843715535477e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.758843715535477e-07,
"step": 435
},
{
"calibration/aurc": 0.11291767152922454,
"calibration/batch_distribution_entropy": 0.5680532576297793,
"calibration/batch_entropy_100bins": 0.3125931234346516,
"calibration/batch_entropy_10bins": 0.5680532576297793,
"calibration/batch_entropy_50bins": 0.3679795671929091,
"calibration/batch_uniqueness": 0.3136267302700734,
"calibration/confidence_entropy": 0.4582802544004433,
"calibration/coverage@0%": 0.1375,
"calibration/coverage@1%": 0.4014249244733641,
"calibration/coverage@10%": 0.5410640270506109,
"calibration/coverage@15%": 0.5890625,
"calibration/coverage@20%": 0.709375,
"calibration/coverage@25%": 0.8984375,
"calibration/coverage@30%": 0.9291666666666668,
"calibration/coverage@5%": 0.49184936736474694,
"calibration/distribution_entropy_10": 0.5680532576297793,
"calibration/distribution_entropy_100": 0.3125931234346516,
"calibration/ece": 0.14599718654737326,
"calibration/mean_confidence": 0.7607226890899895,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.0008680555555555802,
"completions/max_length": 3281.2,
"completions/max_terminated_length": 3281.2,
"completions/mean_length": 996.7343872070312,
"completions/mean_terminated_length": 997.6031860351562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 249.2,
"epoch": 1.0576923076923077,
"grad_norm": 0.0006608100957237184,
"learning_rate": 3.605769230769231e-06,
"loss": 0.0003,
"num_tokens": 1050923655.0,
"reward": 1.2805131673812866,
"reward_std": 0.1213545024394989,
"rewards/accuracy_reward": 0.7131944537162781,
"rewards/brier_reward": 0.8487717032432556,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451335906982,
"rewards/mean_confidence_reward": 0.7509288191795349,
"sampling/batch_mean_priority_error": 0.04343561407930108,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3805555555555556,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.00835284162312746,
"sampling/priority_kl": 0.03000011034309864,
"sampling/priority_scale": 0.7119822085136548,
"sampling/prob_entropy": 10.27895450592041,
"sampling/prob_max": 4.6156414464348926e-05,
"sampling/prob_min": 1.988572548725642e-05,
"sampling/prompt_draws_max": 5.0,
"sampling/prompt_draws_mean": 1.051199984550476,
"sampling/prompt_draws_total": 31536.0,
"sampling/seen_fraction": 0.7209666728973388,
"sampling/unseen_fraction": 0.27903332710266116,
"signal/accuracy_reward/centered_abs_mean": 0.12660590559244156,
"signal/accuracy_reward/group_std_mean": 0.1687839388847351,
"signal/accuracy_reward/group_zero_std_frac": 0.5138888895511627,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06330295279622078,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06330295279622078,
"signal/advantage_abs_mean": 0.08899948745965958,
"signal/advantage_pre_scale_abs_mean": 0.08899948745965958,
"signal/advantage_pre_scale_std": 0.17068710923194885,
"signal/advantage_std": 0.17068710923194885,
"signal/brier_reward/centered_abs_mean": 0.0685460977256298,
"signal/brier_reward/group_std_mean": 0.092455093562603,
"signal/brier_reward/group_zero_std_frac": 0.20833333432674409,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0342730488628149,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0342730488628149,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0018391926772892476,
"signal/format_reward/group_std_mean": 0.0051025690510869024,
"signal/format_reward/group_zero_std_frac": 0.9722222089767456,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0009195963386446238,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0009195963386446238,
"signal/mean_confidence_reward/centered_abs_mean": 0.04989257976412773,
"signal/mean_confidence_reward/group_std_mean": 0.06543896198272706,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23333333432674408,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.989257604393061e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.989257604393061e-07,
"step": 440
},
{
"calibration/aurc": 0.13666503221385767,
"calibration/batch_distribution_entropy": 0.7076334146559488,
"calibration/batch_entropy_100bins": 0.3924286491471299,
"calibration/batch_entropy_10bins": 0.7076334146559488,
"calibration/batch_entropy_50bins": 0.4619606563336549,
"calibration/batch_uniqueness": 0.5822380897384157,
"calibration/confidence_entropy": 0.5120896384533352,
"calibration/coverage@0%": 0.09010416666666667,
"calibration/coverage@1%": 0.19583333333333333,
"calibration/coverage@10%": 0.5733857814593781,
"calibration/coverage@15%": 0.6379691147927115,
"calibration/coverage@20%": 0.7316844271070772,
"calibration/coverage@25%": 0.7901442291088179,
"calibration/coverage@30%": 0.8031732217110807,
"calibration/coverage@5%": 0.4139550264550264,
"calibration/distribution_entropy_10": 0.7076334146559488,
"calibration/distribution_entropy_100": 0.3924286491471299,
"calibration/ece": 0.13634455038542825,
"calibration/mean_confidence": 0.6688398879460401,
"calibration/unique_confidence_per_question": 0.0234375,
"calibration/unique_confidences": 9.0,
"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": 0.00277777777777779,
"completions/max_length": 3636.2,
"completions/max_terminated_length": 3636.2,
"completions/mean_length": 991.1034912109375,
"completions/mean_terminated_length": 993.92783203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 248.6,
"epoch": 1.0697115384615385,
"grad_norm": 0.000556574494112283,
"learning_rate": 3.575721153846154e-06,
"loss": -0.0028,
"num_tokens": 1065407151.0,
"reward": 1.31825795173645,
"reward_std": 0.10573276579380035,
"rewards/accuracy_reward": 0.7677951216697693,
"rewards/brier_reward": 0.8717447876930237,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9969617843627929,
"rewards/mean_confidence_reward": 0.7086805701255798,
"sampling/batch_mean_priority_error": 0.02726226362076916,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.35,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.00838301219046116,
"sampling/priority_kl": 0.029999752715229987,
"sampling/priority_scale": 0.7115830718772486,
"sampling/prob_entropy": 10.278950309753418,
"sampling/prob_max": 4.62854171928484e-05,
"sampling/prob_min": 1.960499575943686e-05,
"sampling/prompt_draws_max": 5.6,
"sampling/prompt_draws_mean": 1.0631999969482422,
"sampling/prompt_draws_total": 31896.0,
"sampling/seen_fraction": 0.7251199960708619,
"sampling/unseen_fraction": 0.2748800039291382,
"signal/accuracy_reward/centered_abs_mean": 0.11357964277267456,
"signal/accuracy_reward/group_std_mean": 0.15543469190597534,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888835906982,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05678982138633728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05678982138633728,
"signal/advantage_abs_mean": 0.07521249949932099,
"signal/advantage_pre_scale_abs_mean": 0.07521249949932099,
"signal/advantage_pre_scale_std": 0.15374746918678284,
"signal/advantage_std": 0.15374746918678284,
"signal/brier_reward/centered_abs_mean": 0.05601532757282257,
"signal/brier_reward/group_std_mean": 0.07705468088388442,
"signal/brier_reward/group_zero_std_frac": 0.14444444626569747,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028007663786411285,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028007663786411285,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004931640531867743,
"signal/format_reward/group_std_mean": 0.009485878376290202,
"signal/format_reward/group_zero_std_frac": 0.9583333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024658202659338714,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0024658202659338714,
"signal/mean_confidence_reward/centered_abs_mean": 0.05166775062680244,
"signal/mean_confidence_reward/group_std_mean": 0.0685989424586296,
"signal/mean_confidence_reward/group_zero_std_frac": 0.14722222536802293,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.166774883491598e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.166774883491598e-07,
"step": 445
},
{
"calibration/aurc": 0.17119985867040605,
"calibration/batch_distribution_entropy": 0.6938781727692345,
"calibration/batch_entropy_100bins": 0.37935902900642526,
"calibration/batch_entropy_10bins": 0.6938781727692345,
"calibration/batch_entropy_50bins": 0.44657531096869957,
"calibration/batch_uniqueness": 0.5243673203869619,
"calibration/confidence_entropy": 0.504338539628425,
"calibration/coverage@0%": 0.017754569190600523,
"calibration/coverage@1%": 0.017754569190600523,
"calibration/coverage@10%": 0.4992860639686684,
"calibration/coverage@15%": 0.5238019473455179,
"calibration/coverage@20%": 0.6404686140121845,
"calibration/coverage@25%": 0.7448759791122714,
"calibration/coverage@30%": 0.7949480526544821,
"calibration/coverage@5%": 0.37684127502175807,
"calibration/distribution_entropy_10": 0.6938781727692345,
"calibration/distribution_entropy_100": 0.37935902900642526,
"calibration/ece": 0.10688887075718015,
"calibration/mean_confidence": 0.6589066579634466,
"calibration/unique_confidence_per_question": 0.022395833333333334,
"calibration/unique_confidences": 8.6,
"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": 0.0006944444444444642,
"completions/max_length": 3458.6,
"completions/max_terminated_length": 3458.6,
"completions/mean_length": 973.0631225585937,
"completions/mean_terminated_length": 973.7551879882812,
"completions/min_length": 139.0,
"completions/min_terminated_length": 275.4,
"epoch": 1.0817307692307692,
"grad_norm": 0.0006742961122654378,
"learning_rate": 3.5456730769230774e-06,
"loss": 0.0009,
"num_tokens": 1079686790.0,
"reward": 1.3066105604171754,
"reward_std": 0.1120601698756218,
"rewards/accuracy_reward": 0.7502604126930237,
"rewards/brier_reward": 0.8643359184265137,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9986111164093018,
"rewards/mean_confidence_reward": 0.6820138692855835,
"sampling/batch_mean_priority_error": 0.03446128472222223,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3777777777777778,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008412830717861652,
"sampling/priority_kl": 0.03000001423060894,
"sampling/priority_scale": 0.7117285073036328,
"sampling/prob_entropy": 10.278955078125,
"sampling/prob_max": 4.643480133381672e-05,
"sampling/prob_min": 1.9433926718193105e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0752000093460083,
"sampling/prompt_draws_total": 32256.0,
"sampling/seen_fraction": 0.7297666668891907,
"sampling/unseen_fraction": 0.2702333331108093,
"signal/accuracy_reward/centered_abs_mean": 0.13707139641046523,
"signal/accuracy_reward/group_std_mean": 0.1787209153175354,
"signal/accuracy_reward/group_zero_std_frac": 0.49166667461395264,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06853569820523261,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06853569820523261,
"signal/advantage_abs_mean": 0.08267289400100708,
"signal/advantage_pre_scale_abs_mean": 0.08267289400100708,
"signal/advantage_pre_scale_std": 0.15754978358745575,
"signal/advantage_std": 0.15754978358745575,
"signal/brier_reward/centered_abs_mean": 0.06549736708402634,
"signal/brier_reward/group_std_mean": 0.08823522627353668,
"signal/brier_reward/group_zero_std_frac": 0.14166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03274868354201317,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03274868354201317,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0026150173507630825,
"signal/format_reward/group_std_mean": 0.006526978686451912,
"signal/format_reward/group_zero_std_frac": 0.9666666626930237,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0013075086753815413,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0013075086753815413,
"signal/mean_confidence_reward/centered_abs_mean": 0.06105956807732582,
"signal/mean_confidence_reward/group_std_mean": 0.07881901264190674,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1472222238779068,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.105956799729029e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.105956799729029e-07,
"step": 450
},
{
"epoch": 1.0817307692307692,
"eval_calibration/aurc": 0.1456926255290758,
"eval_calibration/batch_distribution_entropy": 0.6996202405927568,
"eval_calibration/batch_entropy_100bins": 0.3839213205044793,
"eval_calibration/batch_entropy_10bins": 0.6996202405927568,
"eval_calibration/batch_entropy_50bins": 0.45194596670295084,
"eval_calibration/batch_uniqueness": 0.5007183364839319,
"eval_calibration/confidence_entropy": 0.4720367910302038,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.45304347826086955,
"eval_calibration/coverage@15%": 0.6547826086956522,
"eval_calibration/coverage@20%": 0.7965217391304348,
"eval_calibration/coverage@25%": 0.88,
"eval_calibration/coverage@30%": 0.9817391304347826,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6996202405927568,
"eval_calibration/distribution_entropy_100": 0.3839213205044793,
"eval_calibration/ece": 0.021217391304347813,
"eval_calibration/mean_confidence": 0.6989565217391304,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.001736111111111105,
"eval_completions/max_length": 2455.6666666666665,
"eval_completions/max_terminated_length": 2455.6666666666665,
"eval_completions/mean_length": 979.8416849772135,
"eval_completions/mean_terminated_length": 981.5661824544271,
"eval_completions/min_length": 290.1666666666667,
"eval_completions/min_terminated_length": 340.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1079686790.0,
"eval_reward": 1.2733750542004902,
"eval_reward_std": 0.30162295202414197,
"eval_rewards/accuracy_reward": 0.6979166666666666,
"eval_rewards/brier_reward": 0.850555549065272,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9982638855775198,
"eval_rewards/mean_confidence_reward": 0.6977430780728658,
"eval_runtime": 147.3716,
"eval_samples_per_second": 6.786,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4115668386220932,
"eval_signal/accuracy_reward/group_std_mean": 0.4599919815858205,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2057834193110466,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.2057834193110466,
"eval_signal/advantage_abs_mean": 0.25963140030701953,
"eval_signal/advantage_pre_scale_abs_mean": 0.25963140030701953,
"eval_signal/advantage_pre_scale_std": 0.2986646344264348,
"eval_signal/advantage_std": 0.2986646344264348,
"eval_signal/brier_reward/centered_abs_mean": 0.15869846940040588,
"eval_signal/brier_reward/group_std_mean": 0.21192201475302377,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07934923470020294,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07934923470020294,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/group_std_mean": 0.009820927555362383,
"eval_signal/format_reward/group_zero_std_frac": 0.9444444477558136,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.2031846543153127,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2419320916136106,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.031846577210672e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.031846577210672e-06,
"eval_steps_per_second": 0.041,
"step": 450
},
{
"epoch": 1.0817307692307692,
"step": 450,
"train_probe_calibration/aurc": 0.09937392783360088,
"train_probe_calibration/batch_distribution_entropy": 0.6987087019871145,
"train_probe_calibration/batch_entropy_100bins": 0.38393012567447593,
"train_probe_calibration/batch_entropy_10bins": 0.6987087019871145,
"train_probe_calibration/batch_entropy_50bins": 0.4519563320066044,
"train_probe_calibration/batch_uniqueness": 0.4989943289224953,
"train_probe_calibration/confidence_entropy": 0.4712346490131543,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6556521739130434,
"train_probe_calibration/coverage@15%": 0.7930434782608695,
"train_probe_calibration/coverage@20%": 0.8852173913043478,
"train_probe_calibration/coverage@25%": 0.9808695652173913,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.4539130434782609,
"train_probe_calibration/distribution_entropy_10": 0.6987087019871145,
"train_probe_calibration/distribution_entropy_100": 0.38393012567447593,
"train_probe_calibration/ece": 0.04843478260869559,
"train_probe_calibration/mean_confidence": 0.699391304347826,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.005034722222222232,
"train_probe_completions/max_length": 2869.8333333333335,
"train_probe_completions/max_terminated_length": 2869.8333333333335,
"train_probe_completions/mean_length": 993.7186075846354,
"train_probe_completions/mean_terminated_length": 998.8824666341146,
"train_probe_completions/min_length": 166.16666666666666,
"train_probe_completions/min_terminated_length": 271.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1079686790.0,
"train_probe_reward": 1.3065911928812664,
"train_probe_reward_std": 0.27413928508758545,
"train_probe_rewards/accuracy_reward": 0.7465277711550394,
"train_probe_rewards/brier_reward": 0.8683767517407736,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9982638955116272,
"train_probe_rewards/mean_confidence_reward": 0.6981770694255829,
"train_probe_runtime": 171.2718,
"train_probe_samples_per_second": 5.839,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3681640625,
"train_probe_signal/accuracy_reward/group_std_mean": 0.43382490674654645,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.18408203125,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.18408203125,
"train_probe_signal/advantage_abs_mean": 0.22712557514508566,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.22712557514508566,
"train_probe_signal/advantage_pre_scale_std": 0.27301861842473346,
"train_probe_signal/advantage_std": 0.27301861842473346,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13910971085230509,
"train_probe_signal/brier_reward/group_std_mean": 0.18819688508907953,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06955485542615254,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06955485542615254,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/group_std_mean": 0.009820927555362383,
"train_probe_signal/format_reward/group_zero_std_frac": 0.944444457689921,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.2051540638009707,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2430253947774569,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.051540567056994e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.051540567056994e-06,
"train_probe_steps_per_second": 0.035
},
{
"calibration/aurc": 0.12143333430388896,
"calibration/batch_distribution_entropy": 0.6992091580531643,
"calibration/batch_entropy_100bins": 0.3825769752796119,
"calibration/batch_entropy_10bins": 0.6992091580531643,
"calibration/batch_entropy_50bins": 0.4503634252555604,
"calibration/batch_uniqueness": 0.5406986655035744,
"calibration/confidence_entropy": 0.48598507484551545,
"calibration/coverage@0%": 0.07604166666666666,
"calibration/coverage@1%": 0.10104166666666667,
"calibration/coverage@10%": 0.3617887837249782,
"calibration/coverage@15%": 0.7390489012184508,
"calibration/coverage@20%": 0.8335223563968668,
"calibration/coverage@25%": 0.9101827676240208,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.32266236945169713,
"calibration/distribution_entropy_10": 0.6992091580531643,
"calibration/distribution_entropy_100": 0.3825769752796119,
"calibration/ece": 0.15797976501305483,
"calibration/mean_confidence": 0.6753209312445605,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.001388888888888906,
"completions/max_length": 3361.2,
"completions/max_terminated_length": 3361.2,
"completions/mean_length": 1019.3338500976563,
"completions/mean_terminated_length": 1020.7703857421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 292.8,
"epoch": 1.09375,
"grad_norm": 0.0006985415820963681,
"learning_rate": 3.5156250000000003e-06,
"loss": 0.0013,
"num_tokens": 1094556460.0,
"reward": 1.2889756202697753,
"reward_std": 0.12136874049901962,
"rewards/accuracy_reward": 0.7276909708976745,
"rewards/brier_reward": 0.8517222285270691,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.998524296283722,
"rewards/mean_confidence_reward": 0.6873003482818604,
"sampling/batch_mean_priority_error": 0.04208454441084229,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3555555555555555,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008451656624674796,
"sampling/priority_kl": 0.02999986857175827,
"sampling/priority_scale": 0.7116955935722217,
"sampling/prob_entropy": 10.278954315185548,
"sampling/prob_max": 4.657706740545109e-05,
"sampling/prob_min": 1.9494253501761703e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0871999979019165,
"sampling/prompt_draws_total": 32616.0,
"sampling/seen_fraction": 0.7340999960899353,
"sampling/unseen_fraction": 0.2659000039100647,
"signal/accuracy_reward/centered_abs_mean": 0.14761827141046524,
"signal/accuracy_reward/group_std_mean": 0.1924243152141571,
"signal/accuracy_reward/group_zero_std_frac": 0.46111111640930175,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.07380913570523262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.07380913570523262,
"signal/advantage_abs_mean": 0.08975814431905746,
"signal/advantage_pre_scale_abs_mean": 0.08975814431905746,
"signal/advantage_pre_scale_std": 0.16560624539852142,
"signal/advantage_std": 0.16560624539852142,
"signal/brier_reward/centered_abs_mean": 0.07768742889165878,
"signal/brier_reward/group_std_mean": 0.10408626347780228,
"signal/brier_reward/group_zero_std_frac": 0.21111111640930175,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03884371444582939,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03884371444582939,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0028591578942723573,
"signal/format_reward/group_std_mean": 0.008347788266837597,
"signal/format_reward/group_zero_std_frac": 0.9527777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0014295789471361787,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0014295789471361787,
"signal/mean_confidence_reward/centered_abs_mean": 0.06654405444860459,
"signal/mean_confidence_reward/group_std_mean": 0.0879412516951561,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23055556118488313,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.654405297012999e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.654405297012999e-07,
"step": 455
},
{
"calibration/aurc": 0.10392280207853939,
"calibration/batch_distribution_entropy": 0.5815910248331363,
"calibration/batch_entropy_100bins": 0.3225027846771071,
"calibration/batch_entropy_10bins": 0.5815910248331363,
"calibration/batch_entropy_50bins": 0.37964506007055215,
"calibration/batch_uniqueness": 0.2745551215277778,
"calibration/confidence_entropy": 0.437338301182815,
"calibration/coverage@0%": 0.15208333333333332,
"calibration/coverage@1%": 0.15208333333333332,
"calibration/coverage@10%": 0.6505208333333334,
"calibration/coverage@15%": 0.8020833333333334,
"calibration/coverage@20%": 0.8286458333333334,
"calibration/coverage@25%": 0.8755208333333334,
"calibration/coverage@30%": 0.9072916666666668,
"calibration/coverage@5%": 0.4265625,
"calibration/distribution_entropy_10": 0.5815910248331363,
"calibration/distribution_entropy_100": 0.3225027846771071,
"calibration/ece": 0.11749999999999991,
"calibration/mean_confidence": 0.6855208333333334,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0012152777777777902,
"completions/max_length": 3590.2,
"completions/max_terminated_length": 3590.2,
"completions/mean_length": 1007.80869140625,
"completions/mean_terminated_length": 1009.020703125,
"completions/min_length": 68.8,
"completions/min_terminated_length": 282.2,
"epoch": 1.1057692307692308,
"grad_norm": 0.0006103526102378964,
"learning_rate": 3.4855769230769233e-06,
"loss": -0.0002,
"num_tokens": 1109287856.0,
"reward": 1.2962172985076905,
"reward_std": 0.10616759359836578,
"rewards/accuracy_reward": 0.7383680582046509,
"rewards/brier_reward": 0.8554409980773926,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9986111044883728,
"rewards/mean_confidence_reward": 0.7227604031562805,
"sampling/batch_mean_priority_error": 0.05214705427814047,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.34444444444444444,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008502156287431718,
"sampling/priority_kl": 0.029999998584389686,
"sampling/priority_scale": 0.711798530840315,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 4.672173818107694e-05,
"sampling/prob_min": 1.955234147317242e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.0992000102996826,
"sampling/prompt_draws_total": 32976.0,
"sampling/seen_fraction": 0.7384333372116089,
"sampling/unseen_fraction": 0.26156666278839114,
"signal/accuracy_reward/centered_abs_mean": 0.10821397453546525,
"signal/accuracy_reward/group_std_mean": 0.14976944923400878,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05410698726773262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05410698726773262,
"signal/advantage_abs_mean": 0.0763231635093689,
"signal/advantage_pre_scale_abs_mean": 0.0763231635093689,
"signal/advantage_pre_scale_std": 0.15467407703399658,
"signal/advantage_std": 0.15467407703399658,
"signal/brier_reward/centered_abs_mean": 0.07721022069454193,
"signal/brier_reward/group_std_mean": 0.10425049513578415,
"signal/brier_reward/group_zero_std_frac": 0.2888888865709305,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03860511034727097,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03860511034727097,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0025933159748092295,
"signal/format_reward/group_std_mean": 0.005929096881300211,
"signal/format_reward/group_zero_std_frac": 0.9722222089767456,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0012966579874046148,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0012966579874046148,
"signal/mean_confidence_reward/centered_abs_mean": 0.062055671215057374,
"signal/mean_confidence_reward/group_std_mean": 0.08193774446845055,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3277777761220932,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.205566592143441e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.205566592143441e-07,
"step": 460
},
{
"calibration/aurc": 0.08059233839576513,
"calibration/batch_distribution_entropy": 0.4605690366871271,
"calibration/batch_entropy_100bins": 0.2529573163787063,
"calibration/batch_entropy_10bins": 0.4605690366871271,
"calibration/batch_entropy_50bins": 0.29777726002592436,
"calibration/batch_uniqueness": -0.004366139578864012,
"calibration/confidence_entropy": 0.4054791820657573,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.16510416666666666,
"calibration/coverage@10%": 0.5729166666666667,
"calibration/coverage@15%": 0.9066294060052218,
"calibration/coverage@20%": 0.9269704634464752,
"calibration/coverage@25%": 0.954622225848564,
"calibration/coverage@30%": 0.9963446475195823,
"calibration/coverage@5%": 0.5007316800714924,
"calibration/distribution_entropy_10": 0.4605690366871271,
"calibration/distribution_entropy_100": 0.2529573163787063,
"calibration/ece": 0.10463188608756416,
"calibration/mean_confidence": 0.7691945551062004,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003125,
"completions/max_length": 3376.0,
"completions/max_terminated_length": 3376.0,
"completions/mean_length": 1011.2415893554687,
"completions/mean_terminated_length": 1014.4902587890625,
"completions/min_length": 65.2,
"completions/min_terminated_length": 284.6,
"epoch": 1.1177884615384615,
"grad_norm": 0.000557567982468754,
"learning_rate": 3.4555288461538466e-06,
"loss": -0.0009,
"num_tokens": 1124026927.0,
"reward": 1.2991013526916504,
"reward_std": 0.11327702552080154,
"rewards/accuracy_reward": 0.7419270873069763,
"rewards/brier_reward": 0.8595590353012085,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967013835906983,
"rewards/mean_confidence_reward": 0.7529860854148864,
"sampling/batch_mean_priority_error": 0.053493539116674746,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32499999999999996,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008554048836231232,
"sampling/priority_kl": 0.029999839141964912,
"sampling/priority_scale": 0.711510258843191,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.685054227593355e-05,
"sampling/prob_min": 1.9613161566667258e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1112000226974488,
"sampling/prompt_draws_total": 33336.0,
"sampling/seen_fraction": 0.7422599911689758,
"sampling/unseen_fraction": 0.25774000883102416,
"signal/accuracy_reward/centered_abs_mean": 0.12089301198720932,
"signal/accuracy_reward/group_std_mean": 0.16041539907455443,
"signal/accuracy_reward/group_zero_std_frac": 0.5388889014720917,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06044650599360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06044650599360466,
"signal/advantage_abs_mean": 0.08364090770483017,
"signal/advantage_pre_scale_abs_mean": 0.08364090770483017,
"signal/advantage_pre_scale_std": 0.16810988783836364,
"signal/advantage_std": 0.16810988783836364,
"signal/brier_reward/centered_abs_mean": 0.07260150983929634,
"signal/brier_reward/group_std_mean": 0.09838626384735108,
"signal/brier_reward/group_zero_std_frac": 0.35833334028720853,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03630075491964817,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03630075491964817,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00542534728301689,
"signal/format_reward/group_std_mean": 0.01027821134775877,
"signal/format_reward/group_zero_std_frac": 0.955555546283722,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002712673641508445,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002712673641508445,
"signal/mean_confidence_reward/centered_abs_mean": 0.061685122549533844,
"signal/mean_confidence_reward/group_std_mean": 0.08196858763694763,
"signal/mean_confidence_reward/group_zero_std_frac": 0.397222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.168512015847227e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.168512015847227e-07,
"step": 465
},
{
"calibration/aurc": 0.17243657087985084,
"calibration/batch_distribution_entropy": 0.4785682077586254,
"calibration/batch_entropy_100bins": 0.26017619224627336,
"calibration/batch_entropy_10bins": 0.4785682077586254,
"calibration/batch_entropy_50bins": 0.3062752038967914,
"calibration/batch_uniqueness": 0.013543680388338326,
"calibration/confidence_entropy": 0.4068504298385657,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4390859952770221,
"calibration/coverage@15%": 0.6420314426741135,
"calibration/coverage@20%": 0.7307174703019095,
"calibration/coverage@25%": 0.7581151832460733,
"calibration/coverage@30%": 0.7664921465968586,
"calibration/coverage@5%": 0.13838120104438642,
"calibration/distribution_entropy_10": 0.4785682077586254,
"calibration/distribution_entropy_100": 0.26017619224627336,
"calibration/ece": 0.11765368328499626,
"calibration/mean_confidence": 0.7780121137494821,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.003472222222222232,
"completions/max_length": 3773.8,
"completions/max_terminated_length": 3773.8,
"completions/mean_length": 1097.7507080078126,
"completions/mean_terminated_length": 1101.61884765625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 287.6,
"epoch": 1.1298076923076923,
"grad_norm": 0.0006817008252255619,
"learning_rate": 3.4254807692307695e-06,
"loss": -0.0026,
"num_tokens": 1139782103.0,
"reward": 1.2943272113800048,
"reward_std": 0.1254418209195137,
"rewards/accuracy_reward": 0.7381944417953491,
"rewards/brier_reward": 0.8540902853012085,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541626930237,
"rewards/mean_confidence_reward": 0.7828645825386047,
"sampling/batch_mean_priority_error": 0.05845517749135944,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.38333333333333336,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008615374192595482,
"sampling/priority_kl": 0.030000054463744163,
"sampling/priority_scale": 0.7117458641761913,
"sampling/prob_entropy": 10.278947448730468,
"sampling/prob_max": 4.699580094893463e-05,
"sampling/prob_min": 1.9646077998913824e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.123200011253357,
"sampling/prompt_draws_total": 33696.0,
"sampling/seen_fraction": 0.7464800000190734,
"sampling/unseen_fraction": 0.2535199999809265,
"signal/accuracy_reward/centered_abs_mean": 0.11816406399011611,
"signal/accuracy_reward/group_std_mean": 0.1576286733150482,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05908203199505806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05908203199505806,
"signal/advantage_abs_mean": 0.09323482066392899,
"signal/advantage_pre_scale_abs_mean": 0.09323482066392899,
"signal/advantage_pre_scale_std": 0.18418655097484588,
"signal/advantage_std": 0.18418655097484588,
"signal/brier_reward/centered_abs_mean": 0.08546116352081298,
"signal/brier_reward/group_std_mean": 0.11382831633090973,
"signal/brier_reward/group_zero_std_frac": 0.37777777910232546,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04273058176040649,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04273058176040649,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006152343726716936,
"signal/format_reward/group_std_mean": 0.011012036725878716,
"signal/format_reward/group_zero_std_frac": 0.9555555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003076171863358468,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003076171863358468,
"signal/mean_confidence_reward/centered_abs_mean": 0.0545052208006382,
"signal/mean_confidence_reward/group_std_mean": 0.0735972911119461,
"signal/mean_confidence_reward/group_zero_std_frac": 0.43888888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.450521996408497e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.450521996408497e-07,
"step": 470
},
{
"calibration/aurc": 0.2685868515086937,
"calibration/batch_distribution_entropy": 0.41914427430692525,
"calibration/batch_entropy_100bins": 0.22963451571698973,
"calibration/batch_entropy_10bins": 0.41914427430692525,
"calibration/batch_entropy_50bins": 0.2703220364467047,
"calibration/batch_uniqueness": -0.12729341363433064,
"calibration/confidence_entropy": 0.3964918726873754,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.13368146214099216,
"calibration/coverage@15%": 0.14203655352480418,
"calibration/coverage@20%": 0.30809399477806787,
"calibration/coverage@25%": 0.5036196234712107,
"calibration/coverage@30%": 0.5528253401126838,
"calibration/coverage@5%": 0.12741514360313316,
"calibration/distribution_entropy_10": 0.41914427430692525,
"calibration/distribution_entropy_100": 0.22963451571698973,
"calibration/ece": 0.1977212607645093,
"calibration/mean_confidence": 0.7959806739842428,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.002777777777777812,
"completions/max_length": 4003.8,
"completions/max_terminated_length": 4003.8,
"completions/mean_length": 1113.821875,
"completions/mean_terminated_length": 1116.839892578125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 334.0,
"epoch": 1.1418269230769231,
"grad_norm": 0.0006080775056034327,
"learning_rate": 3.3954326923076925e-06,
"loss": -0.0022,
"num_tokens": 1155714611.0,
"reward": 1.2394050359725952,
"reward_std": 0.12434252202510834,
"rewards/accuracy_reward": 0.6710069417953491,
"rewards/brier_reward": 0.8106519103050231,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9971354246139527,
"rewards/mean_confidence_reward": 0.7947829842567444,
"sampling/batch_mean_priority_error": 0.09056192310100955,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3333333333333333,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.00870037693530321,
"sampling/priority_kl": 0.03000004291534424,
"sampling/priority_scale": 0.7121343434089795,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 4.71480016130954e-05,
"sampling/prob_min": 1.961133639269974e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1351999759674072,
"sampling/prompt_draws_total": 34056.0,
"sampling/seen_fraction": 0.7507666826248169,
"sampling/unseen_fraction": 0.24923331737518312,
"signal/accuracy_reward/centered_abs_mean": 0.10917968600988388,
"signal/accuracy_reward/group_std_mean": 0.1482059821486473,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05458984300494194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05458984300494194,
"signal/advantage_abs_mean": 0.09205847829580308,
"signal/advantage_pre_scale_abs_mean": 0.09205847829580308,
"signal/advantage_pre_scale_std": 0.18328936994075776,
"signal/advantage_std": 0.18328936994075776,
"signal/brier_reward/centered_abs_mean": 0.08727029114961624,
"signal/brier_reward/group_std_mean": 0.1160641372203827,
"signal/brier_reward/group_zero_std_frac": 0.3999999940395355,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.04363514557480812,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.04363514557480812,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00482313372194767,
"signal/format_reward/group_std_mean": 0.009565300587564707,
"signal/format_reward/group_zero_std_frac": 0.9583333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002411566860973835,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002411566860973835,
"signal/mean_confidence_reward/centered_abs_mean": 0.054275186359882356,
"signal/mean_confidence_reward/group_std_mean": 0.0711105227470398,
"signal/mean_confidence_reward/group_zero_std_frac": 0.46111111640930175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.42751837429023e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.42751837429023e-07,
"step": 475
},
{
"calibration/aurc": 0.16509835396682498,
"calibration/batch_distribution_entropy": 0.5817777701404863,
"calibration/batch_entropy_100bins": 0.3238100239509861,
"calibration/batch_entropy_10bins": 0.5817777701404863,
"calibration/batch_entropy_50bins": 0.3811839209928083,
"calibration/batch_uniqueness": 0.2799376063331884,
"calibration/confidence_entropy": 0.44051279536966287,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.46918061174987463,
"calibration/coverage@15%": 0.5476342486155917,
"calibration/coverage@20%": 0.5873400706598004,
"calibration/coverage@25%": 0.7842809230271447,
"calibration/coverage@30%": 0.8664871199194095,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.5817777701404863,
"calibration/distribution_entropy_100": 0.3238100239509861,
"calibration/ece": 0.08974379444828766,
"calibration/mean_confidence": 0.7054703479311134,
"calibration/unique_confidence_per_question": 0.0234375,
"calibration/unique_confidences": 9.0,
"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": 0.004600694444444442,
"completions/max_length": 3721.2,
"completions/max_terminated_length": 3721.2,
"completions/mean_length": 1088.9832641601563,
"completions/mean_terminated_length": 1094.0543090820313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 288.6,
"epoch": 1.1538461538461537,
"grad_norm": 0.0006057985010556877,
"learning_rate": 3.365384615384616e-06,
"loss": -0.0027,
"num_tokens": 1171333938.0,
"reward": 1.311488938331604,
"reward_std": 0.1304037630558014,
"rewards/accuracy_reward": 0.7541666626930237,
"rewards/brier_reward": 0.8734835147857666,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.995312488079071,
"rewards/mean_confidence_reward": 0.7591232657432556,
"sampling/batch_mean_priority_error": 0.0432108694971293,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3416666666666667,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008764292113482952,
"sampling/priority_kl": 0.029999880865216256,
"sampling/priority_scale": 0.7124254643684254,
"sampling/prob_entropy": 10.278950309753418,
"sampling/prob_max": 4.7294204705394804e-05,
"sampling/prob_min": 1.9665095533127896e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1471999883651733,
"sampling/prompt_draws_total": 34416.0,
"sampling/seen_fraction": 0.7548466563224793,
"sampling/unseen_fraction": 0.24515334367752076,
"signal/accuracy_reward/centered_abs_mean": 0.1254557266831398,
"signal/accuracy_reward/group_std_mean": 0.17018299102783202,
"signal/accuracy_reward/group_zero_std_frac": 0.5000000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0627278633415699,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0627278633415699,
"signal/advantage_abs_mean": 0.09320852905511856,
"signal/advantage_pre_scale_abs_mean": 0.09320852905511856,
"signal/advantage_pre_scale_std": 0.18777829706668853,
"signal/advantage_std": 0.18777829706668853,
"signal/brier_reward/centered_abs_mean": 0.07927191704511642,
"signal/brier_reward/group_std_mean": 0.10883116573095322,
"signal/brier_reward/group_zero_std_frac": 0.29166666865348817,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03963595852255821,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03963595852255821,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.008398437546566129,
"signal/format_reward/group_std_mean": 0.017096645571291445,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0041992187732830645,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0041992187732830645,
"signal/mean_confidence_reward/centered_abs_mean": 0.05635145679116249,
"signal/mean_confidence_reward/group_std_mean": 0.07729533314704895,
"signal/mean_confidence_reward/group_zero_std_frac": 0.347222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.635145612359338e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.635145612359338e-07,
"step": 480
},
{
"calibration/aurc": 0.14424101126813588,
"calibration/batch_distribution_entropy": 0.6692289095471307,
"calibration/batch_entropy_100bins": 0.3789358043962861,
"calibration/batch_entropy_10bins": 0.6692289095471307,
"calibration/batch_entropy_50bins": 0.44607709780536053,
"calibration/batch_uniqueness": 0.5106775389038136,
"calibration/confidence_entropy": 0.49288429135846296,
"calibration/coverage@0%": 0.11137752381993904,
"calibration/coverage@1%": 0.11137752381993904,
"calibration/coverage@10%": 0.31678621490802383,
"calibration/coverage@15%": 0.5255120119247787,
"calibration/coverage@20%": 0.7083988676090485,
"calibration/coverage@25%": 0.8282933124057072,
"calibration/coverage@30%": 0.9046870268989673,
"calibration/coverage@5%": 0.2693903815746904,
"calibration/distribution_entropy_10": 0.6692289095471307,
"calibration/distribution_entropy_100": 0.3789358043962861,
"calibration/ece": 0.08697532913278365,
"calibration/mean_confidence": 0.6685785178846475,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003732638888888906,
"completions/max_length": 3593.6,
"completions/max_terminated_length": 3593.6,
"completions/mean_length": 1068.6515747070312,
"completions/mean_terminated_length": 1072.6423828125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 291.6,
"epoch": 1.1658653846153846,
"grad_norm": 0.00042986380867660046,
"learning_rate": 3.3353365384615388e-06,
"loss": -0.0028,
"num_tokens": 1186752388.0,
"reward": 1.291651439666748,
"reward_std": 0.10780009925365448,
"rewards/accuracy_reward": 0.7271701335906983,
"rewards/brier_reward": 0.8598515510559082,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9962673664093018,
"rewards/mean_confidence_reward": 0.695164954662323,
"sampling/batch_mean_priority_error": 0.040754491299291075,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.38888888888888884,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008808174729347229,
"sampling/priority_kl": 0.03000020273029804,
"sampling/priority_scale": 0.7133005202980712,
"sampling/prob_entropy": 10.278949737548828,
"sampling/prob_max": 4.746251870528795e-05,
"sampling/prob_min": 1.9713817891897634e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1592000007629395,
"sampling/prompt_draws_total": 34776.0,
"sampling/seen_fraction": 0.7593999981880188,
"sampling/unseen_fraction": 0.2406000018119812,
"signal/accuracy_reward/centered_abs_mean": 0.11838650405406952,
"signal/accuracy_reward/group_std_mean": 0.16019512414932252,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05919325202703476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05919325202703476,
"signal/advantage_abs_mean": 0.07694000229239464,
"signal/advantage_pre_scale_abs_mean": 0.07694000229239464,
"signal/advantage_pre_scale_std": 0.15797904133796692,
"signal/advantage_std": 0.15797904133796692,
"signal/brier_reward/centered_abs_mean": 0.06782751232385635,
"signal/brier_reward/group_std_mean": 0.09157582372426987,
"signal/brier_reward/group_zero_std_frac": 0.2694444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033913756161928175,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033913756161928175,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006906467000953853,
"signal/format_reward/group_std_mean": 0.015525073930621146,
"signal/format_reward/group_zero_std_frac": 0.9277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0034532335004769264,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0034532335004769264,
"signal/mean_confidence_reward/centered_abs_mean": 0.05602431297302246,
"signal/mean_confidence_reward/group_std_mean": 0.07353781908750534,
"signal/mean_confidence_reward/group_zero_std_frac": 0.29722222983837127,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.602431428997079e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.602431428997079e-07,
"step": 485
},
{
"calibration/aurc": 0.07385052157310985,
"calibration/batch_distribution_entropy": 0.6865127587476862,
"calibration/batch_entropy_100bins": 0.38104507949249855,
"calibration/batch_entropy_10bins": 0.6865127587476862,
"calibration/batch_entropy_50bins": 0.44856010232083676,
"calibration/batch_uniqueness": 0.5568101683217519,
"calibration/confidence_entropy": 0.5090127053537694,
"calibration/coverage@0%": 0.10793520942408377,
"calibration/coverage@1%": 0.17477594049457984,
"calibration/coverage@10%": 0.6945750061244469,
"calibration/coverage@15%": 0.8549916329618883,
"calibration/coverage@20%": 0.9536036771105308,
"calibration/coverage@25%": 0.9838120104438642,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.488348268242808,
"calibration/distribution_entropy_10": 0.6865127587476862,
"calibration/distribution_entropy_100": 0.38104507949249855,
"calibration/ece": 0.15832489957513127,
"calibration/mean_confidence": 0.6925591290156485,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.006250000000000022,
"completions/max_length": 3622.0,
"completions/max_terminated_length": 3622.0,
"completions/mean_length": 1027.5976806640624,
"completions/mean_terminated_length": 1034.0362060546875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 233.6,
"epoch": 1.1778846153846154,
"grad_norm": 0.0005065092700533569,
"learning_rate": 3.3052884615384617e-06,
"loss": -0.0066,
"num_tokens": 1201653385.0,
"reward": 1.2866623878479004,
"reward_std": 0.10978196710348129,
"rewards/accuracy_reward": 0.7243923664093017,
"rewards/brier_reward": 0.8551692605018616,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99375,
"rewards/mean_confidence_reward": 0.6556684136390686,
"sampling/batch_mean_priority_error": 0.031650594283802194,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.28055555555555556,
"sampling/error_ema_max": 0.20249629020690918,
"sampling/error_ema_mean": 0.008839315734803677,
"sampling/priority_kl": 0.029999951645731927,
"sampling/priority_scale": 0.7136028229957446,
"sampling/prob_entropy": 10.278948211669922,
"sampling/prob_max": 4.760698429890908e-05,
"sampling/prob_min": 1.976646854018327e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1711999893188476,
"sampling/prompt_draws_total": 35136.0,
"sampling/seen_fraction": 0.7632866621017456,
"sampling/unseen_fraction": 0.23671333789825438,
"signal/accuracy_reward/centered_abs_mean": 0.11789821982383727,
"signal/accuracy_reward/group_std_mean": 0.16054655611515045,
"signal/accuracy_reward/group_zero_std_frac": 0.5277777850627899,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05894910991191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05894910991191864,
"signal/advantage_abs_mean": 0.07676194682717323,
"signal/advantage_pre_scale_abs_mean": 0.07676194682717323,
"signal/advantage_pre_scale_std": 0.1554905205965042,
"signal/advantage_std": 0.1554905205965042,
"signal/brier_reward/centered_abs_mean": 0.0649530403316021,
"signal/brier_reward/group_std_mean": 0.09048911035060883,
"signal/brier_reward/group_zero_std_frac": 0.1361111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03247652016580105,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03247652016580105,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01105685792863369,
"signal/format_reward/group_std_mean": 0.02208307832479477,
"signal/format_reward/group_zero_std_frac": 0.9055555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005528428964316845,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005528428964316845,
"signal/mean_confidence_reward/centered_abs_mean": 0.06184353157877922,
"signal/mean_confidence_reward/group_std_mean": 0.08153964579105377,
"signal/mean_confidence_reward/group_zero_std_frac": 0.14444444626569747,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 6.184352855598263e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 6.184352855598263e-07,
"step": 490
},
{
"calibration/aurc": 0.06241843717141251,
"calibration/batch_distribution_entropy": 0.6022017445135018,
"calibration/batch_entropy_100bins": 0.32593844547926,
"calibration/batch_entropy_10bins": 0.6022017445135018,
"calibration/batch_entropy_50bins": 0.3836894643783205,
"calibration/batch_uniqueness": 0.3656484331557499,
"calibration/confidence_entropy": 0.46177567770797545,
"calibration/coverage@0%": 0.31406934962374067,
"calibration/coverage@1%": 0.4649504376487631,
"calibration/coverage@10%": 0.7808990655147648,
"calibration/coverage@15%": 0.8241748638517699,
"calibration/coverage@20%": 0.8997835679177838,
"calibration/coverage@25%": 0.9518068588025022,
"calibration/coverage@30%": 0.975,
"calibration/coverage@5%": 0.5931389600144851,
"calibration/distribution_entropy_10": 0.6022017445135018,
"calibration/distribution_entropy_100": 0.32593844547926,
"calibration/ece": 0.11784205514953403,
"calibration/mean_confidence": 0.7561079918549011,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.005555555555555536,
"completions/max_length": 3740.2,
"completions/max_terminated_length": 3740.2,
"completions/mean_length": 1011.887255859375,
"completions/mean_terminated_length": 1017.530224609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 273.2,
"epoch": 1.1899038461538463,
"grad_norm": 0.000543592032045126,
"learning_rate": 3.2752403846153846e-06,
"loss": -0.0054,
"num_tokens": 1216408662.0,
"reward": 1.2978950262069702,
"reward_std": 0.11456762105226517,
"rewards/accuracy_reward": 0.7465277791023255,
"rewards/brier_reward": 0.8549774646759033,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9942708373069763,
"rewards/mean_confidence_reward": 0.6990104198455811,
"sampling/batch_mean_priority_error": 0.03569228065040018,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3388888888888889,
"sampling/error_ema_max": 0.20283221006393432,
"sampling/error_ema_mean": 0.008870593085885049,
"sampling/priority_kl": 0.030000014975667,
"sampling/priority_scale": 0.7138257563346997,
"sampling/prob_entropy": 10.278950691223145,
"sampling/prob_max": 4.774942208314314e-05,
"sampling/prob_min": 1.9820158195216207e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.1832000017166138,
"sampling/prompt_draws_total": 35496.0,
"sampling/seen_fraction": 0.7670066595077515,
"sampling/unseen_fraction": 0.23299334049224854,
"signal/accuracy_reward/centered_abs_mean": 0.1231119766831398,
"signal/accuracy_reward/group_std_mean": 0.16388897597789764,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0615559883415699,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0615559883415699,
"signal/advantage_abs_mean": 0.0827570840716362,
"signal/advantage_pre_scale_abs_mean": 0.0827570840716362,
"signal/advantage_pre_scale_std": 0.1666143447160721,
"signal/advantage_std": 0.1666143447160721,
"signal/brier_reward/centered_abs_mean": 0.06436868607997895,
"signal/brier_reward/group_std_mean": 0.08770309686660767,
"signal/brier_reward/group_zero_std_frac": 0.2194444417953491,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.032184343039989474,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.032184343039989474,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010123697854578496,
"signal/format_reward/group_std_mean": 0.019488899409770964,
"signal/format_reward/group_zero_std_frac": 0.919444465637207,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005061848927289248,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005061848927289248,
"signal/mean_confidence_reward/centered_abs_mean": 0.05872992873191833,
"signal/mean_confidence_reward/group_std_mean": 0.07616192176938057,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2194444417953491,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.872992744571093e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.872992744571093e-07,
"step": 495
},
{
"calibration/aurc": 0.16765371594935913,
"calibration/batch_distribution_entropy": 0.6470495126669717,
"calibration/batch_entropy_100bins": 0.35226262364466754,
"calibration/batch_entropy_10bins": 0.6470495126669717,
"calibration/batch_entropy_50bins": 0.414677860993004,
"calibration/batch_uniqueness": 0.45946199632610396,
"calibration/confidence_entropy": 0.47391215467122316,
"calibration/coverage@0%": 0.09608355091383812,
"calibration/coverage@1%": 0.22036553524804176,
"calibration/coverage@10%": 0.43185378590078327,
"calibration/coverage@15%": 0.44960835509138375,
"calibration/coverage@20%": 0.6095449847693646,
"calibration/coverage@25%": 0.7096945713664056,
"calibration/coverage@30%": 0.8529917319408181,
"calibration/coverage@5%": 0.2710182767624021,
"calibration/distribution_entropy_10": 0.6470495126669717,
"calibration/distribution_entropy_100": 0.35226262364466754,
"calibration/ece": 0.138503590078329,
"calibration/mean_confidence": 0.7416133594429939,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0051215277777778125,
"completions/max_length": 3449.0,
"completions/max_terminated_length": 3449.0,
"completions/mean_length": 971.3076416015625,
"completions/mean_terminated_length": 976.3692626953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 305.8,
"epoch": 1.2019230769230769,
"grad_norm": 0.0005834887851960957,
"learning_rate": 3.245192307692308e-06,
"loss": -0.0049,
"num_tokens": 1230721070.0,
"reward": 1.2864141702651977,
"reward_std": 0.1127560317516327,
"rewards/accuracy_reward": 0.7265624880790711,
"rewards/brier_reward": 0.8513723850250244,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9948784708976746,
"rewards/mean_confidence_reward": 0.7555121302604675,
"sampling/batch_mean_priority_error": 0.04447198323683641,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3277777777777777,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.008906039968132972,
"sampling/priority_kl": 0.03000001348555088,
"sampling/priority_scale": 0.7144269287819043,
"sampling/prob_entropy": 10.278946876525879,
"sampling/prob_max": 4.7904710663715376e-05,
"sampling/prob_min": 1.9869897369062527e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.19520001411438,
"sampling/prompt_draws_total": 35856.0,
"sampling/seen_fraction": 0.7709800004959106,
"sampling/unseen_fraction": 0.22901999950408936,
"signal/accuracy_reward/centered_abs_mean": 0.10986327975988389,
"signal/accuracy_reward/group_std_mean": 0.14956827759742736,
"signal/accuracy_reward/group_zero_std_frac": 0.5527777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05493163987994194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05493163987994194,
"signal/advantage_abs_mean": 0.08156569451093673,
"signal/advantage_pre_scale_abs_mean": 0.08156569451093673,
"signal/advantage_pre_scale_std": 0.1688571810722351,
"signal/advantage_std": 0.1688571810722351,
"signal/brier_reward/centered_abs_mean": 0.06213839054107666,
"signal/brier_reward/group_std_mean": 0.08537867367267608,
"signal/brier_reward/group_zero_std_frac": 0.24166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03106919527053833,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03106919527053833,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007731119659729302,
"signal/format_reward/group_std_mean": 0.013650832884013652,
"signal/format_reward/group_zero_std_frac": 0.9444444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003865559829864651,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003865559829864651,
"signal/mean_confidence_reward/centered_abs_mean": 0.04973362013697624,
"signal/mean_confidence_reward/group_std_mean": 0.06599054262042045,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.973361910742824e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.973361910742824e-07,
"step": 500
},
{
"epoch": 1.2019230769230769,
"eval_calibration/aurc": 0.12491768383573112,
"eval_calibration/batch_distribution_entropy": 0.589795021571096,
"eval_calibration/batch_entropy_100bins": 0.3239887453964619,
"eval_calibration/batch_entropy_10bins": 0.589795021571096,
"eval_calibration/batch_entropy_50bins": 0.3813943089867336,
"eval_calibration/batch_uniqueness": 0.29058405650919084,
"eval_calibration/confidence_entropy": 0.44710805611226434,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5693112467306015,
"eval_calibration/coverage@15%": 0.7471665213600698,
"eval_calibration/coverage@20%": 0.8186573670444638,
"eval_calibration/coverage@25%": 0.939843068875327,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.589795021571096,
"eval_calibration/distribution_entropy_100": 0.3239887453964619,
"eval_calibration/ece": 0.05074106364428948,
"eval_calibration/mean_confidence": 0.760767218831735,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.004340277777777808,
"eval_completions/max_length": 2845.3333333333335,
"eval_completions/max_terminated_length": 2845.3333333333335,
"eval_completions/mean_length": 986.4887390136719,
"eval_completions/mean_terminated_length": 990.8467610677084,
"eval_completions/min_length": 75.0,
"eval_completions/min_terminated_length": 329.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1230721070.0,
"eval_reward": 1.288488507270813,
"eval_reward_std": 0.31713973979155224,
"eval_rewards/accuracy_reward": 0.7265625,
"eval_rewards/brier_reward": 0.8547395765781403,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.995659738779068,
"eval_rewards/mean_confidence_reward": 0.7574652334054311,
"eval_runtime": 207.7192,
"eval_samples_per_second": 4.814,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3863389740387599,
"eval_signal/accuracy_reward/group_std_mean": 0.44475344320138294,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19316948701937994,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19316948701937994,
"eval_signal/advantage_abs_mean": 0.2657373497883479,
"eval_signal/advantage_pre_scale_abs_mean": 0.2657373497883479,
"eval_signal/advantage_pre_scale_std": 0.3151099781195323,
"eval_signal/advantage_std": 0.3151099781195323,
"eval_signal/brier_reward/centered_abs_mean": 0.16554688413937887,
"eval_signal/brier_reward/group_std_mean": 0.22295028964678446,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08277344206968944,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08277344206968944,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.008409287935743729,
"eval_signal/format_reward/group_std_mean": 0.02455231888840596,
"eval_signal/format_reward/group_zero_std_frac": 0.8611111442248026,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.004204643967871864,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.004204643967871864,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.16469181329011917,
"eval_signal/mean_confidence_reward/group_std_mean": 0.20532790819803873,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.646918121878116e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.646918121878116e-06,
"eval_steps_per_second": 0.029,
"step": 500
},
{
"epoch": 1.2019230769230769,
"step": 500,
"train_probe_calibration/aurc": 0.0909489939142326,
"train_probe_calibration/batch_distribution_entropy": 0.5765734888146544,
"train_probe_calibration/batch_entropy_100bins": 0.315147910452945,
"train_probe_calibration/batch_entropy_10bins": 0.5765734888146544,
"train_probe_calibration/batch_entropy_50bins": 0.37098702113473647,
"train_probe_calibration/batch_uniqueness": 0.26795141206308043,
"train_probe_calibration/confidence_entropy": 0.44427701730624236,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6899563318777293,
"train_probe_calibration/coverage@15%": 0.8296943231441049,
"train_probe_calibration/coverage@20%": 0.8951965065502183,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5765734888146544,
"train_probe_calibration/distribution_entropy_100": 0.315147910452945,
"train_probe_calibration/ece": 0.05414847161572035,
"train_probe_calibration/mean_confidence": 0.7689082969432315,
"train_probe_calibration/unique_confidence_per_question": 0.006944444444444444,
"train_probe_calibration/unique_confidences": 8,
"train_probe_completions/clipped_ratio": 0.005208333333333333,
"train_probe_completions/max_length": 3334.8333333333335,
"train_probe_completions/max_terminated_length": 3334.8333333333335,
"train_probe_completions/mean_length": 1006.7073262532552,
"train_probe_completions/mean_terminated_length": 1012.0302327473959,
"train_probe_completions/min_length": 105.16666666666667,
"train_probe_completions/min_terminated_length": 273.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1230721070.0,
"train_probe_reward": 1.3158843914667766,
"train_probe_reward_std": 0.3002795527378718,
"train_probe_rewards/accuracy_reward": 0.7630208233992258,
"train_probe_rewards/brier_reward": 0.8748090267181396,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9939236144224802,
"train_probe_rewards/mean_confidence_reward": 0.7642360925674438,
"train_probe_runtime": 203.9359,
"train_probe_samples_per_second": 4.904,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3511827290058136,
"train_probe_signal/accuracy_reward/group_std_mean": 0.42282938460508984,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1755913645029068,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1755913645029068,
"train_probe_signal/advantage_abs_mean": 0.24092507114013037,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.24092507114013037,
"train_probe_signal/advantage_pre_scale_std": 0.3001803358395894,
"train_probe_signal/advantage_std": 0.3001803358395894,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14480145027240118,
"train_probe_signal/brier_reward/group_std_mean": 0.2011666273077329,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07240072513620059,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07240072513620059,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.011773003110041222,
"train_probe_signal/format_reward/group_std_mean": 0.034373246443768345,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8055555820465088,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.16013995806376138,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.19872909784317017,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.601399503670109e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.601399503670109e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.10340563136783469,
"calibration/batch_distribution_entropy": 0.5784544873790741,
"calibration/batch_entropy_100bins": 0.31161774907468726,
"calibration/batch_entropy_10bins": 0.5784544873790741,
"calibration/batch_entropy_50bins": 0.36683137227778395,
"calibration/batch_uniqueness": 0.2844652792158847,
"calibration/confidence_entropy": 0.4593182530564782,
"calibration/coverage@0%": 0.24305086599319234,
"calibration/coverage@1%": 0.2702759968832447,
"calibration/coverage@10%": 0.6897732052417556,
"calibration/coverage@15%": 0.7698497067630805,
"calibration/coverage@20%": 0.816460947659054,
"calibration/coverage@25%": 0.8677412059784956,
"calibration/coverage@30%": 0.8724409448818898,
"calibration/coverage@5%": 0.5355485582848349,
"calibration/distribution_entropy_10": 0.5784544873790741,
"calibration/distribution_entropy_100": 0.31161774907468726,
"calibration/ece": 0.16430700985154553,
"calibration/mean_confidence": 0.7618425281355379,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.004166666666666674,
"completions/max_length": 3440.4,
"completions/max_terminated_length": 3440.4,
"completions/mean_length": 986.4609375,
"completions/mean_terminated_length": 990.5824584960938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 299.0,
"epoch": 1.2139423076923077,
"grad_norm": 0.0005349395796656609,
"learning_rate": 3.215144230769231e-06,
"loss": -0.0044,
"num_tokens": 1245170988.0,
"reward": 1.2705562591552735,
"reward_std": 0.11584860682487488,
"rewards/accuracy_reward": 0.7032986044883728,
"rewards/brier_reward": 0.8420520901679993,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9957465291023254,
"rewards/mean_confidence_reward": 0.7627951264381408,
"sampling/batch_mean_priority_error": 0.04778075370517789,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32222222222222224,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.008955173008143903,
"sampling/priority_kl": 0.029999982565641403,
"sampling/priority_scale": 0.7150903284782544,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 4.8062007408589125e-05,
"sampling/prob_min": 1.9918856924050488e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.207200002670288,
"sampling/prompt_draws_total": 36216.0,
"sampling/seen_fraction": 0.7749266624450684,
"sampling/unseen_fraction": 0.22507333755493164,
"signal/accuracy_reward/centered_abs_mean": 0.11077474057674408,
"signal/accuracy_reward/group_std_mean": 0.15214249938726426,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05538737028837204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05538737028837204,
"signal/advantage_abs_mean": 0.08241928070783615,
"signal/advantage_pre_scale_abs_mean": 0.08241928070783615,
"signal/advantage_pre_scale_std": 0.1683724880218506,
"signal/advantage_std": 0.1683724880218506,
"signal/brier_reward/centered_abs_mean": 0.06432757750153542,
"signal/brier_reward/group_std_mean": 0.08972443491220475,
"signal/brier_reward/group_zero_std_frac": 0.29166666865348817,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03216378875076771,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03216378875076771,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007557508582249284,
"signal/format_reward/group_std_mean": 0.014655602350831031,
"signal/format_reward/group_zero_std_frac": 0.9388889074325562,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003778754291124642,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003778754291124642,
"signal/mean_confidence_reward/centered_abs_mean": 0.04986003115773201,
"signal/mean_confidence_reward/group_std_mean": 0.06695612743496895,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30833334028720855,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.986002693385671e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.986002693385671e-07,
"step": 505
},
{
"calibration/aurc": 0.12091567329964312,
"calibration/batch_distribution_entropy": 0.6790414307027339,
"calibration/batch_entropy_100bins": 0.38284855021121017,
"calibration/batch_entropy_10bins": 0.6790414307027339,
"calibration/batch_entropy_50bins": 0.45068311887099244,
"calibration/batch_uniqueness": 0.5521023320397787,
"calibration/confidence_entropy": 0.49995264723742083,
"calibration/coverage@0%": 0.11880101149236241,
"calibration/coverage@1%": 0.13916654674040418,
"calibration/coverage@10%": 0.46862419520911713,
"calibration/coverage@15%": 0.7173659117308444,
"calibration/coverage@20%": 0.8121888555950741,
"calibration/coverage@25%": 0.9218670120543025,
"calibration/coverage@30%": 0.9643044619422572,
"calibration/coverage@5%": 0.2675304676952913,
"calibration/distribution_entropy_10": 0.6790414307027339,
"calibration/distribution_entropy_100": 0.38284855021121017,
"calibration/ece": 0.10470707582080965,
"calibration/mean_confidence": 0.6953173891367366,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.00625,
"completions/max_length": 3732.4,
"completions/max_terminated_length": 3732.4,
"completions/mean_length": 1025.9516357421876,
"completions/mean_terminated_length": 1032.6173217773437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 279.8,
"epoch": 1.2259615384615385,
"grad_norm": 0.000531997240614146,
"learning_rate": 3.185096153846154e-06,
"loss": -0.0031,
"num_tokens": 1260136031.0,
"reward": 1.290212059020996,
"reward_std": 0.10709643959999085,
"rewards/accuracy_reward": 0.7328124880790711,
"rewards/brier_reward": 0.8538472175598144,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99375,
"rewards/mean_confidence_reward": 0.7211632013320923,
"sampling/batch_mean_priority_error": 0.03896267267771803,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.35,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.008996725827455521,
"sampling/priority_kl": 0.03000003769993782,
"sampling/priority_scale": 0.7160103619331494,
"sampling/prob_entropy": 10.278948211669922,
"sampling/prob_max": 4.822869086638093e-05,
"sampling/prob_min": 1.9965295723523014e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.2192000150680542,
"sampling/prompt_draws_total": 36576.0,
"sampling/seen_fraction": 0.7789933443069458,
"sampling/unseen_fraction": 0.2210066556930542,
"signal/accuracy_reward/centered_abs_mean": 0.10720485895872116,
"signal/accuracy_reward/group_std_mean": 0.1490975335240364,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05360242947936058,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05360242947936058,
"signal/advantage_abs_mean": 0.07514681965112686,
"signal/advantage_pre_scale_abs_mean": 0.07514681965112686,
"signal/advantage_pre_scale_std": 0.15790832936763763,
"signal/advantage_std": 0.15790832936763763,
"signal/brier_reward/centered_abs_mean": 0.05860713720321655,
"signal/brier_reward/group_std_mean": 0.08091188073158265,
"signal/brier_reward/group_zero_std_frac": 0.23611111789941788,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029303568601608276,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029303568601608276,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005859375,
"signal/format_reward/group_std_mean": 0.012872960604727268,
"signal/format_reward/group_zero_std_frac": 0.9388889074325562,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0029296875,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0029296875,
"signal/mean_confidence_reward/centered_abs_mean": 0.049307186901569364,
"signal/mean_confidence_reward/group_std_mean": 0.06485608667135238,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2500000029802322,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.930718716877891e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.930718716877891e-07,
"step": 510
},
{
"calibration/aurc": 0.127601593118219,
"calibration/batch_distribution_entropy": 0.7003445364316236,
"calibration/batch_entropy_100bins": 0.39342611836657276,
"calibration/batch_entropy_10bins": 0.7003445364316236,
"calibration/batch_entropy_50bins": 0.4631348609598662,
"calibration/batch_uniqueness": 0.6125533215887912,
"calibration/confidence_entropy": 0.5242468094631405,
"calibration/coverage@0%": 0.08802083333333334,
"calibration/coverage@1%": 0.11145833333333335,
"calibration/coverage@10%": 0.4462222584856397,
"calibration/coverage@15%": 0.6676634573542211,
"calibration/coverage@20%": 0.840625,
"calibration/coverage@25%": 0.8557291666666668,
"calibration/coverage@30%": 0.909375,
"calibration/coverage@5%": 0.2366296779808529,
"calibration/distribution_entropy_10": 0.7003445364316236,
"calibration/distribution_entropy_100": 0.39342611836657276,
"calibration/ece": 0.1235987815491731,
"calibration/mean_confidence": 0.6859905352480419,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.003993055555555558,
"completions/max_length": 3653.6,
"completions/max_terminated_length": 3653.6,
"completions/mean_length": 1023.5614624023438,
"completions/mean_terminated_length": 1027.6983764648437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 238.4,
"epoch": 1.2379807692307692,
"grad_norm": 0.0005270320107229054,
"learning_rate": 3.1550480769230772e-06,
"loss": -0.0029,
"num_tokens": 1275026531.0,
"reward": 1.2900764226913453,
"reward_std": 0.10513209253549576,
"rewards/accuracy_reward": 0.7271701335906983,
"rewards/brier_reward": 0.8571354269981384,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9958333373069763,
"rewards/mean_confidence_reward": 0.6929513812065125,
"sampling/batch_mean_priority_error": 0.030280168036501465,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.3638888888888889,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.009028730355203152,
"sampling/priority_kl": 0.030000071972608566,
"sampling/priority_scale": 0.7171281397575513,
"sampling/prob_entropy": 10.278954887390137,
"sampling/prob_max": 4.840152687393129e-05,
"sampling/prob_min": 2.0009274157928304e-05,
"sampling/prompt_draws_max": 6.0,
"sampling/prompt_draws_mean": 1.2312000036239623,
"sampling/prompt_draws_total": 36936.0,
"sampling/seen_fraction": 0.7831333398818969,
"sampling/unseen_fraction": 0.21686666011810302,
"signal/accuracy_reward/centered_abs_mean": 0.11109483391046523,
"signal/accuracy_reward/group_std_mean": 0.15123350024223328,
"signal/accuracy_reward/group_zero_std_frac": 0.5500000059604645,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05554741695523262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05554741695523262,
"signal/advantage_abs_mean": 0.07481730729341507,
"signal/advantage_pre_scale_abs_mean": 0.07481730729341507,
"signal/advantage_pre_scale_std": 0.15349577963352204,
"signal/advantage_std": 0.15349577963352204,
"signal/brier_reward/centered_abs_mean": 0.056370685994625094,
"signal/brier_reward/group_std_mean": 0.07771345749497413,
"signal/brier_reward/group_zero_std_frac": 0.13888889253139497,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028185342997312547,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028185342997312547,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006966145779006183,
"signal/format_reward/group_std_mean": 0.014438749477267265,
"signal/format_reward/group_zero_std_frac": 0.9333333492279052,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0034830728895030917,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0034830728895030917,
"signal/mean_confidence_reward/centered_abs_mean": 0.05178168341517449,
"signal/mean_confidence_reward/group_std_mean": 0.06841083616018295,
"signal/mean_confidence_reward/group_zero_std_frac": 0.1416666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.178168294150964e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.178168294150964e-07,
"step": 515
},
{
"calibration/aurc": 0.11846951228931728,
"calibration/batch_distribution_entropy": 0.6525936085812469,
"calibration/batch_entropy_100bins": 0.3623802260025296,
"calibration/batch_entropy_10bins": 0.6525936085812469,
"calibration/batch_entropy_50bins": 0.42658813878724466,
"calibration/batch_uniqueness": 0.4680648818769151,
"calibration/confidence_entropy": 0.47899836063768664,
"calibration/coverage@0%": 0.15833333333333333,
"calibration/coverage@1%": 0.3026041666666667,
"calibration/coverage@10%": 0.5026150741710296,
"calibration/coverage@15%": 0.6038094458987784,
"calibration/coverage@20%": 0.7969448877694695,
"calibration/coverage@25%": 0.8788396693687659,
"calibration/coverage@30%": 0.8788396693687659,
"calibration/coverage@5%": 0.4316835732984293,
"calibration/distribution_entropy_10": 0.6525936085812469,
"calibration/distribution_entropy_100": 0.3623802260025296,
"calibration/ece": 0.14434328991975715,
"calibration/mean_confidence": 0.6961930389047614,
"calibration/unique_confidence_per_question": 0.021354166666666664,
"calibration/unique_confidences": 8.2,
"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": 0.004774305555555558,
"completions/max_length": 3867.0,
"completions/max_terminated_length": 3867.0,
"completions/mean_length": 1104.88525390625,
"completions/mean_terminated_length": 1110.30390625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 308.2,
"epoch": 1.25,
"grad_norm": 0.0005092016654089093,
"learning_rate": 3.125e-06,
"loss": -0.0034,
"num_tokens": 1290852057.0,
"reward": 1.2876001596450806,
"reward_std": 0.11605858206748962,
"rewards/accuracy_reward": 0.7191840291023255,
"rewards/brier_reward": 0.8607769012451172,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9952256917953491,
"rewards/mean_confidence_reward": 0.6877343535423279,
"sampling/batch_mean_priority_error": 0.029131544333956733,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.33611111111111114,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.009056623838841916,
"sampling/priority_kl": 0.029999974742531778,
"sampling/priority_scale": 0.7184703051811084,
"sampling/prob_entropy": 10.278947067260741,
"sampling/prob_max": 4.858248357777484e-05,
"sampling/prob_min": 1.993556652450934e-05,
"sampling/prompt_draws_max": 6.2,
"sampling/prompt_draws_mean": 1.2431999921798706,
"sampling/prompt_draws_total": 37296.0,
"sampling/seen_fraction": 0.7873533248901368,
"sampling/unseen_fraction": 0.2126466751098633,
"signal/accuracy_reward/centered_abs_mean": 0.12505968064069747,
"signal/accuracy_reward/group_std_mean": 0.16689604818820952,
"signal/accuracy_reward/group_zero_std_frac": 0.5194444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06252984032034874,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06252984032034874,
"signal/advantage_abs_mean": 0.08188586384057998,
"signal/advantage_pre_scale_abs_mean": 0.08188586384057998,
"signal/advantage_pre_scale_std": 0.16324815154075623,
"signal/advantage_std": 0.16324815154075623,
"signal/brier_reward/centered_abs_mean": 0.059866245836019516,
"signal/brier_reward/group_std_mean": 0.08398651629686356,
"signal/brier_reward/group_zero_std_frac": 0.15277777686715127,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029933122918009758,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029933122918009758,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.008783637313172221,
"signal/format_reward/group_std_mean": 0.02044099196791649,
"signal/format_reward/group_zero_std_frac": 0.9000000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004391818656586111,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004391818656586111,
"signal/mean_confidence_reward/centered_abs_mean": 0.054735784977674486,
"signal/mean_confidence_reward/group_std_mean": 0.07394103705883026,
"signal/mean_confidence_reward/group_zero_std_frac": 0.16944444328546523,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.473578539749723e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.473578539749723e-07,
"step": 520
},
{
"calibration/aurc": 0.09678537048996547,
"calibration/batch_distribution_entropy": 0.6171174397568164,
"calibration/batch_entropy_100bins": 0.34609810798263696,
"calibration/batch_entropy_10bins": 0.6171174397568164,
"calibration/batch_entropy_50bins": 0.4074210928966894,
"calibration/batch_uniqueness": 0.4126739037344662,
"calibration/confidence_entropy": 0.4748172772125029,
"calibration/coverage@0%": 0.10976253298153034,
"calibration/coverage@1%": 0.22955419964819698,
"calibration/coverage@10%": 0.6564760495175418,
"calibration/coverage@15%": 0.7377408790067616,
"calibration/coverage@20%": 0.7880261492308591,
"calibration/coverage@25%": 0.8417837844839132,
"calibration/coverage@30%": 0.9145660052441771,
"calibration/coverage@5%": 0.5810012932521524,
"calibration/distribution_entropy_10": 0.6171174397568164,
"calibration/distribution_entropy_100": 0.34609810798263696,
"calibration/ece": 0.13466796804572648,
"calibration/mean_confidence": 0.7060822233364991,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.006510416666666652,
"completions/max_length": 3772.4,
"completions/max_terminated_length": 3772.4,
"completions/mean_length": 1085.634765625,
"completions/mean_terminated_length": 1092.8089599609375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 283.4,
"epoch": 1.2620192307692308,
"grad_norm": 0.0005398774519562721,
"learning_rate": 3.094951923076923e-06,
"loss": -0.0066,
"num_tokens": 1306456009.0,
"reward": 1.2903912544250489,
"reward_std": 0.11569144278764724,
"rewards/accuracy_reward": 0.717187511920929,
"rewards/brier_reward": 0.870177960395813,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9934027791023254,
"rewards/mean_confidence_reward": 0.7156336784362793,
"sampling/batch_mean_priority_error": 0.028870389832563247,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32222222222222224,
"sampling/error_ema_max": 0.2029161900281906,
"sampling/error_ema_mean": 0.009079698100686073,
"sampling/priority_kl": 0.03000027798116207,
"sampling/priority_scale": 0.7196202219231054,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 4.875725426245481e-05,
"sampling/prob_min": 1.9516599422786385e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.2551999807357788,
"sampling/prompt_draws_total": 37656.0,
"sampling/seen_fraction": 0.7913266539573669,
"sampling/unseen_fraction": 0.20867334604263305,
"signal/accuracy_reward/centered_abs_mean": 0.11032986044883727,
"signal/accuracy_reward/group_std_mean": 0.1527843713760376,
"signal/accuracy_reward/group_zero_std_frac": 0.5333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05516493022441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05516493022441864,
"signal/advantage_abs_mean": 0.07982016056776046,
"signal/advantage_pre_scale_abs_mean": 0.07982016056776046,
"signal/advantage_pre_scale_std": 0.16589364111423494,
"signal/advantage_std": 0.16589364111423494,
"signal/brier_reward/centered_abs_mean": 0.0613716222345829,
"signal/brier_reward/group_std_mean": 0.08781401515007019,
"signal/brier_reward/group_zero_std_frac": 0.2194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03068581111729145,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03068581111729145,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010568576585501432,
"signal/format_reward/group_std_mean": 0.020994319766759872,
"signal/format_reward/group_zero_std_frac": 0.9083333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005284288292750716,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005284288292750716,
"signal/mean_confidence_reward/centered_abs_mean": 0.053071295469999315,
"signal/mean_confidence_reward/group_std_mean": 0.07041076719760894,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2416666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.307129413267831e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.307129413267831e-07,
"step": 525
},
{
"calibration/aurc": 0.1391010387250999,
"calibration/batch_distribution_entropy": 0.5834054079074263,
"calibration/batch_entropy_100bins": 0.3178072726414959,
"calibration/batch_entropy_10bins": 0.5834054079074263,
"calibration/batch_entropy_50bins": 0.37411757927497885,
"calibration/batch_uniqueness": 0.3022310683542095,
"calibration/confidence_entropy": 0.4543434565360574,
"calibration/coverage@0%": 0.09109947643979058,
"calibration/coverage@1%": 0.09109947643979058,
"calibration/coverage@10%": 0.5162209256626522,
"calibration/coverage@15%": 0.5621755017452007,
"calibration/coverage@20%": 0.7390406849912741,
"calibration/coverage@25%": 0.8044257198952879,
"calibration/coverage@30%": 0.8922829406631761,
"calibration/coverage@5%": 0.2702445559307206,
"calibration/distribution_entropy_10": 0.5834054079074263,
"calibration/distribution_entropy_100": 0.3178072726414959,
"calibration/ece": 0.11636329422580065,
"calibration/mean_confidence": 0.7494788875712548,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.005902777777777768,
"completions/max_length": 3653.6,
"completions/max_terminated_length": 3653.6,
"completions/mean_length": 1049.12138671875,
"completions/mean_terminated_length": 1055.4906372070313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 298.0,
"epoch": 1.2740384615384617,
"grad_norm": 0.000520919100381434,
"learning_rate": 3.0649038461538464e-06,
"loss": -0.0057,
"num_tokens": 1321664575.0,
"reward": 1.2939966201782227,
"reward_std": 0.11089726239442825,
"rewards/accuracy_reward": 0.7355034589767456,
"rewards/brier_reward": 0.8584644198417664,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9940104126930237,
"rewards/mean_confidence_reward": 0.7482899308204651,
"sampling/batch_mean_priority_error": 0.04006456962425671,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.32222222222222224,
"sampling/error_ema_max": 0.2244912087917328,
"sampling/error_ema_mean": 0.009115009754896163,
"sampling/priority_kl": 0.029999951645731927,
"sampling/priority_scale": 0.7208636105293408,
"sampling/prob_entropy": 10.278956031799316,
"sampling/prob_max": 4.893360455753282e-05,
"sampling/prob_min": 1.955622537934687e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.267199993133545,
"sampling/prompt_draws_total": 38016.0,
"sampling/seen_fraction": 0.7952666640281677,
"sampling/unseen_fraction": 0.20473333597183227,
"signal/accuracy_reward/centered_abs_mean": 0.1049533411860466,
"signal/accuracy_reward/group_std_mean": 0.1418349489569664,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0524766705930233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0524766705930233,
"signal/advantage_abs_mean": 0.07906464189291,
"signal/advantage_pre_scale_abs_mean": 0.07906464189291,
"signal/advantage_pre_scale_std": 0.16779219806194307,
"signal/advantage_std": 0.16779219806194307,
"signal/brier_reward/centered_abs_mean": 0.06269395127892494,
"signal/brier_reward/group_std_mean": 0.0868382677435875,
"signal/brier_reward/group_zero_std_frac": 0.26944445073604584,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03134697563946247,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03134697563946247,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009792751725763082,
"signal/format_reward/group_std_mean": 0.01961270458996296,
"signal/format_reward/group_zero_std_frac": 0.9111111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004896375862881541,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004896375862881541,
"signal/mean_confidence_reward/centered_abs_mean": 0.05228136330842972,
"signal/mean_confidence_reward/group_std_mean": 0.06960389465093612,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2888888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.228136160440045e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.228136160440045e-07,
"step": 530
},
{
"calibration/aurc": 0.14580923682604757,
"calibration/batch_distribution_entropy": 0.5803569578541288,
"calibration/batch_entropy_100bins": 0.31552345276778493,
"calibration/batch_entropy_10bins": 0.5803569578541288,
"calibration/batch_entropy_50bins": 0.3714291034715423,
"calibration/batch_uniqueness": 0.31153982143640346,
"calibration/confidence_entropy": 0.4751440675952664,
"calibration/coverage@0%": 0.20455620412013178,
"calibration/coverage@1%": 0.2254986124970951,
"calibration/coverage@10%": 0.3195822454308094,
"calibration/coverage@15%": 0.6283453002610966,
"calibration/coverage@20%": 0.6971306570931244,
"calibration/coverage@25%": 0.8098125354000028,
"calibration/coverage@30%": 0.8970188017185405,
"calibration/coverage@5%": 0.31382308312714446,
"calibration/distribution_entropy_10": 0.5803569578541288,
"calibration/distribution_entropy_100": 0.31552345276778493,
"calibration/ece": 0.13262880646542285,
"calibration/mean_confidence": 0.73030042415089,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.007118055555555581,
"completions/max_length": 3618.6,
"completions/max_terminated_length": 3618.6,
"completions/mean_length": 1069.8447998046875,
"completions/mean_terminated_length": 1077.4728759765626,
"completions/min_length": 0.0,
"completions/min_terminated_length": 257.0,
"epoch": 1.2860576923076923,
"grad_norm": 0.0005245500360615551,
"learning_rate": 3.0348557692307694e-06,
"loss": -0.007,
"num_tokens": 1337072067.0,
"reward": 1.3126560688018798,
"reward_std": 0.10845615118741989,
"rewards/accuracy_reward": 0.7636284708976746,
"rewards/brier_reward": 0.8687873482704163,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9928819298744201,
"rewards/mean_confidence_reward": 0.7105642557144165,
"sampling/batch_mean_priority_error": 0.02997719025814488,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.33888888888888885,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009147688373923301,
"sampling/priority_kl": 0.030000006034970283,
"sampling/priority_scale": 0.7223210751777515,
"sampling/prob_entropy": 10.278949165344239,
"sampling/prob_max": 4.911954893032089e-05,
"sampling/prob_min": 1.9594168406911194e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.279200005531311,
"sampling/prompt_draws_total": 38376.0,
"sampling/seen_fraction": 0.7992733359336853,
"sampling/unseen_fraction": 0.2007266640663147,
"signal/accuracy_reward/centered_abs_mean": 0.11240776926279068,
"signal/accuracy_reward/group_std_mean": 0.15260213911533355,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05620388463139534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05620388463139534,
"signal/advantage_abs_mean": 0.07656532227993011,
"signal/advantage_pre_scale_abs_mean": 0.07656532227993011,
"signal/advantage_pre_scale_std": 0.16070781648159027,
"signal/advantage_std": 0.16070781648159027,
"signal/brier_reward/centered_abs_mean": 0.05885149613022804,
"signal/brier_reward/group_std_mean": 0.08213269263505936,
"signal/brier_reward/group_zero_std_frac": 0.2444444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02942574806511402,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02942574806511402,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.012185329757630824,
"signal/format_reward/group_std_mean": 0.023869803920388222,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006092664878815412,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006092664878815412,
"signal/mean_confidence_reward/centered_abs_mean": 0.05606608688831329,
"signal/mean_confidence_reward/group_std_mean": 0.07428872585296631,
"signal/mean_confidence_reward/group_zero_std_frac": 0.25277777910232546,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.606608397101809e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.606608397101809e-07,
"step": 535
},
{
"calibration/aurc": 0.10262722858718329,
"calibration/batch_distribution_entropy": 0.6637057506834922,
"calibration/batch_entropy_100bins": 0.3623222900478186,
"calibration/batch_entropy_10bins": 0.6637057506834922,
"calibration/batch_entropy_50bins": 0.4265199375187546,
"calibration/batch_uniqueness": 0.5035236225549949,
"calibration/confidence_entropy": 0.49673250121522416,
"calibration/coverage@0%": 0.11943809834638816,
"calibration/coverage@1%": 0.23612652306353352,
"calibration/coverage@10%": 0.6410227572030598,
"calibration/coverage@15%": 0.7449392062892216,
"calibration/coverage@20%": 0.7798717563327379,
"calibration/coverage@25%": 0.8375965370344922,
"calibration/coverage@30%": 0.8838257523704822,
"calibration/coverage@5%": 0.5576431880124593,
"calibration/distribution_entropy_10": 0.6637057506834922,
"calibration/distribution_entropy_100": 0.3623222900478186,
"calibration/ece": 0.16562619669277623,
"calibration/mean_confidence": 0.7031650834822043,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.006510416666666674,
"completions/max_length": 3626.6,
"completions/max_terminated_length": 3626.6,
"completions/mean_length": 1010.912158203125,
"completions/mean_terminated_length": 1017.4999145507812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 252.2,
"epoch": 1.2980769230769231,
"grad_norm": 0.0005195676349103451,
"learning_rate": 3.0048076923076923e-06,
"loss": -0.006,
"num_tokens": 1351813903.0,
"reward": 1.3025019407272338,
"reward_std": 0.10647796541452408,
"rewards/accuracy_reward": 0.7448784708976746,
"rewards/brier_reward": 0.8666215419769288,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9934895753860473,
"rewards/mean_confidence_reward": 0.7113368153572083,
"sampling/batch_mean_priority_error": 0.029256865188797367,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2722222222222222,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.00917521920055151,
"sampling/priority_kl": 0.03000005558133125,
"sampling/priority_scale": 0.7232611239189282,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 4.928471753373742e-05,
"sampling/prob_min": 1.963656031875871e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.2911999940872192,
"sampling/prompt_draws_total": 38736.0,
"sampling/seen_fraction": 0.802793312072754,
"sampling/unseen_fraction": 0.1972066879272461,
"signal/accuracy_reward/centered_abs_mean": 0.09930012971162797,
"signal/accuracy_reward/group_std_mean": 0.14416081011295317,
"signal/accuracy_reward/group_zero_std_frac": 0.5361111223697662,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04965006485581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04965006485581398,
"signal/advantage_abs_mean": 0.07084471881389617,
"signal/advantage_pre_scale_abs_mean": 0.07084471881389617,
"signal/advantage_pre_scale_std": 0.1556703507900238,
"signal/advantage_std": 0.1556703507900238,
"signal/brier_reward/centered_abs_mean": 0.05297866091132164,
"signal/brier_reward/group_std_mean": 0.07698716819286347,
"signal/brier_reward/group_zero_std_frac": 0.20000000670552254,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02648933045566082,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02648933045566082,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009543185913935303,
"signal/format_reward/group_std_mean": 0.019470962509512902,
"signal/format_reward/group_zero_std_frac": 0.9138889074325561,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004771592956967652,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004771592956967652,
"signal/mean_confidence_reward/centered_abs_mean": 0.05201877728104591,
"signal/mean_confidence_reward/group_std_mean": 0.07003375068306923,
"signal/mean_confidence_reward/group_zero_std_frac": 0.21388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.201877797844645e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.201877797844645e-07,
"step": 540
},
{
"calibration/aurc": 0.10556109814611606,
"calibration/batch_distribution_entropy": 0.600323191977656,
"calibration/batch_entropy_100bins": 0.3348939794410301,
"calibration/batch_entropy_10bins": 0.600323191977656,
"calibration/batch_entropy_50bins": 0.39423177405879073,
"calibration/batch_uniqueness": 0.4230486148288807,
"calibration/confidence_entropy": 0.48495376273374,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.20783289817232378,
"calibration/coverage@10%": 0.6653353459530027,
"calibration/coverage@15%": 0.8357579960835511,
"calibration/coverage@20%": 0.8942123585726719,
"calibration/coverage@25%": 0.9166666666666667,
"calibration/coverage@30%": 0.9479166666666667,
"calibration/coverage@5%": 0.38077948215839863,
"calibration/distribution_entropy_10": 0.600323191977656,
"calibration/distribution_entropy_100": 0.3348939794410301,
"calibration/ece": 0.11420474325500422,
"calibration/mean_confidence": 0.7316595953002613,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.003993055555555558,
"completions/max_length": 3622.2,
"completions/max_terminated_length": 3622.2,
"completions/mean_length": 1049.6030517578124,
"completions/mean_terminated_length": 1053.8172973632813,
"completions/min_length": 0.0,
"completions/min_terminated_length": 226.8,
"epoch": 1.3100961538461537,
"grad_norm": 0.00040970509871840477,
"learning_rate": 2.974759615384616e-06,
"loss": -0.0041,
"num_tokens": 1367008978.0,
"reward": 1.3004966974258423,
"reward_std": 0.09596788734197617,
"rewards/accuracy_reward": 0.7361979246139526,
"rewards/brier_reward": 0.8687743067741394,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9960069417953491,
"rewards/mean_confidence_reward": 0.7135416626930237,
"sampling/batch_mean_priority_error": 0.02854209502281709,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2972222222222222,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009199992194771767,
"sampling/priority_kl": 0.029999776929616927,
"sampling/priority_scale": 0.7241226016776636,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 4.944601314491592e-05,
"sampling/prob_min": 1.967924072232563e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3032000064849854,
"sampling/prompt_draws_total": 39096.0,
"sampling/seen_fraction": 0.806166672706604,
"sampling/unseen_fraction": 0.193833327293396,
"signal/accuracy_reward/centered_abs_mean": 0.09425455927848816,
"signal/accuracy_reward/group_std_mean": 0.1308629497885704,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222208976745,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04712727963924408,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04712727963924408,
"signal/advantage_abs_mean": 0.06631011962890625,
"signal/advantage_pre_scale_abs_mean": 0.06631011962890625,
"signal/advantage_pre_scale_std": 0.1461026519536972,
"signal/advantage_std": 0.1461026519536972,
"signal/brier_reward/centered_abs_mean": 0.0521217867732048,
"signal/brier_reward/group_std_mean": 0.07372160628437996,
"signal/brier_reward/group_zero_std_frac": 0.22500000298023223,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0260608933866024,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0260608933866024,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007367621548473835,
"signal/format_reward/group_std_mean": 0.016247615963220597,
"signal/format_reward/group_zero_std_frac": 0.9250000238418579,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0036838107742369177,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0036838107742369177,
"signal/mean_confidence_reward/centered_abs_mean": 0.0497748501598835,
"signal/mean_confidence_reward/group_std_mean": 0.06694498136639596,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23055556118488313,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.977485048129893e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.977485048129893e-07,
"step": 545
},
{
"calibration/aurc": 0.1477792257131352,
"calibration/batch_distribution_entropy": 0.5479882600039013,
"calibration/batch_entropy_100bins": 0.31210863541520106,
"calibration/batch_entropy_10bins": 0.5479882600039013,
"calibration/batch_entropy_50bins": 0.3674092357353625,
"calibration/batch_uniqueness": 0.2573770453378573,
"calibration/confidence_entropy": 0.47178416347281205,
"calibration/coverage@0%": 0.23201625218150088,
"calibration/coverage@1%": 0.3627454188481675,
"calibration/coverage@10%": 0.520556265545341,
"calibration/coverage@15%": 0.5404515535034561,
"calibration/coverage@20%": 0.6984934546385235,
"calibration/coverage@25%": 0.748784629179206,
"calibration/coverage@30%": 0.7916156719366232,
"calibration/coverage@5%": 0.44763959887867416,
"calibration/distribution_entropy_10": 0.5479882600039013,
"calibration/distribution_entropy_100": 0.31210863541520106,
"calibration/ece": 0.12668608975416018,
"calibration/mean_confidence": 0.7079187279273338,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.003993055555555558,
"completions/max_length": 3873.4,
"completions/max_terminated_length": 3873.4,
"completions/mean_length": 1055.5867553710937,
"completions/mean_terminated_length": 1059.9642700195313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 294.4,
"epoch": 1.3221153846153846,
"grad_norm": 0.0004762877942994237,
"learning_rate": 2.9447115384615386e-06,
"loss": -0.0029,
"num_tokens": 1382259161.0,
"reward": 1.3024785041809082,
"reward_std": 0.10325210392475129,
"rewards/accuracy_reward": 0.7399305582046509,
"rewards/brier_reward": 0.8690920114517212,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9959201455116272,
"rewards/mean_confidence_reward": 0.7217013597488403,
"sampling/batch_mean_priority_error": 0.03352335706747509,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.25,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009228985756635666,
"sampling/priority_kl": 0.03000006005167961,
"sampling/priority_scale": 0.725033014989458,
"sampling/prob_entropy": 10.278955268859864,
"sampling/prob_max": 4.961000668117777e-05,
"sampling/prob_min": 1.972165518964175e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3152000188827515,
"sampling/prompt_draws_total": 39456.0,
"sampling/seen_fraction": 0.809500002861023,
"sampling/unseen_fraction": 0.19049999713897706,
"signal/accuracy_reward/centered_abs_mean": 0.10303819477558136,
"signal/accuracy_reward/group_std_mean": 0.14597719609737397,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888895511628,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05151909738779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05151909738779068,
"signal/advantage_abs_mean": 0.07086191922426224,
"signal/advantage_pre_scale_abs_mean": 0.07086191922426224,
"signal/advantage_pre_scale_std": 0.15113241821527482,
"signal/advantage_std": 0.15113241821527482,
"signal/brier_reward/centered_abs_mean": 0.059060697257518766,
"signal/brier_reward/group_std_mean": 0.0823451228439808,
"signal/brier_reward/group_zero_std_frac": 0.2472222253680229,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029530348628759383,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029530348628759383,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007199435611255467,
"signal/format_reward/group_std_mean": 0.01524608675390482,
"signal/format_reward/group_zero_std_frac": 0.9305555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0035997178056277335,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0035997178056277335,
"signal/mean_confidence_reward/centered_abs_mean": 0.0533691443502903,
"signal/mean_confidence_reward/group_std_mean": 0.07072461247444153,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26111111640930174,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.336914341569354e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.336914341569354e-07,
"step": 550
},
{
"epoch": 1.3221153846153846,
"eval_calibration/aurc": 0.13271904073131,
"eval_calibration/batch_distribution_entropy": 0.5830219432889268,
"eval_calibration/batch_entropy_100bins": 0.32198688591764624,
"eval_calibration/batch_entropy_10bins": 0.5830219432889268,
"eval_calibration/batch_entropy_50bins": 0.37903775239808685,
"eval_calibration/batch_uniqueness": 0.2664253397778066,
"eval_calibration/confidence_entropy": 0.44559934075697216,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5839860748476936,
"eval_calibration/coverage@15%": 0.720626631853786,
"eval_calibration/coverage@20%": 0.7771975630983464,
"eval_calibration/coverage@25%": 0.9155787641427328,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5830219432889268,
"eval_calibration/distribution_entropy_100": 0.32198688591764624,
"eval_calibration/ece": 0.05126196692776324,
"eval_calibration/mean_confidence": 0.7456048738033073,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.002604166666666685,
"eval_completions/max_length": 2827.1666666666665,
"eval_completions/max_terminated_length": 2827.1666666666665,
"eval_completions/mean_length": 1019.587168375651,
"eval_completions/mean_terminated_length": 1022.301767985026,
"eval_completions/min_length": 180.66666666666666,
"eval_completions/min_terminated_length": 328.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 1382259161.0,
"eval_reward": 1.2812618215878804,
"eval_reward_std": 0.3112395455439885,
"eval_rewards/accuracy_reward": 0.7048611144224802,
"eval_rewards/brier_reward": 0.8602517445882162,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9973958432674408,
"eval_rewards/mean_confidence_reward": 0.7436631619930267,
"eval_runtime": 176.3071,
"eval_samples_per_second": 5.672,
"eval_signal/accuracy_reward/centered_abs_mean": 0.4022352397441864,
"eval_signal/accuracy_reward/group_std_mean": 0.45433419446150464,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.2011176198720932,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.2011176198720932,
"eval_signal/advantage_abs_mean": 0.2658406247695287,
"eval_signal/advantage_pre_scale_abs_mean": 0.2658406247695287,
"eval_signal/advantage_pre_scale_std": 0.30921146273612976,
"eval_signal/advantage_std": 0.30921146273612976,
"eval_signal/brier_reward/centered_abs_mean": 0.15962457656860352,
"eval_signal/brier_reward/group_std_mean": 0.2166272054115931,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07981228828430176,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07981228828430176,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0050455727614462376,
"eval_signal/format_reward/group_std_mean": 0.014731391333043575,
"eval_signal/format_reward/group_zero_std_frac": 0.9166666865348816,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025227863807231188,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0025227863807231188,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.18547633538643518,
"eval_signal/mean_confidence_reward/group_std_mean": 0.22163266936937967,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8547631839282985e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8547631839282985e-06,
"eval_steps_per_second": 0.034,
"step": 550
},
{
"epoch": 1.3221153846153846,
"step": 550,
"train_probe_calibration/aurc": 0.10593902445540661,
"train_probe_calibration/batch_distribution_entropy": 0.5521083451073077,
"train_probe_calibration/batch_entropy_100bins": 0.3021480965124056,
"train_probe_calibration/batch_entropy_10bins": 0.5521083451073077,
"train_probe_calibration/batch_entropy_50bins": 0.3556838504991608,
"train_probe_calibration/batch_uniqueness": 0.18839915502191362,
"train_probe_calibration/confidence_entropy": 0.4368206661541801,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6820557491289199,
"train_probe_calibration/coverage@15%": 0.794425087108014,
"train_probe_calibration/coverage@20%": 0.9242160278745645,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5521083451073077,
"train_probe_calibration/distribution_entropy_100": 0.3021480965124056,
"train_probe_calibration/ece": 0.02447735191637619,
"train_probe_calibration/mean_confidence": 0.7603658536585367,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.0034722222222222467,
"train_probe_completions/max_length": 2756.5,
"train_probe_completions/max_terminated_length": 2756.5,
"train_probe_completions/mean_length": 1045.1444905598958,
"train_probe_completions/mean_terminated_length": 1048.7373758951824,
"train_probe_completions/min_length": 75.0,
"train_probe_completions/min_terminated_length": 256.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1382259161.0,
"train_probe_reward": 1.3149598638216655,
"train_probe_reward_std": 0.2965337634086609,
"train_probe_rewards/accuracy_reward": 0.7630208333333334,
"train_probe_rewards/brier_reward": 0.8703559041023254,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9965277910232544,
"train_probe_rewards/mean_confidence_reward": 0.7577256659666697,
"train_probe_runtime": 189.9017,
"train_probe_samples_per_second": 5.266,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3529188384612401,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4245219975709915,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17645941923062006,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17645941923062006,
"train_probe_signal/advantage_abs_mean": 0.2381102442741394,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2381102442741394,
"train_probe_signal/advantage_pre_scale_std": 0.29500773052374524,
"train_probe_signal/advantage_std": 0.29500773052374524,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1511371706922849,
"train_probe_signal/brier_reward/group_std_mean": 0.21034086495637894,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07556858534614246,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07556858534614246,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/group_std_mean": 0.019641855110724766,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888889153798422,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.17651907602945963,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.21432253966728845,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7651907076772961e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7651907076772961e-06,
"train_probe_steps_per_second": 0.032
},
{
"calibration/aurc": 0.11236809342863448,
"calibration/batch_distribution_entropy": 0.4864726383348795,
"calibration/batch_entropy_100bins": 0.265186801412563,
"calibration/batch_entropy_10bins": 0.4864726383348795,
"calibration/batch_entropy_50bins": 0.31217361193637044,
"calibration/batch_uniqueness": 0.06888384482603635,
"calibration/confidence_entropy": 0.4253938572160966,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.19322916666666667,
"calibration/coverage@10%": 0.4472231288076588,
"calibration/coverage@15%": 0.6732267188859878,
"calibration/coverage@20%": 0.7201343559617058,
"calibration/coverage@25%": 0.927412358572672,
"calibration/coverage@30%": 0.953267188859878,
"calibration/coverage@5%": 0.4102208442123586,
"calibration/distribution_entropy_10": 0.4864726383348795,
"calibration/distribution_entropy_100": 0.265186801412563,
"calibration/ece": 0.11251528285465615,
"calibration/mean_confidence": 0.7762872932985205,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.003211805555555558,
"completions/max_length": 3490.8,
"completions/max_terminated_length": 3490.8,
"completions/mean_length": 1027.4213012695313,
"completions/mean_terminated_length": 1030.7630737304687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 277.0,
"epoch": 1.3341346153846154,
"grad_norm": 0.0004955082549713552,
"learning_rate": 2.9146634615384615e-06,
"loss": -0.0026,
"num_tokens": 1397194670.0,
"reward": 1.3129889965057373,
"reward_std": 0.10958507806062698,
"rewards/accuracy_reward": 0.756250011920929,
"rewards/brier_reward": 0.8729244709014893,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967881917953492,
"rewards/mean_confidence_reward": 0.7695572853088379,
"sampling/batch_mean_priority_error": 0.043720418268419124,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24999999999999994,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009267618693411351,
"sampling/priority_kl": 0.030000002682209016,
"sampling/priority_scale": 0.7255671441322192,
"sampling/prob_entropy": 10.278947830200195,
"sampling/prob_max": 4.975697302143089e-05,
"sampling/prob_min": 1.9709531261469238e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3272000074386596,
"sampling/prompt_draws_total": 39816.0,
"sampling/seen_fraction": 0.8124399900436401,
"sampling/unseen_fraction": 0.18756000995635985,
"signal/accuracy_reward/centered_abs_mean": 0.10514322966337204,
"signal/accuracy_reward/group_std_mean": 0.1475824624300003,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05257161483168602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05257161483168602,
"signal/advantage_abs_mean": 0.07649606168270111,
"signal/advantage_pre_scale_abs_mean": 0.07649606168270111,
"signal/advantage_pre_scale_std": 0.16270316243171692,
"signal/advantage_std": 0.16270316243171692,
"signal/brier_reward/centered_abs_mean": 0.06445361748337745,
"signal/brier_reward/group_std_mean": 0.09001707881689072,
"signal/brier_reward/group_zero_std_frac": 0.33055556416511533,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03222680874168873,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03222680874168873,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005517578113358468,
"signal/format_reward/group_std_mean": 0.010547200590372086,
"signal/format_reward/group_zero_std_frac": 0.9555555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002758789056679234,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002758789056679234,
"signal/mean_confidence_reward/centered_abs_mean": 0.04958822429180145,
"signal/mean_confidence_reward/group_std_mean": 0.06706940978765488,
"signal/mean_confidence_reward/group_zero_std_frac": 0.37222222089767454,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.958821762102161e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.958821762102161e-07,
"step": 555
},
{
"calibration/aurc": 0.19349510104636097,
"calibration/batch_distribution_entropy": 0.4805007810185836,
"calibration/batch_entropy_100bins": 0.26217946661122016,
"calibration/batch_entropy_10bins": 0.4805007810185836,
"calibration/batch_entropy_50bins": 0.3086334260664991,
"calibration/batch_uniqueness": 0.03672054455055672,
"calibration/confidence_entropy": 0.4213402318788935,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.28261939730200175,
"calibration/coverage@15%": 0.48259627937336813,
"calibration/coverage@20%": 0.6195370704208987,
"calibration/coverage@25%": 0.6242245704208986,
"calibration/coverage@30%": 0.8592183505802906,
"calibration/coverage@5%": 0.14375,
"calibration/distribution_entropy_10": 0.4805007810185836,
"calibration/distribution_entropy_100": 0.26217946661122016,
"calibration/ece": 0.14547908293804312,
"calibration/mean_confidence": 0.766256093108508,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.0022569444444444863,
"completions/max_length": 3695.0,
"completions/max_terminated_length": 3695.0,
"completions/mean_length": 1026.3934814453125,
"completions/mean_terminated_length": 1028.7162109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 262.4,
"epoch": 1.3461538461538463,
"grad_norm": 0.0004708245978690684,
"learning_rate": 2.8846153846153845e-06,
"loss": -0.0023,
"num_tokens": 1412146755.0,
"reward": 1.2862065076828002,
"reward_std": 0.11520719677209854,
"rewards/accuracy_reward": 0.7211805582046509,
"rewards/brier_reward": 0.853473961353302,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9977430582046509,
"rewards/mean_confidence_reward": 0.7711458206176758,
"sampling/batch_mean_priority_error": 0.053999831678116283,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.30833333333333335,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009317892044782639,
"sampling/priority_kl": 0.030000148713588713,
"sampling/priority_scale": 0.726933139632456,
"sampling/prob_entropy": 10.27895278930664,
"sampling/prob_max": 4.993655747966841e-05,
"sampling/prob_min": 1.96603676158702e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.339199995994568,
"sampling/prompt_draws_total": 40176.0,
"sampling/seen_fraction": 0.8159599900245667,
"sampling/unseen_fraction": 0.18404000997543335,
"signal/accuracy_reward/centered_abs_mean": 0.11303168386220933,
"signal/accuracy_reward/group_std_mean": 0.15187525302171706,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05651584193110466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05651584193110466,
"signal/advantage_abs_mean": 0.08364091664552689,
"signal/advantage_pre_scale_abs_mean": 0.08364091664552689,
"signal/advantage_pre_scale_std": 0.17128926217556,
"signal/advantage_std": 0.17128926217556,
"signal/brier_reward/centered_abs_mean": 0.06988281607627869,
"signal/brier_reward/group_std_mean": 0.09534413069486618,
"signal/brier_reward/group_zero_std_frac": 0.3583333432674408,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034941408038139346,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034941408038139346,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004210069356486201,
"signal/format_reward/group_std_mean": 0.009841996990144253,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0021050346782431006,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0021050346782431006,
"signal/mean_confidence_reward/centered_abs_mean": 0.04588542580604553,
"signal/mean_confidence_reward/group_std_mean": 0.06155051738023758,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4055555582046509,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.588542481087643e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.588542481087643e-07,
"step": 560
},
{
"calibration/aurc": 0.10364681460025409,
"calibration/batch_distribution_entropy": 0.5636566513218076,
"calibration/batch_entropy_100bins": 0.30608440987544927,
"calibration/batch_entropy_10bins": 0.5636566513218076,
"calibration/batch_entropy_50bins": 0.36031761490111924,
"calibration/batch_uniqueness": 0.2434725418054239,
"calibration/confidence_entropy": 0.45280722241919025,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.23514197127937336,
"calibration/coverage@10%": 0.3917482593559617,
"calibration/coverage@15%": 0.799688587902524,
"calibration/coverage@20%": 0.9070604873803308,
"calibration/coverage@25%": 0.9289817232375979,
"calibration/coverage@30%": 0.9775456919060052,
"calibration/coverage@5%": 0.3750380765883377,
"calibration/distribution_entropy_10": 0.5636566513218076,
"calibration/distribution_entropy_100": 0.30608440987544927,
"calibration/ece": 0.10714221605744126,
"calibration/mean_confidence": 0.745126060704961,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.0024305555555555356,
"completions/max_length": 3635.8,
"completions/max_terminated_length": 3635.8,
"completions/mean_length": 1036.5881958007812,
"completions/mean_terminated_length": 1039.1330078125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 266.4,
"epoch": 1.3581730769230769,
"grad_norm": 0.000573323224671185,
"learning_rate": 2.8545673076923082e-06,
"loss": -0.002,
"num_tokens": 1427154971.0,
"reward": 1.3023345947265625,
"reward_std": 0.10061877071857453,
"rewards/accuracy_reward": 0.7288194417953491,
"rewards/brier_reward": 0.878265643119812,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.997569453716278,
"rewards/mean_confidence_reward": 0.7324305295944213,
"sampling/batch_mean_priority_error": 0.032454187780661635,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.29166666666666663,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009358871728181839,
"sampling/priority_kl": 0.030000001564621924,
"sampling/priority_scale": 0.7282448708778247,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 5.011486864532344e-05,
"sampling/prob_min": 1.9697412426467054e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.351199984550476,
"sampling/prompt_draws_total": 40536.0,
"sampling/seen_fraction": 0.8193533301353455,
"sampling/unseen_fraction": 0.18064666986465455,
"signal/accuracy_reward/centered_abs_mean": 0.10447048544883727,
"signal/accuracy_reward/group_std_mean": 0.14341845363378525,
"signal/accuracy_reward/group_zero_std_frac": 0.569444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05223524272441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05223524272441864,
"signal/advantage_abs_mean": 0.0709734357893467,
"signal/advantage_pre_scale_abs_mean": 0.0709734357893467,
"signal/advantage_pre_scale_std": 0.15031934678554534,
"signal/advantage_std": 0.15031934678554534,
"signal/brier_reward/centered_abs_mean": 0.05996907502412796,
"signal/brier_reward/group_std_mean": 0.08309411108493805,
"signal/brier_reward/group_zero_std_frac": 0.31944444477558137,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02998453751206398,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02998453751206398,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0044053818914107975,
"signal/format_reward/group_std_mean": 0.00898937825113535,
"signal/format_reward/group_zero_std_frac": 0.9611111164093018,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0022026909457053987,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0022026909457053987,
"signal/mean_confidence_reward/centered_abs_mean": 0.05105469524860382,
"signal/mean_confidence_reward/group_std_mean": 0.06670601516962052,
"signal/mean_confidence_reward/group_zero_std_frac": 0.35000001192092894,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.105469369937054e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.105469369937054e-07,
"step": 565
},
{
"calibration/aurc": 0.11637660210021865,
"calibration/batch_distribution_entropy": 0.6026897912239121,
"calibration/batch_entropy_100bins": 0.33899787993418273,
"calibration/batch_entropy_10bins": 0.6026897912239121,
"calibration/batch_entropy_50bins": 0.39906281931877646,
"calibration/batch_uniqueness": 0.3477195535831427,
"calibration/confidence_entropy": 0.45724440003584677,
"calibration/coverage@0%": 0.12864583333333332,
"calibration/coverage@1%": 0.12864583333333332,
"calibration/coverage@10%": 0.5159020390070922,
"calibration/coverage@15%": 0.6976897116826843,
"calibration/coverage@20%": 0.8466561464177916,
"calibration/coverage@25%": 0.9232823581560284,
"calibration/coverage@30%": 0.9609375,
"calibration/coverage@5%": 0.2828125,
"calibration/distribution_entropy_10": 0.6026897912239121,
"calibration/distribution_entropy_100": 0.33899787993418273,
"calibration/ece": 0.110054493500361,
"calibration/mean_confidence": 0.7014624187545138,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003211805555555558,
"completions/max_length": 3079.2,
"completions/max_terminated_length": 3079.2,
"completions/mean_length": 1003.11484375,
"completions/mean_terminated_length": 1006.3870239257812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 232.4,
"epoch": 1.3701923076923077,
"grad_norm": 0.0005236753495410085,
"learning_rate": 2.8245192307692307e-06,
"loss": -0.0025,
"num_tokens": 1441808870.0,
"reward": 1.3041511535644532,
"reward_std": 0.09854458272457123,
"rewards/accuracy_reward": 0.7466145753860474,
"rewards/brier_reward": 0.8648854255676269,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967881917953492,
"rewards/mean_confidence_reward": 0.7043923616409302,
"sampling/batch_mean_priority_error": 0.039206801273562374,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.28888888888888886,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009391096048057079,
"sampling/priority_kl": 0.0300001073628664,
"sampling/priority_scale": 0.7299370229942724,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 5.030815664213151e-05,
"sampling/prob_min": 1.973052021639887e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3631999969482422,
"sampling/prompt_draws_total": 40896.0,
"sampling/seen_fraction": 0.8229599952697754,
"sampling/unseen_fraction": 0.1770400047302246,
"signal/accuracy_reward/centered_abs_mean": 0.10590820461511612,
"signal/accuracy_reward/group_std_mean": 0.14967113137245178,
"signal/accuracy_reward/group_zero_std_frac": 0.5305555760860443,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05295410230755806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05295410230755806,
"signal/advantage_abs_mean": 0.06743100956082344,
"signal/advantage_pre_scale_abs_mean": 0.06743100956082344,
"signal/advantage_pre_scale_std": 0.14558007717132568,
"signal/advantage_std": 0.14558007717132568,
"signal/brier_reward/centered_abs_mean": 0.059187816083431245,
"signal/brier_reward/group_std_mean": 0.08274016678333282,
"signal/brier_reward/group_zero_std_frac": 0.2583333343267441,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029593908041715623,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029593908041715623,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005626085097901523,
"signal/format_reward/group_std_mean": 0.011389839416369795,
"signal/format_reward/group_zero_std_frac": 0.950000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0028130425489507616,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0028130425489507616,
"signal/mean_confidence_reward/centered_abs_mean": 0.05424913391470909,
"signal/mean_confidence_reward/group_std_mean": 0.071955007314682,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27777778208255766,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.424913524620934e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.424913524620934e-07,
"step": 570
},
{
"calibration/aurc": 0.16748827581278503,
"calibration/batch_distribution_entropy": 0.633057807141147,
"calibration/batch_entropy_100bins": 0.3477108957075576,
"calibration/batch_entropy_10bins": 0.633057807141147,
"calibration/batch_entropy_50bins": 0.40931964051177916,
"calibration/batch_uniqueness": 0.42872305796571275,
"calibration/confidence_entropy": 0.47054753129181115,
"calibration/coverage@0%": 0.08511749347258486,
"calibration/coverage@1%": 0.08511749347258486,
"calibration/coverage@10%": 0.34360313315926894,
"calibration/coverage@15%": 0.571150199257936,
"calibration/coverage@20%": 0.6690105812834959,
"calibration/coverage@25%": 0.8905162311630107,
"calibration/coverage@30%": 0.9324671230717569,
"calibration/coverage@5%": 0.08511749347258486,
"calibration/distribution_entropy_10": 0.633057807141147,
"calibration/distribution_entropy_100": 0.3477108957075576,
"calibration/ece": 0.12355976349458997,
"calibration/mean_confidence": 0.7184623108726773,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.001996527777777768,
"completions/max_length": 3514.8,
"completions/max_terminated_length": 3514.8,
"completions/mean_length": 1018.8572143554687,
"completions/mean_terminated_length": 1020.90146484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 272.2,
"epoch": 1.3822115384615383,
"grad_norm": 0.0005015527713112533,
"learning_rate": 2.7944711538461537e-06,
"loss": -0.0021,
"num_tokens": 1456619897.0,
"reward": 1.3007010221481323,
"reward_std": 0.0904072642326355,
"rewards/accuracy_reward": 0.7381076335906982,
"rewards/brier_reward": 0.8652769207954407,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9980034708976746,
"rewards/mean_confidence_reward": 0.7014843702316285,
"sampling/batch_mean_priority_error": 0.03784606769545894,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24444444444444446,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009430542774498462,
"sampling/priority_kl": 0.029999880492687224,
"sampling/priority_scale": 0.7312095224158839,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 5.048633538535796e-05,
"sampling/prob_min": 1.9768124184338375e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3751999855041503,
"sampling/prompt_draws_total": 41256.0,
"sampling/seen_fraction": 0.8261666774749756,
"sampling/unseen_fraction": 0.17383332252502443,
"signal/accuracy_reward/centered_abs_mean": 0.10423719584941864,
"signal/accuracy_reward/group_std_mean": 0.13614998161792755,
"signal/accuracy_reward/group_zero_std_frac": 0.6166666865348815,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05211859792470932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05211859792470932,
"signal/advantage_abs_mean": 0.066029342263937,
"signal/advantage_pre_scale_abs_mean": 0.066029342263937,
"signal/advantage_pre_scale_std": 0.13949486911296843,
"signal/advantage_std": 0.13949486911296843,
"signal/brier_reward/centered_abs_mean": 0.0614675983786583,
"signal/brier_reward/group_std_mean": 0.08250565230846404,
"signal/brier_reward/group_zero_std_frac": 0.26944445073604584,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03073379918932915,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03073379918932915,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.003792317700572312,
"signal/format_reward/group_std_mean": 0.009447787888348103,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001896158850286156,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001896158850286156,
"signal/mean_confidence_reward/centered_abs_mean": 0.05299696773290634,
"signal/mean_confidence_reward/group_std_mean": 0.06921513825654983,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2833333432674408,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.299696567817592e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.299696567817592e-07,
"step": 575
},
{
"calibration/aurc": 0.09452517261125645,
"calibration/batch_distribution_entropy": 0.5651422000292813,
"calibration/batch_entropy_100bins": 0.3085382199945976,
"calibration/batch_entropy_10bins": 0.5651422000292813,
"calibration/batch_entropy_50bins": 0.36320620047106555,
"calibration/batch_uniqueness": 0.25278723390567037,
"calibration/confidence_entropy": 0.44761963367691926,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13177083333333334,
"calibration/coverage@10%": 0.7177170583115753,
"calibration/coverage@15%": 0.740958724978242,
"calibration/coverage@20%": 0.8317534268929505,
"calibration/coverage@25%": 0.8920107158398608,
"calibration/coverage@30%": 0.9472584856396867,
"calibration/coverage@5%": 0.6343128916449087,
"calibration/distribution_entropy_10": 0.5651422000292813,
"calibration/distribution_entropy_100": 0.3085382199945976,
"calibration/ece": 0.11731548411662304,
"calibration/mean_confidence": 0.7440300815926894,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.003125,
"completions/max_length": 3870.0,
"completions/max_terminated_length": 3870.0,
"completions/mean_length": 1048.67880859375,
"completions/mean_terminated_length": 1051.9510986328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 249.2,
"epoch": 1.3942307692307692,
"grad_norm": 0.0004841455665882677,
"learning_rate": 2.7644230769230775e-06,
"loss": -0.0025,
"num_tokens": 1471801285.0,
"reward": 1.2834061145782472,
"reward_std": 0.09844976812601089,
"rewards/accuracy_reward": 0.7129340171813965,
"rewards/brier_reward": 0.8569887042045593,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.996874988079071,
"rewards/mean_confidence_reward": 0.7255121350288392,
"sampling/batch_mean_priority_error": 0.04486913942545373,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2111111111111111,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009472385980188847,
"sampling/priority_kl": 0.030000155419111253,
"sampling/priority_scale": 0.7318387806648389,
"sampling/prob_entropy": 10.278948211669922,
"sampling/prob_max": 5.0635512889130044e-05,
"sampling/prob_min": 1.9810545927612112e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3871999979019165,
"sampling/prompt_draws_total": 41616.0,
"sampling/seen_fraction": 0.8288466572761536,
"sampling/unseen_fraction": 0.17115334272384644,
"signal/accuracy_reward/centered_abs_mean": 0.10468207597732544,
"signal/accuracy_reward/group_std_mean": 0.14231548309326172,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05234103798866272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05234103798866272,
"signal/advantage_abs_mean": 0.06959071755409241,
"signal/advantage_pre_scale_abs_mean": 0.06959071755409241,
"signal/advantage_pre_scale_std": 0.1507834017276764,
"signal/advantage_std": 0.1507834017276764,
"signal/brier_reward/centered_abs_mean": 0.06135156080126762,
"signal/brier_reward/group_std_mean": 0.08509978950023651,
"signal/brier_reward/group_zero_std_frac": 0.34166666865348816,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03067578040063381,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03067578040063381,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005696614552289247,
"signal/format_reward/group_std_mean": 0.011707442020997405,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0028483072761446237,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0028483072761446237,
"signal/mean_confidence_reward/centered_abs_mean": 0.05245281159877777,
"signal/mean_confidence_reward/group_std_mean": 0.0707058347761631,
"signal/mean_confidence_reward/group_zero_std_frac": 0.36111111640930177,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.245280874532909e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.245280874532909e-07,
"step": 580
},
{
"calibration/aurc": 0.13932227451395895,
"calibration/batch_distribution_entropy": 0.6448697784807116,
"calibration/batch_entropy_100bins": 0.36058733309465746,
"calibration/batch_entropy_10bins": 0.6448697784807116,
"calibration/batch_entropy_50bins": 0.4244775742648617,
"calibration/batch_uniqueness": 0.42685474793608336,
"calibration/confidence_entropy": 0.4686404208047999,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.46281258641283707,
"calibration/coverage@15%": 0.6135447063646275,
"calibration/coverage@20%": 0.7540883443113853,
"calibration/coverage@25%": 0.8325451428643802,
"calibration/coverage@30%": 0.9068541276502673,
"calibration/coverage@5%": 0.2557474818265518,
"calibration/distribution_entropy_10": 0.6448697784807116,
"calibration/distribution_entropy_100": 0.36058733309465746,
"calibration/ece": 0.09226729368323519,
"calibration/mean_confidence": 0.6993229161116602,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.005468749999999978,
"completions/max_length": 3880.0,
"completions/max_terminated_length": 3880.0,
"completions/mean_length": 1112.0519287109375,
"completions/mean_terminated_length": 1118.2107421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 318.4,
"epoch": 1.40625,
"grad_norm": 0.000712965673301369,
"learning_rate": 2.7343750000000004e-06,
"loss": -0.0057,
"num_tokens": 1487703483.0,
"reward": 1.2871733903884888,
"reward_std": 0.11923541873693466,
"rewards/accuracy_reward": 0.7236979126930236,
"rewards/brier_reward": 0.8561024308204651,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9945312380790711,
"rewards/mean_confidence_reward": 0.760451364517212,
"sampling/batch_mean_priority_error": 0.05240654750019175,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2583333333333334,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009516830556094646,
"sampling/priority_kl": 0.03000010997056961,
"sampling/priority_scale": 0.7327525675529614,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 5.079586335341446e-05,
"sampling/prob_min": 1.9756658366532066e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.3992000102996827,
"sampling/prompt_draws_total": 41976.0,
"sampling/seen_fraction": 0.8316733241081238,
"sampling/unseen_fraction": 0.16832667589187622,
"signal/accuracy_reward/centered_abs_mean": 0.11232096254825592,
"signal/accuracy_reward/group_std_mean": 0.1508895754814148,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05616048127412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05616048127412796,
"signal/advantage_abs_mean": 0.08523141592741013,
"signal/advantage_pre_scale_abs_mean": 0.08523141592741013,
"signal/advantage_pre_scale_std": 0.1773594230413437,
"signal/advantage_std": 0.1773594230413437,
"signal/brier_reward/centered_abs_mean": 0.07225770503282547,
"signal/brier_reward/group_std_mean": 0.09987465739250183,
"signal/brier_reward/group_zero_std_frac": 0.36111111640930177,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.036128852516412735,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.036128852516412735,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009673394076526166,
"signal/format_reward/group_std_mean": 0.01826365813612938,
"signal/format_reward/group_zero_std_frac": 0.9277778029441833,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004836697038263083,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004836697038263083,
"signal/mean_confidence_reward/centered_abs_mean": 0.053898122161626816,
"signal/mean_confidence_reward/group_std_mean": 0.07314420342445374,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4138888895511627,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.389812258727034e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.389812258727034e-07,
"step": 585
},
{
"calibration/aurc": 0.13942999390042368,
"calibration/batch_distribution_entropy": 0.5017374369243515,
"calibration/batch_entropy_100bins": 0.2704270845744181,
"calibration/batch_entropy_10bins": 0.5017374369243515,
"calibration/batch_entropy_50bins": 0.3183423884874351,
"calibration/batch_uniqueness": 0.120202369989494,
"calibration/confidence_entropy": 0.4277014706590796,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13333333333333333,
"calibration/coverage@10%": 0.325,
"calibration/coverage@15%": 0.6423574847693646,
"calibration/coverage@20%": 0.7034336923411663,
"calibration/coverage@25%": 0.8751958224543082,
"calibration/coverage@30%": 0.9394255874673629,
"calibration/coverage@5%": 0.2786458333333333,
"calibration/distribution_entropy_10": 0.5017374369243515,
"calibration/distribution_entropy_100": 0.2704270845744181,
"calibration/ece": 0.1212374891209748,
"calibration/mean_confidence": 0.7874815056570933,
"calibration/unique_confidence_per_question": 0.01927083333333333,
"calibration/unique_confidences": 7.4,
"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": 0.003038194444444464,
"completions/max_length": 3645.8,
"completions/max_terminated_length": 3645.8,
"completions/mean_length": 1017.2987915039063,
"completions/mean_terminated_length": 1020.4353637695312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 272.4,
"epoch": 1.4182692307692308,
"grad_norm": 0.00043438063585199416,
"learning_rate": 2.7043269230769233e-06,
"loss": -0.0041,
"num_tokens": 1502516589.0,
"reward": 1.3097166061401366,
"reward_std": 0.0976711392402649,
"rewards/accuracy_reward": 0.7488715171813964,
"rewards/brier_reward": 0.8735842347145081,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9969618082046509,
"rewards/mean_confidence_reward": 0.7808419942855835,
"sampling/batch_mean_priority_error": 0.044314487949195534,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2555555555555556,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009564095363020896,
"sampling/priority_kl": 0.02999996468424797,
"sampling/priority_scale": 0.7341858922736719,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 5.097756948089227e-05,
"sampling/prob_min": 1.9790743317571467e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4111999988555908,
"sampling/prompt_draws_total": 42336.0,
"sampling/seen_fraction": 0.8347866654396057,
"sampling/unseen_fraction": 0.16521333456039428,
"signal/accuracy_reward/centered_abs_mean": 0.090966796875,
"signal/accuracy_reward/group_std_mean": 0.12698046565055848,
"signal/accuracy_reward/group_zero_std_frac": 0.6083333373069764,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0454833984375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0454833984375,
"signal/advantage_abs_mean": 0.06731701195240021,
"signal/advantage_pre_scale_abs_mean": 0.06731701195240021,
"signal/advantage_pre_scale_std": 0.1525650978088379,
"signal/advantage_std": 0.1525650978088379,
"signal/brier_reward/centered_abs_mean": 0.055035379528999326,
"signal/brier_reward/group_std_mean": 0.07981750667095185,
"signal/brier_reward/group_zero_std_frac": 0.37777777910232546,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027517689764499663,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027517689764499663,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0055501303169876335,
"signal/format_reward/group_std_mean": 0.01122641284018755,
"signal/format_reward/group_zero_std_frac": 0.9527777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0027750651584938168,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0027750651584938168,
"signal/mean_confidence_reward/centered_abs_mean": 0.0433946542441845,
"signal/mean_confidence_reward/group_std_mean": 0.0584526002407074,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4250000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.339465363045747e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.339465363045747e-07,
"step": 590
},
{
"calibration/aurc": 0.11260714434102297,
"calibration/batch_distribution_entropy": 0.5449134079570979,
"calibration/batch_entropy_100bins": 0.29232000449749385,
"calibration/batch_entropy_10bins": 0.5449134079570979,
"calibration/batch_entropy_50bins": 0.34411437959640323,
"calibration/batch_uniqueness": 0.21047108118652816,
"calibration/confidence_entropy": 0.43431141306142684,
"calibration/coverage@0%": 0.1484375,
"calibration/coverage@1%": 0.1484375,
"calibration/coverage@10%": 0.6736571618037136,
"calibration/coverage@15%": 0.7293200154730327,
"calibration/coverage@20%": 0.7685952696728559,
"calibration/coverage@25%": 0.8,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.428125,
"calibration/distribution_entropy_10": 0.5449134079570979,
"calibration/distribution_entropy_100": 0.29232000449749385,
"calibration/ece": 0.137520032051282,
"calibration/mean_confidence": 0.7773144617595051,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.006423611111111139,
"completions/max_length": 3772.2,
"completions/max_terminated_length": 3772.2,
"completions/mean_length": 1045.2372680664062,
"completions/mean_terminated_length": 1052.1338256835938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 278.6,
"epoch": 1.4302884615384617,
"grad_norm": 0.0004948692512698472,
"learning_rate": 2.6742788461538467e-06,
"loss": -0.0076,
"num_tokens": 1517658618.0,
"reward": 1.3219900369644164,
"reward_std": 0.09807792603969574,
"rewards/accuracy_reward": 0.7714409708976746,
"rewards/brier_reward": 0.8789479613304139,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9935763716697693,
"rewards/mean_confidence_reward": 0.7427256941795349,
"sampling/batch_mean_priority_error": 0.03236808832125963,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2555555555555556,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009601323492825031,
"sampling/priority_kl": 0.0300000436604023,
"sampling/priority_scale": 0.7354452908271923,
"sampling/prob_entropy": 10.278949165344239,
"sampling/prob_max": 5.1152726518921555e-05,
"sampling/prob_min": 1.98265362996608e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.423200011253357,
"sampling/prompt_draws_total": 42696.0,
"sampling/seen_fraction": 0.8377133131027221,
"sampling/unseen_fraction": 0.16228668689727782,
"signal/accuracy_reward/centered_abs_mean": 0.09402669221162796,
"signal/accuracy_reward/group_std_mean": 0.12942231297492982,
"signal/accuracy_reward/group_zero_std_frac": 0.6027777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04701334610581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04701334610581398,
"signal/advantage_abs_mean": 0.06896353140473366,
"signal/advantage_pre_scale_abs_mean": 0.06896353140473366,
"signal/advantage_pre_scale_std": 0.15308170765638351,
"signal/advantage_std": 0.15308170765638351,
"signal/brier_reward/centered_abs_mean": 0.055911561101675035,
"signal/brier_reward/group_std_mean": 0.07856666445732116,
"signal/brier_reward/group_zero_std_frac": 0.3083333343267441,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027955780550837517,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027955780550837517,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01025390620343387,
"signal/format_reward/group_std_mean": 0.017292334325611593,
"signal/format_reward/group_zero_std_frac": 0.9333333253860474,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005126953101716935,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005126953101716935,
"signal/mean_confidence_reward/centered_abs_mean": 0.046098099648952486,
"signal/mean_confidence_reward/group_std_mean": 0.06332936584949493,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33333333134651183,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.6098096504465504e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.6098096504465504e-07,
"step": 595
},
{
"calibration/aurc": 0.10463935890886092,
"calibration/batch_distribution_entropy": 0.6619123417933046,
"calibration/batch_entropy_100bins": 0.36687692710750464,
"calibration/batch_entropy_10bins": 0.6619123417933046,
"calibration/batch_entropy_50bins": 0.43188158257200693,
"calibration/batch_uniqueness": 0.4892413688617815,
"calibration/confidence_entropy": 0.4763627433511208,
"calibration/coverage@0%": 0.11780104712041886,
"calibration/coverage@1%": 0.11780104712041886,
"calibration/coverage@10%": 0.5687064055814526,
"calibration/coverage@15%": 0.7911625669918922,
"calibration/coverage@20%": 0.8315940634877009,
"calibration/coverage@25%": 0.9298859420090697,
"calibration/coverage@30%": 0.954569190600522,
"calibration/coverage@5%": 0.3743518721036731,
"calibration/distribution_entropy_10": 0.6619123417933046,
"calibration/distribution_entropy_100": 0.36687692710750464,
"calibration/ece": 0.15183055306638016,
"calibration/mean_confidence": 0.7065397890530327,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.003993055555555558,
"completions/max_length": 3725.4,
"completions/max_terminated_length": 3725.4,
"completions/mean_length": 1078.1233642578125,
"completions/mean_terminated_length": 1082.4756103515624,
"completions/min_length": 0.0,
"completions/min_terminated_length": 256.6,
"epoch": 1.4423076923076923,
"grad_norm": 0.0004698413540609181,
"learning_rate": 2.6442307692307696e-06,
"loss": -0.0045,
"num_tokens": 1533203047.0,
"reward": 1.3211289644241333,
"reward_std": 0.0979426309466362,
"rewards/accuracy_reward": 0.7806423664093017,
"rewards/brier_reward": 0.8656814217567443,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9959201335906982,
"rewards/mean_confidence_reward": 0.7000954866409301,
"sampling/batch_mean_priority_error": 0.032977938594612934,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2833333333333333,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009629619680345059,
"sampling/priority_kl": 0.030000071227550506,
"sampling/priority_scale": 0.7374177158577367,
"sampling/prob_entropy": 10.278955268859864,
"sampling/prob_max": 5.13555969519075e-05,
"sampling/prob_min": 1.9854630227200688e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.435200023651123,
"sampling/prompt_draws_total": 43056.0,
"sampling/seen_fraction": 0.8410533308982849,
"sampling/unseen_fraction": 0.1589466691017151,
"signal/accuracy_reward/centered_abs_mean": 0.10369466096162797,
"signal/accuracy_reward/group_std_mean": 0.14326806515455245,
"signal/accuracy_reward/group_zero_std_frac": 0.5694444656372071,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05184733048081398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05184733048081398,
"signal/advantage_abs_mean": 0.06741442307829856,
"signal/advantage_pre_scale_abs_mean": 0.06741442307829856,
"signal/advantage_pre_scale_std": 0.14682429730892183,
"signal/advantage_std": 0.14682429730892183,
"signal/brier_reward/centered_abs_mean": 0.05674092769622803,
"signal/brier_reward/group_std_mean": 0.07852299660444259,
"signal/brier_reward/group_zero_std_frac": 0.22500000298023223,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028370463848114015,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028370463848114015,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0074707032181322575,
"signal/format_reward/group_std_mean": 0.016295704618096353,
"signal/format_reward/group_zero_std_frac": 0.9250000238418579,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0037353516090661287,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0037353516090661287,
"signal/mean_confidence_reward/centered_abs_mean": 0.051773006469011305,
"signal/mean_confidence_reward/group_std_mean": 0.06856385469436646,
"signal/mean_confidence_reward/group_zero_std_frac": 0.24166667461395264,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.177300749892311e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.177300749892311e-07,
"step": 600
},
{
"epoch": 1.4423076923076923,
"eval_calibration/aurc": 0.11749798048665533,
"eval_calibration/batch_distribution_entropy": 0.7077334731137986,
"eval_calibration/batch_entropy_100bins": 0.3875815454761296,
"eval_calibration/batch_entropy_10bins": 0.7077334731137986,
"eval_calibration/batch_entropy_50bins": 0.4562547243176337,
"eval_calibration/batch_uniqueness": 0.5498560286798497,
"eval_calibration/confidence_entropy": 0.4879134814798048,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5772925764192139,
"eval_calibration/coverage@15%": 0.7458515283842795,
"eval_calibration/coverage@20%": 0.8148471615720524,
"eval_calibration/coverage@25%": 0.9065502183406113,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.7077334731137986,
"eval_calibration/distribution_entropy_100": 0.3875815454761296,
"eval_calibration/ece": 0.03554585152838413,
"eval_calibration/mean_confidence": 0.7082096069868996,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.006076388888888895,
"eval_completions/max_length": 3122.0,
"eval_completions/max_terminated_length": 3122.0,
"eval_completions/mean_length": 1089.3387451171875,
"eval_completions/mean_terminated_length": 1096.1107584635417,
"eval_completions/min_length": 155.83333333333334,
"eval_completions/min_terminated_length": 358.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1533203047.0,
"eval_reward": 1.2864523927370708,
"eval_reward_std": 0.2984392096598943,
"eval_rewards/accuracy_reward": 0.7230902910232544,
"eval_rewards/brier_reward": 0.8558767636617025,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9939236144224802,
"eval_rewards/mean_confidence_reward": 0.7039062678813934,
"eval_runtime": 191.414,
"eval_samples_per_second": 5.224,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3887261301279068,
"eval_signal/accuracy_reward/group_std_mean": 0.44594499965508777,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1943630650639534,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.1943630650639534,
"eval_signal/advantage_abs_mean": 0.24910660833120346,
"eval_signal/advantage_pre_scale_abs_mean": 0.24910660833120346,
"eval_signal/advantage_pre_scale_std": 0.29744037985801697,
"eval_signal/advantage_std": 0.29744037985801697,
"eval_signal/brier_reward/centered_abs_mean": 0.1458252047499021,
"eval_signal/brier_reward/group_std_mean": 0.19646542519330978,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07291260237495105,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07291260237495105,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.011664496424297491,
"eval_signal/format_reward/group_std_mean": 0.031383837262789406,
"eval_signal/format_reward/group_zero_std_frac": 0.8333333532015482,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0058322482121487456,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0058322482121487456,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.18724499146143594,
"eval_signal/mean_confidence_reward/group_std_mean": 0.22393441945314407,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8724499189678074e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8724499189678074e-06,
"eval_steps_per_second": 0.031,
"step": 600
},
{
"epoch": 1.4423076923076923,
"step": 600,
"train_probe_calibration/aurc": 0.0811248529110322,
"train_probe_calibration/batch_distribution_entropy": 0.6947460190964702,
"train_probe_calibration/batch_entropy_100bins": 0.37999471276504704,
"train_probe_calibration/batch_entropy_10bins": 0.6947460190964702,
"train_probe_calibration/batch_entropy_50bins": 0.44732362760407207,
"train_probe_calibration/batch_uniqueness": 0.5262528608423931,
"train_probe_calibration/confidence_entropy": 0.4823398318438881,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7523975588491717,
"train_probe_calibration/coverage@15%": 0.8116826503923278,
"train_probe_calibration/coverage@20%": 0.8962510897994769,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.4219703574542284,
"train_probe_calibration/distribution_entropy_10": 0.6947460190964702,
"train_probe_calibration/distribution_entropy_100": 0.37999471276504704,
"train_probe_calibration/ece": 0.07541412380122044,
"train_probe_calibration/mean_confidence": 0.7123801220575415,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.005902777777777775,
"train_probe_completions/max_length": 3296.6666666666665,
"train_probe_completions/max_terminated_length": 3296.6666666666665,
"train_probe_completions/mean_length": 1121.4295450846355,
"train_probe_completions/mean_terminated_length": 1128.2926839192708,
"train_probe_completions/min_length": 188.33333333333334,
"train_probe_completions/min_terminated_length": 276.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1533203047.0,
"train_probe_reward": 1.3208882013956706,
"train_probe_reward_std": 0.2697082112232844,
"train_probe_rewards/accuracy_reward": 0.7690972288449606,
"train_probe_rewards/brier_reward": 0.8770051995913187,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.995659718910853,
"train_probe_rewards/mean_confidence_reward": 0.7092882096767426,
"train_probe_runtime": 190.0224,
"train_probe_samples_per_second": 5.263,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3442925363779068,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4188603013753891,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1721462681889534,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1721462681889534,
"train_probe_signal/advantage_abs_mean": 0.2152716045578321,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2152716045578321,
"train_probe_signal/advantage_pre_scale_std": 0.2695246562361717,
"train_probe_signal/advantage_std": 0.2695246562361717,
"train_probe_signal/brier_reward/centered_abs_mean": 0.12592938169836998,
"train_probe_signal/brier_reward/group_std_mean": 0.1717534065246582,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06296469084918499,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06296469084918499,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.00830078125,
"train_probe_signal/format_reward/group_std_mean": 0.021562909707427025,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888888955116272,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.004150390625,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.004150390625,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.18776040027538934,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.22326893856128058,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8776040064949484e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8776040064949484e-06,
"train_probe_steps_per_second": 0.032
},
{
"calibration/aurc": 0.1625902676601865,
"calibration/batch_distribution_entropy": 0.688178501134466,
"calibration/batch_entropy_100bins": 0.3773364125369895,
"calibration/batch_entropy_10bins": 0.688178501134466,
"calibration/batch_entropy_50bins": 0.44419431958654004,
"calibration/batch_uniqueness": 0.5459930237340862,
"calibration/confidence_entropy": 0.4974271541450772,
"calibration/coverage@0%": 0.08655913978494624,
"calibration/coverage@1%": 0.1685434739886016,
"calibration/coverage@10%": 0.328105061226429,
"calibration/coverage@15%": 0.4577343120676735,
"calibration/coverage@20%": 0.7047311792905762,
"calibration/coverage@25%": 0.8369028378426489,
"calibration/coverage@30%": 0.952465335711409,
"calibration/coverage@5%": 0.23120665936719167,
"calibration/distribution_entropy_10": 0.688178501134466,
"calibration/distribution_entropy_100": 0.3773364125369895,
"calibration/ece": 0.1463983525444193,
"calibration/mean_confidence": 0.6849986304418707,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.008854166666666653,
"completions/max_length": 3869.8,
"completions/max_terminated_length": 3869.8,
"completions/mean_length": 1103.256103515625,
"completions/mean_terminated_length": 1113.0990234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 276.2,
"epoch": 1.4543269230769231,
"grad_norm": 0.00044460591743700206,
"learning_rate": 2.6141826923076926e-06,
"loss": -0.0097,
"num_tokens": 1548997005.0,
"reward": 1.2924524068832397,
"reward_std": 0.10913865864276887,
"rewards/accuracy_reward": 0.745312488079071,
"rewards/brier_reward": 0.8484323143959045,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9911458253860473,
"rewards/mean_confidence_reward": 0.7084201455116272,
"sampling/batch_mean_priority_error": 0.03887890530603154,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.26666666666666666,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009661880135536195,
"sampling/priority_kl": 0.030000338703393935,
"sampling/priority_scale": 0.7392990172142163,
"sampling/prob_entropy": 10.27895565032959,
"sampling/prob_max": 5.1556043763412164e-05,
"sampling/prob_min": 1.9883858840330504e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4471999883651734,
"sampling/prompt_draws_total": 43416.0,
"sampling/seen_fraction": 0.8442333340644836,
"sampling/unseen_fraction": 0.15576666593551636,
"signal/accuracy_reward/centered_abs_mean": 0.10305989682674407,
"signal/accuracy_reward/group_std_mean": 0.14260239899158478,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05152994841337204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05152994841337204,
"signal/advantage_abs_mean": 0.075344218313694,
"signal/advantage_pre_scale_abs_mean": 0.075344218313694,
"signal/advantage_pre_scale_std": 0.16476980745792388,
"signal/advantage_std": 0.16476980745792388,
"signal/brier_reward/centered_abs_mean": 0.06039917916059494,
"signal/brier_reward/group_std_mean": 0.08526878505945205,
"signal/brier_reward/group_zero_std_frac": 0.20833333432674409,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03019958958029747,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03019958958029747,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.014333767257630825,
"signal/format_reward/group_std_mean": 0.02507205940783024,
"signal/format_reward/group_zero_std_frac": 0.9027777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.007166883628815413,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.007166883628815413,
"signal/mean_confidence_reward/centered_abs_mean": 0.05270670726895332,
"signal/mean_confidence_reward/group_std_mean": 0.07211160734295845,
"signal/mean_confidence_reward/group_zero_std_frac": 0.21388888955116273,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.270670556001277e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.270670556001277e-07,
"step": 605
},
{
"calibration/aurc": 0.14458147639177027,
"calibration/batch_distribution_entropy": 0.6089697186869654,
"calibration/batch_entropy_100bins": 0.3300116115365298,
"calibration/batch_entropy_10bins": 0.6089697186869654,
"calibration/batch_entropy_50bins": 0.38848432955766393,
"calibration/batch_uniqueness": 0.4002591418256035,
"calibration/confidence_entropy": 0.47229101300975074,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.19109662073490813,
"calibration/coverage@10%": 0.4357939632545932,
"calibration/coverage@15%": 0.5927083333333334,
"calibration/coverage@20%": 0.703125,
"calibration/coverage@25%": 0.8250909391534391,
"calibration/coverage@30%": 0.8649801587301587,
"calibration/coverage@5%": 0.32838336614173225,
"calibration/distribution_entropy_10": 0.6089697186869654,
"calibration/distribution_entropy_100": 0.3300116115365298,
"calibration/ece": 0.15325526366495854,
"calibration/mean_confidence": 0.7555445738553516,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.003211805555555558,
"completions/max_length": 3788.4,
"completions/max_terminated_length": 3788.4,
"completions/mean_length": 1117.24462890625,
"completions/mean_terminated_length": 1120.8237548828124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 310.0,
"epoch": 1.4663461538461537,
"grad_norm": 0.00047830064431764185,
"learning_rate": 2.584134615384616e-06,
"loss": -0.0029,
"num_tokens": 1564966735.0,
"reward": 1.2964518308639525,
"reward_std": 0.10213772356510162,
"rewards/accuracy_reward": 0.7368055582046509,
"rewards/brier_reward": 0.8592951416969299,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967882037162781,
"rewards/mean_confidence_reward": 0.735937488079071,
"sampling/batch_mean_priority_error": 0.03690880815930133,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2722222222222222,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009698636457324028,
"sampling/priority_kl": 0.02999996729195118,
"sampling/priority_scale": 0.7415312468772755,
"sampling/prob_entropy": 10.278956413269043,
"sampling/prob_max": 5.177144994377159e-05,
"sampling/prob_min": 1.990957753150724e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4591999769210815,
"sampling/prompt_draws_total": 43776.0,
"sampling/seen_fraction": 0.8475733399391174,
"sampling/unseen_fraction": 0.15242666006088257,
"signal/accuracy_reward/centered_abs_mean": 0.09847005158662796,
"signal/accuracy_reward/group_std_mean": 0.13745512515306474,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04923502579331398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04923502579331398,
"signal/advantage_abs_mean": 0.07168345451354981,
"signal/advantage_pre_scale_abs_mean": 0.07168345451354981,
"signal/advantage_pre_scale_std": 0.1520264506340027,
"signal/advantage_std": 0.1520264506340027,
"signal/brier_reward/centered_abs_mean": 0.05799070745706558,
"signal/brier_reward/group_std_mean": 0.08116589337587357,
"signal/brier_reward/group_zero_std_frac": 0.20833333432674409,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02899535372853279,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02899535372853279,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005799696128815412,
"signal/format_reward/group_std_mean": 0.012281404808163643,
"signal/format_reward/group_zero_std_frac": 0.9444444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002899848064407706,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002899848064407706,
"signal/mean_confidence_reward/centered_abs_mean": 0.048818358778953554,
"signal/mean_confidence_reward/group_std_mean": 0.0655674695968628,
"signal/mean_confidence_reward/group_zero_std_frac": 0.22500000298023223,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.881835764081189e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.881835764081189e-07,
"step": 610
},
{
"calibration/aurc": 0.10717646892595933,
"calibration/batch_distribution_entropy": 0.5944995185996648,
"calibration/batch_entropy_100bins": 0.3223670980698781,
"calibration/batch_entropy_10bins": 0.5944995185996648,
"calibration/batch_entropy_50bins": 0.3794853319919132,
"calibration/batch_uniqueness": 0.3876516859568565,
"calibration/confidence_entropy": 0.47987866269041757,
"calibration/coverage@0%": 0.27700645890043146,
"calibration/coverage@1%": 0.4954637668277688,
"calibration/coverage@10%": 0.6712006962576152,
"calibration/coverage@15%": 0.7573625761531767,
"calibration/coverage@20%": 0.7733625761531766,
"calibration/coverage@25%": 0.8140936466492603,
"calibration/coverage@30%": 0.8203599651871192,
"calibration/coverage@5%": 0.5960749921300668,
"calibration/distribution_entropy_10": 0.5944995185996648,
"calibration/distribution_entropy_100": 0.3223670980698781,
"calibration/ece": 0.13857618576745728,
"calibration/mean_confidence": 0.7310599092643002,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.00659722222222221,
"completions/max_length": 3889.6,
"completions/max_terminated_length": 3889.6,
"completions/mean_length": 1076.1330932617188,
"completions/mean_terminated_length": 1083.3836547851563,
"completions/min_length": 0.0,
"completions/min_terminated_length": 246.2,
"epoch": 1.4783653846153846,
"grad_norm": 0.0006141048506833613,
"learning_rate": 2.554086538461539e-06,
"loss": -0.0066,
"num_tokens": 1580488364.0,
"reward": 1.314866876602173,
"reward_std": 0.10772739797830581,
"rewards/accuracy_reward": 0.7642361044883728,
"rewards/brier_reward": 0.8720798373222352,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9934027910232544,
"rewards/mean_confidence_reward": 0.7517708301544189,
"sampling/batch_mean_priority_error": 0.03643278677661457,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2361111111111111,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009736244194209575,
"sampling/priority_kl": 0.03000003136694431,
"sampling/priority_scale": 0.7432866037590429,
"sampling/prob_entropy": 10.278955078125,
"sampling/prob_max": 5.196688580326736e-05,
"sampling/prob_min": 1.9939581397920848e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4711999893188477,
"sampling/prompt_draws_total": 44136.0,
"sampling/seen_fraction": 0.8505399942398071,
"sampling/unseen_fraction": 0.14946000576019286,
"signal/accuracy_reward/centered_abs_mean": 0.10202908217906952,
"signal/accuracy_reward/group_std_mean": 0.14247817397117615,
"signal/accuracy_reward/group_zero_std_frac": 0.5638888835906982,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05101454108953476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05101454108953476,
"signal/advantage_abs_mean": 0.07561420053243637,
"signal/advantage_pre_scale_abs_mean": 0.07561420053243637,
"signal/advantage_pre_scale_std": 0.1636774629354477,
"signal/advantage_std": 0.1636774629354477,
"signal/brier_reward/centered_abs_mean": 0.056111641228199005,
"signal/brier_reward/group_std_mean": 0.07825350314378739,
"signal/brier_reward/group_zero_std_frac": 0.26666667461395266,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028055820614099503,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028055820614099503,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.011002604011446237,
"signal/format_reward/group_std_mean": 0.019482703506946565,
"signal/format_reward/group_zero_std_frac": 0.9222222447395325,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005501302005723119,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005501302005723119,
"signal/mean_confidence_reward/centered_abs_mean": 0.04708550795912743,
"signal/mean_confidence_reward/group_std_mean": 0.063015665858984,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2750000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.7085508754207693e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.7085508754207693e-07,
"step": 615
},
{
"calibration/aurc": 0.08026650067991369,
"calibration/batch_distribution_entropy": 0.6396196020074065,
"calibration/batch_entropy_100bins": 0.35237163811846184,
"calibration/batch_entropy_10bins": 0.6396196020074065,
"calibration/batch_entropy_50bins": 0.41480619106771527,
"calibration/batch_uniqueness": 0.4768779100385724,
"calibration/confidence_entropy": 0.5046711555324517,
"calibration/coverage@0%": 0.23890747389033945,
"calibration/coverage@1%": 0.4128658072236727,
"calibration/coverage@10%": 0.6864931591311476,
"calibration/coverage@15%": 0.803647817535496,
"calibration/coverage@20%": 0.8913060083220238,
"calibration/coverage@25%": 0.9221932114882506,
"calibration/coverage@30%": 0.9221932114882506,
"calibration/coverage@5%": 0.5839131007001614,
"calibration/distribution_entropy_10": 0.6396196020074065,
"calibration/distribution_entropy_100": 0.35237163811846184,
"calibration/ece": 0.14547514358016939,
"calibration/mean_confidence": 0.6918397988901672,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.00590277777777779,
"completions/max_length": 3942.4,
"completions/max_terminated_length": 3942.4,
"completions/mean_length": 1063.0981201171876,
"completions/mean_terminated_length": 1069.5595336914062,
"completions/min_length": 0.0,
"completions/min_terminated_length": 269.0,
"epoch": 1.4903846153846154,
"grad_norm": 0.0004827018128708005,
"learning_rate": 2.5240384615384618e-06,
"loss": -0.007,
"num_tokens": 1595827062.0,
"reward": 1.3161508083343505,
"reward_std": 0.09967431724071503,
"rewards/accuracy_reward": 0.7645833253860473,
"rewards/brier_reward": 0.8736067652702332,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9940972208976746,
"rewards/mean_confidence_reward": 0.7170052051544189,
"sampling/batch_mean_priority_error": 0.03028266689573435,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.19166666666666668,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009760693646967411,
"sampling/priority_kl": 0.02999991215765476,
"sampling/priority_scale": 0.7443806945579127,
"sampling/prob_entropy": 10.278950119018555,
"sampling/prob_max": 5.213724070927128e-05,
"sampling/prob_min": 1.9976762996520846e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.4832000017166138,
"sampling/prompt_draws_total": 44496.0,
"sampling/seen_fraction": 0.8530266761779786,
"sampling/unseen_fraction": 0.1469733238220215,
"signal/accuracy_reward/centered_abs_mean": 0.09691840261220933,
"signal/accuracy_reward/group_std_mean": 0.1350296199321747,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04845920130610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04845920130610466,
"signal/advantage_abs_mean": 0.06861151233315468,
"signal/advantage_pre_scale_abs_mean": 0.06861151233315468,
"signal/advantage_pre_scale_std": 0.15253279954195023,
"signal/advantage_std": 0.15253279954195023,
"signal/brier_reward/centered_abs_mean": 0.054378128796815875,
"signal/brier_reward/group_std_mean": 0.07744768559932709,
"signal/brier_reward/group_zero_std_frac": 0.20555555820465088,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027189064398407937,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027189064398407937,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010091146035119892,
"signal/format_reward/group_std_mean": 0.01992493961006403,
"signal/format_reward/group_zero_std_frac": 0.9138888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005045573017559946,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005045573017559946,
"signal/mean_confidence_reward/centered_abs_mean": 0.049541018158197406,
"signal/mean_confidence_reward/group_std_mean": 0.0657179981470108,
"signal/mean_confidence_reward/group_zero_std_frac": 0.225,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.954101825660473e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.954101825660473e-07,
"step": 620
},
{
"calibration/aurc": 0.1692123429059063,
"calibration/batch_distribution_entropy": 0.7073127475286846,
"calibration/batch_entropy_100bins": 0.3959757774730325,
"calibration/batch_entropy_10bins": 0.7073127475286846,
"calibration/batch_entropy_50bins": 0.4661362784068519,
"calibration/batch_uniqueness": 0.6192726904672575,
"calibration/confidence_entropy": 0.5314451492589153,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.10704960835509139,
"calibration/coverage@10%": 0.19200119669277632,
"calibration/coverage@15%": 0.34942235081495865,
"calibration/coverage@20%": 0.651106819781827,
"calibration/coverage@25%": 0.861975514219057,
"calibration/coverage@30%": 0.9502932210572362,
"calibration/coverage@5%": 0.17059127502175805,
"calibration/distribution_entropy_10": 0.7073127475286846,
"calibration/distribution_entropy_100": 0.3959757774730325,
"calibration/ece": 0.09932219899275038,
"calibration/mean_confidence": 0.6648237434156267,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.00425347222222221,
"completions/max_length": 3901.8,
"completions/max_terminated_length": 3901.8,
"completions/mean_length": 1117.1639892578125,
"completions/mean_terminated_length": 1121.9173828125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 303.2,
"epoch": 1.5024038461538463,
"grad_norm": 0.0005235833232291043,
"learning_rate": 2.4939903846153847e-06,
"loss": -0.0035,
"num_tokens": 1611828215.0,
"reward": 1.3013794660568236,
"reward_std": 0.09948191493749618,
"rewards/accuracy_reward": 0.74296875,
"rewards/brier_reward": 0.8642031192779541,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9955729246139526,
"rewards/mean_confidence_reward": 0.7044878482818604,
"sampling/batch_mean_priority_error": 0.031117468382855173,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.24722222222222223,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009788672253489495,
"sampling/priority_kl": 0.029999787732958794,
"sampling/priority_scale": 0.7461758435005322,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 5.233678166405298e-05,
"sampling/prob_min": 2.0006873819511382e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.495199990272522,
"sampling/prompt_draws_total": 44856.0,
"sampling/seen_fraction": 0.8558666706085205,
"sampling/unseen_fraction": 0.1441333293914795,
"signal/accuracy_reward/centered_abs_mean": 0.0996365025639534,
"signal/accuracy_reward/group_std_mean": 0.14100307822227479,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0498182512819767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0498182512819767,
"signal/advantage_abs_mean": 0.06826261132955551,
"signal/advantage_pre_scale_abs_mean": 0.06826261132955551,
"signal/advantage_pre_scale_std": 0.14662763476371765,
"signal/advantage_std": 0.14662763476371765,
"signal/brier_reward/centered_abs_mean": 0.055239399522542955,
"signal/brier_reward/group_std_mean": 0.07736771106719971,
"signal/brier_reward/group_zero_std_frac": 0.23611111044883729,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027619699761271477,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027619699761271477,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007948133535683155,
"signal/format_reward/group_std_mean": 0.01642947643995285,
"signal/format_reward/group_zero_std_frac": 0.9277778029441833,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003974066767841578,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003974066767841578,
"signal/mean_confidence_reward/centered_abs_mean": 0.04968967735767364,
"signal/mean_confidence_reward/group_std_mean": 0.06586038321256638,
"signal/mean_confidence_reward/group_zero_std_frac": 0.24166666567325593,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.968967687091208e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.968967687091208e-07,
"step": 625
},
{
"calibration/aurc": 0.07472276214057484,
"calibration/batch_distribution_entropy": 0.6541581603664168,
"calibration/batch_entropy_100bins": 0.3582094950133639,
"calibration/batch_entropy_10bins": 0.6541581603664168,
"calibration/batch_entropy_50bins": 0.42167842174866077,
"calibration/batch_uniqueness": 0.45456095770444616,
"calibration/confidence_entropy": 0.4816627977491092,
"calibration/coverage@0%": 0.20175853018372703,
"calibration/coverage@1%": 0.3404983955128461,
"calibration/coverage@10%": 0.6757976022952985,
"calibration/coverage@15%": 0.7468123990545482,
"calibration/coverage@20%": 0.8864184404933099,
"calibration/coverage@25%": 0.9407666134167011,
"calibration/coverage@30%": 0.9884210526315791,
"calibration/coverage@5%": 0.54803349468375,
"calibration/distribution_entropy_10": 0.6541581603664168,
"calibration/distribution_entropy_100": 0.3582094950133639,
"calibration/ece": 0.1292614681630037,
"calibration/mean_confidence": 0.70898345643133,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0019097222222222321,
"completions/max_length": 3805.6,
"completions/max_terminated_length": 3805.6,
"completions/mean_length": 1087.827685546875,
"completions/mean_terminated_length": 1089.9354614257813,
"completions/min_length": 43.6,
"completions/min_terminated_length": 284.4,
"epoch": 1.5144230769230769,
"grad_norm": 0.0004652974021155387,
"learning_rate": 2.463942307692308e-06,
"loss": -0.0006,
"num_tokens": 1627494742.0,
"reward": 1.3149696588516235,
"reward_std": 0.08956841826438904,
"rewards/accuracy_reward": 0.75234375,
"rewards/brier_reward": 0.8796649217605591,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166626930237,
"rewards/mean_confidence_reward": 0.6975868105888366,
"sampling/batch_mean_priority_error": 0.025660975930045727,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2333333333333333,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009810564666986465,
"sampling/priority_kl": 0.02999999262392521,
"sampling/priority_scale": 0.7481342852348462,
"sampling/prob_entropy": 10.27895278930664,
"sampling/prob_max": 5.254092757240869e-05,
"sampling/prob_min": 2.0034283807035536e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.507200002670288,
"sampling/prompt_draws_total": 45216.0,
"sampling/seen_fraction": 0.8587466716766358,
"sampling/unseen_fraction": 0.14125332832336426,
"signal/accuracy_reward/centered_abs_mean": 0.099853515625,
"signal/accuracy_reward/group_std_mean": 0.13629501461982726,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0499267578125,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0499267578125,
"signal/advantage_abs_mean": 0.0632256880402565,
"signal/advantage_pre_scale_abs_mean": 0.0632256880402565,
"signal/advantage_pre_scale_std": 0.13528273105621338,
"signal/advantage_std": 0.13528273105621338,
"signal/brier_reward/centered_abs_mean": 0.05339196249842644,
"signal/brier_reward/group_std_mean": 0.07426847070455551,
"signal/brier_reward/group_zero_std_frac": 0.22222222685813903,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02669598124921322,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02669598124921322,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0038845486007630823,
"signal/format_reward/group_std_mean": 0.00885537788271904,
"signal/format_reward/group_zero_std_frac": 0.9583333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019422743003815412,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019422743003815412,
"signal/mean_confidence_reward/centered_abs_mean": 0.05010091587901115,
"signal/mean_confidence_reward/group_std_mean": 0.06616592630743981,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23611111640930177,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.010091399526573e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.010091399526573e-07,
"step": 630
},
{
"calibration/aurc": 0.0869278000999156,
"calibration/batch_distribution_entropy": 0.6093055797850038,
"calibration/batch_entropy_100bins": 0.34342417809934345,
"calibration/batch_entropy_10bins": 0.6093055797850038,
"calibration/batch_entropy_50bins": 0.4042733859018993,
"calibration/batch_uniqueness": 0.37829591462204915,
"calibration/confidence_entropy": 0.46445287192138196,
"calibration/coverage@0%": 0.17471310903252826,
"calibration/coverage@1%": 0.17471310903252826,
"calibration/coverage@10%": 0.6890747385742918,
"calibration/coverage@15%": 0.8484273941592011,
"calibration/coverage@20%": 0.8900041805536268,
"calibration/coverage@25%": 0.9237453276165347,
"calibration/coverage@30%": 0.9583333333333333,
"calibration/coverage@5%": 0.4357315831097729,
"calibration/distribution_entropy_10": 0.6093055797850038,
"calibration/distribution_entropy_100": 0.34342417809934345,
"calibration/ece": 0.136922299089084,
"calibration/mean_confidence": 0.6978689054156199,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.005381944444444442,
"completions/max_length": 3956.2,
"completions/max_terminated_length": 3956.2,
"completions/mean_length": 1087.4025390625,
"completions/mean_terminated_length": 1093.3152587890625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 282.6,
"epoch": 1.5264423076923077,
"grad_norm": 0.0005978142144158483,
"learning_rate": 2.433894230769231e-06,
"loss": -0.0044,
"num_tokens": 1643114099.0,
"reward": 1.3020598888397217,
"reward_std": 0.10506750345230102,
"rewards/accuracy_reward": 0.7394965171813965,
"rewards/brier_reward": 0.8700781345367432,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.994531261920929,
"rewards/mean_confidence_reward": 0.692760419845581,
"sampling/batch_mean_priority_error": 0.035493932026362335,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.2416666666666667,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009836449101567269,
"sampling/priority_kl": 0.030000242963433267,
"sampling/priority_scale": 0.7501380623085424,
"sampling/prob_entropy": 10.278949165344239,
"sampling/prob_max": 5.274732830002904e-05,
"sampling/prob_min": 2.006111608352512e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5192000150680542,
"sampling/prompt_draws_total": 45576.0,
"sampling/seen_fraction": 0.8615866661071777,
"sampling/unseen_fraction": 0.13841333389282226,
"signal/accuracy_reward/centered_abs_mean": 0.10477973222732544,
"signal/accuracy_reward/group_std_mean": 0.14873329997062684,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888895511628,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05238986611366272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05238986611366272,
"signal/advantage_abs_mean": 0.07023787051439286,
"signal/advantage_pre_scale_abs_mean": 0.07023787051439286,
"signal/advantage_pre_scale_std": 0.15317056477069854,
"signal/advantage_std": 0.15317056477069854,
"signal/brier_reward/centered_abs_mean": 0.05945446789264679,
"signal/brier_reward/group_std_mean": 0.08587936162948609,
"signal/brier_reward/group_zero_std_frac": 0.275,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029727233946323393,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029727233946323393,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009977213479578495,
"signal/format_reward/group_std_mean": 0.02117712702602148,
"signal/format_reward/group_zero_std_frac": 0.9055555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004988606739789248,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004988606739789248,
"signal/mean_confidence_reward/centered_abs_mean": 0.0510736845433712,
"signal/mean_confidence_reward/group_std_mean": 0.06932668089866638,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3083333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.107368110657262e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.107368110657262e-07,
"step": 635
},
{
"calibration/aurc": 0.16204915459909533,
"calibration/batch_distribution_entropy": 0.5274177386820045,
"calibration/batch_entropy_100bins": 0.2888359937240352,
"calibration/batch_entropy_10bins": 0.5274177386820045,
"calibration/batch_entropy_50bins": 0.3400130584847098,
"calibration/batch_uniqueness": 0.1655676973886686,
"calibration/confidence_entropy": 0.42929724407108766,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.29937611653153773,
"calibration/coverage@15%": 0.3413906829737529,
"calibration/coverage@20%": 0.6711885039790432,
"calibration/coverage@25%": 0.8699130660510619,
"calibration/coverage@30%": 0.9429319371727749,
"calibration/coverage@5%": 0.1368421052631579,
"calibration/distribution_entropy_10": 0.5274177386820045,
"calibration/distribution_entropy_100": 0.2888359937240352,
"calibration/ece": 0.09777153337267655,
"calibration/mean_confidence": 0.752498526092933,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.006857638888888884,
"completions/max_length": 3726.6,
"completions/max_terminated_length": 3726.6,
"completions/mean_length": 1074.8276123046876,
"completions/mean_terminated_length": 1082.3476684570312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 275.2,
"epoch": 1.5384615384615383,
"grad_norm": 0.0006733312038704753,
"learning_rate": 2.403846153846154e-06,
"loss": -0.0059,
"num_tokens": 1658593169.0,
"reward": 1.3056791067123412,
"reward_std": 0.10761468857526779,
"rewards/accuracy_reward": 0.7495659708976745,
"rewards/brier_reward": 0.8687222003936768,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9930555701255799,
"rewards/mean_confidence_reward": 0.7232986450195312,
"sampling/batch_mean_priority_error": 0.03791013955058041,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.23055555555555554,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009872362203896045,
"sampling/priority_kl": 0.03000005818903446,
"sampling/priority_scale": 0.7523871720070019,
"sampling/prob_entropy": 10.278955078125,
"sampling/prob_max": 5.2965757640777154e-05,
"sampling/prob_min": 2.008588744502049e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5312000036239624,
"sampling/prompt_draws_total": 45936.0,
"sampling/seen_fraction": 0.8645066618919373,
"sampling/unseen_fraction": 0.13549333810806274,
"signal/accuracy_reward/centered_abs_mean": 0.1015896275639534,
"signal/accuracy_reward/group_std_mean": 0.14180152118206024,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0507948137819767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0507948137819767,
"signal/advantage_abs_mean": 0.0746236190199852,
"signal/advantage_pre_scale_abs_mean": 0.0746236190199852,
"signal/advantage_pre_scale_std": 0.16241733729839325,
"signal/advantage_std": 0.16241733729839325,
"signal/brier_reward/centered_abs_mean": 0.06814387738704682,
"signal/brier_reward/group_std_mean": 0.09454997181892395,
"signal/brier_reward/group_zero_std_frac": 0.3111111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03407193869352341,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03407193869352341,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010449218843132258,
"signal/format_reward/group_std_mean": 0.019220804050564765,
"signal/format_reward/group_zero_std_frac": 0.9222222447395325,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005224609421566129,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005224609421566129,
"signal/mean_confidence_reward/centered_abs_mean": 0.05597711279988289,
"signal/mean_confidence_reward/group_std_mean": 0.07454514056444168,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3444444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.597710924121202e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.597710924121202e-07,
"step": 640
},
{
"calibration/aurc": 0.14557671748218712,
"calibration/batch_distribution_entropy": 0.48910598486412377,
"calibration/batch_entropy_100bins": 0.2652488436322299,
"calibration/batch_entropy_10bins": 0.48910598486412377,
"calibration/batch_entropy_50bins": 0.31224664703352756,
"calibration/batch_uniqueness": 0.052291835380526994,
"calibration/confidence_entropy": 0.42219030296825294,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.36709317585301837,
"calibration/coverage@15%": 0.6915279590948651,
"calibration/coverage@20%": 0.7566579634464752,
"calibration/coverage@25%": 0.7796344647519582,
"calibration/coverage@30%": 0.9255072345517842,
"calibration/coverage@5%": 0.16614583333333333,
"calibration/distribution_entropy_10": 0.48910598486412377,
"calibration/distribution_entropy_100": 0.2652488436322299,
"calibration/ece": 0.09933202463799398,
"calibration/mean_confidence": 0.7654084570715379,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.008506944444444442,
"completions/max_length": 3880.8,
"completions/max_terminated_length": 3880.8,
"completions/mean_length": 1150.5470458984375,
"completions/mean_terminated_length": 1160.5489013671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 293.2,
"epoch": 1.5504807692307692,
"grad_norm": 0.0005169560317881405,
"learning_rate": 2.373798076923077e-06,
"loss": -0.0113,
"num_tokens": 1674952815.0,
"reward": 1.288080620765686,
"reward_std": 0.12121982872486115,
"rewards/accuracy_reward": 0.7271701455116272,
"rewards/brier_reward": 0.8579175591468811,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9910590291023255,
"rewards/mean_confidence_reward": 0.7332204699516296,
"sampling/batch_mean_priority_error": 0.04735698801884604,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.21666666666666665,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009915192052721978,
"sampling/priority_kl": 0.029999776929616927,
"sampling/priority_scale": 0.7540454684989527,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 5.3160951938480136e-05,
"sampling/prob_min": 2.0116871382924728e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5432000160217285,
"sampling/prompt_draws_total": 46296.0,
"sampling/seen_fraction": 0.8670066714286804,
"sampling/unseen_fraction": 0.13299332857131957,
"signal/accuracy_reward/centered_abs_mean": 0.10897894948720932,
"signal/accuracy_reward/group_std_mean": 0.1503289371728897,
"signal/accuracy_reward/group_zero_std_frac": 0.5472222447395325,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05448947474360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05448947474360466,
"signal/advantage_abs_mean": 0.08551755398511887,
"signal/advantage_pre_scale_abs_mean": 0.08551755398511887,
"signal/advantage_pre_scale_std": 0.18074164986610414,
"signal/advantage_std": 0.18074164986610414,
"signal/brier_reward/centered_abs_mean": 0.07495963871479035,
"signal/brier_reward/group_std_mean": 0.1039071187376976,
"signal/brier_reward/group_zero_std_frac": 0.3388888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.037479819357395174,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.037479819357395174,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.014284939505159855,
"signal/format_reward/group_std_mean": 0.025171923637390136,
"signal/format_reward/group_zero_std_frac": 0.9000000119209289,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0071424697525799274,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0071424697525799274,
"signal/mean_confidence_reward/centered_abs_mean": 0.05712837353348732,
"signal/mean_confidence_reward/group_std_mean": 0.0764322578907013,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3777777850627899,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.712837264582049e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.712837264582049e-07,
"step": 645
},
{
"calibration/aurc": 0.14833079176694414,
"calibration/batch_distribution_entropy": 0.3771737415299348,
"calibration/batch_entropy_100bins": 0.20520349746532088,
"calibration/batch_entropy_10bins": 0.3771737415299348,
"calibration/batch_entropy_50bins": 0.24156223705140362,
"calibration/batch_uniqueness": -0.19533839640881734,
"calibration/confidence_entropy": 0.3984146293875352,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.30994764397905755,
"calibration/coverage@15%": 0.5146911268163985,
"calibration/coverage@20%": 0.7292323419304039,
"calibration/coverage@25%": 0.9198952879581151,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.1863874345549738,
"calibration/distribution_entropy_10": 0.3771737415299348,
"calibration/distribution_entropy_100": 0.20520349746532088,
"calibration/ece": 0.10669382387807795,
"calibration/mean_confidence": 0.8033982003309867,
"calibration/unique_confidence_per_question": 0.019791666666666662,
"calibration/unique_confidences": 7.6,
"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": 0.007638888888888862,
"completions/max_length": 3855.8,
"completions/max_terminated_length": 3855.8,
"completions/mean_length": 1063.4618896484376,
"completions/mean_terminated_length": 1071.8775146484375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 302.8,
"epoch": 1.5625,
"grad_norm": 0.0005790924187749624,
"learning_rate": 2.3437500000000002e-06,
"loss": -0.0066,
"num_tokens": 1690267352.0,
"reward": 1.3241550207138062,
"reward_std": 0.11605097949504853,
"rewards/accuracy_reward": 0.7796875,
"rewards/brier_reward": 0.8764192819595337,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9921875119209289,
"rewards/mean_confidence_reward": 0.792578113079071,
"sampling/batch_mean_priority_error": 0.04947610722316871,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20277777777777778,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.009960375353693962,
"sampling/priority_kl": 0.029999787360429762,
"sampling/priority_scale": 0.7557832061545924,
"sampling/prob_entropy": 10.278946876525879,
"sampling/prob_max": 5.335809182724915e-05,
"sampling/prob_min": 2.0122729256399908e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5552000045776366,
"sampling/prompt_draws_total": 46656.0,
"sampling/seen_fraction": 0.869486677646637,
"sampling/unseen_fraction": 0.13051332235336305,
"signal/accuracy_reward/centered_abs_mean": 0.10058593600988389,
"signal/accuracy_reward/group_std_mean": 0.13794552087783812,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05029296800494194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05029296800494194,
"signal/advantage_abs_mean": 0.08049516677856446,
"signal/advantage_pre_scale_abs_mean": 0.08049516677856446,
"signal/advantage_pre_scale_std": 0.1770169109106064,
"signal/advantage_std": 0.1770169109106064,
"signal/brier_reward/centered_abs_mean": 0.06743035167455673,
"signal/brier_reward/group_std_mean": 0.09736959785223007,
"signal/brier_reward/group_zero_std_frac": 0.4250000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033715175837278365,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033715175837278365,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.011990017257630825,
"signal/format_reward/group_std_mean": 0.02398403510451317,
"signal/format_reward/group_zero_std_frac": 0.8944444537162781,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0059950086288154125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0059950086288154125,
"signal/mean_confidence_reward/centered_abs_mean": 0.049835624545812605,
"signal/mean_confidence_reward/group_std_mean": 0.06977761834859848,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4694444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.983562519100815e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.983562519100815e-07,
"step": 650
},
{
"epoch": 1.5625,
"eval_calibration/aurc": 0.1593088521292015,
"eval_calibration/batch_distribution_entropy": 0.45063243173571826,
"eval_calibration/batch_entropy_100bins": 0.2465860221535046,
"eval_calibration/batch_entropy_10bins": 0.45063243173571826,
"eval_calibration/batch_entropy_50bins": 0.2902770755506939,
"eval_calibration/batch_uniqueness": -0.048974735640835654,
"eval_calibration/confidence_entropy": 0.4098930569864548,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.7151567944250871,
"eval_calibration/coverage@20%": 0.8153310104529616,
"eval_calibration/coverage@25%": 0.936411149825784,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.45063243173571826,
"eval_calibration/distribution_entropy_100": 0.2465860221535046,
"eval_calibration/ece": 0.06210801393728245,
"eval_calibration/mean_confidence": 0.784233449477352,
"eval_calibration/unique_confidence_per_question": 0.006944444444444444,
"eval_calibration/unique_confidences": 8,
"eval_completions/clipped_ratio": 0.0034722222222222285,
"eval_completions/max_length": 2993.1666666666665,
"eval_completions/max_terminated_length": 2993.1666666666665,
"eval_completions/mean_length": 1073.7899576822917,
"eval_completions/mean_terminated_length": 1077.5393676757812,
"eval_completions/min_length": 175.5,
"eval_completions/min_terminated_length": 332.0,
"eval_loss": 0.0,
"eval_num_tokens": 1690267352.0,
"eval_reward": 1.2808280785878499,
"eval_reward_std": 0.33280307054519653,
"eval_rewards/accuracy_reward": 0.7196180621782938,
"eval_rewards/brier_reward": 0.845494786898295,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.996527781089147,
"eval_rewards/mean_confidence_reward": 0.7815104027589163,
"eval_runtime": 195.6129,
"eval_samples_per_second": 5.112,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3918728331724803,
"eval_signal/accuracy_reward/group_std_mean": 0.4482126832008362,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19593641658624014,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19593641658624014,
"eval_signal/advantage_abs_mean": 0.27938560644785565,
"eval_signal/advantage_pre_scale_abs_mean": 0.27938560644785565,
"eval_signal/advantage_pre_scale_std": 0.3301011423269908,
"eval_signal/advantage_std": 0.3301011423269908,
"eval_signal/brier_reward/centered_abs_mean": 0.19026151796181998,
"eval_signal/brier_reward/group_std_mean": 0.25656550129254657,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09513075898090999,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09513075898090999,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"eval_signal/format_reward/group_std_mean": 0.019641855110724766,
"eval_signal/format_reward/group_zero_std_frac": 0.8888889054457346,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.16597218811511993,
"eval_signal/mean_confidence_reward/group_std_mean": 0.21074802925189337,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6597218556550313e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6597218556550313e-06,
"eval_steps_per_second": 0.031,
"step": 650
},
{
"epoch": 1.5625,
"step": 650,
"train_probe_calibration/aurc": 0.11914299116697188,
"train_probe_calibration/batch_distribution_entropy": 0.4031877624643556,
"train_probe_calibration/batch_entropy_100bins": 0.22108942808918203,
"train_probe_calibration/batch_entropy_10bins": 0.4031877624643556,
"train_probe_calibration/batch_entropy_50bins": 0.26026289754960896,
"train_probe_calibration/batch_uniqueness": -0.15853625979672398,
"train_probe_calibration/confidence_entropy": 0.39954392713722525,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.0,
"train_probe_calibration/coverage@15%": 0.8393013100436681,
"train_probe_calibration/coverage@20%": 0.9502183406113537,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.4031877624643556,
"train_probe_calibration/distribution_entropy_100": 0.22108942808918203,
"train_probe_calibration/ece": 0.03397379912663767,
"train_probe_calibration/mean_confidence": 0.8000873362445416,
"train_probe_calibration/unique_confidence_per_question": 0.006944444444444444,
"train_probe_calibration/unique_confidences": 8,
"train_probe_completions/clipped_ratio": 0.005208333333333333,
"train_probe_completions/max_length": 3387.8333333333335,
"train_probe_completions/max_terminated_length": 3387.8333333333335,
"train_probe_completions/mean_length": 1080.039082845052,
"train_probe_completions/mean_terminated_length": 1085.6224975585938,
"train_probe_completions/min_length": 100.0,
"train_probe_completions/min_terminated_length": 281.0,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1690267352.0,
"train_probe_reward": 1.3141876459121704,
"train_probe_reward_std": 0.31840459009011585,
"train_probe_rewards/accuracy_reward": 0.7708333333333334,
"train_probe_rewards/brier_reward": 0.8636024494965872,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9939236144224802,
"train_probe_rewards/mean_confidence_reward": 0.7952256699403127,
"train_probe_runtime": 205.8787,
"train_probe_samples_per_second": 4.857,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3425564269224803,
"train_probe_signal/accuracy_reward/group_std_mean": 0.41801586250464123,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17127821346124014,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17127821346124014,
"train_probe_signal/advantage_abs_mean": 0.25017442802588147,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.25017442802588147,
"train_probe_signal/advantage_pre_scale_std": 0.31694228450457257,
"train_probe_signal/advantage_std": 0.31694228450457257,
"train_probe_signal/brier_reward/centered_abs_mean": 0.17618762950102487,
"train_probe_signal/brier_reward/group_std_mean": 0.24327529221773148,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08809381475051244,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.08809381475051244,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.011773003110041222,
"train_probe_signal/format_reward/group_std_mean": 0.034373246443768345,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8055555820465088,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.15472543984651566,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.20226836701234183,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.5472543850592047e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.5472543850592047e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.1108042678066182,
"calibration/batch_distribution_entropy": 0.42072037376674637,
"calibration/batch_entropy_100bins": 0.22931455418855728,
"calibration/batch_entropy_10bins": 0.42072037376674637,
"calibration/batch_entropy_50bins": 0.26994538291236825,
"calibration/batch_uniqueness": -0.08188383227497045,
"calibration/confidence_entropy": 0.4100209457284782,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.11510416666666667,
"calibration/coverage@10%": 0.5703043407310705,
"calibration/coverage@15%": 0.7349573490813649,
"calibration/coverage@20%": 0.8836239710873837,
"calibration/coverage@25%": 0.9465157823858406,
"calibration/coverage@30%": 0.9564895356666803,
"calibration/coverage@5%": 0.3140625,
"calibration/distribution_entropy_10": 0.42072037376674637,
"calibration/distribution_entropy_100": 0.22931455418855728,
"calibration/ece": 0.10658886146536148,
"calibration/mean_confidence": 0.7913267767461489,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.006770833333333348,
"completions/max_length": 3753.2,
"completions/max_terminated_length": 3753.2,
"completions/mean_length": 1044.0681518554688,
"completions/mean_terminated_length": 1051.2261840820313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 249.8,
"epoch": 1.5745192307692308,
"grad_norm": 0.0008380069630220532,
"learning_rate": 2.3137019230769236e-06,
"loss": -0.0067,
"num_tokens": 1705374633.0,
"reward": 1.3213003873825073,
"reward_std": 0.129642491042614,
"rewards/accuracy_reward": 0.7828993201255798,
"rewards/brier_reward": 0.8664566040039062,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9932291626930236,
"rewards/mean_confidence_reward": 0.7906076312065125,
"sampling/batch_mean_priority_error": 0.052881742346931083,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.25555555555555554,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010013969987630844,
"sampling/priority_kl": 0.029999995976686476,
"sampling/priority_scale": 0.7580755293602124,
"sampling/prob_entropy": 10.278956985473632,
"sampling/prob_max": 5.357732443371788e-05,
"sampling/prob_min": 2.012997429119423e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.567199993133545,
"sampling/prompt_draws_total": 47016.0,
"sampling/seen_fraction": 0.8722466707229615,
"sampling/unseen_fraction": 0.12775332927703859,
"signal/accuracy_reward/centered_abs_mean": 0.1139485701918602,
"signal/accuracy_reward/group_std_mean": 0.1550443172454834,
"signal/accuracy_reward/group_zero_std_frac": 0.5444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0569742850959301,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0569742850959301,
"signal/advantage_abs_mean": 0.0928182914853096,
"signal/advantage_pre_scale_abs_mean": 0.0928182914853096,
"signal/advantage_pre_scale_std": 0.1907956838607788,
"signal/advantage_std": 0.1907956838607788,
"signal/brier_reward/centered_abs_mean": 0.07859473824501037,
"signal/brier_reward/group_std_mean": 0.10812324136495591,
"signal/brier_reward/group_zero_std_frac": 0.3944444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.039297369122505185,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.039297369122505185,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01182725690305233,
"signal/format_reward/group_std_mean": 0.023676491528749465,
"signal/format_reward/group_zero_std_frac": 0.8972222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005913628451526165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005913628451526165,
"signal/mean_confidence_reward/centered_abs_mean": 0.05477920100092888,
"signal/mean_confidence_reward/group_std_mean": 0.07536292672157288,
"signal/mean_confidence_reward/group_zero_std_frac": 0.45833333730697634,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.477919728491543e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.477919728491543e-07,
"step": 655
},
{
"calibration/aurc": 0.11131981175442647,
"calibration/batch_distribution_entropy": 0.41437273865924407,
"calibration/batch_entropy_100bins": 0.2257903019472393,
"calibration/batch_entropy_10bins": 0.41437273865924407,
"calibration/batch_entropy_50bins": 0.26579669019581226,
"calibration/batch_uniqueness": -0.08864372136822053,
"calibration/confidence_entropy": 0.40409001678171536,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.12637075718015667,
"calibration/coverage@10%": 0.53459153114999,
"calibration/coverage@15%": 0.7111558835824373,
"calibration/coverage@20%": 0.8782477048970255,
"calibration/coverage@25%": 0.9270458882451414,
"calibration/coverage@30%": 0.9323229331000228,
"calibration/coverage@5%": 0.34516971279373365,
"calibration/distribution_entropy_10": 0.41437273865924407,
"calibration/distribution_entropy_100": 0.2257903019472393,
"calibration/ece": 0.08919565324662408,
"calibration/mean_confidence": 0.7744310496497983,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.00625,
"completions/max_length": 3810.6,
"completions/max_terminated_length": 3810.6,
"completions/mean_length": 1064.3376220703126,
"completions/mean_terminated_length": 1071.087841796875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 260.8,
"epoch": 1.5865384615384617,
"grad_norm": 0.0007543113315477967,
"learning_rate": 2.283653846153846e-06,
"loss": -0.0074,
"num_tokens": 1720727802.0,
"reward": 1.3003161668777465,
"reward_std": 0.11281631141901016,
"rewards/accuracy_reward": 0.7461805582046509,
"rewards/brier_reward": 0.8606866598129272,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9937500119209289,
"rewards/mean_confidence_reward": 0.7538802027702332,
"sampling/batch_mean_priority_error": 0.05363177499702718,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20833333333333334,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010069727897644043,
"sampling/priority_kl": 0.03000015877187252,
"sampling/priority_scale": 0.7605190575355664,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 5.380468719522469e-05,
"sampling/prob_min": 2.0151718490524218e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.579199981689453,
"sampling/prompt_draws_total": 47376.0,
"sampling/seen_fraction": 0.875,
"sampling/unseen_fraction": 0.125,
"signal/accuracy_reward/centered_abs_mean": 0.10485026091337205,
"signal/accuracy_reward/group_std_mean": 0.14478610157966615,
"signal/accuracy_reward/group_zero_std_frac": 0.5555555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05242513045668602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05242513045668602,
"signal/advantage_abs_mean": 0.0782653659582138,
"signal/advantage_pre_scale_abs_mean": 0.0782653659582138,
"signal/advantage_pre_scale_std": 0.1688857764005661,
"signal/advantage_std": 0.1688857764005661,
"signal/brier_reward/centered_abs_mean": 0.0728747896850109,
"signal/brier_reward/group_std_mean": 0.10191280543804168,
"signal/brier_reward/group_zero_std_frac": 0.375,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03643739484250545,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03643739484250545,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010872395848855377,
"signal/format_reward/group_std_mean": 0.022299006581306458,
"signal/format_reward/group_zero_std_frac": 0.9,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.005436197924427688,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.005436197924427688,
"signal/mean_confidence_reward/centered_abs_mean": 0.0543424591422081,
"signal/mean_confidence_reward/group_std_mean": 0.07400039434432984,
"signal/mean_confidence_reward/group_zero_std_frac": 0.42777777910232545,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.434245792912407e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.434245792912407e-07,
"step": 660
},
{
"calibration/aurc": 0.1159017932887628,
"calibration/batch_distribution_entropy": 0.46562924757054425,
"calibration/batch_entropy_100bins": 0.2589695955178345,
"calibration/batch_entropy_10bins": 0.46562924757054425,
"calibration/batch_entropy_50bins": 0.3048548177506447,
"calibration/batch_uniqueness": 0.05487002083086965,
"calibration/confidence_entropy": 0.4198923602522945,
"calibration/coverage@0%": 0.1328125,
"calibration/coverage@1%": 0.13385416666666666,
"calibration/coverage@10%": 0.6289555662145125,
"calibration/coverage@15%": 0.6962895778364115,
"calibration/coverage@20%": 0.7218997361477573,
"calibration/coverage@25%": 0.8915332440011603,
"calibration/coverage@30%": 0.9278412465982401,
"calibration/coverage@5%": 0.4299216502074764,
"calibration/distribution_entropy_10": 0.46562924757054425,
"calibration/distribution_entropy_100": 0.2589695955178345,
"calibration/ece": 0.1093648627357036,
"calibration/mean_confidence": 0.7479002819167239,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.004947916666666652,
"completions/max_length": 3819.2,
"completions/max_terminated_length": 3819.2,
"completions/mean_length": 1034.9790893554687,
"completions/mean_terminated_length": 1040.2152709960938,
"completions/min_length": 0.0,
"completions/min_terminated_length": 284.6,
"epoch": 1.5985576923076923,
"grad_norm": 0.0005559017299674451,
"learning_rate": 2.2536057692307694e-06,
"loss": -0.0048,
"num_tokens": 1735753353.0,
"reward": 1.3110567331314087,
"reward_std": 0.11565322428941727,
"rewards/accuracy_reward": 0.7572048664093017,
"rewards/brier_reward": 0.86984201669693,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9950520873069764,
"rewards/mean_confidence_reward": 0.7291840076446533,
"sampling/batch_mean_priority_error": 0.041489098082864787,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18055555555555552,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010115215368568898,
"sampling/priority_kl": 0.02999992445111275,
"sampling/priority_scale": 0.7622721730964258,
"sampling/prob_entropy": 10.278946685791016,
"sampling/prob_max": 5.400420050136745e-05,
"sampling/prob_min": 2.0180710271233694e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.5911999940872192,
"sampling/prompt_draws_total": 47736.0,
"sampling/seen_fraction": 0.8773200035095214,
"sampling/unseen_fraction": 0.12267999649047852,
"signal/accuracy_reward/centered_abs_mean": 0.12019856721162796,
"signal/accuracy_reward/group_std_mean": 0.15842821300029755,
"signal/accuracy_reward/group_zero_std_frac": 0.547222226858139,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.06009928360581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.06009928360581398,
"signal/advantage_abs_mean": 0.08310452848672867,
"signal/advantage_pre_scale_abs_mean": 0.08310452848672867,
"signal/advantage_pre_scale_std": 0.1731024533510208,
"signal/advantage_std": 0.1731024533510208,
"signal/brier_reward/centered_abs_mean": 0.07321913093328476,
"signal/brier_reward/group_std_mean": 0.10007321387529373,
"signal/brier_reward/group_zero_std_frac": 0.33333333730697634,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03660956546664238,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03660956546664238,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009087456623092294,
"signal/format_reward/group_std_mean": 0.01993839703500271,
"signal/format_reward/group_zero_std_frac": 0.9083333253860474,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004543728311546147,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004543728311546147,
"signal/mean_confidence_reward/centered_abs_mean": 0.05739041939377785,
"signal/mean_confidence_reward/group_std_mean": 0.07699377685785294,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3777777850627899,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.739041739616368e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.739041739616368e-07,
"step": 665
},
{
"calibration/aurc": 0.11042005183808701,
"calibration/batch_distribution_entropy": 0.5913382170007677,
"calibration/batch_entropy_100bins": 0.32877258800773546,
"calibration/batch_entropy_10bins": 0.5913382170007677,
"calibration/batch_entropy_50bins": 0.38702577110679987,
"calibration/batch_uniqueness": 0.34693221792977613,
"calibration/confidence_entropy": 0.45987446598557097,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.12180851063829787,
"calibration/coverage@10%": 0.3584779722885766,
"calibration/coverage@15%": 0.7167577612497782,
"calibration/coverage@20%": 0.8871506519359545,
"calibration/coverage@25%": 0.9662269129287597,
"calibration/coverage@30%": 0.9662269129287597,
"calibration/coverage@5%": 0.3091463958020851,
"calibration/distribution_entropy_10": 0.5913382170007677,
"calibration/distribution_entropy_100": 0.32877258800773546,
"calibration/ece": 0.14242507122515471,
"calibration/mean_confidence": 0.6710677370417025,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.010677083333333327,
"completions/max_length": 3976.8,
"completions/max_terminated_length": 3976.8,
"completions/mean_length": 1062.3335083007812,
"completions/mean_terminated_length": 1073.8999389648438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 208.4,
"epoch": 1.6105769230769231,
"grad_norm": 0.0004333420074544847,
"learning_rate": 2.2235576923076924e-06,
"loss": -0.0123,
"num_tokens": 1751071755.0,
"reward": 1.285237693786621,
"reward_std": 0.11045846790075302,
"rewards/accuracy_reward": 0.7262152910232544,
"rewards/brier_reward": 0.8550104260444641,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9892361044883728,
"rewards/mean_confidence_reward": 0.6805728912353516,
"sampling/batch_mean_priority_error": 0.038647688258623844,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20833333333333334,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010153127089142799,
"sampling/priority_kl": 0.030000197887420654,
"sampling/priority_scale": 0.7641624867683277,
"sampling/prob_entropy": 10.278944778442384,
"sampling/prob_max": 5.420961242634803e-05,
"sampling/prob_min": 2.0208075875416397e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6032000064849854,
"sampling/prompt_draws_total": 48096.0,
"sampling/seen_fraction": 0.8796666622161865,
"sampling/unseen_fraction": 0.12033333778381347,
"signal/accuracy_reward/centered_abs_mean": 0.11707899421453476,
"signal/accuracy_reward/group_std_mean": 0.15186972320079803,
"signal/accuracy_reward/group_zero_std_frac": 0.5694444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05853949710726738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05853949710726738,
"signal/advantage_abs_mean": 0.07833757400512695,
"signal/advantage_pre_scale_abs_mean": 0.07833757400512695,
"signal/advantage_pre_scale_std": 0.1668864071369171,
"signal/advantage_std": 0.1668864071369171,
"signal/brier_reward/centered_abs_mean": 0.0744941309094429,
"signal/brier_reward/group_std_mean": 0.10006592720746994,
"signal/brier_reward/group_zero_std_frac": 0.3027777850627899,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03724706545472145,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03724706545472145,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.017892795242369176,
"signal/format_reward/group_std_mean": 0.03308853507041931,
"signal/format_reward/group_zero_std_frac": 0.8638888835906983,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.008946397621184588,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.008946397621184588,
"signal/mean_confidence_reward/centered_abs_mean": 0.05739258825778961,
"signal/mean_confidence_reward/group_std_mean": 0.07705257311463357,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32222222685813906,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.73925871094616e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.73925871094616e-07,
"step": 670
},
{
"calibration/aurc": 0.10770899593223338,
"calibration/batch_distribution_entropy": 0.6491505336506754,
"calibration/batch_entropy_100bins": 0.36866915529152067,
"calibration/batch_entropy_10bins": 0.6491505336506754,
"calibration/batch_entropy_50bins": 0.43399136459222154,
"calibration/batch_uniqueness": 0.4690463544560851,
"calibration/confidence_entropy": 0.48343109218125313,
"calibration/coverage@0%": 0.1550098340674257,
"calibration/coverage@1%": 0.1550098340674257,
"calibration/coverage@10%": 0.5382968708318092,
"calibration/coverage@15%": 0.7641439630800948,
"calibration/coverage@20%": 0.8426342540515769,
"calibration/coverage@25%": 0.9148564762737991,
"calibration/coverage@30%": 0.9496062992125985,
"calibration/coverage@5%": 0.368785838942907,
"calibration/distribution_entropy_10": 0.6491505336506754,
"calibration/distribution_entropy_100": 0.36866915529152067,
"calibration/ece": 0.12161472797602957,
"calibration/mean_confidence": 0.6472128889170813,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.007465277777777768,
"completions/max_length": 3823.4,
"completions/max_terminated_length": 3823.4,
"completions/mean_length": 1071.7802368164062,
"completions/mean_terminated_length": 1079.799658203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 252.0,
"epoch": 1.6225961538461537,
"grad_norm": 0.00043507353984750807,
"learning_rate": 2.1935096153846157e-06,
"loss": -0.0083,
"num_tokens": 1766519559.0,
"reward": 1.2930452346801757,
"reward_std": 0.09925612062215805,
"rewards/accuracy_reward": 0.7403645873069763,
"rewards/brier_reward": 0.8531779527664185,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9925347328186035,
"rewards/mean_confidence_reward": 0.6609287977218627,
"sampling/batch_mean_priority_error": 0.04003342984423544,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16388888888888892,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010190530680119991,
"sampling/priority_kl": 0.02999986708164215,
"sampling/priority_scale": 0.7658457219367847,
"sampling/prob_entropy": 10.278948974609374,
"sampling/prob_max": 5.440502281999216e-05,
"sampling/prob_min": 2.023689194174949e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6151999950408935,
"sampling/prompt_draws_total": 48456.0,
"sampling/seen_fraction": 0.8818799972534179,
"sampling/unseen_fraction": 0.11812000274658203,
"signal/accuracy_reward/centered_abs_mean": 0.10931532233953475,
"signal/accuracy_reward/group_std_mean": 0.14885615557432175,
"signal/accuracy_reward/group_zero_std_frac": 0.55277778506279,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05465766116976738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05465766116976738,
"signal/advantage_abs_mean": 0.06904994621872902,
"signal/advantage_pre_scale_abs_mean": 0.06904994621872902,
"signal/advantage_pre_scale_std": 0.15002830922603608,
"signal/advantage_std": 0.15002830922603608,
"signal/brier_reward/centered_abs_mean": 0.06792017593979835,
"signal/brier_reward/group_std_mean": 0.09253233224153519,
"signal/brier_reward/group_zero_std_frac": 0.25,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033960087969899175,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033960087969899175,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.01263020820915699,
"signal/format_reward/group_std_mean": 0.02473071329295635,
"signal/format_reward/group_zero_std_frac": 0.8944444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.006315104104578495,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.006315104104578495,
"signal/mean_confidence_reward/centered_abs_mean": 0.05845487043261528,
"signal/mean_confidence_reward/group_std_mean": 0.07675684094429017,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2694444447755814,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.845486839461955e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.845486839461955e-07,
"step": 675
},
{
"calibration/aurc": 0.09314557605787317,
"calibration/batch_distribution_entropy": 0.6623083902603728,
"calibration/batch_entropy_100bins": 0.3666467877218496,
"calibration/batch_entropy_10bins": 0.6623083902603728,
"calibration/batch_entropy_50bins": 0.4316106662108378,
"calibration/batch_uniqueness": 0.49281141493055564,
"calibration/confidence_entropy": 0.4771051935748657,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5671875,
"calibration/coverage@15%": 0.8802083333333333,
"calibration/coverage@20%": 0.9526041666666666,
"calibration/coverage@25%": 0.9796875,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.31927083333333334,
"calibration/distribution_entropy_10": 0.6623083902603728,
"calibration/distribution_entropy_100": 0.3666467877218496,
"calibration/ece": 0.16416666666666666,
"calibration/mean_confidence": 0.68875,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.0073784722222222324,
"completions/max_length": 3977.4,
"completions/max_terminated_length": 3977.4,
"completions/mean_length": 1054.840185546875,
"completions/mean_terminated_length": 1062.661328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 238.8,
"epoch": 1.6346153846153846,
"grad_norm": 0.0005089513724669814,
"learning_rate": 2.1634615384615387e-06,
"loss": -0.0086,
"num_tokens": 1781786774.0,
"reward": 1.3072033882141114,
"reward_std": 0.09771093726158142,
"rewards/accuracy_reward": 0.7621527791023255,
"rewards/brier_reward": 0.8596189141273498,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9926215291023255,
"rewards/mean_confidence_reward": 0.6796440839767456,
"sampling/batch_mean_priority_error": 0.03946241023398036,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18611111111111112,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010225803218781948,
"sampling/priority_kl": 0.029999976232647895,
"sampling/priority_scale": 0.7673097312217578,
"sampling/prob_entropy": 10.278958702087403,
"sampling/prob_max": 5.459113890537992e-05,
"sampling/prob_min": 2.0267764557502233e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6272000074386597,
"sampling/prompt_draws_total": 48816.0,
"sampling/seen_fraction": 0.8839199900627136,
"sampling/unseen_fraction": 0.11608000993728637,
"signal/accuracy_reward/centered_abs_mean": 0.09992404282093048,
"signal/accuracy_reward/group_std_mean": 0.1337385058403015,
"signal/accuracy_reward/group_zero_std_frac": 0.6166666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04996202141046524,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04996202141046524,
"signal/advantage_abs_mean": 0.06958887130022048,
"signal/advantage_pre_scale_abs_mean": 0.06958887130022048,
"signal/advantage_pre_scale_std": 0.1529184877872467,
"signal/advantage_std": 0.1529184877872467,
"signal/brier_reward/centered_abs_mean": 0.06304295957088471,
"signal/brier_reward/group_std_mean": 0.08608659505844116,
"signal/brier_reward/group_zero_std_frac": 0.2888888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031521479785442355,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031521479785442355,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.012353515625,
"signal/format_reward/group_std_mean": 0.023333887942135333,
"signal/format_reward/group_zero_std_frac": 0.9055555582046508,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0061767578125,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0061767578125,
"signal/mean_confidence_reward/centered_abs_mean": 0.0535801962018013,
"signal/mean_confidence_reward/group_std_mean": 0.07002054750919343,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30277777910232545,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.358019279810832e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.358019279810832e-07,
"step": 680
},
{
"calibration/aurc": 0.1198504663022937,
"calibration/batch_distribution_entropy": 0.5066540096082259,
"calibration/batch_entropy_100bins": 0.2788550145350276,
"calibration/batch_entropy_10bins": 0.5066540096082259,
"calibration/batch_entropy_50bins": 0.32826361127430026,
"calibration/batch_uniqueness": 0.1365229156710747,
"calibration/confidence_entropy": 0.440284206232496,
"calibration/coverage@0%": 0.11614583333333332,
"calibration/coverage@1%": 0.12135416666666668,
"calibration/coverage@10%": 0.4124090608465608,
"calibration/coverage@15%": 0.7188620071684587,
"calibration/coverage@20%": 0.8839299048472435,
"calibration/coverage@25%": 0.9078125,
"calibration/coverage@30%": 0.94375,
"calibration/coverage@5%": 0.2625,
"calibration/distribution_entropy_10": 0.5066540096082259,
"calibration/distribution_entropy_100": 0.2788550145350276,
"calibration/ece": 0.10422848395630649,
"calibration/mean_confidence": 0.7424086341525858,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.00590277777777779,
"completions/max_length": 3630.6,
"completions/max_terminated_length": 3630.6,
"completions/mean_length": 1009.5178955078125,
"completions/mean_terminated_length": 1015.4686645507812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 224.0,
"epoch": 1.6466346153846154,
"grad_norm": 0.0005279097822494805,
"learning_rate": 2.1334134615384616e-06,
"loss": -0.0054,
"num_tokens": 1796476292.0,
"reward": 1.3186150312423706,
"reward_std": 0.10911149829626084,
"rewards/accuracy_reward": 0.7713541626930237,
"rewards/brier_reward": 0.8717638969421386,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9940972328186035,
"rewards/mean_confidence_reward": 0.7347743034362793,
"sampling/batch_mean_priority_error": 0.035970672211721816,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16111111111111112,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010261932387948036,
"sampling/priority_kl": 0.029999835789203642,
"sampling/priority_scale": 0.7691675721900537,
"sampling/prob_entropy": 10.278951835632324,
"sampling/prob_max": 5.479661704157479e-05,
"sampling/prob_min": 2.0295308058848604e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6392000198364258,
"sampling/prompt_draws_total": 49176.0,
"sampling/seen_fraction": 0.8860866785049438,
"sampling/unseen_fraction": 0.11391332149505615,
"signal/accuracy_reward/centered_abs_mean": 0.10825737714767455,
"signal/accuracy_reward/group_std_mean": 0.14899571537971495,
"signal/accuracy_reward/group_zero_std_frac": 0.55277778506279,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05412868857383728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05412868857383728,
"signal/advantage_abs_mean": 0.07697880044579505,
"signal/advantage_pre_scale_abs_mean": 0.07697880044579505,
"signal/advantage_pre_scale_std": 0.16402750015258788,
"signal/advantage_std": 0.16402750015258788,
"signal/brier_reward/centered_abs_mean": 0.06315912455320358,
"signal/brier_reward/group_std_mean": 0.08676011264324188,
"signal/brier_reward/group_zero_std_frac": 0.325,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03157956227660179,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03157956227660179,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.010112847201526165,
"signal/format_reward/group_std_mean": 0.020026170834898948,
"signal/format_reward/group_zero_std_frac": 0.9138889074325561,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0050564236007630825,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0050564236007630825,
"signal/mean_confidence_reward/centered_abs_mean": 0.056768130511045456,
"signal/mean_confidence_reward/group_std_mean": 0.07474597916007042,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3555555522441864,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.676812747879012e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.676812747879012e-07,
"step": 685
},
{
"calibration/aurc": 0.15312435512897443,
"calibration/batch_distribution_entropy": 0.5166774911978564,
"calibration/batch_entropy_100bins": 0.2830634113269603,
"calibration/batch_entropy_10bins": 0.5166774911978564,
"calibration/batch_entropy_50bins": 0.33321766788647383,
"calibration/batch_uniqueness": 0.13921276205743438,
"calibration/confidence_entropy": 0.4343256009444184,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1671875,
"calibration/coverage@10%": 0.49479173193536485,
"calibration/coverage@15%": 0.506932057895946,
"calibration/coverage@20%": 0.5360039544128602,
"calibration/coverage@25%": 0.5819628647214854,
"calibration/coverage@30%": 0.8712576153176675,
"calibration/coverage@5%": 0.4500052214958542,
"calibration/distribution_entropy_10": 0.5166774911978564,
"calibration/distribution_entropy_100": 0.2830634113269603,
"calibration/ece": 0.14271067610468618,
"calibration/mean_confidence": 0.7531193116142556,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.005642361111111094,
"completions/max_length": 3827.8,
"completions/max_terminated_length": 3827.8,
"completions/mean_length": 974.21796875,
"completions/mean_terminated_length": 979.8821655273438,
"completions/min_length": 0.0,
"completions/min_terminated_length": 238.0,
"epoch": 1.6586538461538463,
"grad_norm": 0.0005919505492784083,
"learning_rate": 2.103365384615385e-06,
"loss": -0.003,
"num_tokens": 1810795411.0,
"reward": 1.2960154294967652,
"reward_std": 0.10875263512134552,
"rewards/accuracy_reward": 0.7316840171813965,
"rewards/brier_reward": 0.865973949432373,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9943576574325561,
"rewards/mean_confidence_reward": 0.7592534780502319,
"sampling/batch_mean_priority_error": 0.046497556586661125,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.18055555555555555,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010300593636929988,
"sampling/priority_kl": 0.029999887570738792,
"sampling/priority_scale": 0.7711004435783252,
"sampling/prob_entropy": 10.278952980041504,
"sampling/prob_max": 5.5005505419103426e-05,
"sampling/prob_min": 2.032188313023653e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.651200008392334,
"sampling/prompt_draws_total": 49536.0,
"sampling/seen_fraction": 0.8882599949836731,
"sampling/unseen_fraction": 0.1117400050163269,
"signal/accuracy_reward/centered_abs_mean": 0.10711262971162797,
"signal/accuracy_reward/group_std_mean": 0.14350715726614,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05355631485581398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05355631485581398,
"signal/advantage_abs_mean": 0.07843385115265847,
"signal/advantage_pre_scale_abs_mean": 0.07843385115265847,
"signal/advantage_pre_scale_std": 0.16597414314746856,
"signal/advantage_std": 0.16597414314746856,
"signal/brier_reward/centered_abs_mean": 0.06248611286282539,
"signal/brier_reward/group_std_mean": 0.08682104349136352,
"signal/brier_reward/group_zero_std_frac": 0.3777777850627899,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031243056431412695,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031243056431412695,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00728624127805233,
"signal/format_reward/group_std_mean": 0.014419934712350368,
"signal/format_reward/group_zero_std_frac": 0.9388888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003643120639026165,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003643120639026165,
"signal/mean_confidence_reward/centered_abs_mean": 0.052331282198429106,
"signal/mean_confidence_reward/group_std_mean": 0.06940097361803055,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4111111104488373,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.233128092640982e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.233128092640982e-07,
"step": 690
},
{
"calibration/aurc": 0.1597328193560566,
"calibration/batch_distribution_entropy": 0.48849890794341916,
"calibration/batch_entropy_100bins": 0.2646663572293887,
"calibration/batch_entropy_10bins": 0.48849890794341916,
"calibration/batch_entropy_50bins": 0.31156095346465396,
"calibration/batch_uniqueness": 0.06236602230248702,
"calibration/confidence_entropy": 0.4200712640087535,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.2,
"calibration/coverage@15%": 0.3395833333333333,
"calibration/coverage@20%": 0.8581456701479546,
"calibration/coverage@25%": 0.954569190600522,
"calibration/coverage@30%": 0.9765013054830287,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.48849890794341916,
"calibration/distribution_entropy_100": 0.2646663572293887,
"calibration/ece": 0.11059222693646661,
"calibration/mean_confidence": 0.785867194299391,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.003472222222222232,
"completions/max_length": 3532.2,
"completions/max_terminated_length": 3532.2,
"completions/mean_length": 981.27109375,
"completions/mean_terminated_length": 984.7066528320313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 239.4,
"epoch": 1.6706730769230769,
"grad_norm": 0.000510729500092566,
"learning_rate": 2.073317307692308e-06,
"loss": -0.0034,
"num_tokens": 1825216070.0,
"reward": 1.3108274698257447,
"reward_std": 0.10949984639883041,
"rewards/accuracy_reward": 0.7611111164093017,
"rewards/brier_reward": 0.8640008687973022,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9965277910232544,
"rewards/mean_confidence_reward": 0.7624218702316284,
"sampling/batch_mean_priority_error": 0.04515728146589985,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.19444444444444448,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010345915518701076,
"sampling/priority_kl": 0.03000008873641491,
"sampling/priority_scale": 0.7729870260460302,
"sampling/prob_entropy": 10.278949356079101,
"sampling/prob_max": 5.521025886991993e-05,
"sampling/prob_min": 2.0347893951111473e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6631999969482423,
"sampling/prompt_draws_total": 49896.0,
"sampling/seen_fraction": 0.8903666734695435,
"sampling/unseen_fraction": 0.10963332653045654,
"signal/accuracy_reward/centered_abs_mean": 0.10617404580116271,
"signal/accuracy_reward/group_std_mean": 0.14624143689870833,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111223697662,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05308702290058136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05308702290058136,
"signal/advantage_abs_mean": 0.0775064080953598,
"signal/advantage_pre_scale_abs_mean": 0.0775064080953598,
"signal/advantage_pre_scale_std": 0.16172997355461122,
"signal/advantage_std": 0.16172997355461122,
"signal/brier_reward/centered_abs_mean": 0.06556255593895913,
"signal/brier_reward/group_std_mean": 0.09077477753162384,
"signal/brier_reward/group_zero_std_frac": 0.37500000596046446,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.032781277969479564,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.032781277969479564,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006076388945803046,
"signal/format_reward/group_std_mean": 0.011436474788933993,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003038194472901523,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003038194472901523,
"signal/mean_confidence_reward/centered_abs_mean": 0.04954320043325424,
"signal/mean_confidence_reward/group_std_mean": 0.06611797288060188,
"signal/mean_confidence_reward/group_zero_std_frac": 0.425,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.954319933858642e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.954319933858642e-07,
"step": 695
},
{
"calibration/aurc": 0.17845096277736253,
"calibration/batch_distribution_entropy": 0.507448634925201,
"calibration/batch_entropy_100bins": 0.27393716460227274,
"calibration/batch_entropy_10bins": 0.507448634925201,
"calibration/batch_entropy_50bins": 0.32247439790360655,
"calibration/batch_uniqueness": 0.1000203556561402,
"calibration/confidence_entropy": 0.4250286446102992,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1671875,
"calibration/coverage@10%": 0.1875,
"calibration/coverage@15%": 0.37232375979112275,
"calibration/coverage@20%": 0.5614093777197563,
"calibration/coverage@25%": 0.7432210073977371,
"calibration/coverage@30%": 0.9024490248226951,
"calibration/coverage@5%": 0.17291666666666666,
"calibration/distribution_entropy_10": 0.507448634925201,
"calibration/distribution_entropy_100": 0.27393716460227274,
"calibration/ece": 0.10796262186822965,
"calibration/mean_confidence": 0.769326342402274,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.00277777777777779,
"completions/max_length": 3776.8,
"completions/max_terminated_length": 3776.8,
"completions/mean_length": 1013.50947265625,
"completions/mean_terminated_length": 1016.4000854492188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 249.0,
"epoch": 1.6826923076923077,
"grad_norm": 0.0004915227764286101,
"learning_rate": 2.043269230769231e-06,
"loss": -0.001,
"num_tokens": 1840023155.0,
"reward": 1.2889822721481323,
"reward_std": 0.10611613094806671,
"rewards/accuracy_reward": 0.7234375,
"rewards/brier_reward": 0.8572899460792541,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9972222208976745,
"rewards/mean_confidence_reward": 0.748107624053955,
"sampling/batch_mean_priority_error": 0.04979654861793659,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.175,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01039465218782425,
"sampling/priority_kl": 0.029999758675694467,
"sampling/priority_scale": 0.77528671619948,
"sampling/prob_entropy": 10.278947257995606,
"sampling/prob_max": 5.543693623621948e-05,
"sampling/prob_min": 2.0364622105262243e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6751999855041504,
"sampling/prompt_draws_total": 50256.0,
"sampling/seen_fraction": 0.8926133275032043,
"sampling/unseen_fraction": 0.10738667249679565,
"signal/accuracy_reward/centered_abs_mean": 0.11032986044883727,
"signal/accuracy_reward/group_std_mean": 0.14479590356349945,
"signal/accuracy_reward/group_zero_std_frac": 0.5888888835906982,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05516493022441864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05516493022441864,
"signal/advantage_abs_mean": 0.07801882773637772,
"signal/advantage_pre_scale_abs_mean": 0.07801882773637772,
"signal/advantage_pre_scale_std": 0.1612319380044937,
"signal/advantage_std": 0.1612319380044937,
"signal/brier_reward/centered_abs_mean": 0.06465163081884384,
"signal/brier_reward/group_std_mean": 0.08930426687002183,
"signal/brier_reward/group_zero_std_frac": 0.3027777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03232581540942192,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03232581540942192,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005088975746184587,
"signal/format_reward/group_std_mean": 0.010972938034683466,
"signal/format_reward/group_zero_std_frac": 0.950000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025444878730922936,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0025444878730922936,
"signal/mean_confidence_reward/centered_abs_mean": 0.053876417130231856,
"signal/mean_confidence_reward/group_std_mean": 0.0717897079885006,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3305555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.387641522247577e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.387641522247577e-07,
"step": 700
},
{
"epoch": 1.6826923076923077,
"eval_calibration/aurc": 0.1281310126651566,
"eval_calibration/batch_distribution_entropy": 0.6122337191613557,
"eval_calibration/batch_entropy_100bins": 0.3384483209155802,
"eval_calibration/batch_entropy_10bins": 0.6122337191613557,
"eval_calibration/batch_entropy_50bins": 0.3984158873338679,
"eval_calibration/batch_uniqueness": 0.3394313964103316,
"eval_calibration/confidence_entropy": 0.45993146248134936,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.6172624237140366,
"eval_calibration/coverage@15%": 0.7314734088927637,
"eval_calibration/coverage@20%": 0.8134263295553618,
"eval_calibration/coverage@25%": 0.9145597210113339,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6122337191613557,
"eval_calibration/distribution_entropy_100": 0.3384483209155802,
"eval_calibration/ece": 0.027114210985178678,
"eval_calibration/mean_confidence": 0.7259808195292067,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.00434027777777779,
"eval_completions/max_length": 2841.1666666666665,
"eval_completions/max_terminated_length": 2841.1666666666665,
"eval_completions/mean_length": 965.6305847167969,
"eval_completions/mean_terminated_length": 969.8583780924479,
"eval_completions/min_length": 118.66666666666667,
"eval_completions/min_terminated_length": 306.3333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 1840023155.0,
"eval_reward": 1.2856452663739522,
"eval_reward_std": 0.3019138028224309,
"eval_rewards/accuracy_reward": 0.71875,
"eval_rewards/brier_reward": 0.8568663497765859,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9956597288449606,
"eval_rewards/mean_confidence_reward": 0.7228298286596934,
"eval_runtime": 187.0543,
"eval_samples_per_second": 5.346,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3912760466337204,
"eval_signal/accuracy_reward/group_std_mean": 0.447576567530632,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1956380233168602,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.1956380233168602,
"eval_signal/advantage_abs_mean": 0.25326357036828995,
"eval_signal/advantage_pre_scale_abs_mean": 0.25326357036828995,
"eval_signal/advantage_pre_scale_std": 0.30079903701941174,
"eval_signal/advantage_std": 0.30079903701941174,
"eval_signal/brier_reward/centered_abs_mean": 0.15465549876292548,
"eval_signal/brier_reward/group_std_mean": 0.2081133077541987,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07732774938146274,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07732774938146274,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.008409287935743729,
"eval_signal/format_reward/group_std_mean": 0.02455231888840596,
"eval_signal/format_reward/group_zero_std_frac": 0.8611111342906952,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.004204643967871864,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.004204643967871864,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.19646265357732773,
"eval_signal/mean_confidence_reward/group_std_mean": 0.22983714193105698,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9646265248714676e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9646265248714676e-06,
"eval_steps_per_second": 0.032,
"step": 700
},
{
"epoch": 1.6826923076923077,
"step": 700,
"train_probe_calibration/aurc": 0.09242685611624851,
"train_probe_calibration/batch_distribution_entropy": 0.5871877959379022,
"train_probe_calibration/batch_entropy_100bins": 0.3208042085735073,
"train_probe_calibration/batch_entropy_10bins": 0.5871877959379022,
"train_probe_calibration/batch_entropy_50bins": 0.3776455237641256,
"train_probe_calibration/batch_uniqueness": 0.28527984253508726,
"train_probe_calibration/confidence_entropy": 0.45388621485494296,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.6966783216783217,
"train_probe_calibration/coverage@15%": 0.8277972027972028,
"train_probe_calibration/coverage@20%": 0.9388111888111889,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5871877959379022,
"train_probe_calibration/distribution_entropy_100": 0.3208042085735073,
"train_probe_calibration/ece": 0.05061188811188807,
"train_probe_calibration/mean_confidence": 0.7408216783216783,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.006944444444444457,
"train_probe_completions/max_length": 3206.6666666666665,
"train_probe_completions/max_terminated_length": 3206.6666666666665,
"train_probe_completions/mean_length": 970.924326578776,
"train_probe_completions/mean_terminated_length": 977.7750854492188,
"train_probe_completions/min_length": 49.333333333333336,
"train_probe_completions/min_terminated_length": 250.33333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1840023155.0,
"train_probe_reward": 1.3135360479354858,
"train_probe_reward_std": 0.2896604041258494,
"train_probe_rewards/accuracy_reward": 0.7638888855775198,
"train_probe_rewards/brier_reward": 0.8701128363609314,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9930555621782938,
"train_probe_rewards/mean_confidence_reward": 0.7356770634651184,
"train_probe_runtime": 201.7941,
"train_probe_samples_per_second": 4.956,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3512369791666667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.42336805164813995,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.17561848958333334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.17561848958333334,
"train_probe_signal/advantage_abs_mean": 0.23088564723730087,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.23088564723730087,
"train_probe_signal/advantage_pre_scale_std": 0.2906094044446945,
"train_probe_signal/advantage_std": 0.2906094044446945,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14325847725073496,
"train_probe_signal/brier_reward/group_std_mean": 0.19320360322793326,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07162923862536748,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07162923862536748,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.013346354011446238,
"train_probe_signal/format_reward/group_std_mean": 0.0362943010404706,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8055555820465088,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.006673177005723119,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.006673177005723119,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.19063582768042883,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2232448955376943,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9063582499256881e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9063582499256881e-06,
"train_probe_steps_per_second": 0.03
},
{
"calibration/aurc": 0.08404616910375391,
"calibration/batch_distribution_entropy": 0.5988151606281764,
"calibration/batch_entropy_100bins": 0.3335366281759386,
"calibration/batch_entropy_10bins": 0.5988151606281764,
"calibration/batch_entropy_50bins": 0.39263392211128456,
"calibration/batch_uniqueness": 0.3433787299809475,
"calibration/confidence_entropy": 0.466011845904414,
"calibration/coverage@0%": 0.1418848167539267,
"calibration/coverage@1%": 0.36977537346023875,
"calibration/coverage@10%": 0.5878690372809116,
"calibration/coverage@15%": 0.803157923515746,
"calibration/coverage@20%": 0.9260925874896666,
"calibration/coverage@25%": 0.9564590796362633,
"calibration/coverage@30%": 0.9879581151832462,
"calibration/coverage@5%": 0.40542044899406937,
"calibration/distribution_entropy_10": 0.5988151606281764,
"calibration/distribution_entropy_100": 0.3335366281759386,
"calibration/ece": 0.11570963956581257,
"calibration/mean_confidence": 0.7057089920404749,
"calibration/unique_confidence_per_question": 0.021354166666666664,
"calibration/unique_confidences": 8.2,
"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": 0.00425347222222221,
"completions/max_length": 3435.0,
"completions/max_terminated_length": 3435.0,
"completions/mean_length": 980.7009643554687,
"completions/mean_terminated_length": 984.8880493164063,
"completions/min_length": 0.0,
"completions/min_terminated_length": 281.6,
"epoch": 1.6947115384615383,
"grad_norm": 0.00038545418647117913,
"learning_rate": 2.013221153846154e-06,
"loss": -0.0039,
"num_tokens": 1854408702.0,
"reward": 1.307462787628174,
"reward_std": 0.09673023074865342,
"rewards/accuracy_reward": 0.7457465171813965,
"rewards/brier_reward": 0.8734184384346009,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9957465410232544,
"rewards/mean_confidence_reward": 0.7106597065925598,
"sampling/batch_mean_priority_error": 0.032918609107163224,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.20555555555555555,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010429654642939568,
"sampling/priority_kl": 0.029999804869294166,
"sampling/priority_scale": 0.7776785313850268,
"sampling/prob_entropy": 10.278947067260741,
"sampling/prob_max": 5.5668204731773584e-05,
"sampling/prob_min": 2.038492457359098e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6871999979019165,
"sampling/prompt_draws_total": 50616.0,
"sampling/seen_fraction": 0.8948533415794373,
"sampling/unseen_fraction": 0.10514665842056274,
"signal/accuracy_reward/centered_abs_mean": 0.10552842915058136,
"signal/accuracy_reward/group_std_mean": 0.14385080635547637,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777969837189,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05276421457529068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05276421457529068,
"signal/advantage_abs_mean": 0.06741584911942482,
"signal/advantage_pre_scale_abs_mean": 0.06741584911942482,
"signal/advantage_pre_scale_std": 0.1482072502374649,
"signal/advantage_std": 0.1482072502374649,
"signal/brier_reward/centered_abs_mean": 0.05303824469447136,
"signal/brier_reward/group_std_mean": 0.07423444986343383,
"signal/brier_reward/group_zero_std_frac": 0.3277777791023254,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02651912234723568,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02651912234723568,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007796223927289248,
"signal/format_reward/group_std_mean": 0.016708648204803465,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003898111963644624,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003898111963644624,
"signal/mean_confidence_reward/centered_abs_mean": 0.04716092497110367,
"signal/mean_confidence_reward/group_std_mean": 0.06453502401709557,
"signal/mean_confidence_reward/group_zero_std_frac": 0.34166666865348816,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.716092234957614e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.716092234957614e-07,
"step": 705
},
{
"calibration/aurc": 0.056411707079725336,
"calibration/batch_distribution_entropy": 0.5792413277835646,
"calibration/batch_entropy_100bins": 0.31917351489767043,
"calibration/batch_entropy_10bins": 0.5792413277835646,
"calibration/batch_entropy_50bins": 0.3757258975533331,
"calibration/batch_uniqueness": 0.2996425228471575,
"calibration/confidence_entropy": 0.4706598405371064,
"calibration/coverage@0%": 0.26875,
"calibration/coverage@1%": 0.440625,
"calibration/coverage@10%": 0.8707297120418849,
"calibration/coverage@15%": 0.8801047120418849,
"calibration/coverage@20%": 0.9183246073298429,
"calibration/coverage@25%": 0.9183246073298429,
"calibration/coverage@30%": 0.9183246073298429,
"calibration/coverage@5%": 0.7516458333333333,
"calibration/distribution_entropy_10": 0.5792413277835646,
"calibration/distribution_entropy_100": 0.31917351489767043,
"calibration/ece": 0.14625667539266998,
"calibration/mean_confidence": 0.7274407286212916,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.004253472222222232,
"completions/max_length": 3389.8,
"completions/max_terminated_length": 3389.8,
"completions/mean_length": 966.7477416992188,
"completions/mean_terminated_length": 970.8892822265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 237.2,
"epoch": 1.7067307692307692,
"grad_norm": 0.0004834532446693629,
"learning_rate": 1.983173076923077e-06,
"loss": -0.0035,
"num_tokens": 1868619428.0,
"reward": 1.3241670846939086,
"reward_std": 0.0868344023823738,
"rewards/accuracy_reward": 0.7802951455116272,
"rewards/brier_reward": 0.87227863073349,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9957465171813965,
"rewards/mean_confidence_reward": 0.6925434112548828,
"sampling/batch_mean_priority_error": 0.030951319032566225,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17777777777777778,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010457353666424751,
"sampling/priority_kl": 0.030000027641654013,
"sampling/priority_scale": 0.7802383721107617,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 5.590729124378413e-05,
"sampling/prob_min": 2.0404893803060986e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.6991999864578247,
"sampling/prompt_draws_total": 50976.0,
"sampling/seen_fraction": 0.8971199989318848,
"sampling/unseen_fraction": 0.10288000106811523,
"signal/accuracy_reward/centered_abs_mean": 0.09852973222732545,
"signal/accuracy_reward/group_std_mean": 0.13320806324481965,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04926486611366272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04926486611366272,
"signal/advantage_abs_mean": 0.06283641681075096,
"signal/advantage_pre_scale_abs_mean": 0.06283641681075096,
"signal/advantage_pre_scale_std": 0.13592168539762498,
"signal/advantage_std": 0.13592168539762498,
"signal/brier_reward/centered_abs_mean": 0.05235719159245491,
"signal/brier_reward/group_std_mean": 0.0704245388507843,
"signal/brier_reward/group_zero_std_frac": 0.26388888955116274,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026178595796227454,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026178595796227454,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007145182462409139,
"signal/format_reward/group_std_mean": 0.012697589956223965,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0035725912312045693,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0035725912312045693,
"signal/mean_confidence_reward/centered_abs_mean": 0.05213705077767372,
"signal/mean_confidence_reward/group_std_mean": 0.06859569475054741,
"signal/mean_confidence_reward/group_zero_std_frac": 0.26666666865348815,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.2137048669465e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.2137048669465e-07,
"step": 710
},
{
"calibration/aurc": 0.10144029251164566,
"calibration/batch_distribution_entropy": 0.5868865138582311,
"calibration/batch_entropy_100bins": 0.3227534884668084,
"calibration/batch_entropy_10bins": 0.5868865138582311,
"calibration/batch_entropy_50bins": 0.3799401845154352,
"calibration/batch_uniqueness": 0.3160996147466604,
"calibration/confidence_entropy": 0.4645166674578388,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13125,
"calibration/coverage@10%": 0.5943580933682373,
"calibration/coverage@15%": 0.6554482984293194,
"calibration/coverage@20%": 0.7467822862129144,
"calibration/coverage@25%": 0.9282722513089006,
"calibration/coverage@30%": 0.9429319371727749,
"calibration/coverage@5%": 0.5207624345549738,
"calibration/distribution_entropy_10": 0.5868865138582311,
"calibration/distribution_entropy_100": 0.3227534884668084,
"calibration/ece": 0.14183191535776607,
"calibration/mean_confidence": 0.7273729275741712,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.003125,
"completions/max_length": 3701.4,
"completions/max_terminated_length": 3701.4,
"completions/mean_length": 996.6374145507813,
"completions/mean_terminated_length": 999.7630126953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 315.6,
"epoch": 1.71875,
"grad_norm": 0.0005059548420831561,
"learning_rate": 1.953125e-06,
"loss": -0.0017,
"num_tokens": 1883175795.0,
"reward": 1.3099539518356322,
"reward_std": 0.09733439832925797,
"rewards/accuracy_reward": 0.7560763835906983,
"rewards/brier_reward": 0.8669427156448364,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.996875011920929,
"rewards/mean_confidence_reward": 0.6936632037162781,
"sampling/batch_mean_priority_error": 0.03457258739857514,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17222222222222222,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01048535518348217,
"sampling/priority_kl": 0.030000003799796105,
"sampling/priority_scale": 0.7830242693657056,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 5.615855916403234e-05,
"sampling/prob_min": 2.0421350927790628e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7111999988555908,
"sampling/prompt_draws_total": 51336.0,
"sampling/seen_fraction": 0.8994266510009765,
"sampling/unseen_fraction": 0.10057334899902344,
"signal/accuracy_reward/centered_abs_mean": 0.10865885317325592,
"signal/accuracy_reward/group_std_mean": 0.14986452460289001,
"signal/accuracy_reward/group_zero_std_frac": 0.550000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05432942658662796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05432942658662796,
"signal/advantage_abs_mean": 0.06795400306582451,
"signal/advantage_pre_scale_abs_mean": 0.06795400306582451,
"signal/advantage_pre_scale_std": 0.14416785538196564,
"signal/advantage_std": 0.14416785538196564,
"signal/brier_reward/centered_abs_mean": 0.05352667644619942,
"signal/brier_reward/group_std_mean": 0.07452422827482223,
"signal/brier_reward/group_zero_std_frac": 0.2277777761220932,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02676333822309971,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02676333822309971,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005642361147329212,
"signal/format_reward/group_std_mean": 0.011255127005279064,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002821180573664606,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002821180573664606,
"signal/mean_confidence_reward/centered_abs_mean": 0.04913303405046463,
"signal/mean_confidence_reward/group_std_mean": 0.06512986496090889,
"signal/mean_confidence_reward/group_zero_std_frac": 0.23888888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.913303087050735e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.913303087050735e-07,
"step": 715
},
{
"calibration/aurc": 0.11201574671393137,
"calibration/batch_distribution_entropy": 0.6315005836708136,
"calibration/batch_entropy_100bins": 0.35334694917426485,
"calibration/batch_entropy_10bins": 0.6315005836708136,
"calibration/batch_entropy_50bins": 0.41595431146220596,
"calibration/batch_uniqueness": 0.4524251302083334,
"calibration/confidence_entropy": 0.4942181505118133,
"calibration/coverage@0%": 0.0859375,
"calibration/coverage@1%": 0.196875,
"calibration/coverage@10%": 0.47916666666666663,
"calibration/coverage@15%": 0.71875,
"calibration/coverage@20%": 0.7713541666666667,
"calibration/coverage@25%": 0.8442708333333334,
"calibration/coverage@30%": 0.9651041666666668,
"calibration/coverage@5%": 0.33437500000000003,
"calibration/distribution_entropy_10": 0.6315005836708136,
"calibration/distribution_entropy_100": 0.35334694917426485,
"calibration/ece": 0.11661458333333326,
"calibration/mean_confidence": 0.6916145833333335,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.002951388888888862,
"completions/max_length": 3609.6,
"completions/max_terminated_length": 3609.6,
"completions/mean_length": 992.2913208007812,
"completions/mean_terminated_length": 995.2805786132812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 267.0,
"epoch": 1.7307692307692308,
"grad_norm": 0.0004260011774022132,
"learning_rate": 1.9230769230769234e-06,
"loss": -0.0018,
"num_tokens": 1897715887.0,
"reward": 1.3017566204071045,
"reward_std": 0.095692677795887,
"rewards/accuracy_reward": 0.7407986044883728,
"rewards/brier_reward": 0.8656519293785095,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9970486044883728,
"rewards/mean_confidence_reward": 0.705946171283722,
"sampling/batch_mean_priority_error": 0.03343948360511164,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16666666666666669,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010514728166162968,
"sampling/priority_kl": 0.029999824613332747,
"sampling/priority_scale": 0.7848560034064576,
"sampling/prob_entropy": 10.27894992828369,
"sampling/prob_max": 5.6361448514508083e-05,
"sampling/prob_min": 2.0444238180061804e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.723200011253357,
"sampling/prompt_draws_total": 51696.0,
"sampling/seen_fraction": 0.9012800097465515,
"sampling/unseen_fraction": 0.09871999025344849,
"signal/accuracy_reward/centered_abs_mean": 0.11094835102558136,
"signal/accuracy_reward/group_std_mean": 0.14612414538860322,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05547417551279068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05547417551279068,
"signal/advantage_abs_mean": 0.06998142004013061,
"signal/advantage_pre_scale_abs_mean": 0.06998142004013061,
"signal/advantage_pre_scale_std": 0.1458520144224167,
"signal/advantage_std": 0.1458520144224167,
"signal/brier_reward/centered_abs_mean": 0.05215005427598953,
"signal/brier_reward/group_std_mean": 0.07294831350445748,
"signal/brier_reward/group_zero_std_frac": 0.272222226858139,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026075027137994766,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026075027137994766,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004926215298473835,
"signal/format_reward/group_std_mean": 0.01056935265660286,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024631076492369177,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0024631076492369177,
"signal/mean_confidence_reward/centered_abs_mean": 0.04923503249883652,
"signal/mean_confidence_reward/group_std_mean": 0.06458318158984185,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2805555611848831,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.923503070131118e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.923503070131118e-07,
"step": 720
},
{
"calibration/aurc": 0.06508023949155624,
"calibration/batch_distribution_entropy": 0.5558616334714455,
"calibration/batch_entropy_100bins": 0.3077710333755689,
"calibration/batch_entropy_10bins": 0.5558616334714455,
"calibration/batch_entropy_50bins": 0.3623030807961205,
"calibration/batch_uniqueness": 0.27050792753929254,
"calibration/confidence_entropy": 0.4731002437303237,
"calibration/coverage@0%": 0.2578125,
"calibration/coverage@1%": 0.2838541666666667,
"calibration/coverage@10%": 0.6793594160104987,
"calibration/coverage@15%": 0.8896448490813649,
"calibration/coverage@20%": 0.9484375,
"calibration/coverage@25%": 0.9901041666666666,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5234375,
"calibration/distribution_entropy_10": 0.5558616334714455,
"calibration/distribution_entropy_100": 0.3077710333755689,
"calibration/ece": 0.1668618766404199,
"calibration/mean_confidence": 0.7121259842519685,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.002256944444444442,
"completions/max_length": 3377.8,
"completions/max_terminated_length": 3377.8,
"completions/mean_length": 1015.0033813476563,
"completions/mean_terminated_length": 1017.2540283203125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 301.0,
"epoch": 1.7427884615384617,
"grad_norm": 0.00045864039566367865,
"learning_rate": 1.8930288461538463e-06,
"loss": -0.0012,
"num_tokens": 1912523734.0,
"reward": 1.330811357498169,
"reward_std": 0.08866008371114731,
"rewards/accuracy_reward": 0.7939236044883728,
"rewards/brier_reward": 0.8700286626815796,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99765625,
"rewards/mean_confidence_reward": 0.7115191102027894,
"sampling/batch_mean_priority_error": 0.03423864334196256,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.15555555555555553,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01054015289992094,
"sampling/priority_kl": 0.030000202357769012,
"sampling/priority_scale": 0.787105053733103,
"sampling/prob_entropy": 10.278961372375488,
"sampling/prob_max": 5.6586982100270686e-05,
"sampling/prob_min": 2.0466489513637497e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.735199999809265,
"sampling/prompt_draws_total": 52056.0,
"sampling/seen_fraction": 0.9032733321189881,
"sampling/unseen_fraction": 0.09672666788101196,
"signal/accuracy_reward/centered_abs_mean": 0.09814453125,
"signal/accuracy_reward/group_std_mean": 0.13914075791835784,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.049072265625,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.049072265625,
"signal/advantage_abs_mean": 0.06016180217266083,
"signal/advantage_pre_scale_abs_mean": 0.06016180217266083,
"signal/advantage_pre_scale_std": 0.13105083703994752,
"signal/advantage_std": 0.13105083703994752,
"signal/brier_reward/centered_abs_mean": 0.05176843777298927,
"signal/brier_reward/group_std_mean": 0.0720445230603218,
"signal/brier_reward/group_zero_std_frac": 0.2833333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.025884218886494635,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.025884218886494635,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004258897516410798,
"signal/format_reward/group_std_mean": 0.009134303964674472,
"signal/format_reward/group_zero_std_frac": 0.9583333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002129448758205399,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002129448758205399,
"signal/mean_confidence_reward/centered_abs_mean": 0.04849230349063873,
"signal/mean_confidence_reward/group_std_mean": 0.06421271786093712,
"signal/mean_confidence_reward/group_zero_std_frac": 0.2916666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.849230208492372e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.849230208492372e-07,
"step": 725
},
{
"calibration/aurc": 0.13226417419465478,
"calibration/batch_distribution_entropy": 0.556962817758184,
"calibration/batch_entropy_100bins": 0.3047709072838795,
"calibration/batch_entropy_10bins": 0.556962817758184,
"calibration/batch_entropy_50bins": 0.35877138090261723,
"calibration/batch_uniqueness": 0.2528834304460806,
"calibration/confidence_entropy": 0.46571681160407047,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.6160899461925753,
"calibration/coverage@15%": 0.7093813230942497,
"calibration/coverage@20%": 0.7668666416350868,
"calibration/coverage@25%": 0.8565204872604044,
"calibration/coverage@30%": 0.9248680918189731,
"calibration/coverage@5%": 0.19345354656222802,
"calibration/distribution_entropy_10": 0.556962817758184,
"calibration/distribution_entropy_100": 0.3047709072838795,
"calibration/ece": 0.10036003188724493,
"calibration/mean_confidence": 0.7258346391964574,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.001996527777777768,
"completions/max_length": 3585.6,
"completions/max_terminated_length": 3585.6,
"completions/mean_length": 1024.8122680664062,
"completions/mean_terminated_length": 1026.8658447265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 247.4,
"epoch": 1.7548076923076923,
"grad_norm": 0.000421989883761853,
"learning_rate": 1.8629807692307695e-06,
"loss": -0.0009,
"num_tokens": 1927461891.0,
"reward": 1.310717487335205,
"reward_std": 0.10098863393068314,
"rewards/accuracy_reward": 0.7563368201255798,
"rewards/brier_reward": 0.8670798420906067,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9980034708976746,
"rewards/mean_confidence_reward": 0.7394097208976745,
"sampling/batch_mean_priority_error": 0.03910652136909175,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17222222222222222,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010574315674602985,
"sampling/priority_kl": 0.02999994680285454,
"sampling/priority_scale": 0.7895619689719752,
"sampling/prob_entropy": 10.278949356079101,
"sampling/prob_max": 5.68221221328713e-05,
"sampling/prob_min": 2.0486395442276262e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7472000122070312,
"sampling/prompt_draws_total": 52416.0,
"sampling/seen_fraction": 0.9052866697311401,
"sampling/unseen_fraction": 0.09471333026885986,
"signal/accuracy_reward/centered_abs_mean": 0.10135091096162796,
"signal/accuracy_reward/group_std_mean": 0.1404233992099762,
"signal/accuracy_reward/group_zero_std_frac": 0.575000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05067545548081398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05067545548081398,
"signal/advantage_abs_mean": 0.07134843468666077,
"signal/advantage_pre_scale_abs_mean": 0.07134843468666077,
"signal/advantage_pre_scale_std": 0.15252363979816436,
"signal/advantage_std": 0.15252363979816436,
"signal/brier_reward/centered_abs_mean": 0.05917214304208755,
"signal/brier_reward/group_std_mean": 0.08254666775465011,
"signal/brier_reward/group_zero_std_frac": 0.2888888895511627,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029586071521043776,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029586071521043776,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.003759765566792339,
"signal/format_reward/group_std_mean": 0.009313787892460823,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0018798827833961696,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0018798827833961696,
"signal/mean_confidence_reward/centered_abs_mean": 0.04908203333616257,
"signal/mean_confidence_reward/group_std_mean": 0.06572242081165314,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3083333343267441,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.908203493414476e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.908203493414476e-07,
"step": 730
},
{
"calibration/aurc": 0.16262073369685778,
"calibration/batch_distribution_entropy": 0.38419717906836637,
"calibration/batch_entropy_100bins": 0.20703974933536848,
"calibration/batch_entropy_10bins": 0.38419717906836637,
"calibration/batch_entropy_50bins": 0.2437238430425174,
"calibration/batch_uniqueness": -0.13750190241657567,
"calibration/confidence_entropy": 0.4000716698967516,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.15104166666666669,
"calibration/coverage@10%": 0.3713541666666667,
"calibration/coverage@15%": 0.5270833333333333,
"calibration/coverage@20%": 0.5270833333333333,
"calibration/coverage@25%": 0.7703125,
"calibration/coverage@30%": 0.9191103677110531,
"calibration/coverage@5%": 0.3458333333333333,
"calibration/distribution_entropy_10": 0.38419717906836637,
"calibration/distribution_entropy_100": 0.20703974933536848,
"calibration/ece": 0.13378821257615323,
"calibration/mean_confidence": 0.8116920963881638,
"calibration/unique_confidence_per_question": 0.017708333333333333,
"calibration/unique_confidences": 6.8,
"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": 0.0021701388888888838,
"completions/max_length": 3682.4,
"completions/max_terminated_length": 3682.4,
"completions/mean_length": 980.3084350585938,
"completions/mean_terminated_length": 982.4520629882812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 243.6,
"epoch": 1.7668269230769231,
"grad_norm": 0.0006200639181770384,
"learning_rate": 1.8329326923076924e-06,
"loss": -0.0005,
"num_tokens": 1941838276.0,
"reward": 1.3009915828704834,
"reward_std": 0.10750979483127594,
"rewards/accuracy_reward": 0.7427083373069763,
"rewards/brier_reward": 0.8618637204170227,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9973958253860473,
"rewards/mean_confidence_reward": 0.769019079208374,
"sampling/batch_mean_priority_error": 0.05118026096658691,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010620946437120438,
"sampling/priority_kl": 0.030000010132789613,
"sampling/priority_scale": 0.791469556116499,
"sampling/prob_entropy": 10.278951644897461,
"sampling/prob_max": 5.703187925973907e-05,
"sampling/prob_min": 2.051140763796866e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7592000246047974,
"sampling/prompt_draws_total": 52776.0,
"sampling/seen_fraction": 0.9070466756820679,
"sampling/unseen_fraction": 0.09295332431793213,
"signal/accuracy_reward/centered_abs_mean": 0.11150173842906952,
"signal/accuracy_reward/group_std_mean": 0.14786103665828704,
"signal/accuracy_reward/group_zero_std_frac": 0.5722222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05575086921453476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05575086921453476,
"signal/advantage_abs_mean": 0.07811176404356956,
"signal/advantage_pre_scale_abs_mean": 0.07811176404356956,
"signal/advantage_pre_scale_std": 0.1644335836172104,
"signal/advantage_std": 0.1644335836172104,
"signal/brier_reward/centered_abs_mean": 0.062073732167482375,
"signal/brier_reward/group_std_mean": 0.08578675836324692,
"signal/brier_reward/group_zero_std_frac": 0.3694444537162781,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031036866083741187,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031036866083741187,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.0048936631763353946,
"signal/format_reward/group_std_mean": 0.011555349640548229,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024468315881676973,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0024468315881676973,
"signal/mean_confidence_reward/centered_abs_mean": 0.047903656214475634,
"signal/mean_confidence_reward/group_std_mean": 0.06507696658372879,
"signal/mean_confidence_reward/group_zero_std_frac": 0.40277777910232543,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.790365494500292e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.790365494500292e-07,
"step": 735
},
{
"calibration/aurc": 0.0984268474633593,
"calibration/batch_distribution_entropy": 0.37509171035558264,
"calibration/batch_entropy_100bins": 0.20162115419693408,
"calibration/batch_entropy_10bins": 0.37509171035558264,
"calibration/batch_entropy_50bins": 0.23734516051768728,
"calibration/batch_uniqueness": -0.18610338479294536,
"calibration/confidence_entropy": 0.40682152374552005,
"calibration/coverage@0%": 0.18385416666666668,
"calibration/coverage@1%": 0.2,
"calibration/coverage@10%": 0.5096891318537858,
"calibration/coverage@15%": 0.8533126631853787,
"calibration/coverage@20%": 0.900262456483899,
"calibration/coverage@25%": 0.9075718015665796,
"calibration/coverage@30%": 0.9556135770234988,
"calibration/coverage@5%": 0.47422758920800695,
"calibration/distribution_entropy_10": 0.37509171035558264,
"calibration/distribution_entropy_100": 0.20162115419693408,
"calibration/ece": 0.11171507832898168,
"calibration/mean_confidence": 0.7978674390774588,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.00234375,
"completions/max_length": 3268.0,
"completions/max_terminated_length": 3268.0,
"completions/mean_length": 974.5008911132812,
"completions/mean_terminated_length": 976.8037719726562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 246.2,
"epoch": 1.7788461538461537,
"grad_norm": 0.0006327030714601278,
"learning_rate": 1.8028846153846156e-06,
"loss": -0.0003,
"num_tokens": 1956136878.0,
"reward": 1.286976194381714,
"reward_std": 0.11908978521823883,
"rewards/accuracy_reward": 0.7282986164093017,
"rewards/brier_reward": 0.8480685830116272,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.997569453716278,
"rewards/mean_confidence_reward": 0.7898784637451172,
"sampling/batch_mean_priority_error": 0.061700169462697484,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16111111111111112,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010678479447960854,
"sampling/priority_kl": 0.03000004291534424,
"sampling/priority_scale": 0.7937021076912061,
"sampling/prob_entropy": 10.278962516784668,
"sampling/prob_max": 5.7258721790276466e-05,
"sampling/prob_min": 2.0533668430289253e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7711999893188477,
"sampling/prompt_draws_total": 53136.0,
"sampling/seen_fraction": 0.908893346786499,
"sampling/unseen_fraction": 0.09110665321350098,
"signal/accuracy_reward/centered_abs_mean": 0.10706380307674408,
"signal/accuracy_reward/group_std_mean": 0.1492036297917366,
"signal/accuracy_reward/group_zero_std_frac": 0.5527777910232544,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05353190153837204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05353190153837204,
"signal/advantage_abs_mean": 0.0839640349149704,
"signal/advantage_pre_scale_abs_mean": 0.0839640349149704,
"signal/advantage_pre_scale_std": 0.1742776095867157,
"signal/advantage_std": 0.1742776095867157,
"signal/brier_reward/centered_abs_mean": 0.07154948264360428,
"signal/brier_reward/group_std_mean": 0.1000126764178276,
"signal/brier_reward/group_zero_std_frac": 0.38611111640930174,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03577474132180214,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03577474132180214,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004459635436069221,
"signal/format_reward/group_std_mean": 0.0099914675578475,
"signal/format_reward/group_zero_std_frac": 0.9527777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0022298177180346103,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0022298177180346103,
"signal/mean_confidence_reward/centered_abs_mean": 0.04505589380860329,
"signal/mean_confidence_reward/group_std_mean": 0.05993512943387032,
"signal/mean_confidence_reward/group_zero_std_frac": 0.43611112236976624,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.505589231484919e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.505589231484919e-07,
"step": 740
},
{
"calibration/aurc": 0.05148233310008995,
"calibration/batch_distribution_entropy": 0.3488501157020723,
"calibration/batch_entropy_100bins": 0.1841267741588303,
"calibration/batch_entropy_10bins": 0.3488501157020723,
"calibration/batch_entropy_50bins": 0.2167510593935289,
"calibration/batch_uniqueness": -0.2663174130925411,
"calibration/confidence_entropy": 0.3887985954627028,
"calibration/coverage@0%": 0.1712041884816754,
"calibration/coverage@1%": 0.1806282722513089,
"calibration/coverage@10%": 0.8939155664687857,
"calibration/coverage@15%": 0.9435818569553807,
"calibration/coverage@20%": 0.9755208333333334,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6470718503937007,
"calibration/distribution_entropy_10": 0.3488501157020723,
"calibration/distribution_entropy_100": 0.1841267741588303,
"calibration/ece": 0.08471214065355696,
"calibration/mean_confidence": 0.8317367237635873,
"calibration/unique_confidence_per_question": 0.01927083333333333,
"calibration/unique_confidences": 7.4,
"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": 0.0035590277777777903,
"completions/max_length": 3580.2,
"completions/max_terminated_length": 3580.2,
"completions/mean_length": 981.0261352539062,
"completions/mean_terminated_length": 984.6888549804687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 272.6,
"epoch": 1.7908653846153846,
"grad_norm": 0.0005542809376493096,
"learning_rate": 1.7728365384615387e-06,
"loss": -0.0019,
"num_tokens": 1970542395.0,
"reward": 1.3384529113769532,
"reward_std": 0.09226970225572587,
"rewards/accuracy_reward": 0.7962673544883728,
"rewards/brier_reward": 0.8842682242393494,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541746139526,
"rewards/mean_confidence_reward": 0.8027343511581421,
"sampling/batch_mean_priority_error": 0.0450249363335027,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.175,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01073062587529421,
"sampling/priority_kl": 0.030000034347176553,
"sampling/priority_scale": 0.7963388859992847,
"sampling/prob_entropy": 10.278950309753418,
"sampling/prob_max": 5.750622949562967e-05,
"sampling/prob_min": 2.0552280693664217e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.7831999778747558,
"sampling/prompt_draws_total": 53496.0,
"sampling/seen_fraction": 0.9108466625213623,
"sampling/unseen_fraction": 0.0891533374786377,
"signal/accuracy_reward/centered_abs_mean": 0.08584526777267457,
"signal/accuracy_reward/group_std_mean": 0.11938865035772324,
"signal/accuracy_reward/group_zero_std_frac": 0.6388889074325561,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04292263388633728,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04292263388633728,
"signal/advantage_abs_mean": 0.06494638547301293,
"signal/advantage_pre_scale_abs_mean": 0.06494638547301293,
"signal/advantage_pre_scale_std": 0.15016201436519622,
"signal/advantage_std": 0.15016201436519622,
"signal/brier_reward/centered_abs_mean": 0.05143366307020188,
"signal/brier_reward/group_std_mean": 0.07305691987276078,
"signal/brier_reward/group_zero_std_frac": 0.4916666805744171,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02571683153510094,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02571683153510094,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006011284794658423,
"signal/format_reward/group_std_mean": 0.010725349932909013,
"signal/format_reward/group_zero_std_frac": 0.9583333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0030056423973292114,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0030056423973292114,
"signal/mean_confidence_reward/centered_abs_mean": 0.04237794876098633,
"signal/mean_confidence_reward/group_std_mean": 0.05708230137825012,
"signal/mean_confidence_reward/group_zero_std_frac": 0.5194444537162781,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.2377944851068607e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.2377944851068607e-07,
"step": 745
},
{
"calibration/aurc": 0.12513098641408765,
"calibration/batch_distribution_entropy": 0.4386680736185493,
"calibration/batch_entropy_100bins": 0.23583337034985075,
"calibration/batch_entropy_10bins": 0.4386680736185493,
"calibration/batch_entropy_50bins": 0.2776192278238811,
"calibration/batch_uniqueness": -0.0454765110770183,
"calibration/confidence_entropy": 0.42076816317473975,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5476555700609226,
"calibration/coverage@15%": 0.7570740861618799,
"calibration/coverage@20%": 0.8138571583986074,
"calibration/coverage@25%": 0.8925655461270671,
"calibration/coverage@30%": 0.8925655461270671,
"calibration/coverage@5%": 0.19166666666666668,
"calibration/distribution_entropy_10": 0.4386680736185493,
"calibration/distribution_entropy_100": 0.23583337034985075,
"calibration/ece": 0.09808841927763264,
"calibration/mean_confidence": 0.7692097748041777,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0019097222222222321,
"completions/max_length": 3707.2,
"completions/max_terminated_length": 3707.2,
"completions/mean_length": 1036.6534790039063,
"completions/mean_terminated_length": 1038.6652465820312,
"completions/min_length": 0.0,
"completions/min_terminated_length": 335.2,
"epoch": 1.8028846153846154,
"grad_norm": 0.0005948901525698602,
"learning_rate": 1.7427884615384616e-06,
"loss": -0.001,
"num_tokens": 1985591043.0,
"reward": 1.3137980937957763,
"reward_std": 0.10690182149410248,
"rewards/accuracy_reward": 0.753125,
"rewards/brier_reward": 0.8765390992164612,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166626930237,
"rewards/mean_confidence_reward": 0.7790190696716308,
"sampling/batch_mean_priority_error": 0.04252574058302948,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.17777777777777776,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010769122280180454,
"sampling/priority_kl": 0.02999994307756424,
"sampling/priority_scale": 0.7995206534629687,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 5.777966289315373e-05,
"sampling/prob_min": 2.0565289742080494e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.795199990272522,
"sampling/prompt_draws_total": 53856.0,
"sampling/seen_fraction": 0.9129800081253052,
"sampling/unseen_fraction": 0.08701999187469482,
"signal/accuracy_reward/centered_abs_mean": 0.10002170205116272,
"signal/accuracy_reward/group_std_mean": 0.13846919387578965,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05001085102558136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05001085102558136,
"signal/advantage_abs_mean": 0.07599502354860306,
"signal/advantage_pre_scale_abs_mean": 0.07599502354860306,
"signal/advantage_pre_scale_std": 0.16599199771881104,
"signal/advantage_std": 0.16599199771881104,
"signal/brier_reward/centered_abs_mean": 0.06395492181181908,
"signal/brier_reward/group_std_mean": 0.08838129937648773,
"signal/brier_reward/group_zero_std_frac": 0.4138888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03197746090590954,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03197746090590954,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.003971354104578495,
"signal/format_reward/group_std_mean": 0.010237774625420571,
"signal/format_reward/group_zero_std_frac": 0.9472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019856770522892475,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019856770522892475,
"signal/mean_confidence_reward/centered_abs_mean": 0.045269110798835756,
"signal/mean_confidence_reward/group_std_mean": 0.06089982837438583,
"signal/mean_confidence_reward/group_zero_std_frac": 0.45000000596046447,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.5269109705259325e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.5269109705259325e-07,
"step": 750
},
{
"epoch": 1.8028846153846154,
"eval_calibration/aurc": 0.13990344378862302,
"eval_calibration/batch_distribution_entropy": 0.49701023927115834,
"eval_calibration/batch_entropy_100bins": 0.2746262363397858,
"eval_calibration/batch_entropy_10bins": 0.49701023927115834,
"eval_calibration/batch_entropy_50bins": 0.3232855620039196,
"eval_calibration/batch_uniqueness": 0.08372214988425933,
"eval_calibration/confidence_entropy": 0.42793329476976205,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.7326388888888888,
"eval_calibration/coverage@20%": 0.8489583333333334,
"eval_calibration/coverage@25%": 0.9348958333333334,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.49701023927115834,
"eval_calibration/distribution_entropy_100": 0.2746262363397858,
"eval_calibration/ece": 0.03541666666666681,
"eval_calibration/mean_confidence": 0.763888888888889,
"eval_calibration/unique_confidence_per_question": 0.006944444444444444,
"eval_calibration/unique_confidences": 8,
"eval_completions/clipped_ratio": 0.0,
"eval_completions/max_length": 2543.8333333333335,
"eval_completions/max_terminated_length": 2543.8333333333335,
"eval_completions/mean_length": 1003.161122639974,
"eval_completions/mean_terminated_length": 1003.161122639974,
"eval_completions/min_length": 340.5,
"eval_completions/min_terminated_length": 340.5,
"eval_loss": 0.0,
"eval_num_tokens": 1985591043.0,
"eval_reward": 1.2930545409520466,
"eval_reward_std": 0.30748553077379864,
"eval_rewards/accuracy_reward": 0.7300347089767456,
"eval_rewards/brier_reward": 0.8560590147972107,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 1.0,
"eval_rewards/mean_confidence_reward": 0.7638888756434122,
"eval_runtime": 149.0266,
"eval_samples_per_second": 6.71,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3828667551279068,
"eval_signal/accuracy_reward/group_std_mean": 0.4424740672111511,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1914333775639534,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.1914333775639534,
"eval_signal/advantage_abs_mean": 0.2585411916176478,
"eval_signal/advantage_pre_scale_abs_mean": 0.2585411916176478,
"eval_signal/advantage_pre_scale_std": 0.3062230596939723,
"eval_signal/advantage_std": 0.3062230596939723,
"eval_signal/brier_reward/centered_abs_mean": 0.16724719355503717,
"eval_signal/brier_reward/group_std_mean": 0.2248398462931315,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08362359677751859,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08362359677751859,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0,
"eval_signal/format_reward/group_std_mean": 0.0,
"eval_signal/format_reward/group_zero_std_frac": 1.0,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.1763671636581421,
"eval_signal/mean_confidence_reward/group_std_mean": 0.21374296893676123,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.7636715862560475e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.7636715862560475e-06,
"eval_steps_per_second": 0.04,
"step": 750
},
{
"epoch": 1.8028846153846154,
"step": 750,
"train_probe_calibration/aurc": 0.09669643522423507,
"train_probe_calibration/batch_distribution_entropy": 0.4769103370021153,
"train_probe_calibration/batch_entropy_100bins": 0.2612872498607003,
"train_probe_calibration/batch_entropy_10bins": 0.4769103370021153,
"train_probe_calibration/batch_entropy_50bins": 0.30758312294373313,
"train_probe_calibration/batch_uniqueness": 0.033334749723803854,
"train_probe_calibration/confidence_entropy": 0.4227911882444281,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7543554006968641,
"train_probe_calibration/coverage@15%": 0.8658536585365854,
"train_probe_calibration/coverage@20%": 0.9538327526132404,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.4769103370021153,
"train_probe_calibration/distribution_entropy_100": 0.2612872498607003,
"train_probe_calibration/ece": 0.03911149825783945,
"train_probe_calibration/mean_confidence": 0.7772648083623694,
"train_probe_calibration/unique_confidence_per_question": 0.006944444444444444,
"train_probe_calibration/unique_confidences": 8,
"train_probe_completions/clipped_ratio": 0.005902777777777794,
"train_probe_completions/max_length": 2893.0,
"train_probe_completions/max_terminated_length": 2893.0,
"train_probe_completions/mean_length": 1008.5927327473959,
"train_probe_completions/mean_terminated_length": 1014.651611328125,
"train_probe_completions/min_length": 121.33333333333333,
"train_probe_completions/min_terminated_length": 291.1666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 1985591043.0,
"train_probe_reward": 1.3298992117245991,
"train_probe_reward_std": 0.2911212046941121,
"train_probe_rewards/accuracy_reward": 0.7821180621782938,
"train_probe_rewards/brier_reward": 0.8811371823151907,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.996527781089147,
"train_probe_rewards/mean_confidence_reward": 0.7745659252007803,
"train_probe_runtime": 179.8887,
"train_probe_samples_per_second": 5.559,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3321940104166667,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4113796005646388,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16609700520833334,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16609700520833334,
"train_probe_signal/advantage_abs_mean": 0.22670255601406097,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.22670255601406097,
"train_probe_signal/advantage_pre_scale_std": 0.28947464128335315,
"train_probe_signal/advantage_std": 0.28947464128335315,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14373591418067613,
"train_probe_signal/brier_reward/group_std_mean": 0.20871607214212418,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07186795709033807,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07186795709033807,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/group_std_mean": 0.019641855110724766,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888889054457346,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.16852210462093353,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.20680046826601028,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.6852209986003193e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.6852209986003193e-06,
"train_probe_steps_per_second": 0.033
},
{
"calibration/aurc": 0.09349917406678779,
"calibration/batch_distribution_entropy": 0.44226068472105906,
"calibration/batch_entropy_100bins": 0.24420624657005888,
"calibration/batch_entropy_10bins": 0.44226068472105906,
"calibration/batch_entropy_50bins": 0.2874756422383079,
"calibration/batch_uniqueness": -0.006146544850521329,
"calibration/confidence_entropy": 0.43103626804361667,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5820917645778938,
"calibration/coverage@15%": 0.7831674281984334,
"calibration/coverage@20%": 0.9045882288946909,
"calibration/coverage@25%": 0.9618798955613578,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.35444816144473457,
"calibration/distribution_entropy_10": 0.44226068472105906,
"calibration/distribution_entropy_100": 0.24420624657005888,
"calibration/ece": 0.09561221714534365,
"calibration/mean_confidence": 0.7642822563098348,
"calibration/unique_confidence_per_question": 0.020312499999999997,
"calibration/unique_confidences": 7.8,
"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": 0.001649305555555558,
"completions/max_length": 3528.4,
"completions/max_terminated_length": 3528.4,
"completions/mean_length": 991.83125,
"completions/mean_terminated_length": 993.4691162109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 271.6,
"epoch": 1.8149038461538463,
"grad_norm": 0.000718863622751087,
"learning_rate": 1.7127403846153848e-06,
"loss": -0.0004,
"num_tokens": 2000126795.0,
"reward": 1.3270102977752685,
"reward_std": 0.09794525802135468,
"rewards/accuracy_reward": 0.7756944537162781,
"rewards/brier_reward": 0.8801336884498596,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9981770753860474,
"rewards/mean_confidence_reward": 0.7755208492279053,
"sampling/batch_mean_priority_error": 0.04309707667727515,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010810686834156513,
"sampling/priority_kl": 0.029999897256493567,
"sampling/priority_scale": 0.8023889839882031,
"sampling/prob_entropy": 10.278952598571777,
"sampling/prob_max": 5.8038592396769675e-05,
"sampling/prob_min": 2.058104109892156e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8072000026702881,
"sampling/prompt_draws_total": 54216.0,
"sampling/seen_fraction": 0.9149400115013122,
"sampling/unseen_fraction": 0.08505998849868775,
"signal/accuracy_reward/centered_abs_mean": 0.09638671875,
"signal/accuracy_reward/group_std_mean": 0.12863395363092422,
"signal/accuracy_reward/group_zero_std_frac": 0.6277777791023255,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.048193359375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.048193359375,
"signal/advantage_abs_mean": 0.07153508141636848,
"signal/advantage_pre_scale_abs_mean": 0.07153508141636848,
"signal/advantage_pre_scale_std": 0.15592266619205475,
"signal/advantage_std": 0.15592266619205475,
"signal/brier_reward/centered_abs_mean": 0.06087609231472015,
"signal/brier_reward/group_std_mean": 0.08291507065296173,
"signal/brier_reward/group_zero_std_frac": 0.42777777910232545,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030438046157360076,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030438046157360076,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0034559461520984767,
"signal/format_reward/group_std_mean": 0.008465695334598422,
"signal/format_reward/group_zero_std_frac": 0.9583333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0017279730760492384,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0017279730760492384,
"signal/mean_confidence_reward/centered_abs_mean": 0.046497957408428194,
"signal/mean_confidence_reward/group_std_mean": 0.06168990060687065,
"signal/mean_confidence_reward/group_zero_std_frac": 0.475,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.649795300792903e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.649795300792903e-07,
"step": 755
},
{
"calibration/aurc": 0.14914645317799086,
"calibration/batch_distribution_entropy": 0.4790925837613993,
"calibration/batch_entropy_100bins": 0.26074205573889475,
"calibration/batch_entropy_10bins": 0.4790925837613993,
"calibration/batch_entropy_50bins": 0.30694132924471074,
"calibration/batch_uniqueness": 0.06379253474589286,
"calibration/confidence_entropy": 0.4314014961182906,
"calibration/coverage@0%": 0.17604166666666668,
"calibration/coverage@1%": 0.19479166666666667,
"calibration/coverage@10%": 0.3557291666666667,
"calibration/coverage@15%": 0.5234375,
"calibration/coverage@20%": 0.5401041666666667,
"calibration/coverage@25%": 0.8331320713664055,
"calibration/coverage@30%": 0.8889809073107051,
"calibration/coverage@5%": 0.3557291666666667,
"calibration/distribution_entropy_10": 0.4790925837613993,
"calibration/distribution_entropy_100": 0.26074205573889475,
"calibration/ece": 0.1447995539599652,
"calibration/mean_confidence": 0.7540081048738035,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.002430555555555558,
"completions/max_length": 3893.0,
"completions/max_terminated_length": 3893.0,
"completions/mean_length": 1057.9886352539063,
"completions/mean_terminated_length": 1060.56220703125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 274.0,
"epoch": 1.8269230769230769,
"grad_norm": 0.0006375846569426358,
"learning_rate": 1.682692307692308e-06,
"loss": -0.0004,
"num_tokens": 2015390120.0,
"reward": 1.3015799283981324,
"reward_std": 0.10563021302223205,
"rewards/accuracy_reward": 0.73125,
"rewards/brier_reward": 0.8744123339653015,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9974826455116272,
"rewards/mean_confidence_reward": 0.742907989025116,
"sampling/batch_mean_priority_error": 0.04129415914572554,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010846528597176075,
"sampling/priority_kl": 0.02999998964369297,
"sampling/priority_scale": 0.8044263661140576,
"sampling/prob_entropy": 10.278950119018555,
"sampling/prob_max": 5.825789921800606e-05,
"sampling/prob_min": 2.0604497694876046e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8191999912261962,
"sampling/prompt_draws_total": 54576.0,
"sampling/seen_fraction": 0.9165599942207336,
"sampling/unseen_fraction": 0.08344000577926636,
"signal/accuracy_reward/centered_abs_mean": 0.1109917551279068,
"signal/accuracy_reward/group_std_mean": 0.14689186215400696,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0554958775639534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0554958775639534,
"signal/advantage_abs_mean": 0.07737808376550674,
"signal/advantage_pre_scale_abs_mean": 0.07737808376550674,
"signal/advantage_pre_scale_std": 0.15911754965782166,
"signal/advantage_std": 0.15911754965782166,
"signal/brier_reward/centered_abs_mean": 0.06404606327414512,
"signal/brier_reward/group_std_mean": 0.08760401755571365,
"signal/brier_reward/group_zero_std_frac": 0.36111111640930177,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03202303163707256,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03202303163707256,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004616970429196954,
"signal/format_reward/group_std_mean": 0.00960054025053978,
"signal/format_reward/group_zero_std_frac": 0.9583333373069763,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002308485214598477,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002308485214598477,
"signal/mean_confidence_reward/centered_abs_mean": 0.05342720374464989,
"signal/mean_confidence_reward/group_std_mean": 0.0703733816742897,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.342720271528379e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.342720271528379e-07,
"step": 760
},
{
"calibration/aurc": 0.13740616837098713,
"calibration/batch_distribution_entropy": 0.5073045708548226,
"calibration/batch_entropy_100bins": 0.27755860925825987,
"calibration/batch_entropy_10bins": 0.5073045708548226,
"calibration/batch_entropy_50bins": 0.32673750395815093,
"calibration/batch_uniqueness": 0.11712409118522035,
"calibration/confidence_entropy": 0.437041303338047,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5658283694220788,
"calibration/coverage@15%": 0.5992897448954018,
"calibration/coverage@20%": 0.6863547761199131,
"calibration/coverage@25%": 0.7145042539267015,
"calibration/coverage@30%": 0.9077553851174935,
"calibration/coverage@5%": 0.3893215532286213,
"calibration/distribution_entropy_10": 0.5073045708548226,
"calibration/distribution_entropy_100": 0.27755860925825987,
"calibration/ece": 0.10657638926861046,
"calibration/mean_confidence": 0.733388931702505,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.005642361111111094,
"completions/max_length": 3775.6,
"completions/max_terminated_length": 3775.6,
"completions/mean_length": 1080.2004516601562,
"completions/mean_terminated_length": 1086.2259155273437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 301.2,
"epoch": 1.8389423076923077,
"grad_norm": 0.00046238666982389987,
"learning_rate": 1.6526442307692309e-06,
"loss": -0.0029,
"num_tokens": 2030915565.0,
"reward": 1.3130412101745605,
"reward_std": 0.09514122605323791,
"rewards/accuracy_reward": 0.7511284708976745,
"rewards/brier_reward": 0.8805816173553467,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9943576455116272,
"rewards/mean_confidence_reward": 0.7333853960037231,
"sampling/batch_mean_priority_error": 0.03588200018134492,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.175,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010885423980653286,
"sampling/priority_kl": 0.030000252276659013,
"sampling/priority_scale": 0.8076436937553808,
"sampling/prob_entropy": 10.278957939147949,
"sampling/prob_max": 5.85400739510078e-05,
"sampling/prob_min": 2.0618402777472512e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8312000036239624,
"sampling/prompt_draws_total": 54936.0,
"sampling/seen_fraction": 0.9185333371162414,
"sampling/unseen_fraction": 0.08146666288375855,
"signal/accuracy_reward/centered_abs_mean": 0.08905707374215126,
"signal/accuracy_reward/group_std_mean": 0.1289079800248146,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666805744171,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04452853687107563,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04452853687107563,
"signal/advantage_abs_mean": 0.06434951797127723,
"signal/advantage_pre_scale_abs_mean": 0.06434951797127723,
"signal/advantage_pre_scale_std": 0.14531429409980773,
"signal/advantage_std": 0.14531429409980773,
"signal/brier_reward/centered_abs_mean": 0.05801150500774384,
"signal/brier_reward/group_std_mean": 0.08372148722410203,
"signal/brier_reward/group_zero_std_frac": 0.34166666865348816,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02900575250387192,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02900575250387192,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007872178684920073,
"signal/format_reward/group_std_mean": 0.01580634731799364,
"signal/format_reward/group_zero_std_frac": 0.9277778029441833,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003936089342460036,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003936089342460036,
"signal/mean_confidence_reward/centered_abs_mean": 0.05199273899197578,
"signal/mean_confidence_reward/group_std_mean": 0.06956614181399345,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3694444537162781,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.199273573452956e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.199273573452956e-07,
"step": 765
},
{
"calibration/aurc": 0.1058074407278425,
"calibration/batch_distribution_entropy": 0.5508122068033574,
"calibration/batch_entropy_100bins": 0.30632343612808477,
"calibration/batch_entropy_10bins": 0.5508122068033574,
"calibration/batch_entropy_50bins": 0.36059899273830937,
"calibration/batch_uniqueness": 0.24586654214986564,
"calibration/confidence_entropy": 0.4377355294125821,
"calibration/coverage@0%": 0.11830238726790451,
"calibration/coverage@1%": 0.2645164864846147,
"calibration/coverage@10%": 0.49349614293690847,
"calibration/coverage@15%": 0.6703091203709031,
"calibration/coverage@20%": 0.8773928715829582,
"calibration/coverage@25%": 0.9046154478731934,
"calibration/coverage@30%": 0.9287624295004899,
"calibration/coverage@5%": 0.4709064704031346,
"calibration/distribution_entropy_10": 0.5508122068033574,
"calibration/distribution_entropy_100": 0.30632343612808477,
"calibration/ece": 0.124238298273329,
"calibration/mean_confidence": 0.7130441251754973,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.003645833333333326,
"completions/max_length": 3689.8,
"completions/max_terminated_length": 3689.8,
"completions/mean_length": 1063.927783203125,
"completions/mean_terminated_length": 1067.8088623046874,
"completions/min_length": 0.0,
"completions/min_terminated_length": 272.4,
"epoch": 1.8509615384615383,
"grad_norm": 0.0005264647770673037,
"learning_rate": 1.622596153846154e-06,
"loss": -0.0032,
"num_tokens": 2046281389.0,
"reward": 1.2954176902770995,
"reward_std": 0.10103552639484406,
"rewards/accuracy_reward": 0.7413194298744201,
"rewards/brier_reward": 0.8531475901603699,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541626930237,
"rewards/mean_confidence_reward": 0.7126562595367432,
"sampling/batch_mean_priority_error": 0.05059813466010541,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.14444444444444446,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010923280008137226,
"sampling/priority_kl": 0.02999989278614521,
"sampling/priority_scale": 0.8107421695487573,
"sampling/prob_entropy": 10.278948402404785,
"sampling/prob_max": 5.8815087686525655e-05,
"sampling/prob_min": 2.0632501764339396e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8432000160217286,
"sampling/prompt_draws_total": 55296.0,
"sampling/seen_fraction": 0.9204333186149597,
"sampling/unseen_fraction": 0.07956668138504028,
"signal/accuracy_reward/centered_abs_mean": 0.10087890475988388,
"signal/accuracy_reward/group_std_mean": 0.14286032617092131,
"signal/accuracy_reward/group_zero_std_frac": 0.5611111164093018,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05043945237994194,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05043945237994194,
"signal/advantage_abs_mean": 0.06797776594758034,
"signal/advantage_pre_scale_abs_mean": 0.06797776594758034,
"signal/advantage_pre_scale_std": 0.15065972805023192,
"signal/advantage_std": 0.15065972805023192,
"signal/brier_reward/centered_abs_mean": 0.06256764978170395,
"signal/brier_reward/group_std_mean": 0.08869975209236144,
"signal/brier_reward/group_zero_std_frac": 0.3194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03128382489085198,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03128382489085198,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006759982544463128,
"signal/format_reward/group_std_mean": 0.015655810199677943,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003379991272231564,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003379991272231564,
"signal/mean_confidence_reward/centered_abs_mean": 0.049361446499824525,
"signal/mean_confidence_reward/group_std_mean": 0.06687793955206871,
"signal/mean_confidence_reward/group_zero_std_frac": 0.35833333134651185,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.936144591738412e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.936144591738412e-07,
"step": 770
},
{
"calibration/aurc": 0.10568492932644938,
"calibration/batch_distribution_entropy": 0.58030989898059,
"calibration/batch_entropy_100bins": 0.3294897567753045,
"calibration/batch_entropy_10bins": 0.58030989898059,
"calibration/batch_entropy_50bins": 0.3878700105762888,
"calibration/batch_uniqueness": 0.33344796337384425,
"calibration/confidence_entropy": 0.4726254435927121,
"calibration/coverage@0%": 0.10837696335078535,
"calibration/coverage@1%": 0.1778963436381015,
"calibration/coverage@10%": 0.5566325895366886,
"calibration/coverage@15%": 0.7742783533429363,
"calibration/coverage@20%": 0.8180596376602803,
"calibration/coverage@25%": 0.8840599455040872,
"calibration/coverage@30%": 0.934741144414169,
"calibration/coverage@5%": 0.4609998417910961,
"calibration/distribution_entropy_10": 0.58030989898059,
"calibration/distribution_entropy_100": 0.3294897567753045,
"calibration/ece": 0.11717980862289623,
"calibration/mean_confidence": 0.7064151465581333,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.005208333333333348,
"completions/max_length": 3900.6,
"completions/max_terminated_length": 3900.6,
"completions/mean_length": 1092.501904296875,
"completions/mean_terminated_length": 1098.26162109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 316.2,
"epoch": 1.8629807692307692,
"grad_norm": 0.00046158998156897724,
"learning_rate": 1.592548076923077e-06,
"loss": -0.0033,
"num_tokens": 2061994467.0,
"reward": 1.3138395309448243,
"reward_std": 0.10530554354190827,
"rewards/accuracy_reward": 0.7564236044883728,
"rewards/brier_reward": 0.8764496445655823,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9947916507720947,
"rewards/mean_confidence_reward": 0.7057465314865112,
"sampling/batch_mean_priority_error": 0.029372394873767048,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.14166666666666666,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010955540835857392,
"sampling/priority_kl": 0.030000030994415283,
"sampling/priority_scale": 0.8133450329536572,
"sampling/prob_entropy": 10.278959846496582,
"sampling/prob_max": 5.906694204895757e-05,
"sampling/prob_min": 2.065128392132465e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8552000045776367,
"sampling/prompt_draws_total": 55656.0,
"sampling/seen_fraction": 0.9221133351325989,
"sampling/unseen_fraction": 0.07788666486740112,
"signal/accuracy_reward/centered_abs_mean": 0.1110568568110466,
"signal/accuracy_reward/group_std_mean": 0.15447987318038942,
"signal/accuracy_reward/group_zero_std_frac": 0.5305555701255799,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0555284284055233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0555284284055233,
"signal/advantage_abs_mean": 0.07142583802342414,
"signal/advantage_pre_scale_abs_mean": 0.07142583802342414,
"signal/advantage_pre_scale_std": 0.1522584453225136,
"signal/advantage_std": 0.1522584453225136,
"signal/brier_reward/centered_abs_mean": 0.05967013090848923,
"signal/brier_reward/group_std_mean": 0.08542276471853257,
"signal/brier_reward/group_zero_std_frac": 0.2611111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029835065454244615,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029835065454244615,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.008289930713362991,
"signal/format_reward/group_std_mean": 0.017449943348765373,
"signal/format_reward/group_zero_std_frac": 0.919444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0041449653566814956,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0041449653566814956,
"signal/mean_confidence_reward/centered_abs_mean": 0.052318794280290605,
"signal/mean_confidence_reward/group_std_mean": 0.07218771576881408,
"signal/mean_confidence_reward/group_zero_std_frac": 0.27222222089767456,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.231879185885191e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.231879185885191e-07,
"step": 775
},
{
"calibration/aurc": 0.12650340119612366,
"calibration/batch_distribution_entropy": 0.6069324227226676,
"calibration/batch_entropy_100bins": 0.3344824618210083,
"calibration/batch_entropy_10bins": 0.6069324227226676,
"calibration/batch_entropy_50bins": 0.3937473421748004,
"calibration/batch_uniqueness": 0.3424567241129418,
"calibration/confidence_entropy": 0.463840319597579,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5450485483621087,
"calibration/coverage@15%": 0.7180795302311593,
"calibration/coverage@20%": 0.7525376109546658,
"calibration/coverage@25%": 0.8271407543664193,
"calibration/coverage@30%": 0.8626050167685081,
"calibration/coverage@5%": 0.28555501112964154,
"calibration/distribution_entropy_10": 0.6069324227226676,
"calibration/distribution_entropy_100": 0.3344824618210083,
"calibration/ece": 0.1031593465351158,
"calibration/mean_confidence": 0.7074604731122898,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.002430555555555558,
"completions/max_length": 3569.0,
"completions/max_terminated_length": 3569.0,
"completions/mean_length": 1031.5218994140625,
"completions/mean_terminated_length": 1034.0815063476562,
"completions/min_length": 0.0,
"completions/min_terminated_length": 288.8,
"epoch": 1.875,
"grad_norm": 0.0005521697457879782,
"learning_rate": 1.5625e-06,
"loss": -0.0004,
"num_tokens": 2076971551.0,
"reward": 1.3163609266281129,
"reward_std": 0.09743509739637375,
"rewards/accuracy_reward": 0.7552951455116272,
"rewards/brier_reward": 0.8798429131507873,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9975694417953491,
"rewards/mean_confidence_reward": 0.7181510210037232,
"sampling/batch_mean_priority_error": 0.033419626889365825,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11944444444444444,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.010983368568122386,
"sampling/priority_kl": 0.03000000976026058,
"sampling/priority_scale": 0.8154753029579297,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 5.929811304667964e-05,
"sampling/prob_min": 2.067512359644752e-05,
"sampling/prompt_draws_max": 7.0,
"sampling/prompt_draws_mean": 1.8672000169754028,
"sampling/prompt_draws_total": 56016.0,
"sampling/seen_fraction": 0.9235666632652283,
"sampling/unseen_fraction": 0.07643333673477173,
"signal/accuracy_reward/centered_abs_mean": 0.10843641459941863,
"signal/accuracy_reward/group_std_mean": 0.14697835743427276,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05421820729970932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05421820729970932,
"signal/advantage_abs_mean": 0.06979118883609772,
"signal/advantage_pre_scale_abs_mean": 0.06979118883609772,
"signal/advantage_pre_scale_std": 0.14512248635292052,
"signal/advantage_std": 0.14512248635292052,
"signal/brier_reward/centered_abs_mean": 0.054565370082855225,
"signal/brier_reward/group_std_mean": 0.07615428417921066,
"signal/brier_reward/group_zero_std_frac": 0.3194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027282685041427612,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027282685041427612,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0044596354942768816,
"signal/format_reward/group_std_mean": 0.009759350214153528,
"signal/format_reward/group_zero_std_frac": 0.9555555582046509,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0022298177471384408,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0022298177471384408,
"signal/mean_confidence_reward/centered_abs_mean": 0.05360189527273178,
"signal/mean_confidence_reward/group_std_mean": 0.07146198004484176,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32222222089767455,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.360189334169263e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.360189334169263e-07,
"step": 780
},
{
"calibration/aurc": 0.08575687912745038,
"calibration/batch_distribution_entropy": 0.6169860530156981,
"calibration/batch_entropy_100bins": 0.34476735909655454,
"calibration/batch_entropy_10bins": 0.6169860530156981,
"calibration/batch_entropy_50bins": 0.40585455683932975,
"calibration/batch_uniqueness": 0.4083399131337545,
"calibration/confidence_entropy": 0.477630784825506,
"calibration/coverage@0%": 0.238339044821584,
"calibration/coverage@1%": 0.238339044821584,
"calibration/coverage@10%": 0.7210201090192845,
"calibration/coverage@15%": 0.7512832669140214,
"calibration/coverage@20%": 0.8411735605331867,
"calibration/coverage@25%": 0.85213961797444,
"calibration/coverage@30%": 0.9281544592551876,
"calibration/coverage@5%": 0.6155446669873117,
"calibration/distribution_entropy_10": 0.6169860530156981,
"calibration/distribution_entropy_100": 0.34476735909655454,
"calibration/ece": 0.13663992573633807,
"calibration/mean_confidence": 0.6788759719550181,
"calibration/unique_confidence_per_question": 0.021354166666666664,
"calibration/unique_confidences": 8.2,
"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": 0.003211805555555558,
"completions/max_length": 3779.6,
"completions/max_terminated_length": 3779.6,
"completions/mean_length": 1086.3251708984376,
"completions/mean_terminated_length": 1089.8551513671875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 324.8,
"epoch": 1.8870192307692308,
"grad_norm": 0.0005370154976844788,
"learning_rate": 1.5324519230769232e-06,
"loss": -0.0026,
"num_tokens": 2092606657.0,
"reward": 1.2937704801559449,
"reward_std": 0.09794444888830185,
"rewards/accuracy_reward": 0.7217881798744201,
"rewards/brier_reward": 0.8689505338668824,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967881917953492,
"rewards/mean_confidence_reward": 0.7033767580986023,
"sampling/batch_mean_priority_error": 0.0366663303187036,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11388888888888889,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011012093536555767,
"sampling/priority_kl": 0.029999838769435884,
"sampling/priority_scale": 0.8172821938758716,
"sampling/prob_entropy": 10.278954315185548,
"sampling/prob_max": 5.9512741427170115e-05,
"sampling/prob_min": 2.0624500029953198e-05,
"sampling/prompt_draws_max": 7.2,
"sampling/prompt_draws_mean": 1.8791999816894531,
"sampling/prompt_draws_total": 56376.0,
"sampling/seen_fraction": 0.9249000072479248,
"sampling/unseen_fraction": 0.0750999927520752,
"signal/accuracy_reward/centered_abs_mean": 0.10648328810930252,
"signal/accuracy_reward/group_std_mean": 0.14075976759195327,
"signal/accuracy_reward/group_zero_std_frac": 0.5916666626930237,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05324164405465126,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05324164405465126,
"signal/advantage_abs_mean": 0.07044520229101181,
"signal/advantage_pre_scale_abs_mean": 0.07044520229101181,
"signal/advantage_pre_scale_std": 0.14834425449371338,
"signal/advantage_std": 0.14834425449371338,
"signal/brier_reward/centered_abs_mean": 0.06095951944589615,
"signal/brier_reward/group_std_mean": 0.08446675837039948,
"signal/brier_reward/group_zero_std_frac": 0.2888888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030479759722948075,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030479759722948075,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00603841149713844,
"signal/format_reward/group_std_mean": 0.014342158287763595,
"signal/format_reward/group_zero_std_frac": 0.9305555701255799,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00301920574856922,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00301920574856922,
"signal/mean_confidence_reward/centered_abs_mean": 0.055245233327150346,
"signal/mean_confidence_reward/group_std_mean": 0.07242218255996705,
"signal/mean_confidence_reward/group_zero_std_frac": 0.30833333134651186,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.524522975974832e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.524522975974832e-07,
"step": 785
},
{
"calibration/aurc": 0.09134358211133084,
"calibration/batch_distribution_entropy": 0.5184122528279234,
"calibration/batch_entropy_100bins": 0.28326808197942577,
"calibration/batch_entropy_10bins": 0.5184122528279234,
"calibration/batch_entropy_50bins": 0.3334586028670128,
"calibration/batch_uniqueness": 0.1333611226039658,
"calibration/confidence_entropy": 0.43366741381902807,
"calibration/coverage@0%": 0.13507853403141362,
"calibration/coverage@1%": 0.39809936736474694,
"calibration/coverage@10%": 0.5081451788830715,
"calibration/coverage@15%": 0.5822916666666667,
"calibration/coverage@20%": 0.9163585296684118,
"calibration/coverage@25%": 0.9555219240837696,
"calibration/coverage@30%": 0.9910994764397906,
"calibration/coverage@5%": 0.49872109511343804,
"calibration/distribution_entropy_10": 0.5184122528279234,
"calibration/distribution_entropy_100": 0.28326808197942577,
"calibration/ece": 0.13128326788830713,
"calibration/mean_confidence": 0.750606457242583,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003819444444444442,
"completions/max_length": 3623.6,
"completions/max_terminated_length": 3623.6,
"completions/mean_length": 1019.901220703125,
"completions/mean_terminated_length": 1023.8256713867188,
"completions/min_length": 0.0,
"completions/min_terminated_length": 314.8,
"epoch": 1.8990384615384617,
"grad_norm": 0.0005847892607562244,
"learning_rate": 1.5024038461538462e-06,
"loss": -0.0026,
"num_tokens": 2107456751.0,
"reward": 1.3047247171401977,
"reward_std": 0.09398773610591889,
"rewards/accuracy_reward": 0.7427083373069763,
"rewards/brier_reward": 0.8705459713935852,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9961805462837219,
"rewards/mean_confidence_reward": 0.726623260974884,
"sampling/batch_mean_priority_error": 0.03973864163030571,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.16666666666666669,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011049842275679112,
"sampling/priority_kl": 0.030000048875808715,
"sampling/priority_scale": 0.8210367739433423,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 5.9824989875778556e-05,
"sampling/prob_min": 2.0322882119216956e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.8911999940872193,
"sampling/prompt_draws_total": 56736.0,
"sampling/seen_fraction": 0.926853334903717,
"sampling/unseen_fraction": 0.07314666509628295,
"signal/accuracy_reward/centered_abs_mean": 0.09969618171453476,
"signal/accuracy_reward/group_std_mean": 0.1376244455575943,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04984809085726738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04984809085726738,
"signal/advantage_abs_mean": 0.06447783783078194,
"signal/advantage_pre_scale_abs_mean": 0.06447783783078194,
"signal/advantage_pre_scale_std": 0.1427443265914917,
"signal/advantage_std": 0.1427443265914917,
"signal/brier_reward/centered_abs_mean": 0.052842124551534655,
"signal/brier_reward/group_std_mean": 0.07576082348823547,
"signal/brier_reward/group_zero_std_frac": 0.36944444179534913,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026421062275767328,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026421062275767328,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006987847248092293,
"signal/format_reward/group_std_mean": 0.015508862398564815,
"signal/format_reward/group_zero_std_frac": 0.9277777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0034939236240461467,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0034939236240461467,
"signal/mean_confidence_reward/centered_abs_mean": 0.04846192449331284,
"signal/mean_confidence_reward/group_std_mean": 0.064261145144701,
"signal/mean_confidence_reward/group_zero_std_frac": 0.39166667461395266,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.846192496188451e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.846192496188451e-07,
"step": 790
},
{
"calibration/aurc": 0.11700290971783592,
"calibration/batch_distribution_entropy": 0.3694812329975735,
"calibration/batch_entropy_100bins": 0.19917089474001431,
"calibration/batch_entropy_10bins": 0.3694812329975735,
"calibration/batch_entropy_50bins": 0.2344607547298671,
"calibration/batch_uniqueness": -0.22014402630499075,
"calibration/confidence_entropy": 0.39032257527187286,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1828125,
"calibration/coverage@10%": 0.5609375,
"calibration/coverage@15%": 0.7394718233246301,
"calibration/coverage@20%": 0.7785900783289816,
"calibration/coverage@25%": 0.7885117493472584,
"calibration/coverage@30%": 0.9650130548302872,
"calibration/coverage@5%": 0.5229166666666666,
"calibration/distribution_entropy_10": 0.3694812329975735,
"calibration/distribution_entropy_100": 0.19917089474001431,
"calibration/ece": 0.11351827676240207,
"calibration/mean_confidence": 0.8136020452567451,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0012152777777777902,
"completions/max_length": 3386.6,
"completions/max_terminated_length": 3386.6,
"completions/mean_length": 1010.8636352539063,
"completions/mean_terminated_length": 1012.0790161132812,
"completions/min_length": 0.0,
"completions/min_terminated_length": 291.4,
"epoch": 1.9110576923076923,
"grad_norm": 0.0005554448580369353,
"learning_rate": 1.4723557692307693e-06,
"loss": -0.0008,
"num_tokens": 2122168588.0,
"reward": 1.321057105064392,
"reward_std": 0.09689114838838578,
"rewards/accuracy_reward": 0.7663194417953492,
"rewards/brier_reward": 0.8771684169769287,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9986111164093018,
"rewards/mean_confidence_reward": 0.7633159756660461,
"sampling/batch_mean_priority_error": 0.04098942885678015,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011086820624768734,
"sampling/priority_kl": 0.029999766498804092,
"sampling/priority_scale": 0.8243461310630664,
"sampling/prob_entropy": 10.278958702087403,
"sampling/prob_max": 6.011749355820939e-05,
"sampling/prob_min": 2.0333567590569146e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9031999826431274,
"sampling/prompt_draws_total": 57096.0,
"sampling/seen_fraction": 0.9286133170127868,
"sampling/unseen_fraction": 0.07138668298721314,
"signal/accuracy_reward/centered_abs_mean": 0.0936957448720932,
"signal/accuracy_reward/group_std_mean": 0.1331271708011627,
"signal/accuracy_reward/group_zero_std_frac": 0.580555546283722,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0468478724360466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0468478724360466,
"signal/advantage_abs_mean": 0.06651908457279206,
"signal/advantage_pre_scale_abs_mean": 0.06651908457279206,
"signal/advantage_pre_scale_std": 0.14965884685516356,
"signal/advantage_std": 0.14965884685516356,
"signal/brier_reward/centered_abs_mean": 0.05872168764472008,
"signal/brier_reward/group_std_mean": 0.08186099231243134,
"signal/brier_reward/group_zero_std_frac": 0.4083333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02936084382236004,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02936084382236004,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0026692707906477153,
"signal/format_reward/group_std_mean": 0.007258860021829605,
"signal/format_reward/group_zero_std_frac": 0.9611111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0013346353953238577,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0013346353953238577,
"signal/mean_confidence_reward/centered_abs_mean": 0.043116336315870286,
"signal/mean_confidence_reward/group_std_mean": 0.057510948926210406,
"signal/mean_confidence_reward/group_zero_std_frac": 0.45555556416511533,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.311633404086024e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.311633404086024e-07,
"step": 795
},
{
"calibration/aurc": 0.0531468315919163,
"calibration/batch_distribution_entropy": 0.39870880529696723,
"calibration/batch_entropy_100bins": 0.2124436541205837,
"calibration/batch_entropy_10bins": 0.39870880529696723,
"calibration/batch_entropy_50bins": 0.2500852323212258,
"calibration/batch_uniqueness": -0.1439639447652643,
"calibration/confidence_entropy": 0.4015295033087928,
"calibration/coverage@0%": 0.13263707571801567,
"calibration/coverage@1%": 0.32384954308093994,
"calibration/coverage@10%": 0.6923566545279648,
"calibration/coverage@15%": 0.9225445896889746,
"calibration/coverage@20%": 0.9765013054830287,
"calibration/coverage@25%": 0.9926892950391645,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6614465668086665,
"calibration/distribution_entropy_10": 0.39870880529696723,
"calibration/distribution_entropy_100": 0.2124436541205837,
"calibration/ece": 0.11906999650726022,
"calibration/mean_confidence": 0.8056827060716415,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.0014756944444444641,
"completions/max_length": 3871.0,
"completions/max_terminated_length": 3871.0,
"completions/mean_length": 1022.5205078125,
"completions/mean_terminated_length": 1024.047265625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 298.8,
"epoch": 1.9230769230769231,
"grad_norm": 0.0007274962845258415,
"learning_rate": 1.4423076923076922e-06,
"loss": -0.0007,
"num_tokens": 2137044056.0,
"reward": 1.3031952381134033,
"reward_std": 0.11121167093515397,
"rewards/accuracy_reward": 0.7424479246139526,
"rewards/brier_reward": 0.8657499551773071,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9981770873069763,
"rewards/mean_confidence_reward": 0.7726735830307007,
"sampling/batch_mean_priority_error": 0.05381679004621047,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.125,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011131892912089825,
"sampling/priority_kl": 0.030000000819563867,
"sampling/priority_scale": 0.8267765342490747,
"sampling/prob_entropy": 10.278957557678222,
"sampling/prob_max": 6.0364877572283146e-05,
"sampling/prob_min": 2.0352091814856978e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9151999950408936,
"sampling/prompt_draws_total": 57456.0,
"sampling/seen_fraction": 0.9300533413887024,
"sampling/unseen_fraction": 0.0699466586112976,
"signal/accuracy_reward/centered_abs_mean": 0.10494249016046524,
"signal/accuracy_reward/group_std_mean": 0.1474722906947136,
"signal/accuracy_reward/group_zero_std_frac": 0.544444453716278,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05247124508023262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05247124508023262,
"signal/advantage_abs_mean": 0.07756313532590867,
"signal/advantage_pre_scale_abs_mean": 0.07756313532590867,
"signal/advantage_pre_scale_std": 0.16611856520175933,
"signal/advantage_std": 0.16611856520175933,
"signal/brier_reward/centered_abs_mean": 0.06642551645636559,
"signal/brier_reward/group_std_mean": 0.09464680701494217,
"signal/brier_reward/group_zero_std_frac": 0.3694444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.033212758228182794,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.033212758228182794,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0033691405667923393,
"signal/format_reward/group_std_mean": 0.007633072696626187,
"signal/format_reward/group_zero_std_frac": 0.9638888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016845702833961696,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0016845702833961696,
"signal/mean_confidence_reward/centered_abs_mean": 0.0432563029229641,
"signal/mean_confidence_reward/group_std_mean": 0.060444879531860354,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4222222328186035,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.3256300159555393e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.3256300159555393e-07,
"step": 800
},
{
"epoch": 1.9230769230769231,
"eval_calibration/aurc": 0.15189747488764554,
"eval_calibration/batch_distribution_entropy": 0.452406431145014,
"eval_calibration/batch_entropy_100bins": 0.24782939662713582,
"eval_calibration/batch_entropy_10bins": 0.452406431145014,
"eval_calibration/batch_entropy_50bins": 0.2917407558634221,
"eval_calibration/batch_uniqueness": -0.05053761814744801,
"eval_calibration/confidence_entropy": 0.4086416396546222,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.7634782608695653,
"eval_calibration/coverage@20%": 0.82,
"eval_calibration/coverage@25%": 0.9434782608695652,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.452406431145014,
"eval_calibration/distribution_entropy_100": 0.24782939662713582,
"eval_calibration/ece": 0.06547826086956546,
"eval_calibration/mean_confidence": 0.7889565217391306,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.0017361111111111234,
"eval_completions/max_length": 2835.6666666666665,
"eval_completions/max_terminated_length": 2835.6666666666665,
"eval_completions/mean_length": 1001.4951680501302,
"eval_completions/mean_terminated_length": 1003.2116902669271,
"eval_completions/min_length": 236.66666666666666,
"eval_completions/min_terminated_length": 330.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 2137044056.0,
"eval_reward": 1.2835538585980732,
"eval_reward_std": 0.32838855187098187,
"eval_rewards/accuracy_reward": 0.7222222288449606,
"eval_rewards/brier_reward": 0.8466059267520905,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9982638955116272,
"eval_rewards/mean_confidence_reward": 0.7875867386658987,
"eval_runtime": 181.7797,
"eval_samples_per_second": 5.501,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3897569427887599,
"eval_signal/accuracy_reward/group_std_mean": 0.4464738667011261,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19487847139437994,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19487847139437994,
"eval_signal/advantage_abs_mean": 0.27775069574515027,
"eval_signal/advantage_pre_scale_abs_mean": 0.27775069574515027,
"eval_signal/advantage_pre_scale_std": 0.3263417234023412,
"eval_signal/advantage_std": 0.3263417234023412,
"eval_signal/brier_reward/centered_abs_mean": 0.1897987350821495,
"eval_signal/brier_reward/group_std_mean": 0.25104477753241855,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.09489936754107475,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.09489936754107475,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/group_std_mean": 0.009820927555362383,
"eval_signal/format_reward/group_zero_std_frac": 0.944444457689921,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0016818575871487458,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.15858287115891775,
"eval_signal/mean_confidence_reward/group_std_mean": 0.20427095144987106,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.585828651210856e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.585828651210856e-06,
"eval_steps_per_second": 0.033,
"step": 800
},
{
"epoch": 1.9230769230769231,
"step": 800,
"train_probe_calibration/aurc": 0.09918878298184919,
"train_probe_calibration/batch_distribution_entropy": 0.4207500172518936,
"train_probe_calibration/batch_entropy_100bins": 0.2282892016326656,
"train_probe_calibration/batch_entropy_10bins": 0.4207500172518936,
"train_probe_calibration/batch_entropy_50bins": 0.2687383544736379,
"train_probe_calibration/batch_uniqueness": -0.12686578449905483,
"train_probe_calibration/confidence_entropy": 0.4009151837695988,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7878260869565218,
"train_probe_calibration/coverage@15%": 0.8956521739130435,
"train_probe_calibration/coverage@20%": 0.9808695652173913,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.4207500172518936,
"train_probe_calibration/distribution_entropy_100": 0.2282892016326656,
"train_probe_calibration/ece": 0.03947826086956502,
"train_probe_calibration/mean_confidence": 0.8020869565217393,
"train_probe_calibration/unique_confidence_per_question": 0.006944444444444444,
"train_probe_calibration/unique_confidences": 8,
"train_probe_completions/clipped_ratio": 0.0017361111111111234,
"train_probe_completions/max_length": 2715.1666666666665,
"train_probe_completions/max_terminated_length": 2715.1666666666665,
"train_probe_completions/mean_length": 1000.1415100097656,
"train_probe_completions/mean_terminated_length": 1001.8871358235677,
"train_probe_completions/min_length": 197.66666666666666,
"train_probe_completions/min_terminated_length": 261.6666666666667,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2137044056.0,
"train_probe_reward": 1.3348864515622456,
"train_probe_reward_std": 0.29340456426143646,
"train_probe_rewards/accuracy_reward": 0.7881944378217062,
"train_probe_rewards/brier_reward": 0.8832986156145731,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9982638955116272,
"train_probe_rewards/mean_confidence_reward": 0.8006944358348846,
"train_probe_runtime": 168.1813,
"train_probe_samples_per_second": 5.946,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3246527711550395,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4065794100364049,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16232638557751974,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16232638557751974,
"train_probe_signal/advantage_abs_mean": 0.2283279299736023,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2283279299736023,
"train_probe_signal/advantage_pre_scale_std": 0.2924281011025111,
"train_probe_signal/advantage_std": 0.2924281011025111,
"train_probe_signal/brier_reward/centered_abs_mean": 0.15074980755647024,
"train_probe_signal/brier_reward/group_std_mean": 0.21790056924025217,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07537490377823512,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07537490377823512,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/group_std_mean": 0.009820927555362383,
"train_probe_signal/format_reward/group_zero_std_frac": 0.944444457689921,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0016818575871487458,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.1441731428106626,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.19188025345404944,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.4417313991543779e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.4417313991543779e-06,
"train_probe_steps_per_second": 0.036
},
{
"calibration/aurc": 0.18385633368569074,
"calibration/batch_distribution_entropy": 0.44733582079772455,
"calibration/batch_entropy_100bins": 0.24339174978119438,
"calibration/batch_entropy_10bins": 0.44733582079772455,
"calibration/batch_entropy_50bins": 0.2865168297969042,
"calibration/batch_uniqueness": -0.033245663910445734,
"calibration/confidence_entropy": 0.4115919705930738,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.35045963881636205,
"calibration/coverage@15%": 0.3666476283724978,
"calibration/coverage@20%": 0.5364583333333333,
"calibration/coverage@25%": 0.7046875,
"calibration/coverage@30%": 0.8723958333333334,
"calibration/coverage@5%": 0.31394827023498695,
"calibration/distribution_entropy_10": 0.44733582079772455,
"calibration/distribution_entropy_100": 0.24339174978119438,
"calibration/ece": 0.13457503807658838,
"calibration/mean_confidence": 0.779022655570061,
"calibration/unique_confidence_per_question": 0.0203125,
"calibration/unique_confidences": 7.8,
"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": 0.0006944444444444642,
"completions/max_length": 3408.8,
"completions/max_terminated_length": 3408.8,
"completions/mean_length": 983.4182373046875,
"completions/mean_terminated_length": 984.1210205078125,
"completions/min_length": 64.2,
"completions/min_terminated_length": 303.0,
"epoch": 1.9350961538461537,
"grad_norm": 0.0008335274760611355,
"learning_rate": 1.4122596153846154e-06,
"loss": 0.0011,
"num_tokens": 2151477610.0,
"reward": 1.3270015478134156,
"reward_std": 0.11657609343528748,
"rewards/accuracy_reward": 0.7857638955116272,
"rewards/brier_reward": 0.8691779732704162,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451216697693,
"rewards/mean_confidence_reward": 0.8054861068725586,
"sampling/batch_mean_priority_error": 0.054985726086469545,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.10833333333333332,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.0111892344430089,
"sampling/priority_kl": 0.02999982424080372,
"sampling/priority_scale": 0.8299192130332813,
"sampling/prob_entropy": 10.278948402404785,
"sampling/prob_max": 6.065202906029299e-05,
"sampling/prob_min": 2.03645795409102e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9272000074386597,
"sampling/prompt_draws_total": 57816.0,
"sampling/seen_fraction": 0.9316666722297668,
"sampling/unseen_fraction": 0.06833332777023315,
"signal/accuracy_reward/centered_abs_mean": 0.11029731184244156,
"signal/accuracy_reward/group_std_mean": 0.15226011127233505,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666686534881,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05514865592122078,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05514865592122078,
"signal/advantage_abs_mean": 0.08358793258666992,
"signal/advantage_pre_scale_abs_mean": 0.08358793258666992,
"signal/advantage_pre_scale_std": 0.1721019297838211,
"signal/advantage_std": 0.1721019297838211,
"signal/brier_reward/centered_abs_mean": 0.07183887735009194,
"signal/brier_reward/group_std_mean": 0.09882517009973527,
"signal/brier_reward/group_zero_std_frac": 0.40555556416511535,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03591943867504597,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03591943867504597,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0018500433885492384,
"signal/format_reward/group_std_mean": 0.005401509813964367,
"signal/format_reward/group_zero_std_frac": 0.9694444417953492,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0009250216942746192,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0009250216942746192,
"signal/mean_confidence_reward/centered_abs_mean": 0.04548124074935913,
"signal/mean_confidence_reward/group_std_mean": 0.060777968168258666,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4666666746139526,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.548124081793503e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.548124081793503e-07,
"step": 805
},
{
"calibration/aurc": 0.17190273395851366,
"calibration/batch_distribution_entropy": 0.4751986455352262,
"calibration/batch_entropy_100bins": 0.2599899300001808,
"calibration/batch_entropy_10bins": 0.4751986455352262,
"calibration/batch_entropy_50bins": 0.30605593899438904,
"calibration/batch_uniqueness": 0.04454210069444448,
"calibration/confidence_entropy": 0.4235046630804894,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.2958333333333333,
"calibration/coverage@15%": 0.484375,
"calibration/coverage@20%": 0.646875,
"calibration/coverage@25%": 0.8208333333333334,
"calibration/coverage@30%": 0.9036458333333334,
"calibration/coverage@5%": 0.0,
"calibration/distribution_entropy_10": 0.4751986455352262,
"calibration/distribution_entropy_100": 0.2599899300001808,
"calibration/ece": 0.11010416666666667,
"calibration/mean_confidence": 0.7810416666666667,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.0009548611111111161,
"completions/max_length": 3496.4,
"completions/max_terminated_length": 3496.4,
"completions/mean_length": 987.4782958984375,
"completions/mean_terminated_length": 988.4582153320313,
"completions/min_length": 49.8,
"completions/min_terminated_length": 274.6,
"epoch": 1.9471153846153846,
"grad_norm": 0.0006270384765230119,
"learning_rate": 1.3822115384615387e-06,
"loss": 0.0001,
"num_tokens": 2165978512.0,
"reward": 1.3170762062072754,
"reward_std": 0.10379236936569214,
"rewards/accuracy_reward": 0.7657117962837219,
"rewards/brier_reward": 0.8693793535232544,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451335906982,
"rewards/mean_confidence_reward": 0.8073871612548829,
"sampling/batch_mean_priority_error": 0.055473721058377626,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.12222222222222223,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011242944747209549,
"sampling/priority_kl": 0.029999969899654387,
"sampling/priority_scale": 0.8319486677879467,
"sampling/prob_entropy": 10.278951454162598,
"sampling/prob_max": 6.088290538173169e-05,
"sampling/prob_min": 2.0387604672578162e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9391999959945678,
"sampling/prompt_draws_total": 58176.0,
"sampling/seen_fraction": 0.9329066634178161,
"sampling/unseen_fraction": 0.06709333658218383,
"signal/accuracy_reward/centered_abs_mean": 0.09194336086511612,
"signal/accuracy_reward/group_std_mean": 0.12984519004821776,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04597168043255806,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04597168043255806,
"signal/advantage_abs_mean": 0.07197851687669754,
"signal/advantage_pre_scale_abs_mean": 0.07197851687669754,
"signal/advantage_pre_scale_std": 0.1576433524489403,
"signal/advantage_std": 0.1576433524489403,
"signal/brier_reward/centered_abs_mean": 0.06384159326553344,
"signal/brier_reward/group_std_mean": 0.09090788215398789,
"signal/brier_reward/group_zero_std_frac": 0.45555556416511533,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03192079663276672,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03192079663276672,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0018066406017169356,
"signal/format_reward/group_std_mean": 0.004452054016292095,
"signal/format_reward/group_zero_std_frac": 0.9777777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0009033203008584678,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0009033203008584678,
"signal/mean_confidence_reward/centered_abs_mean": 0.04468643441796303,
"signal/mean_confidence_reward/group_std_mean": 0.05934521555900574,
"signal/mean_confidence_reward/group_zero_std_frac": 0.522222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.4686433966489856e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.4686433966489856e-07,
"step": 810
},
{
"calibration/aurc": 0.19579915894519634,
"calibration/batch_distribution_entropy": 0.4647881882032324,
"calibration/batch_entropy_100bins": 0.24840480950407517,
"calibration/batch_entropy_10bins": 0.4647881882032324,
"calibration/batch_entropy_50bins": 0.29241812259205274,
"calibration/batch_uniqueness": 0.00656296614868152,
"calibration/confidence_entropy": 0.41222186254121224,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.37643979057591626,
"calibration/coverage@15%": 0.4,
"calibration/coverage@20%": 0.4,
"calibration/coverage@25%": 0.6817708333333333,
"calibration/coverage@30%": 0.6921875,
"calibration/coverage@5%": 0.13802083333333331,
"calibration/distribution_entropy_10": 0.4647881882032324,
"calibration/distribution_entropy_100": 0.24840480950407517,
"calibration/ece": 0.1426440068247372,
"calibration/mean_confidence": 0.8000327132573739,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0011284722222222542,
"completions/max_length": 3592.4,
"completions/max_terminated_length": 3592.4,
"completions/mean_length": 1039.9433349609376,
"completions/mean_terminated_length": 1041.126171875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 300.8,
"epoch": 1.9591346153846154,
"grad_norm": 0.0007290610228665173,
"learning_rate": 1.3521634615384617e-06,
"loss": 0.0003,
"num_tokens": 2181050851.0,
"reward": 1.3023419618606566,
"reward_std": 0.11788448095321655,
"rewards/accuracy_reward": 0.745225703716278,
"rewards/brier_reward": 0.8608315944671631,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9986111044883728,
"rewards/mean_confidence_reward": 0.7790451526641846,
"sampling/batch_mean_priority_error": 0.05542332055228545,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.125,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01129895206540823,
"sampling/priority_kl": 0.0299999862909317,
"sampling/priority_scale": 0.8350183546775952,
"sampling/prob_entropy": 10.278951072692871,
"sampling/prob_max": 6.1165435181465e-05,
"sampling/prob_min": 2.0399702407303266e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.951200008392334,
"sampling/prompt_draws_total": 58536.0,
"sampling/seen_fraction": 0.9344400048255921,
"sampling/unseen_fraction": 0.06555999517440796,
"signal/accuracy_reward/centered_abs_mean": 0.10779622346162795,
"signal/accuracy_reward/group_std_mean": 0.15116337835788726,
"signal/accuracy_reward/group_zero_std_frac": 0.5388888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05389811173081398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05389811173081398,
"signal/advantage_abs_mean": 0.08279699683189393,
"signal/advantage_pre_scale_abs_mean": 0.08279699683189393,
"signal/advantage_pre_scale_std": 0.171286541223526,
"signal/advantage_std": 0.171286541223526,
"signal/brier_reward/centered_abs_mean": 0.07373405247926712,
"signal/brier_reward/group_std_mean": 0.10261615216732026,
"signal/brier_reward/group_zero_std_frac": 0.3638888895511627,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03686702623963356,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03686702623963356,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.002604166604578495,
"signal/format_reward/group_std_mean": 0.005957829533144832,
"signal/format_reward/group_zero_std_frac": 0.9722222089767456,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0013020833022892476,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0013020833022892476,
"signal/mean_confidence_reward/centered_abs_mean": 0.0520480789244175,
"signal/mean_confidence_reward/group_std_mean": 0.06866604089736938,
"signal/mean_confidence_reward/group_zero_std_frac": 0.42222222685813904,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.204807735026406e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.204807735026406e-07,
"step": 815
},
{
"calibration/aurc": 0.09833870413235231,
"calibration/batch_distribution_entropy": 0.4614352913132178,
"calibration/batch_entropy_100bins": 0.2502082935466154,
"calibration/batch_entropy_10bins": 0.4614352913132178,
"calibration/batch_entropy_50bins": 0.2945411548268037,
"calibration/batch_uniqueness": 0.04135342083208417,
"calibration/confidence_entropy": 0.4304910644689368,
"calibration/coverage@0%": 0.29010416666666666,
"calibration/coverage@1%": 0.303125,
"calibration/coverage@10%": 0.5684670275590551,
"calibration/coverage@15%": 0.8559728953797551,
"calibration/coverage@20%": 0.9035079288391824,
"calibration/coverage@25%": 0.9196850393700787,
"calibration/coverage@30%": 0.9417322834645668,
"calibration/coverage@5%": 0.3265625,
"calibration/distribution_entropy_10": 0.4614352913132178,
"calibration/distribution_entropy_100": 0.2502082935466154,
"calibration/ece": 0.09957528756604497,
"calibration/mean_confidence": 0.7683144042577251,
"calibration/unique_confidence_per_question": 0.019791666666666666,
"calibration/unique_confidences": 7.6,
"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": 0.0013020833333333482,
"completions/max_length": 3667.0,
"completions/max_terminated_length": 3667.0,
"completions/mean_length": 1022.9057495117188,
"completions/mean_terminated_length": 1024.2632934570313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 271.4,
"epoch": 1.9711538461538463,
"grad_norm": 0.0005470445030368865,
"learning_rate": 1.3221153846153848e-06,
"loss": -0.0004,
"num_tokens": 2195936517.0,
"reward": 1.323672842979431,
"reward_std": 0.09248801916837693,
"rewards/accuracy_reward": 0.7696180462837219,
"rewards/brier_reward": 0.8791007280349732,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9986111044883728,
"rewards/mean_confidence_reward": 0.7945225596427917,
"sampling/batch_mean_priority_error": 0.048565969888739534,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09166666666666666,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011350595764815808,
"sampling/priority_kl": 0.030000149831175806,
"sampling/priority_scale": 0.8369162738090381,
"sampling/prob_entropy": 10.278959846496582,
"sampling/prob_max": 6.138679746072739e-05,
"sampling/prob_min": 2.042250198428519e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9632000207901001,
"sampling/prompt_draws_total": 58896.0,
"sampling/seen_fraction": 0.9355800032615662,
"sampling/unseen_fraction": 0.06441999673843384,
"signal/accuracy_reward/centered_abs_mean": 0.07965494841337203,
"signal/accuracy_reward/group_std_mean": 0.1155205637216568,
"signal/accuracy_reward/group_zero_std_frac": 0.6333333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.03982747420668602,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.03982747420668602,
"signal/advantage_abs_mean": 0.06276309192180633,
"signal/advantage_pre_scale_abs_mean": 0.06276309192180633,
"signal/advantage_pre_scale_std": 0.14800895154476165,
"signal/advantage_std": 0.14800895154476165,
"signal/brier_reward/centered_abs_mean": 0.05834457501769066,
"signal/brier_reward/group_std_mean": 0.08316372632980347,
"signal/brier_reward/group_zero_std_frac": 0.4250000059604645,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02917228750884533,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02917228750884533,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0026692708022892477,
"signal/format_reward/group_std_mean": 0.0072588598355650905,
"signal/format_reward/group_zero_std_frac": 0.9611111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0013346354011446238,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0013346354011446238,
"signal/mean_confidence_reward/centered_abs_mean": 0.042954657971858975,
"signal/mean_confidence_reward/group_std_mean": 0.05850115194916725,
"signal/mean_confidence_reward/group_zero_std_frac": 0.48055556416511536,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.295465771519957e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.295465771519957e-07,
"step": 820
},
{
"calibration/aurc": 0.15048849581866025,
"calibration/batch_distribution_entropy": 0.401514669951322,
"calibration/batch_entropy_100bins": 0.2161378706493462,
"calibration/batch_entropy_10bins": 0.401514669951322,
"calibration/batch_entropy_50bins": 0.25443400424696244,
"calibration/batch_uniqueness": -0.13833550347222218,
"calibration/confidence_entropy": 0.4026604023740436,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.3026041666666667,
"calibration/coverage@15%": 0.6947916666666667,
"calibration/coverage@20%": 0.7494791666666667,
"calibration/coverage@25%": 0.7885416666666667,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.2635416666666667,
"calibration/distribution_entropy_10": 0.401514669951322,
"calibration/distribution_entropy_100": 0.2161378706493462,
"calibration/ece": 0.12052083333333338,
"calibration/mean_confidence": 0.8017708333333335,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.000868055555555558,
"completions/max_length": 3711.8,
"completions/max_terminated_length": 3711.8,
"completions/mean_length": 1036.2858642578126,
"completions/mean_terminated_length": 1037.209228515625,
"completions/min_length": 118.2,
"completions/min_terminated_length": 307.6,
"epoch": 1.9831730769230769,
"grad_norm": 0.0007763904868625104,
"learning_rate": 1.292067307692308e-06,
"loss": 0.0011,
"num_tokens": 2210964898.0,
"reward": 1.2921491622924806,
"reward_std": 0.12069564908742905,
"rewards/accuracy_reward": 0.7279513955116272,
"rewards/brier_reward": 0.8572864413261414,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9990451335906982,
"rewards/mean_confidence_reward": 0.7701562523841858,
"sampling/batch_mean_priority_error": 0.0555428616098097,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.12777777777777777,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011398331820964813,
"sampling/priority_kl": 0.03000003807246685,
"sampling/priority_scale": 0.8398089825874194,
"sampling/prob_entropy": 10.27895736694336,
"sampling/prob_max": 6.166202801978215e-05,
"sampling/prob_min": 2.0436170962057078e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9752000093460083,
"sampling/prompt_draws_total": 59256.0,
"sampling/seen_fraction": 0.9369800090789795,
"sampling/unseen_fraction": 0.06301999092102051,
"signal/accuracy_reward/centered_abs_mean": 0.11948784738779068,
"signal/accuracy_reward/group_std_mean": 0.1649380922317505,
"signal/accuracy_reward/group_zero_std_frac": 0.5055555701255798,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05974392369389534,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05974392369389534,
"signal/advantage_abs_mean": 0.08622851222753525,
"signal/advantage_pre_scale_abs_mean": 0.08622851222753525,
"signal/advantage_pre_scale_std": 0.17191420793533324,
"signal/advantage_std": 0.17191420793533324,
"signal/brier_reward/centered_abs_mean": 0.0750839278101921,
"signal/brier_reward/group_std_mean": 0.10442855954170227,
"signal/brier_reward/group_zero_std_frac": 0.3444444537162781,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03754196390509605,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03754196390509605,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0017740885028615593,
"signal/format_reward/group_std_mean": 0.00407174676656723,
"signal/format_reward/group_zero_std_frac": 0.9805555462837219,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0008870442514307797,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0008870442514307797,
"signal/mean_confidence_reward/centered_abs_mean": 0.05379992350935936,
"signal/mean_confidence_reward/group_std_mean": 0.07070956826210022,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4111111104488373,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.379991762310965e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.379991762310965e-07,
"step": 825
},
{
"calibration/aurc": 0.08829125696869047,
"calibration/batch_distribution_entropy": 0.48584406488715803,
"calibration/batch_entropy_100bins": 0.26703580379083747,
"calibration/batch_entropy_10bins": 0.48584406488715803,
"calibration/batch_entropy_50bins": 0.3143502276194673,
"calibration/batch_uniqueness": 0.05997741525988515,
"calibration/confidence_entropy": 0.4198865761959188,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.17552083333333332,
"calibration/coverage@10%": 0.5196665578764142,
"calibration/coverage@15%": 0.8503182114882506,
"calibration/coverage@20%": 0.9415483572671889,
"calibration/coverage@25%": 0.9634464751958225,
"calibration/coverage@30%": 0.9765013054830287,
"calibration/coverage@5%": 0.3269160683202785,
"calibration/distribution_entropy_10": 0.48584406488715803,
"calibration/distribution_entropy_100": 0.26703580379083747,
"calibration/ece": 0.0851956864664926,
"calibration/mean_confidence": 0.7715115045691908,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.0005208333333333482,
"completions/max_length": 3806.4,
"completions/max_terminated_length": 3806.4,
"completions/mean_length": 1043.68623046875,
"completions/mean_terminated_length": 1044.231591796875,
"completions/min_length": 48.0,
"completions/min_terminated_length": 267.8,
"epoch": 1.9951923076923077,
"grad_norm": 0.0005658183363266289,
"learning_rate": 1.2620192307692309e-06,
"loss": 0.0008,
"num_tokens": 2226050883.0,
"reward": 1.335673975944519,
"reward_std": 0.09520231038331986,
"rewards/accuracy_reward": 0.7828993082046509,
"rewards/brier_reward": 0.8890407800674438,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9993923544883728,
"rewards/mean_confidence_reward": 0.7782551884651184,
"sampling/batch_mean_priority_error": 0.03993952312948028,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.11111111111111112,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.01144140474498272,
"sampling/priority_kl": 0.029999879747629167,
"sampling/priority_scale": 0.8430393754737452,
"sampling/prob_entropy": 10.278953552246094,
"sampling/prob_max": 6.195996684255078e-05,
"sampling/prob_min": 2.0447872157092206e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9871999740600585,
"sampling/prompt_draws_total": 59616.0,
"sampling/seen_fraction": 0.9384400010108948,
"sampling/unseen_fraction": 0.061559998989105226,
"signal/accuracy_reward/centered_abs_mean": 0.09328884482383729,
"signal/accuracy_reward/group_std_mean": 0.13105592578649522,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04664442241191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04664442241191864,
"signal/advantage_abs_mean": 0.06662840321660042,
"signal/advantage_pre_scale_abs_mean": 0.06662840321660042,
"signal/advantage_pre_scale_std": 0.1486303061246872,
"signal/advantage_std": 0.1486303061246872,
"signal/brier_reward/centered_abs_mean": 0.05784967392683029,
"signal/brier_reward/group_std_mean": 0.08218971043825149,
"signal/brier_reward/group_zero_std_frac": 0.41111111640930176,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028924836963415145,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028924836963415145,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.001177300326526165,
"signal/format_reward/group_std_mean": 0.0034373244270682335,
"signal/format_reward/group_zero_std_frac": 0.9805555462837219,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0005886501632630825,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0005886501632630825,
"signal/mean_confidence_reward/centered_abs_mean": 0.04647516682744026,
"signal/mean_confidence_reward/group_std_mean": 0.061248787492513654,
"signal/mean_confidence_reward/group_zero_std_frac": 0.46111111640930175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.647516277600516e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.647516277600516e-07,
"step": 830
},
{
"calibration/aurc": 0.07549365578168557,
"calibration/batch_distribution_entropy": 0.5254938552647503,
"calibration/batch_entropy_100bins": 0.2960007588278585,
"calibration/batch_entropy_10bins": 0.5254938552647503,
"calibration/batch_entropy_50bins": 0.34844730404000235,
"calibration/batch_uniqueness": 0.1832330232638631,
"calibration/confidence_entropy": 0.43629637896002427,
"calibration/coverage@0%": 0.10885416666666667,
"calibration/coverage@1%": 0.11458333333333333,
"calibration/coverage@10%": 0.7351716958601454,
"calibration/coverage@15%": 0.8265016909603009,
"calibration/coverage@20%": 0.8688489605819507,
"calibration/coverage@25%": 0.9425647965879265,
"calibration/coverage@30%": 0.9734375,
"calibration/coverage@5%": 0.5424786745406823,
"calibration/distribution_entropy_10": 0.5254938552647503,
"calibration/distribution_entropy_100": 0.2960007588278585,
"calibration/ece": 0.10865095482720322,
"calibration/mean_confidence": 0.7365200278999885,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.000694444444444442,
"completions/max_length": 3274.4,
"completions/max_terminated_length": 3274.4,
"completions/mean_length": 973.68837890625,
"completions/mean_terminated_length": 974.4301025390625,
"completions/min_length": 162.8,
"completions/min_terminated_length": 360.8,
"epoch": 2.0072115384615383,
"grad_norm": 0.0005269041866995394,
"learning_rate": 1.231971153846154e-06,
"loss": 0.0003,
"num_tokens": 2241097670.0,
"reward": 1.3262059450149537,
"reward_std": 0.09740138053894043,
"rewards/accuracy_reward": 0.7748263955116272,
"rewards/brier_reward": 0.8787855863571167,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9987847089767456,
"rewards/mean_confidence_reward": 0.7612066030502319,
"sampling/batch_mean_priority_error": 0.043829235399801214,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09722222222222224,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011479710228741169,
"sampling/priority_kl": 0.029999543726444245,
"sampling/priority_scale": 0.8459879100089893,
"sampling/prob_entropy": 10.278957176208497,
"sampling/prob_max": 6.224302342161536e-05,
"sampling/prob_min": 2.0461800886550918e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 1.9991999864578247,
"sampling/prompt_draws_total": 59976.0,
"sampling/seen_fraction": 0.9398066639900208,
"sampling/unseen_fraction": 0.06019333600997925,
"signal/accuracy_reward/centered_abs_mean": 0.10342881977558135,
"signal/accuracy_reward/group_std_mean": 0.14174398630857468,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05171440988779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05171440988779068,
"signal/advantage_abs_mean": 0.06915109902620316,
"signal/advantage_pre_scale_abs_mean": 0.06915109902620316,
"signal/advantage_pre_scale_std": 0.14984942972660065,
"signal/advantage_std": 0.14984942972660065,
"signal/brier_reward/centered_abs_mean": 0.0605738990008831,
"signal/brier_reward/group_std_mean": 0.08339389562606811,
"signal/brier_reward/group_zero_std_frac": 0.3916666626930237,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03028694950044155,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03028694950044155,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.002267795195803046,
"signal/format_reward/group_std_mean": 0.0049757368862628935,
"signal/format_reward/group_zero_std_frac": 0.9777777671813965,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001133897597901523,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001133897597901523,
"signal/mean_confidence_reward/centered_abs_mean": 0.04679037928581238,
"signal/mean_confidence_reward/group_std_mean": 0.06196520924568176,
"signal/mean_confidence_reward/group_zero_std_frac": 0.43055555820465086,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.679037715504819e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.679037715504819e-07,
"step": 835
},
{
"calibration/aurc": 0.147330218464356,
"calibration/batch_distribution_entropy": 0.5416599200734422,
"calibration/batch_entropy_100bins": 0.2980373510942677,
"calibration/batch_entropy_10bins": 0.5416599200734422,
"calibration/batch_entropy_50bins": 0.35084474750423256,
"calibration/batch_uniqueness": 0.22830577188595483,
"calibration/confidence_entropy": 0.44884432769273436,
"calibration/coverage@0%": 0.11770833333333333,
"calibration/coverage@1%": 0.12291666666666667,
"calibration/coverage@10%": 0.31079607267188863,
"calibration/coverage@15%": 0.3739365752828546,
"calibration/coverage@20%": 0.8098958333333333,
"calibration/coverage@25%": 0.8765625,
"calibration/coverage@30%": 0.921875,
"calibration/coverage@5%": 0.2774409812880766,
"calibration/distribution_entropy_10": 0.5416599200734422,
"calibration/distribution_entropy_100": 0.2980373510942677,
"calibration/ece": 0.11948107049608356,
"calibration/mean_confidence": 0.7425214860748478,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.001128472222222232,
"completions/max_length": 3801.6,
"completions/max_terminated_length": 3801.6,
"completions/mean_length": 1044.3095581054688,
"completions/mean_terminated_length": 1045.5166015625,
"completions/min_length": 86.6,
"completions/min_terminated_length": 236.6,
"epoch": 2.019230769230769,
"grad_norm": 0.0006535506690852344,
"learning_rate": 1.201923076923077e-06,
"loss": 0.0007,
"num_tokens": 2256206260.0,
"reward": 1.3112290143966674,
"reward_std": 0.09778019338846207,
"rewards/accuracy_reward": 0.7542534708976746,
"rewards/brier_reward": 0.8694054126739502,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9987847208976746,
"rewards/mean_confidence_reward": 0.729296863079071,
"sampling/batch_mean_priority_error": 0.04321197037989741,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09166666666666667,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011516995541751385,
"sampling/priority_kl": 0.029999730363488197,
"sampling/priority_scale": 0.8477530895499512,
"sampling/prob_entropy": 10.278942108154297,
"sampling/prob_max": 6.246392877073959e-05,
"sampling/prob_min": 2.0486806897679342e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.0112000465393067,
"sampling/prompt_draws_total": 60336.0,
"sampling/seen_fraction": 0.9407866597175598,
"sampling/unseen_fraction": 0.059213340282440186,
"signal/accuracy_reward/centered_abs_mean": 0.10748155415058136,
"signal/accuracy_reward/group_std_mean": 0.14893923699855804,
"signal/accuracy_reward/group_zero_std_frac": 0.5416666865348816,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05374077707529068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05374077707529068,
"signal/advantage_abs_mean": 0.06958878040313721,
"signal/advantage_pre_scale_abs_mean": 0.06958878040313721,
"signal/advantage_pre_scale_std": 0.14752740263938904,
"signal/advantage_std": 0.14752740263938904,
"signal/brier_reward/centered_abs_mean": 0.06264480128884316,
"signal/brier_reward/group_std_mean": 0.08568576574325562,
"signal/brier_reward/group_zero_std_frac": 0.325,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03132240064442158,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03132240064442158,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.002289496548473835,
"signal/format_reward/group_std_mean": 0.005327311344444752,
"signal/format_reward/group_zero_std_frac": 0.9749999880790711,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0011447482742369176,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0011447482742369176,
"signal/mean_confidence_reward/centered_abs_mean": 0.0524658277630806,
"signal/mean_confidence_reward/group_std_mean": 0.06992786824703216,
"signal/mean_confidence_reward/group_zero_std_frac": 0.35555556416511536,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.246582588824821e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.246582588824821e-07,
"step": 840
},
{
"calibration/aurc": 0.0911913957220775,
"calibration/batch_distribution_entropy": 0.5457449115728301,
"calibration/batch_entropy_100bins": 0.29925961674891743,
"calibration/batch_entropy_10bins": 0.5457449115728301,
"calibration/batch_entropy_50bins": 0.35228357885679357,
"calibration/batch_uniqueness": 0.18566876717572223,
"calibration/confidence_entropy": 0.4356012844937179,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.13141361256544504,
"calibration/coverage@10%": 0.5806670234010172,
"calibration/coverage@15%": 0.802144124295063,
"calibration/coverage@20%": 0.9019497212523969,
"calibration/coverage@25%": 0.9921465968586387,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.3580437721464573,
"calibration/distribution_entropy_10": 0.5457449115728301,
"calibration/distribution_entropy_100": 0.29925961674891743,
"calibration/ece": 0.12006665841862763,
"calibration/mean_confidence": 0.732603944620393,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.003038194444444442,
"completions/max_length": 3588.8,
"completions/max_terminated_length": 3588.8,
"completions/mean_length": 1094.13447265625,
"completions/mean_terminated_length": 1097.5215087890624,
"completions/min_length": 0.0,
"completions/min_terminated_length": 286.6,
"epoch": 2.03125,
"grad_norm": 0.0005947959143668413,
"learning_rate": 1.1718750000000001e-06,
"loss": -0.0032,
"num_tokens": 2271920513.0,
"reward": 1.2724606037139892,
"reward_std": 0.10289801359176635,
"rewards/accuracy_reward": 0.6971354007720947,
"rewards/brier_reward": 0.8509834766387939,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967882037162781,
"rewards/mean_confidence_reward": 0.7071788072586059,
"sampling/batch_mean_priority_error": 0.04964866880445866,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555556,
"sampling/error_ema_max": 0.23887455463409424,
"sampling/error_ema_mean": 0.011563451029360294,
"sampling/priority_kl": 0.029999909549951555,
"sampling/priority_scale": 0.8497347890632228,
"sampling/prob_entropy": 10.278948402404785,
"sampling/prob_max": 6.269553414313122e-05,
"sampling/prob_min": 2.0509256501100026e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.023200035095215,
"sampling/prompt_draws_total": 60696.0,
"sampling/seen_fraction": 0.9418333411216736,
"sampling/unseen_fraction": 0.058166658878326415,
"signal/accuracy_reward/centered_abs_mean": 0.102685546875,
"signal/accuracy_reward/group_std_mean": 0.14314813315868377,
"signal/accuracy_reward/group_zero_std_frac": 0.5638888835906982,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0513427734375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0513427734375,
"signal/advantage_abs_mean": 0.07161518633365631,
"signal/advantage_pre_scale_abs_mean": 0.07161518633365631,
"signal/advantage_pre_scale_std": 0.15404508113861085,
"signal/advantage_std": 0.15404508113861085,
"signal/brier_reward/centered_abs_mean": 0.0655230537056923,
"signal/brier_reward/group_std_mean": 0.09098667353391647,
"signal/brier_reward/group_zero_std_frac": 0.30277777910232545,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03276152685284615,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03276152685284615,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005810546781867742,
"signal/format_reward/group_std_mean": 0.012965610064566136,
"signal/format_reward/group_zero_std_frac": 0.9388889074325562,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002905273390933871,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002905273390933871,
"signal/mean_confidence_reward/centered_abs_mean": 0.05646376684308052,
"signal/mean_confidence_reward/group_std_mean": 0.07490446716547013,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32500000596046447,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.646376564527599e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.646376564527599e-07,
"step": 845
},
{
"calibration/aurc": 0.0449775823876252,
"calibration/batch_distribution_entropy": 0.4243073755508937,
"calibration/batch_entropy_100bins": 0.23311381496007616,
"calibration/batch_entropy_10bins": 0.4243073755508937,
"calibration/batch_entropy_50bins": 0.27441781122107606,
"calibration/batch_uniqueness": -0.08864126831024444,
"calibration/confidence_entropy": 0.402103323049886,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.16614583333333333,
"calibration/coverage@10%": 0.8826737924281984,
"calibration/coverage@15%": 0.9274899369016538,
"calibration/coverage@20%": 0.9473332789382072,
"calibration/coverage@25%": 0.9848563968668408,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.7732267188859877,
"calibration/distribution_entropy_10": 0.4243073755508937,
"calibration/distribution_entropy_100": 0.23311381496007616,
"calibration/ece": 0.10496110748476917,
"calibration/mean_confidence": 0.7842422758920802,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.001736111111111116,
"completions/max_length": 3912.0,
"completions/max_terminated_length": 3912.0,
"completions/mean_length": 1051.2964477539062,
"completions/mean_terminated_length": 1053.1206787109375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 271.6,
"epoch": 2.043269230769231,
"grad_norm": 0.0006041685701347888,
"learning_rate": 1.141826923076923e-06,
"loss": -0.0014,
"num_tokens": 2287123224.0,
"reward": 1.3067851781845092,
"reward_std": 0.09317826330661774,
"rewards/accuracy_reward": 0.7391493082046509,
"rewards/brier_reward": 0.8764895796775818,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166746139526,
"rewards/mean_confidence_reward": 0.7414149165153503,
"sampling/batch_mean_priority_error": 0.04238674908527479,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09166666666666666,
"sampling/error_ema_max": 0.25062440037727357,
"sampling/error_ema_mean": 0.011596980690956115,
"sampling/priority_kl": 0.029999906942248345,
"sampling/priority_scale": 0.8517146884696558,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 6.292867037700489e-05,
"sampling/prob_min": 2.0531993141048587e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.035200023651123,
"sampling/prompt_draws_total": 61056.0,
"sampling/seen_fraction": 0.9428666591644287,
"sampling/unseen_fraction": 0.05713334083557129,
"signal/accuracy_reward/centered_abs_mean": 0.10021158754825592,
"signal/accuracy_reward/group_std_mean": 0.13549640774726868,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05010579377412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05010579377412796,
"signal/advantage_abs_mean": 0.06468015760183335,
"signal/advantage_pre_scale_abs_mean": 0.06468015760183335,
"signal/advantage_pre_scale_std": 0.14103153496980667,
"signal/advantage_std": 0.14103153496980667,
"signal/brier_reward/centered_abs_mean": 0.06078602746129036,
"signal/brier_reward/group_std_mean": 0.08554978594183922,
"signal/brier_reward/group_zero_std_frac": 0.37500000596046446,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03039301373064518,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03039301373064518,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.003982204874046147,
"signal/format_reward/group_std_mean": 0.010536715947091579,
"signal/format_reward/group_zero_std_frac": 0.9444444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019911024370230733,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019911024370230733,
"signal/mean_confidence_reward/centered_abs_mean": 0.05210667550563812,
"signal/mean_confidence_reward/group_std_mean": 0.06843070611357689,
"signal/mean_confidence_reward/group_zero_std_frac": 0.41666666865348817,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.210667325172835e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.210667325172835e-07,
"step": 850
},
{
"epoch": 2.043269230769231,
"eval_calibration/aurc": 0.13891627434276235,
"eval_calibration/batch_distribution_entropy": 0.5077117283467684,
"eval_calibration/batch_entropy_100bins": 0.2810605358504689,
"eval_calibration/batch_entropy_10bins": 0.5077117283467684,
"eval_calibration/batch_entropy_50bins": 0.3308599152818019,
"eval_calibration/batch_uniqueness": 0.07801649305555561,
"eval_calibration/confidence_entropy": 0.4140622911715569,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.0,
"eval_calibration/coverage@15%": 0.7335069444444444,
"eval_calibration/coverage@20%": 0.8194444444444444,
"eval_calibration/coverage@25%": 0.9210069444444444,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.5077117283467684,
"eval_calibration/distribution_entropy_100": 0.2810605358504689,
"eval_calibration/ece": 0.03428819444444463,
"eval_calibration/mean_confidence": 0.7492187500000002,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.0,
"eval_completions/max_length": 2704.8333333333335,
"eval_completions/max_terminated_length": 2704.8333333333335,
"eval_completions/mean_length": 1028.1000366210938,
"eval_completions/mean_terminated_length": 1028.1000366210938,
"eval_completions/min_length": 319.0,
"eval_completions/min_terminated_length": 319.0,
"eval_loss": 0.0,
"eval_num_tokens": 2287123224.0,
"eval_reward": 1.2898990114529927,
"eval_reward_std": 0.30779418845971424,
"eval_rewards/accuracy_reward": 0.7204861144224802,
"eval_rewards/brier_reward": 0.8592968980471293,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 1.0,
"eval_rewards/mean_confidence_reward": 0.7492187122503916,
"eval_runtime": 146.1672,
"eval_samples_per_second": 6.841,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3921440889437993,
"eval_signal/accuracy_reward/group_std_mean": 0.44856879611810047,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19607204447189966,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19607204447189966,
"eval_signal/advantage_abs_mean": 0.2605537722508113,
"eval_signal/advantage_pre_scale_abs_mean": 0.2605537722508113,
"eval_signal/advantage_pre_scale_std": 0.3053098718325297,
"eval_signal/advantage_std": 0.3053098718325297,
"eval_signal/brier_reward/centered_abs_mean": 0.1747602125008901,
"eval_signal/brier_reward/group_std_mean": 0.236327663064003,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.08738010625044505,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.08738010625044505,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.0,
"eval_signal/format_reward/group_std_mean": 0.0,
"eval_signal/format_reward/group_zero_std_frac": 1.0,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.19914276897907257,
"eval_signal/mean_confidence_reward/group_std_mean": 0.24361870189507803,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9914276663257624e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9914276663257624e-06,
"eval_steps_per_second": 0.041,
"step": 850
},
{
"epoch": 2.043269230769231,
"step": 850,
"train_probe_calibration/aurc": 0.0901579760393589,
"train_probe_calibration/batch_distribution_entropy": 0.49480410216785375,
"train_probe_calibration/batch_entropy_100bins": 0.2721646004371472,
"train_probe_calibration/batch_entropy_10bins": 0.49480410216785375,
"train_probe_calibration/batch_entropy_50bins": 0.3203877640482687,
"train_probe_calibration/batch_uniqueness": 0.03985419271813424,
"train_probe_calibration/confidence_entropy": 0.41224762261760145,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7621951219512195,
"train_probe_calibration/coverage@15%": 0.8484320557491289,
"train_probe_calibration/coverage@20%": 0.9451219512195121,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.49480410216785375,
"train_probe_calibration/distribution_entropy_100": 0.2721646004371472,
"train_probe_calibration/ece": 0.0388501742160276,
"train_probe_calibration/mean_confidence": 0.7668989547038328,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.002604166666666685,
"train_probe_completions/max_length": 3103.3333333333335,
"train_probe_completions/max_terminated_length": 3103.3333333333335,
"train_probe_completions/mean_length": 1027.0469258626301,
"train_probe_completions/mean_terminated_length": 1029.6766560872395,
"train_probe_completions/min_length": 111.33333333333333,
"train_probe_completions/min_terminated_length": 249.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2287123224.0,
"train_probe_reward": 1.3318913181622822,
"train_probe_reward_std": 0.28379150728384656,
"train_probe_rewards/accuracy_reward": 0.7838541666666666,
"train_probe_rewards/brier_reward": 0.883385439713796,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9965277910232544,
"train_probe_rewards/mean_confidence_reward": 0.7642360826333364,
"train_probe_runtime": 187.9598,
"train_probe_samples_per_second": 5.32,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3283962706724803,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4082522888978322,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16419813533624014,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16419813533624014,
"train_probe_signal/advantage_abs_mean": 0.2219469373424848,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.2219469373424848,
"train_probe_signal/advantage_pre_scale_std": 0.2837086667617162,
"train_probe_signal/advantage_std": 0.2837086667617162,
"train_probe_signal/brier_reward/centered_abs_mean": 0.14791234831015268,
"train_probe_signal/brier_reward/group_std_mean": 0.20902411391337714,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07395617415507634,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.07395617415507634,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/group_std_mean": 0.019641855110724766,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888889153798422,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.18338213860988617,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.23027842243512472,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8338212726121128e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8338212726121128e-06,
"train_probe_steps_per_second": 0.032
},
{
"calibration/aurc": 0.11605278446534488,
"calibration/batch_distribution_entropy": 0.5040144933010529,
"calibration/batch_entropy_100bins": 0.2786357475316114,
"calibration/batch_entropy_10bins": 0.5040144933010529,
"calibration/batch_entropy_50bins": 0.3280054937055892,
"calibration/batch_uniqueness": 0.0996175187728016,
"calibration/confidence_entropy": 0.4256967141929806,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.666072549820796,
"calibration/coverage@15%": 0.6953868542827382,
"calibration/coverage@20%": 0.8389342324376555,
"calibration/coverage@25%": 0.8938741665124758,
"calibration/coverage@30%": 0.9623774182274213,
"calibration/coverage@5%": 0.13229166666666667,
"calibration/distribution_entropy_10": 0.5040144933010529,
"calibration/distribution_entropy_100": 0.2786357475316114,
"calibration/ece": 0.08465455025938327,
"calibration/mean_confidence": 0.7348673337136711,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002517361111111094,
"completions/max_length": 3681.6,
"completions/max_terminated_length": 3681.6,
"completions/mean_length": 1033.7828491210937,
"completions/mean_terminated_length": 1036.364990234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 298.4,
"epoch": 2.0552884615384617,
"grad_norm": 0.0007877013413235545,
"learning_rate": 1.1117788461538462e-06,
"loss": -0.0017,
"num_tokens": 2302117906.0,
"reward": 1.3229143142700195,
"reward_std": 0.11778022795915603,
"rewards/accuracy_reward": 0.7722222208976746,
"rewards/brier_reward": 0.876629376411438,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9969618082046509,
"rewards/mean_confidence_reward": 0.7630728840827942,
"sampling/batch_mean_priority_error": 0.0449851466011066,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09444444444444446,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011639406159520149,
"sampling/priority_kl": 0.029999953508377076,
"sampling/priority_scale": 0.8540645777946339,
"sampling/prob_entropy": 10.278949356079101,
"sampling/prob_max": 6.31784278084524e-05,
"sampling/prob_min": 2.054988290183246e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.0471999645233154,
"sampling/prompt_draws_total": 61416.0,
"sampling/seen_fraction": 0.943993330001831,
"sampling/unseen_fraction": 0.056006669998168945,
"signal/accuracy_reward/centered_abs_mean": 0.11839192807674408,
"signal/accuracy_reward/group_std_mean": 0.1599825292825699,
"signal/accuracy_reward/group_zero_std_frac": 0.5305555582046508,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05919596403837204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05919596403837204,
"signal/advantage_abs_mean": 0.08527102023363113,
"signal/advantage_pre_scale_abs_mean": 0.08527102023363113,
"signal/advantage_pre_scale_std": 0.17474955320358276,
"signal/advantage_std": 0.17474955320358276,
"signal/brier_reward/centered_abs_mean": 0.07458995878696442,
"signal/brier_reward/group_std_mean": 0.10067926496267318,
"signal/brier_reward/group_zero_std_frac": 0.38333333730697633,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03729497939348221,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03729497939348221,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005441623320803046,
"signal/format_reward/group_std_mean": 0.011849517934024334,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002720811660401523,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002720811660401523,
"signal/mean_confidence_reward/centered_abs_mean": 0.05360949635505676,
"signal/mean_confidence_reward/group_std_mean": 0.07186432927846909,
"signal/mean_confidence_reward/group_zero_std_frac": 0.43055555820465086,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.36094944436627e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.36094944436627e-07,
"step": 855
},
{
"calibration/aurc": 0.10894603231881803,
"calibration/batch_distribution_entropy": 0.4515398122960118,
"calibration/batch_entropy_100bins": 0.2452527890133044,
"calibration/batch_entropy_10bins": 0.4515398122960118,
"calibration/batch_entropy_50bins": 0.28870761506958165,
"calibration/batch_uniqueness": -0.017896347537314682,
"calibration/confidence_entropy": 0.41349993638707916,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.43682414055700614,
"calibration/coverage@15%": 0.8484234193718605,
"calibration/coverage@20%": 0.9314967482850545,
"calibration/coverage@25%": 0.9775456919060052,
"calibration/coverage@30%": 0.9911227154046998,
"calibration/coverage@5%": 0.1515625,
"calibration/distribution_entropy_10": 0.4515398122960118,
"calibration/distribution_entropy_100": 0.2452527890133044,
"calibration/ece": 0.08027731020812341,
"calibration/mean_confidence": 0.7775777486242746,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003385416666666652,
"completions/max_length": 3677.6,
"completions/max_terminated_length": 3677.6,
"completions/mean_length": 1044.5763061523437,
"completions/mean_terminated_length": 1048.1059814453124,
"completions/min_length": 52.6,
"completions/min_terminated_length": 261.6,
"epoch": 2.0673076923076925,
"grad_norm": 0.0007250626222230494,
"learning_rate": 1.0817307692307693e-06,
"loss": -0.0027,
"num_tokens": 2317230017.0,
"reward": 1.3033494710922242,
"reward_std": 0.1059815376996994,
"rewards/accuracy_reward": 0.7505208253860474,
"rewards/brier_reward": 0.8599826455116272,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9961805701255798,
"rewards/mean_confidence_reward": 0.7483333230018616,
"sampling/batch_mean_priority_error": 0.052143241468603406,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011684058792889118,
"sampling/priority_kl": 0.030000266805291175,
"sampling/priority_scale": 0.8561827362282202,
"sampling/prob_entropy": 10.278953361511231,
"sampling/prob_max": 6.341807602439076e-05,
"sampling/prob_min": 2.0570450215018354e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.0591999530792235,
"sampling/prompt_draws_total": 61776.0,
"sampling/seen_fraction": 0.9450266599655152,
"sampling/unseen_fraction": 0.05497334003448486,
"signal/accuracy_reward/centered_abs_mean": 0.1040798619389534,
"signal/accuracy_reward/group_std_mean": 0.1432553857564926,
"signal/accuracy_reward/group_zero_std_frac": 0.575,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0520399309694767,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0520399309694767,
"signal/advantage_abs_mean": 0.07406439036130905,
"signal/advantage_pre_scale_abs_mean": 0.07406439036130905,
"signal/advantage_pre_scale_std": 0.15997668206691742,
"signal/advantage_std": 0.15997668206691742,
"signal/brier_reward/centered_abs_mean": 0.06960970610380172,
"signal/brier_reward/group_std_mean": 0.0965530276298523,
"signal/brier_reward/group_zero_std_frac": 0.38333333730697633,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03480485305190086,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03480485305190086,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005631510308012367,
"signal/format_reward/group_std_mean": 0.0119167635217309,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0028157551540061833,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0028157551540061833,
"signal/mean_confidence_reward/centered_abs_mean": 0.05349176675081253,
"signal/mean_confidence_reward/group_std_mean": 0.07224587947130204,
"signal/mean_confidence_reward/group_zero_std_frac": 0.42777777910232545,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.349176603886008e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.349176603886008e-07,
"step": 860
},
{
"calibration/aurc": 0.1575388353347249,
"calibration/batch_distribution_entropy": 0.5194917173691976,
"calibration/batch_entropy_100bins": 0.2839643083536398,
"calibration/batch_entropy_10bins": 0.5194917173691976,
"calibration/batch_entropy_50bins": 0.33427818928988917,
"calibration/batch_uniqueness": 0.1407109572592486,
"calibration/confidence_entropy": 0.4323033214600315,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.3244043733681462,
"calibration/coverage@15%": 0.3395085400348129,
"calibration/coverage@20%": 0.8030466040923361,
"calibration/coverage@25%": 0.8850067666397823,
"calibration/coverage@30%": 0.9274151436031332,
"calibration/coverage@5%": 0.14255874673629243,
"calibration/distribution_entropy_10": 0.5194917173691976,
"calibration/distribution_entropy_100": 0.2839643083536398,
"calibration/ece": 0.10152744006283639,
"calibration/mean_confidence": 0.735262485675001,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.002430555555555558,
"completions/max_length": 3597.2,
"completions/max_terminated_length": 3597.2,
"completions/mean_length": 1077.8300537109376,
"completions/mean_terminated_length": 1080.4748779296874,
"completions/min_length": 73.6,
"completions/min_terminated_length": 315.8,
"epoch": 2.079326923076923,
"grad_norm": 0.0005686726071871817,
"learning_rate": 1.0516826923076925e-06,
"loss": -0.0021,
"num_tokens": 2332757787.0,
"reward": 1.3053359508514404,
"reward_std": 0.09737915992736816,
"rewards/accuracy_reward": 0.7456597328186035,
"rewards/brier_reward": 0.8675147771835328,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9974826335906982,
"rewards/mean_confidence_reward": 0.7421961665153504,
"sampling/batch_mean_priority_error": 0.04871301590709213,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011732461489737033,
"sampling/priority_kl": 0.029999954253435136,
"sampling/priority_scale": 0.8586566388374195,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 6.367937166942283e-05,
"sampling/prob_min": 2.058812024188228e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.0711999893188477,
"sampling/prompt_draws_total": 62136.0,
"sampling/seen_fraction": 0.9461133360862732,
"sampling/unseen_fraction": 0.053886663913726804,
"signal/accuracy_reward/centered_abs_mean": 0.10179036408662796,
"signal/accuracy_reward/group_std_mean": 0.1395990341901779,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333253860473,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05089518204331398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05089518204331398,
"signal/advantage_abs_mean": 0.06680571734905243,
"signal/advantage_pre_scale_abs_mean": 0.06680571734905243,
"signal/advantage_pre_scale_std": 0.14744619727134706,
"signal/advantage_std": 0.14744619727134706,
"signal/brier_reward/centered_abs_mean": 0.06380403712391854,
"signal/brier_reward/group_std_mean": 0.08844312876462937,
"signal/brier_reward/group_zero_std_frac": 0.3694444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03190201856195927,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03190201856195927,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004725477506872267,
"signal/format_reward/group_std_mean": 0.011614077165722847,
"signal/format_reward/group_zero_std_frac": 0.9416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0023627387534361334,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0023627387534361334,
"signal/mean_confidence_reward/centered_abs_mean": 0.05154569447040558,
"signal/mean_confidence_reward/group_std_mean": 0.06966388821601868,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4083333373069763,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.154569066689874e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.154569066689874e-07,
"step": 865
},
{
"calibration/aurc": 0.08651263729876864,
"calibration/batch_distribution_entropy": 0.5327170361876993,
"calibration/batch_entropy_100bins": 0.28920555960383576,
"calibration/batch_entropy_10bins": 0.5327170361876993,
"calibration/batch_entropy_50bins": 0.34044810545888504,
"calibration/batch_uniqueness": 0.1694106040126655,
"calibration/confidence_entropy": 0.430168374973589,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.6377406984334203,
"calibration/coverage@15%": 0.6711148281114012,
"calibration/coverage@20%": 0.8978078764142733,
"calibration/coverage@25%": 0.9056203764142733,
"calibration/coverage@30%": 0.9625,
"calibration/coverage@5%": 0.5829457680591819,
"calibration/distribution_entropy_10": 0.5327170361876993,
"calibration/distribution_entropy_100": 0.28920555960383576,
"calibration/ece": 0.11157011531766743,
"calibration/mean_confidence": 0.7343701044386424,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.002517361111111116,
"completions/max_length": 3796.6,
"completions/max_terminated_length": 3796.6,
"completions/mean_length": 1075.367041015625,
"completions/mean_terminated_length": 1078.114306640625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 211.0,
"epoch": 2.0913461538461537,
"grad_norm": 0.0004986703279428184,
"learning_rate": 1.0216346153846154e-06,
"loss": -0.0021,
"num_tokens": 2348222079.0,
"reward": 1.2846902847290038,
"reward_std": 0.08881339430809021,
"rewards/accuracy_reward": 0.7044270753860473,
"rewards/brier_reward": 0.867456591129303,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9974826335906982,
"rewards/mean_confidence_reward": 0.7102951288223267,
"sampling/batch_mean_priority_error": 0.0457432536502849,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0861111111111111,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011776031740009785,
"sampling/priority_kl": 0.03000001348555088,
"sampling/priority_scale": 0.8612726032966748,
"sampling/prob_entropy": 10.278960800170898,
"sampling/prob_max": 6.394868105417118e-05,
"sampling/prob_min": 2.060419101326261e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.083199977874756,
"sampling/prompt_draws_total": 62496.0,
"sampling/seen_fraction": 0.9472266674041748,
"sampling/unseen_fraction": 0.052773332595825194,
"signal/accuracy_reward/centered_abs_mean": 0.09128146767616271,
"signal/accuracy_reward/group_std_mean": 0.12983390986919402,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04564073383808136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04564073383808136,
"signal/advantage_abs_mean": 0.05929494649171829,
"signal/advantage_pre_scale_abs_mean": 0.05929494649171829,
"signal/advantage_pre_scale_std": 0.13471581488847734,
"signal/advantage_std": 0.13471581488847734,
"signal/brier_reward/centered_abs_mean": 0.05900585949420929,
"signal/brier_reward/group_std_mean": 0.0838774248957634,
"signal/brier_reward/group_zero_std_frac": 0.3361111134290695,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029502929747104646,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029502929747104646,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004682074580341577,
"signal/format_reward/group_std_mean": 0.011498793587088585,
"signal/format_reward/group_zero_std_frac": 0.9416666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0023410372901707887,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0023410372901707887,
"signal/mean_confidence_reward/centered_abs_mean": 0.051914072781801227,
"signal/mean_confidence_reward/group_std_mean": 0.06983351185917855,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3666666686534882,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.191406785343134e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.191406785343134e-07,
"step": 870
},
{
"calibration/aurc": 0.1437440651473768,
"calibration/batch_distribution_entropy": 0.5340035675488405,
"calibration/batch_entropy_100bins": 0.29457401150496026,
"calibration/batch_entropy_10bins": 0.5340035675488405,
"calibration/batch_entropy_50bins": 0.34676776017606487,
"calibration/batch_uniqueness": 0.16744557511469094,
"calibration/confidence_entropy": 0.43602664249424683,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1625,
"calibration/coverage@10%": 0.2,
"calibration/coverage@15%": 0.6401785714285715,
"calibration/coverage@20%": 0.8595238095238095,
"calibration/coverage@25%": 0.9212549603174605,
"calibration/coverage@30%": 0.9822916666666668,
"calibration/coverage@5%": 0.17135416666666667,
"calibration/distribution_entropy_10": 0.5340035675488405,
"calibration/distribution_entropy_100": 0.29457401150496026,
"calibration/ece": 0.1247387566137567,
"calibration/mean_confidence": 0.7314996693121695,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0035590277777777903,
"completions/max_length": 3852.8,
"completions/max_terminated_length": 3852.8,
"completions/mean_length": 1053.2385498046874,
"completions/mean_terminated_length": 1056.9432495117187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 266.8,
"epoch": 2.1033653846153846,
"grad_norm": 0.0005591860972344875,
"learning_rate": 9.915865384615386e-07,
"loss": -0.0038,
"num_tokens": 2363445243.0,
"reward": 1.3051762104034423,
"reward_std": 0.09470634311437606,
"rewards/accuracy_reward": 0.7506076455116272,
"rewards/brier_reward": 0.8633758664131165,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541746139526,
"rewards/mean_confidence_reward": 0.7379427194595337,
"sampling/batch_mean_priority_error": 0.0498768817204301,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011819823831319808,
"sampling/priority_kl": 0.03000010848045349,
"sampling/priority_scale": 0.8642309963935986,
"sampling/prob_entropy": 10.278946876525879,
"sampling/prob_max": 6.423962040571495e-05,
"sampling/prob_min": 2.0617561676772312e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.095199966430664,
"sampling/prompt_draws_total": 62856.0,
"sampling/seen_fraction": 0.9483733415603638,
"sampling/unseen_fraction": 0.05162665843963623,
"signal/accuracy_reward/centered_abs_mean": 0.09739040732383727,
"signal/accuracy_reward/group_std_mean": 0.1317952111363411,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04869520366191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04869520366191864,
"signal/advantage_abs_mean": 0.06749209389090538,
"signal/advantage_pre_scale_abs_mean": 0.06749209389090538,
"signal/advantage_pre_scale_std": 0.148349529504776,
"signal/advantage_std": 0.148349529504776,
"signal/brier_reward/centered_abs_mean": 0.06189800500869751,
"signal/brier_reward/group_std_mean": 0.08536944091320038,
"signal/brier_reward/group_zero_std_frac": 0.35555556416511536,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.030949002504348753,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.030949002504348753,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005902777658775449,
"signal/format_reward/group_std_mean": 0.012171501480042934,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0029513888293877244,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0029513888293877244,
"signal/mean_confidence_reward/centered_abs_mean": 0.05171496197581291,
"signal/mean_confidence_reward/group_std_mean": 0.0686635747551918,
"signal/mean_confidence_reward/group_zero_std_frac": 0.38611111640930174,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.171495786271407e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.171495786271407e-07,
"step": 875
},
{
"calibration/aurc": 0.14732994929428506,
"calibration/batch_distribution_entropy": 0.6149295123568695,
"calibration/batch_entropy_100bins": 0.3384844063253013,
"calibration/batch_entropy_10bins": 0.6149295123568695,
"calibration/batch_entropy_50bins": 0.3984583664943346,
"calibration/batch_uniqueness": 0.361323969608235,
"calibration/confidence_entropy": 0.45064628866814543,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.35448069172373886,
"calibration/coverage@15%": 0.6689921152698226,
"calibration/coverage@20%": 0.7378617845474553,
"calibration/coverage@25%": 0.8281974224114755,
"calibration/coverage@30%": 0.8766648397308836,
"calibration/coverage@5%": 0.16811463787085515,
"calibration/distribution_entropy_10": 0.6149295123568695,
"calibration/distribution_entropy_100": 0.3384844063253013,
"calibration/ece": 0.10045409393098485,
"calibration/mean_confidence": 0.6772221021352508,
"calibration/unique_confidence_per_question": 0.0234375,
"calibration/unique_confidences": 9.0,
"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": 0.001996527777777768,
"completions/max_length": 3781.6,
"completions/max_terminated_length": 3781.6,
"completions/mean_length": 1093.8227783203124,
"completions/mean_terminated_length": 1096.025732421875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 290.2,
"epoch": 2.1153846153846154,
"grad_norm": 0.0006124480860307813,
"learning_rate": 9.615384615384617e-07,
"loss": -0.0012,
"num_tokens": 2379156705.0,
"reward": 1.2899389505386352,
"reward_std": 0.10513612329959869,
"rewards/accuracy_reward": 0.7279513955116272,
"rewards/brier_reward": 0.8540824890136719,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9978298544883728,
"rewards/mean_confidence_reward": 0.7083072662353516,
"sampling/batch_mean_priority_error": 0.04806160534274193,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.01186504978686571,
"sampling/priority_kl": 0.030000089108943938,
"sampling/priority_scale": 0.8671992361778393,
"sampling/prob_entropy": 10.278956985473632,
"sampling/prob_max": 6.45339023321867e-05,
"sampling/prob_min": 2.063132560579106e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.107200002670288,
"sampling/prompt_draws_total": 63216.0,
"sampling/seen_fraction": 0.9494999885559082,
"sampling/unseen_fraction": 0.050500011444091795,
"signal/accuracy_reward/centered_abs_mean": 0.11345485895872116,
"signal/accuracy_reward/group_std_mean": 0.15268072187900544,
"signal/accuracy_reward/group_zero_std_frac": 0.5444444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05672742947936058,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05672742947936058,
"signal/advantage_abs_mean": 0.07435817122459412,
"signal/advantage_pre_scale_abs_mean": 0.07435817122459412,
"signal/advantage_pre_scale_std": 0.15249708890914918,
"signal/advantage_std": 0.15249708890914918,
"signal/brier_reward/centered_abs_mean": 0.07330766320228577,
"signal/brier_reward/group_std_mean": 0.0994098201394081,
"signal/brier_reward/group_zero_std_frac": 0.2833333402872086,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03665383160114288,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03665383160114288,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004161241254769266,
"signal/format_reward/group_std_mean": 0.011080396175384522,
"signal/format_reward/group_zero_std_frac": 0.9416666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002080620627384633,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002080620627384633,
"signal/mean_confidence_reward/centered_abs_mean": 0.0575765036046505,
"signal/mean_confidence_reward/group_std_mean": 0.07623594254255295,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32222222685813906,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.757650228588318e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.757650228588318e-07,
"step": 880
},
{
"calibration/aurc": 0.06487215245338737,
"calibration/batch_distribution_entropy": 0.5959776527718124,
"calibration/batch_entropy_100bins": 0.32927574902598833,
"calibration/batch_entropy_10bins": 0.5959776527718124,
"calibration/batch_entropy_50bins": 0.38761808411641013,
"calibration/batch_uniqueness": 0.3361763733804129,
"calibration/confidence_entropy": 0.45123354706391766,
"calibration/coverage@0%": 0.1094240837696335,
"calibration/coverage@1%": 0.3497292018798697,
"calibration/coverage@10%": 0.651924708090328,
"calibration/coverage@15%": 0.8676686385365221,
"calibration/coverage@20%": 0.958231143117849,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5625324542038649,
"calibration/distribution_entropy_10": 0.5959776527718124,
"calibration/distribution_entropy_100": 0.32927574902598833,
"calibration/ece": 0.12659779976726399,
"calibration/mean_confidence": 0.7134376814647153,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.0035590277777777677,
"completions/max_length": 3849.2,
"completions/max_terminated_length": 3849.2,
"completions/mean_length": 1097.7361328125,
"completions/mean_terminated_length": 1101.774560546875,
"completions/min_length": 58.2,
"completions/min_terminated_length": 258.6,
"epoch": 2.1274038461538463,
"grad_norm": 0.000581674394197762,
"learning_rate": 9.314903846153847e-07,
"loss": -0.0013,
"num_tokens": 2394883937.0,
"reward": 1.3241001129150392,
"reward_std": 0.09584866911172867,
"rewards/accuracy_reward": 0.7738715291023255,
"rewards/brier_reward": 0.8779600858688354,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9963541626930237,
"rewards/mean_confidence_reward": 0.7227430582046509,
"sampling/batch_mean_priority_error": 0.039053923312061485,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555555,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011903371289372444,
"sampling/priority_kl": 0.030000071600079538,
"sampling/priority_scale": 0.8695811093086376,
"sampling/prob_entropy": 10.27895450592041,
"sampling/prob_max": 6.479611329268664e-05,
"sampling/prob_min": 2.0650375154218638e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.1191999912261963,
"sampling/prompt_draws_total": 63576.0,
"sampling/seen_fraction": 0.9504666686058044,
"sampling/unseen_fraction": 0.04953333139419556,
"signal/accuracy_reward/centered_abs_mean": 0.10227321982383727,
"signal/accuracy_reward/group_std_mean": 0.1415991008281708,
"signal/accuracy_reward/group_zero_std_frac": 0.5666666746139526,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05113660991191864,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05113660991191864,
"signal/advantage_abs_mean": 0.06633429676294327,
"signal/advantage_pre_scale_abs_mean": 0.06633429676294327,
"signal/advantage_pre_scale_std": 0.14381582289934158,
"signal/advantage_std": 0.14381582289934158,
"signal/brier_reward/centered_abs_mean": 0.05829312801361084,
"signal/brier_reward/group_std_mean": 0.08310620188713073,
"signal/brier_reward/group_zero_std_frac": 0.32222222685813906,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02914656400680542,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02914656400680542,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0064561630017124115,
"signal/format_reward/group_std_mean": 0.013242663256824017,
"signal/format_reward/group_zero_std_frac": 0.9416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0032280815008562057,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0032280815008562057,
"signal/mean_confidence_reward/centered_abs_mean": 0.05448948219418526,
"signal/mean_confidence_reward/group_std_mean": 0.07169951945543289,
"signal/mean_confidence_reward/group_zero_std_frac": 0.347222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.448948002140241e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.448948002140241e-07,
"step": 885
},
{
"calibration/aurc": 0.11436532695088564,
"calibration/batch_distribution_entropy": 0.54461853581717,
"calibration/batch_entropy_100bins": 0.30666049338838297,
"calibration/batch_entropy_10bins": 0.54461853581717,
"calibration/batch_entropy_50bins": 0.36099577109159164,
"calibration/batch_uniqueness": 0.21106072782428095,
"calibration/confidence_entropy": 0.43878352734476334,
"calibration/coverage@0%": 0.13769633507853402,
"calibration/coverage@1%": 0.15130890052356022,
"calibration/coverage@10%": 0.6097540357766144,
"calibration/coverage@15%": 0.653125,
"calibration/coverage@20%": 0.8551226610095736,
"calibration/coverage@25%": 0.8858967036553524,
"calibration/coverage@30%": 0.9453125,
"calibration/coverage@5%": 0.34565063263525303,
"calibration/distribution_entropy_10": 0.54461853581717,
"calibration/distribution_entropy_100": 0.30666049338838297,
"calibration/ece": 0.12622460302949523,
"calibration/mean_confidence": 0.7298314014519798,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.002256944444444442,
"completions/max_length": 3788.2,
"completions/max_terminated_length": 3788.2,
"completions/mean_length": 1056.4462036132813,
"completions/mean_terminated_length": 1058.8260498046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 211.0,
"epoch": 2.139423076923077,
"grad_norm": 0.0004920501378364861,
"learning_rate": 9.014423076923078e-07,
"loss": -0.0018,
"num_tokens": 2410149109.0,
"reward": 1.3044902086257935,
"reward_std": 0.08810825049877166,
"rewards/accuracy_reward": 0.7372395753860473,
"rewards/brier_reward": 0.8739834904670716,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9977430582046509,
"rewards/mean_confidence_reward": 0.7171787977218628,
"sampling/batch_mean_priority_error": 0.038951622423835125,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011934378184378147,
"sampling/priority_kl": 0.029999952390789987,
"sampling/priority_scale": 0.87255815861281,
"sampling/prob_entropy": 10.278945159912109,
"sampling/prob_max": 6.509524682769552e-05,
"sampling/prob_min": 2.0664640032919124e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.1311999797821044,
"sampling/prompt_draws_total": 63936.0,
"sampling/seen_fraction": 0.95152667760849,
"sampling/unseen_fraction": 0.04847332239151001,
"signal/accuracy_reward/centered_abs_mean": 0.09378797709941863,
"signal/accuracy_reward/group_std_mean": 0.12828219532966614,
"signal/accuracy_reward/group_zero_std_frac": 0.6166666865348815,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04689398854970932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04689398854970932,
"signal/advantage_abs_mean": 0.06113003343343735,
"signal/advantage_pre_scale_abs_mean": 0.06113003343343735,
"signal/advantage_pre_scale_std": 0.13656453937292098,
"signal/advantage_std": 0.13656453937292098,
"signal/brier_reward/centered_abs_mean": 0.05816308483481407,
"signal/brier_reward/group_std_mean": 0.08063021898269654,
"signal/brier_reward/group_zero_std_frac": 0.32222222685813906,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029081542417407034,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029081542417407034,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004101562453433872,
"signal/format_reward/group_std_mean": 0.009300821647047996,
"signal/format_reward/group_zero_std_frac": 0.9555555701255798,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002050781226716936,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002050781226716936,
"signal/mean_confidence_reward/centered_abs_mean": 0.052949226647615436,
"signal/mean_confidence_reward/group_std_mean": 0.06989138722419738,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33611111640930175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.294922686971404e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.294922686971404e-07,
"step": 890
},
{
"calibration/aurc": 0.057605759494229726,
"calibration/batch_distribution_entropy": 0.5504742943258529,
"calibration/batch_entropy_100bins": 0.2990651170025676,
"calibration/batch_entropy_10bins": 0.5504742943258529,
"calibration/batch_entropy_50bins": 0.35205461690237005,
"calibration/batch_uniqueness": 0.17962814847079142,
"calibration/confidence_entropy": 0.42971862399724775,
"calibration/coverage@0%": 0.12063492063492062,
"calibration/coverage@1%": 0.5436541135483065,
"calibration/coverage@10%": 0.7180638878473524,
"calibration/coverage@15%": 0.8899470899470898,
"calibration/coverage@20%": 0.959755291005291,
"calibration/coverage@25%": 0.9796875,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6211786026746656,
"calibration/distribution_entropy_10": 0.5504742943258529,
"calibration/distribution_entropy_100": 0.2990651170025676,
"calibration/ece": 0.12365266190684492,
"calibration/mean_confidence": 0.7442941962983795,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.005034722222222188,
"completions/max_length": 3986.4,
"completions/max_terminated_length": 3986.4,
"completions/mean_length": 1097.265283203125,
"completions/mean_terminated_length": 1102.853125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 281.8,
"epoch": 2.1514423076923075,
"grad_norm": 0.000565341382753104,
"learning_rate": 8.713942307692308e-07,
"loss": -0.0045,
"num_tokens": 2425879621.0,
"reward": 1.309638214111328,
"reward_std": 0.0923440083861351,
"rewards/accuracy_reward": 0.7471354126930236,
"rewards/brier_reward": 0.8773350715637207,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9947916626930237,
"rewards/mean_confidence_reward": 0.7131249904632568,
"sampling/batch_mean_priority_error": 0.03646324516079373,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.1388888888888889,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.011969214305281639,
"sampling/priority_kl": 0.02999997101724148,
"sampling/priority_scale": 0.8769719779258593,
"sampling/prob_entropy": 10.278958511352538,
"sampling/prob_max": 6.547861557919532e-05,
"sampling/prob_min": 2.0666010095737873e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.1432000160217286,
"sampling/prompt_draws_total": 64296.0,
"sampling/seen_fraction": 0.9529199957847595,
"sampling/unseen_fraction": 0.04708000421524048,
"signal/accuracy_reward/centered_abs_mean": 0.0927463099360466,
"signal/accuracy_reward/group_std_mean": 0.13175542801618575,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0463731549680233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0463731549680233,
"signal/advantage_abs_mean": 0.06347616985440255,
"signal/advantage_pre_scale_abs_mean": 0.06347616985440255,
"signal/advantage_pre_scale_std": 0.14331293106079102,
"signal/advantage_std": 0.14331293106079102,
"signal/brier_reward/centered_abs_mean": 0.054565589874982834,
"signal/brier_reward/group_std_mean": 0.07603787034749984,
"signal/brier_reward/group_zero_std_frac": 0.3222222179174423,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027282794937491417,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027282794937491417,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.008865017257630826,
"signal/format_reward/group_std_mean": 0.0172776997089386,
"signal/format_reward/group_zero_std_frac": 0.9250000238418579,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004432508628815413,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004432508628815413,
"signal/mean_confidence_reward/centered_abs_mean": 0.05359158664941788,
"signal/mean_confidence_reward/group_std_mean": 0.07256897538900375,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3277777761220932,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.359158478768223e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.359158478768223e-07,
"step": 895
},
{
"calibration/aurc": 0.13548690825922524,
"calibration/batch_distribution_entropy": 0.6432022400707609,
"calibration/batch_entropy_100bins": 0.35657053840889996,
"calibration/batch_entropy_10bins": 0.6432022400707609,
"calibration/batch_entropy_50bins": 0.4197490685519815,
"calibration/batch_uniqueness": 0.42259828402414107,
"calibration/confidence_entropy": 0.45781386644283095,
"calibration/coverage@0%": 0.08753315649867374,
"calibration/coverage@1%": 0.2184471331315664,
"calibration/coverage@10%": 0.6113231951206562,
"calibration/coverage@15%": 0.6401218026054255,
"calibration/coverage@20%": 0.6745257012787038,
"calibration/coverage@25%": 0.6969257012787039,
"calibration/coverage@30%": 0.8731083994033257,
"calibration/coverage@5%": 0.4549450201189825,
"calibration/distribution_entropy_10": 0.6432022400707609,
"calibration/distribution_entropy_100": 0.35657053840889996,
"calibration/ece": 0.1357104909707279,
"calibration/mean_confidence": 0.6763972364317309,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.004079861111111138,
"completions/max_length": 3848.0,
"completions/max_terminated_length": 3848.0,
"completions/mean_length": 1101.91591796875,
"completions/mean_terminated_length": 1106.378564453125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 263.0,
"epoch": 2.1634615384615383,
"grad_norm": 0.000522150716278702,
"learning_rate": 8.41346153846154e-07,
"loss": -0.0037,
"num_tokens": 2441668316.0,
"reward": 1.2985861778259278,
"reward_std": 0.09206021130084992,
"rewards/accuracy_reward": 0.7346354126930237,
"rewards/brier_reward": 0.8666892528533936,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9958333373069763,
"rewards/mean_confidence_reward": 0.7220485806465149,
"sampling/batch_mean_priority_error": 0.04377663136510779,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.10277777777777777,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012008808366954327,
"sampling/priority_kl": 0.029999631643295287,
"sampling/priority_scale": 0.8818507730262354,
"sampling/prob_entropy": 10.27894229888916,
"sampling/prob_max": 6.588545220438391e-05,
"sampling/prob_min": 2.0431984739843757e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.1552000045776367,
"sampling/prompt_draws_total": 64656.0,
"sampling/seen_fraction": 0.9543733358383178,
"sampling/unseen_fraction": 0.04562666416168213,
"signal/accuracy_reward/centered_abs_mean": 0.08727756142616272,
"signal/accuracy_reward/group_std_mean": 0.1237468108534813,
"signal/accuracy_reward/group_zero_std_frac": 0.6138888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04363878071308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04363878071308136,
"signal/advantage_abs_mean": 0.06414745002985,
"signal/advantage_pre_scale_abs_mean": 0.06414745002985,
"signal/advantage_pre_scale_std": 0.14303337037563324,
"signal/advantage_std": 0.14303337037563324,
"signal/brier_reward/centered_abs_mean": 0.059873423725366595,
"signal/brier_reward/group_std_mean": 0.08284576088190079,
"signal/brier_reward/group_zero_std_frac": 0.325,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029936711862683298,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029936711862683298,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0073242187965661286,
"signal/format_reward/group_std_mean": 0.014999254047870636,
"signal/format_reward/group_zero_std_frac": 0.9333333492279052,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0036621093982830643,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0036621093982830643,
"signal/mean_confidence_reward/centered_abs_mean": 0.0572992704808712,
"signal/mean_confidence_reward/group_std_mean": 0.07567873746156692,
"signal/mean_confidence_reward/group_zero_std_frac": 0.34166666865348816,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.729926783715201e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.729926783715201e-07,
"step": 900
},
{
"epoch": 2.1634615384615383,
"eval_calibration/aurc": 0.1267359752025158,
"eval_calibration/batch_distribution_entropy": 0.6113133851546924,
"eval_calibration/batch_entropy_100bins": 0.3379810892749218,
"eval_calibration/batch_entropy_10bins": 0.6113133851546924,
"eval_calibration/batch_entropy_50bins": 0.3978658698062295,
"eval_calibration/batch_uniqueness": 0.31794023677469857,
"eval_calibration/confidence_entropy": 0.44411410109398053,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.6387434554973822,
"eval_calibration/coverage@15%": 0.7050610820244329,
"eval_calibration/coverage@20%": 0.7827225130890052,
"eval_calibration/coverage@25%": 0.9153577661431065,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.6113133851546924,
"eval_calibration/distribution_entropy_100": 0.3379810892749218,
"eval_calibration/ece": 0.027225130890052296,
"eval_calibration/mean_confidence": 0.7122164048865621,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.005208333333333352,
"eval_completions/max_length": 2968.0,
"eval_completions/max_terminated_length": 2968.0,
"eval_completions/mean_length": 1082.8380432128906,
"eval_completions/mean_terminated_length": 1088.5561625162761,
"eval_completions/min_length": 123.16666666666667,
"eval_completions/min_terminated_length": 327.1666666666667,
"eval_loss": 0.0,
"eval_num_tokens": 2441668316.0,
"eval_reward": 1.2883838415145874,
"eval_reward_std": 0.2966083884239197,
"eval_rewards/accuracy_reward": 0.7239583233992258,
"eval_rewards/brier_reward": 0.8580034772555033,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9947916766007742,
"eval_rewards/mean_confidence_reward": 0.7085069219271342,
"eval_runtime": 198.2183,
"eval_samples_per_second": 5.045,
"eval_signal/accuracy_reward/centered_abs_mean": 0.384006068110466,
"eval_signal/accuracy_reward/group_std_mean": 0.44363180299599964,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.192003034055233,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.192003034055233,
"eval_signal/advantage_abs_mean": 0.24637722472349802,
"eval_signal/advantage_pre_scale_abs_mean": 0.24637722472349802,
"eval_signal/advantage_pre_scale_std": 0.2960158785184224,
"eval_signal/advantage_std": 0.2960158785184224,
"eval_signal/brier_reward/centered_abs_mean": 0.1627978632847468,
"eval_signal/brier_reward/group_std_mean": 0.2185137296716372,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0813989316423734,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.0813989316423734,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.009982638837148746,
"eval_signal/format_reward/group_std_mean": 0.026473373485108215,
"eval_signal/format_reward/group_zero_std_frac": 0.8611111342906952,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.004991319418574373,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.004991319418574373,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.22192923724651337,
"eval_signal/mean_confidence_reward/group_std_mean": 0.2581739276647568,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.219292279429889e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.219292279429889e-06,
"eval_steps_per_second": 0.03,
"step": 900
},
{
"epoch": 2.1634615384615383,
"step": 900,
"train_probe_calibration/aurc": 0.07629200313769308,
"train_probe_calibration/batch_distribution_entropy": 0.5928192578652658,
"train_probe_calibration/batch_entropy_100bins": 0.3262678262764245,
"train_probe_calibration/batch_entropy_10bins": 0.5928192578652658,
"train_probe_calibration/batch_entropy_50bins": 0.3840772061234059,
"train_probe_calibration/batch_uniqueness": 0.2719269274041304,
"train_probe_calibration/confidence_entropy": 0.4367140091328539,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7510917030567685,
"train_probe_calibration/coverage@15%": 0.8864628820960698,
"train_probe_calibration/coverage@20%": 0.9877729257641922,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.5807860262008734,
"train_probe_calibration/distribution_entropy_10": 0.5928192578652658,
"train_probe_calibration/distribution_entropy_100": 0.3262678262764245,
"train_probe_calibration/ece": 0.062270742358078564,
"train_probe_calibration/mean_confidence": 0.7316157205240175,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.009375000000000022,
"train_probe_completions/max_length": 3233.8333333333335,
"train_probe_completions/max_terminated_length": 3233.8333333333335,
"train_probe_completions/mean_length": 1095.05419921875,
"train_probe_completions/mean_terminated_length": 1105.2533569335938,
"train_probe_completions/min_length": 0.0,
"train_probe_completions/min_terminated_length": 253.33333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2441668316.0,
"train_probe_reward": 1.3332928816477458,
"train_probe_reward_std": 0.2719416369994481,
"train_probe_rewards/accuracy_reward": 0.7890625,
"train_probe_rewards/brier_reward": 0.8835850258668264,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.9939236243565878,
"train_probe_rewards/mean_confidence_reward": 0.7271700998147329,
"train_probe_runtime": 209.3306,
"train_probe_samples_per_second": 4.777,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3241644948720932,
"train_probe_signal/accuracy_reward/group_std_mean": 0.4060422231753667,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.1620822474360466,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.1620822474360466,
"train_probe_signal/advantage_abs_mean": 0.21062554667393366,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.21062554667393366,
"train_probe_signal/advantage_pre_scale_std": 0.2726934055487315,
"train_probe_signal/advantage_std": 0.2726934055487315,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13750815515716872,
"train_probe_signal/brier_reward/group_std_mean": 0.1956165755788485,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06875407757858436,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06875407757858436,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.011773003110041222,
"train_probe_signal/format_reward/group_std_mean": 0.034373246443768345,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8055555919806162,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.005886501555020611,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.20697697500387827,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2474315663178762,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.069769796738304e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.069769796738304e-06,
"train_probe_steps_per_second": 0.029
},
{
"calibration/aurc": 0.1144883544149797,
"calibration/batch_distribution_entropy": 0.5319967062043351,
"calibration/batch_entropy_100bins": 0.2956503117019777,
"calibration/batch_entropy_10bins": 0.5319967062043351,
"calibration/batch_entropy_50bins": 0.3480347633536025,
"calibration/batch_uniqueness": 0.19064984252784528,
"calibration/confidence_entropy": 0.42376670423195695,
"calibration/coverage@0%": 0.27033204379861386,
"calibration/coverage@1%": 0.30852733312372693,
"calibration/coverage@10%": 0.6185998121936696,
"calibration/coverage@15%": 0.6751399529339014,
"calibration/coverage@20%": 0.8704714912280702,
"calibration/coverage@25%": 0.944248903508772,
"calibration/coverage@30%": 0.9478947368421053,
"calibration/coverage@5%": 0.41888998328665156,
"calibration/distribution_entropy_10": 0.5319967062043351,
"calibration/distribution_entropy_100": 0.2956503117019777,
"calibration/ece": 0.13160386432329638,
"calibration/mean_confidence": 0.7213281817296651,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.005295138888888884,
"completions/max_length": 3856.8,
"completions/max_terminated_length": 3856.8,
"completions/mean_length": 1053.6728515625,
"completions/mean_terminated_length": 1059.3394897460937,
"completions/min_length": 0.0,
"completions/min_terminated_length": 294.4,
"epoch": 2.175480769230769,
"grad_norm": 0.0004946622648276389,
"learning_rate": 8.11298076923077e-07,
"loss": -0.0064,
"num_tokens": 2456922115.0,
"reward": 1.3134698629379273,
"reward_std": 0.08974100053310394,
"rewards/accuracy_reward": 0.7703125,
"rewards/brier_reward": 0.8621683955192566,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9944444417953491,
"rewards/mean_confidence_reward": 0.7224652528762817,
"sampling/batch_mean_priority_error": 0.046409821812790915,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.09444444444444444,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012050377577543259,
"sampling/priority_kl": 0.029999915137887002,
"sampling/priority_scale": 0.8856883347267285,
"sampling/prob_entropy": 10.278950691223145,
"sampling/prob_max": 6.623360386583954e-05,
"sampling/prob_min": 2.037807607848663e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.167199993133545,
"sampling/prompt_draws_total": 65016.0,
"sampling/seen_fraction": 0.9555866718292236,
"sampling/unseen_fraction": 0.04441332817077637,
"signal/accuracy_reward/centered_abs_mean": 0.09505208283662796,
"signal/accuracy_reward/group_std_mean": 0.12743898183107377,
"signal/accuracy_reward/group_zero_std_frac": 0.625,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04752604141831398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04752604141831398,
"signal/advantage_abs_mean": 0.0632380336523056,
"signal/advantage_pre_scale_abs_mean": 0.0632380336523056,
"signal/advantage_pre_scale_std": 0.14537188410758972,
"signal/advantage_std": 0.14537188410758972,
"signal/brier_reward/centered_abs_mean": 0.05927918329834938,
"signal/brier_reward/group_std_mean": 0.08104176968336105,
"signal/brier_reward/group_zero_std_frac": 0.3555555582046509,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02963959164917469,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02963959164917469,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009722222108393908,
"signal/format_reward/group_std_mean": 0.018346155434846877,
"signal/format_reward/group_zero_std_frac": 0.925000011920929,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004861111054196954,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004861111054196954,
"signal/mean_confidence_reward/centered_abs_mean": 0.0496441125869751,
"signal/mean_confidence_reward/group_std_mean": 0.06583335101604462,
"signal/mean_confidence_reward/group_zero_std_frac": 0.37222222089767454,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.964410720731393e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.964410720731393e-07,
"step": 905
},
{
"calibration/aurc": 0.17540491350213605,
"calibration/batch_distribution_entropy": 0.5520565151511247,
"calibration/batch_entropy_100bins": 0.30991420207134707,
"calibration/batch_entropy_10bins": 0.5520565151511247,
"calibration/batch_entropy_50bins": 0.3648259843086116,
"calibration/batch_uniqueness": 0.25779624289484376,
"calibration/confidence_entropy": 0.448527313953052,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.2859375,
"calibration/coverage@15%": 0.4619791666666667,
"calibration/coverage@20%": 0.7005208333333333,
"calibration/coverage@25%": 0.7848958333333333,
"calibration/coverage@30%": 0.8,
"calibration/coverage@5%": 0.22239583333333335,
"calibration/distribution_entropy_10": 0.5520565151511247,
"calibration/distribution_entropy_100": 0.30991420207134707,
"calibration/ece": 0.12975155026109664,
"calibration/mean_confidence": 0.734895425369887,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.0043402777777777676,
"completions/max_length": 4015.4,
"completions/max_terminated_length": 4015.4,
"completions/mean_length": 1089.65390625,
"completions/mean_terminated_length": 1094.3900268554687,
"completions/min_length": 0.0,
"completions/min_terminated_length": 292.8,
"epoch": 2.1875,
"grad_norm": 0.000545768765732646,
"learning_rate": 7.8125e-07,
"loss": -0.0041,
"num_tokens": 2472536048.0,
"reward": 1.3016733169555663,
"reward_std": 0.09393489807844162,
"rewards/accuracy_reward": 0.74921875,
"rewards/brier_reward": 0.8584540009498596,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9956597208976745,
"rewards/mean_confidence_reward": 0.7101996660232544,
"sampling/batch_mean_priority_error": 0.0466008042900245,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06944444444444445,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.01208706684410572,
"sampling/priority_kl": 0.03000006526708603,
"sampling/priority_scale": 0.8887015282874927,
"sampling/prob_entropy": 10.278955459594727,
"sampling/prob_max": 6.653684540651739e-05,
"sampling/prob_min": 2.038945494859945e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.179200029373169,
"sampling/prompt_draws_total": 65376.0,
"sampling/seen_fraction": 0.9565733432769775,
"sampling/unseen_fraction": 0.04342665672302246,
"signal/accuracy_reward/centered_abs_mean": 0.09839952141046523,
"signal/accuracy_reward/group_std_mean": 0.13703945875167847,
"signal/accuracy_reward/group_zero_std_frac": 0.5777777791023254,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04919976070523262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04919976070523262,
"signal/advantage_abs_mean": 0.06482230573892593,
"signal/advantage_pre_scale_abs_mean": 0.06482230573892593,
"signal/advantage_pre_scale_std": 0.14661589562892913,
"signal/advantage_std": 0.14661589562892913,
"signal/brier_reward/centered_abs_mean": 0.06251252591609954,
"signal/brier_reward/group_std_mean": 0.08707078844308853,
"signal/brier_reward/group_zero_std_frac": 0.31388888955116273,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03125626295804977,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03125626295804977,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0076388888992369175,
"signal/format_reward/group_std_mean": 0.014412016607820988,
"signal/format_reward/group_zero_std_frac": 0.9416666865348816,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0038194444496184587,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0038194444496184587,
"signal/mean_confidence_reward/centered_abs_mean": 0.05177355259656906,
"signal/mean_confidence_reward/group_std_mean": 0.07010893672704696,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3444444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.177355092200742e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.177355092200742e-07,
"step": 910
},
{
"calibration/aurc": 0.10931295708207929,
"calibration/batch_distribution_entropy": 0.6055723892265433,
"calibration/batch_entropy_100bins": 0.33135697825150656,
"calibration/batch_entropy_10bins": 0.6055723892265433,
"calibration/batch_entropy_50bins": 0.3900680734866835,
"calibration/batch_uniqueness": 0.3495132621356931,
"calibration/confidence_entropy": 0.451201528098253,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.07172774869109948,
"calibration/coverage@10%": 0.46993073734729496,
"calibration/coverage@15%": 0.6534385907504363,
"calibration/coverage@20%": 0.8523969240837698,
"calibration/coverage@25%": 0.978125,
"calibration/coverage@30%": 0.99375,
"calibration/coverage@5%": 0.41461878272251307,
"calibration/distribution_entropy_10": 0.6055723892265433,
"calibration/distribution_entropy_100": 0.33135697825150656,
"calibration/ece": 0.11808518760907508,
"calibration/mean_confidence": 0.7256222731239093,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002256944444444442,
"completions/max_length": 3932.4,
"completions/max_terminated_length": 3932.4,
"completions/mean_length": 1129.2636474609376,
"completions/mean_terminated_length": 1131.8522216796875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 258.6,
"epoch": 2.199519230769231,
"grad_norm": 0.0006008677300997078,
"learning_rate": 7.512019230769231e-07,
"loss": -0.0007,
"num_tokens": 2488636077.0,
"reward": 1.3290406227111817,
"reward_std": 0.09223742932081222,
"rewards/accuracy_reward": 0.7696180462837219,
"rewards/brier_reward": 0.8907925248146057,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99765625,
"rewards/mean_confidence_reward": 0.719887125492096,
"sampling/batch_mean_priority_error": 0.029617764136904756,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888889,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012118054181337356,
"sampling/priority_kl": 0.030000020191073416,
"sampling/priority_scale": 0.8916841447120533,
"sampling/prob_entropy": 10.278943252563476,
"sampling/prob_max": 6.68402062729001e-05,
"sampling/prob_min": 2.040126273641363e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.191200017929077,
"sampling/prompt_draws_total": 65736.0,
"sampling/seen_fraction": 0.9575266718864441,
"sampling/unseen_fraction": 0.042473328113555905,
"signal/accuracy_reward/centered_abs_mean": 0.09581163227558136,
"signal/accuracy_reward/group_std_mean": 0.13503813445568086,
"signal/accuracy_reward/group_zero_std_frac": 0.5861111283302307,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04790581613779068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04790581613779068,
"signal/advantage_abs_mean": 0.06331820115447044,
"signal/advantage_pre_scale_abs_mean": 0.06331820115447044,
"signal/advantage_pre_scale_std": 0.1418493792414665,
"signal/advantage_std": 0.1418493792414665,
"signal/brier_reward/centered_abs_mean": 0.0544912040233612,
"signal/brier_reward/group_std_mean": 0.07728255391120911,
"signal/brier_reward/group_zero_std_frac": 0.31666666865348814,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0272456020116806,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.0272456020116806,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004443359316792339,
"signal/format_reward/group_std_mean": 0.011330607160925866,
"signal/format_reward/group_zero_std_frac": 0.9416666746139526,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0022216796583961695,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0022216796583961695,
"signal/mean_confidence_reward/centered_abs_mean": 0.051905933767557144,
"signal/mean_confidence_reward/group_std_mean": 0.07016979679465293,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3361111104488373,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.190593242332397e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.190593242332397e-07,
"step": 915
},
{
"calibration/aurc": 0.055508667022656444,
"calibration/batch_distribution_entropy": 0.5734453863656327,
"calibration/batch_entropy_100bins": 0.3165289962959326,
"calibration/batch_entropy_10bins": 0.5734453863656327,
"calibration/batch_entropy_50bins": 0.3726128130433198,
"calibration/batch_uniqueness": 0.24339880295676136,
"calibration/confidence_entropy": 0.4383371403850176,
"calibration/coverage@0%": 0.20964709762532982,
"calibration/coverage@1%": 0.48854029243623565,
"calibration/coverage@10%": 0.7930299032541777,
"calibration/coverage@15%": 0.9140625,
"calibration/coverage@20%": 0.9484375,
"calibration/coverage@25%": 0.984375,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.5647399956024627,
"calibration/distribution_entropy_10": 0.5734453863656327,
"calibration/distribution_entropy_100": 0.3165289962959326,
"calibration/ece": 0.11776247801231299,
"calibration/mean_confidence": 0.7259812005277047,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.003472222222222232,
"completions/max_length": 3983.6,
"completions/max_terminated_length": 3983.6,
"completions/mean_length": 1101.6866455078125,
"completions/mean_terminated_length": 1105.601123046875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 276.0,
"epoch": 2.2115384615384617,
"grad_norm": 0.0005841179518029094,
"learning_rate": 7.211538461538461e-07,
"loss": -0.0028,
"num_tokens": 2504442323.0,
"reward": 1.3245359659194946,
"reward_std": 0.0880805030465126,
"rewards/accuracy_reward": 0.7708333373069763,
"rewards/brier_reward": 0.8817830443382263,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9964409828186035,
"rewards/mean_confidence_reward": 0.7301736116409302,
"sampling/batch_mean_priority_error": 0.037492953171225175,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.07777777777777778,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012144725769758224,
"sampling/priority_kl": 0.029999976605176927,
"sampling/priority_scale": 0.8953496039146558,
"sampling/prob_entropy": 10.278950691223145,
"sampling/prob_max": 6.718820804962888e-05,
"sampling/prob_min": 2.0407688498380595e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2032000064849853,
"sampling/prompt_draws_total": 66096.0,
"sampling/seen_fraction": 0.9585866570472718,
"sampling/unseen_fraction": 0.04141334295272827,
"signal/accuracy_reward/centered_abs_mean": 0.08761935830116271,
"signal/accuracy_reward/group_std_mean": 0.12568885535001756,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04380967915058136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04380967915058136,
"signal/advantage_abs_mean": 0.05975606217980385,
"signal/advantage_pre_scale_abs_mean": 0.05975606217980385,
"signal/advantage_pre_scale_std": 0.1377297580242157,
"signal/advantage_std": 0.1377297580242157,
"signal/brier_reward/centered_abs_mean": 0.057367566227912906,
"signal/brier_reward/group_std_mean": 0.08021015971899033,
"signal/brier_reward/group_zero_std_frac": 0.3444444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028683783113956453,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028683783113956453,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005560980923473835,
"signal/format_reward/group_std_mean": 0.011116351373493672,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0027804904617369174,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0027804904617369174,
"signal/mean_confidence_reward/centered_abs_mean": 0.051011838018894196,
"signal/mean_confidence_reward/group_std_mean": 0.06827212944626808,
"signal/mean_confidence_reward/group_zero_std_frac": 0.37222222089767454,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.101183717215463e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.101183717215463e-07,
"step": 920
},
{
"calibration/aurc": 0.09897701541147466,
"calibration/batch_distribution_entropy": 0.5075802874075744,
"calibration/batch_entropy_100bins": 0.27116894314853723,
"calibration/batch_entropy_10bins": 0.5075802874075744,
"calibration/batch_entropy_50bins": 0.319215692397717,
"calibration/batch_uniqueness": 0.07614247393287564,
"calibration/confidence_entropy": 0.4189456828768733,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5289939621409921,
"calibration/coverage@15%": 0.8849040498602905,
"calibration/coverage@20%": 0.9418345472035179,
"calibration/coverage@25%": 0.9769781503366772,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.44502420583115754,
"calibration/distribution_entropy_10": 0.5075802874075744,
"calibration/distribution_entropy_100": 0.27116894314853723,
"calibration/ece": 0.11424696961889054,
"calibration/mean_confidence": 0.7663294956140352,
"calibration/unique_confidence_per_question": 0.022395833333333334,
"calibration/unique_confidences": 8.6,
"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": 0.0021701388888888838,
"completions/max_length": 3859.6,
"completions/max_terminated_length": 3859.6,
"completions/mean_length": 1110.97822265625,
"completions/mean_terminated_length": 1113.5282958984376,
"completions/min_length": 0.0,
"completions/min_terminated_length": 310.4,
"epoch": 2.2235576923076925,
"grad_norm": 0.0005271662957966328,
"learning_rate": 6.911057692307694e-07,
"loss": -0.0005,
"num_tokens": 2520327480.0,
"reward": 1.349275279045105,
"reward_std": 0.09004013389348983,
"rewards/accuracy_reward": 0.8153645753860473,
"rewards/brier_reward": 0.8853411555290223,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9978298425674439,
"rewards/mean_confidence_reward": 0.7483246564865113,
"sampling/batch_mean_priority_error": 0.03861666706209427,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.07222222222222222,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012177638709545135,
"sampling/priority_kl": 0.030000193789601327,
"sampling/priority_scale": 0.8984230221016333,
"sampling/prob_entropy": 10.27895221710205,
"sampling/prob_max": 6.749814347131177e-05,
"sampling/prob_min": 2.0418150961631908e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2152000427246095,
"sampling/prompt_draws_total": 66456.0,
"sampling/seen_fraction": 0.9595400094985962,
"sampling/unseen_fraction": 0.04045999050140381,
"signal/accuracy_reward/centered_abs_mean": 0.09408094733953476,
"signal/accuracy_reward/group_std_mean": 0.1315217986702919,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444537162781,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04704047366976738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04704047366976738,
"signal/advantage_abs_mean": 0.06250200793147087,
"signal/advantage_pre_scale_abs_mean": 0.06250200793147087,
"signal/advantage_pre_scale_std": 0.14030635058879853,
"signal/advantage_std": 0.14030635058879853,
"signal/brier_reward/centered_abs_mean": 0.054391111433506015,
"signal/brier_reward/group_std_mean": 0.07733864039182663,
"signal/brier_reward/group_zero_std_frac": 0.30555555820465086,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.027195555716753007,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.027195555716753007,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004031033010687679,
"signal/format_reward/group_std_mean": 0.009265058301389217,
"signal/format_reward/group_zero_std_frac": 0.955555546283722,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0020155165053438394,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0020155165053438394,
"signal/mean_confidence_reward/centered_abs_mean": 0.050693362951278687,
"signal/mean_confidence_reward/group_std_mean": 0.06779844909906388,
"signal/mean_confidence_reward/group_zero_std_frac": 0.32500000596046447,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.069336168617156e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.069336168617156e-07,
"step": 925
},
{
"calibration/aurc": 0.04602142129085677,
"calibration/batch_distribution_entropy": 0.5179350186667527,
"calibration/batch_entropy_100bins": 0.2832849481746148,
"calibration/batch_entropy_10bins": 0.5179350186667527,
"calibration/batch_entropy_50bins": 0.33347845747909655,
"calibration/batch_uniqueness": 0.13834077088192218,
"calibration/confidence_entropy": 0.41921585119226645,
"calibration/coverage@0%": 0.3161458333333333,
"calibration/coverage@1%": 0.32552083333333337,
"calibration/coverage@10%": 0.8563836169284468,
"calibration/coverage@15%": 0.9204461169284468,
"calibration/coverage@20%": 1.0,
"calibration/coverage@25%": 1.0,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.6081315445026177,
"calibration/distribution_entropy_10": 0.5179350186667527,
"calibration/distribution_entropy_100": 0.2832849481746148,
"calibration/ece": 0.12141388525305397,
"calibration/mean_confidence": 0.7557981566317629,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0038194444444444196,
"completions/max_length": 3871.4,
"completions/max_terminated_length": 3871.4,
"completions/mean_length": 1080.3007934570312,
"completions/mean_terminated_length": 1084.4209838867187,
"completions/min_length": 0.0,
"completions/min_terminated_length": 275.2,
"epoch": 2.235576923076923,
"grad_norm": 0.0005985578172840178,
"learning_rate": 6.610576923076924e-07,
"loss": -0.0034,
"num_tokens": 2535861281.0,
"reward": 1.3190537452697755,
"reward_std": 0.09621107131242752,
"rewards/accuracy_reward": 0.7712673544883728,
"rewards/brier_reward": 0.8706449508666992,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9961805582046509,
"rewards/mean_confidence_reward": 0.7308419942855835,
"sampling/batch_mean_priority_error": 0.04382285855867425,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08055555555555556,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012211223132908344,
"sampling/priority_kl": 0.030000171065330504,
"sampling/priority_scale": 0.90102642190177,
"sampling/prob_entropy": 10.278956604003906,
"sampling/prob_max": 6.77827192703262e-05,
"sampling/prob_min": 2.0410302022355607e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2272000312805176,
"sampling/prompt_draws_total": 66816.0,
"sampling/seen_fraction": 0.9603600144386292,
"sampling/unseen_fraction": 0.03963998556137085,
"signal/accuracy_reward/centered_abs_mean": 0.10550672709941863,
"signal/accuracy_reward/group_std_mean": 0.14143348932266236,
"signal/accuracy_reward/group_zero_std_frac": 0.5888889074325562,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05275336354970932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05275336354970932,
"signal/advantage_abs_mean": 0.0693376250565052,
"signal/advantage_pre_scale_abs_mean": 0.0693376250565052,
"signal/advantage_pre_scale_std": 0.15195687115192413,
"signal/advantage_std": 0.15195687115192413,
"signal/brier_reward/centered_abs_mean": 0.06214165687561035,
"signal/brier_reward/group_std_mean": 0.08420437872409821,
"signal/brier_reward/group_zero_std_frac": 0.34722222089767457,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031070828437805176,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031070828437805176,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00675998255610466,
"signal/format_reward/group_std_mean": 0.013127930089831352,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00337999127805233,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00337999127805233,
"signal/mean_confidence_reward/centered_abs_mean": 0.054438484460115434,
"signal/mean_confidence_reward/group_std_mean": 0.07193322405219078,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3611111223697662,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.443848181130306e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.443848181130306e-07,
"step": 930
},
{
"calibration/aurc": 0.08520450435441346,
"calibration/batch_distribution_entropy": 0.5897925504729927,
"calibration/batch_entropy_100bins": 0.3332355642178477,
"calibration/batch_entropy_10bins": 0.5897925504729927,
"calibration/batch_entropy_50bins": 0.39227951449099396,
"calibration/batch_uniqueness": 0.31753901729647843,
"calibration/confidence_entropy": 0.455288827241832,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.19614746518711926,
"calibration/coverage@10%": 0.7087902523933856,
"calibration/coverage@15%": 0.8317531549173195,
"calibration/coverage@20%": 0.8734375,
"calibration/coverage@25%": 0.9416666666666668,
"calibration/coverage@30%": 0.9640625,
"calibration/coverage@5%": 0.47904971714534383,
"calibration/distribution_entropy_10": 0.5897925504729927,
"calibration/distribution_entropy_100": 0.3332355642178477,
"calibration/ece": 0.13071597584856381,
"calibration/mean_confidence": 0.6868101338120105,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0019097222222222321,
"completions/max_length": 3690.6,
"completions/max_terminated_length": 3690.6,
"completions/mean_length": 1090.0173095703126,
"completions/mean_terminated_length": 1092.1841064453124,
"completions/min_length": 0.0,
"completions/min_terminated_length": 297.6,
"epoch": 2.2475961538461537,
"grad_norm": 0.0005806403933092952,
"learning_rate": 6.310096153846154e-07,
"loss": -0.0005,
"num_tokens": 2551520232.0,
"reward": 1.3334928274154663,
"reward_std": 0.09118890315294266,
"rewards/accuracy_reward": 0.7927083373069763,
"rewards/brier_reward": 0.8762595653533936,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9980034589767456,
"rewards/mean_confidence_reward": 0.7182552218437195,
"sampling/batch_mean_priority_error": 0.035575774458760756,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08611111111111111,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012244446761906147,
"sampling/priority_kl": 0.030000387132167815,
"sampling/priority_scale": 0.9053085506660864,
"sampling/prob_entropy": 10.278958320617676,
"sampling/prob_max": 6.8172843020875e-05,
"sampling/prob_min": 2.041101433860604e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2392000198364257,
"sampling/prompt_draws_total": 67176.0,
"sampling/seen_fraction": 0.9614533305168151,
"sampling/unseen_fraction": 0.03854666948318482,
"signal/accuracy_reward/centered_abs_mean": 0.0939453125,
"signal/accuracy_reward/group_std_mean": 0.1325065717101097,
"signal/accuracy_reward/group_zero_std_frac": 0.594444465637207,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04697265625,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04697265625,
"signal/advantage_abs_mean": 0.06348500028252602,
"signal/advantage_pre_scale_abs_mean": 0.06348500028252602,
"signal/advantage_pre_scale_std": 0.13785288780927657,
"signal/advantage_std": 0.13785288780927657,
"signal/brier_reward/centered_abs_mean": 0.059427947551012036,
"signal/brier_reward/group_std_mean": 0.0822067528963089,
"signal/brier_reward/group_zero_std_frac": 0.3222222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029713973775506018,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029713973775506018,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0036729600047692656,
"signal/format_reward/group_std_mean": 0.007964649982750416,
"signal/format_reward/group_zero_std_frac": 0.9638888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0018364800023846328,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0018364800023846328,
"signal/mean_confidence_reward/centered_abs_mean": 0.052813592553138736,
"signal/mean_confidence_reward/group_std_mean": 0.07078111544251442,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3361111134290695,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.281359051423351e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.281359051423351e-07,
"step": 935
},
{
"calibration/aurc": 0.17390562365566864,
"calibration/batch_distribution_entropy": 0.5340300874562619,
"calibration/batch_entropy_100bins": 0.2991392944776172,
"calibration/batch_entropy_10bins": 0.5340300874562619,
"calibration/batch_entropy_50bins": 0.352141937425817,
"calibration/batch_uniqueness": 0.20358551385327642,
"calibration/confidence_entropy": 0.4358217392634711,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.15729166666666666,
"calibration/coverage@10%": 0.321875,
"calibration/coverage@15%": 0.3317708333333333,
"calibration/coverage@20%": 0.5727249238468234,
"calibration/coverage@25%": 0.8056516536118364,
"calibration/coverage@30%": 0.8374578437771977,
"calibration/coverage@5%": 0.16666666666666669,
"calibration/distribution_entropy_10": 0.5340300874562619,
"calibration/distribution_entropy_100": 0.2991392944776172,
"calibration/ece": 0.11295827893820713,
"calibration/mean_confidence": 0.7305735966057442,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.00321180555555558,
"completions/max_length": 3717.4,
"completions/max_terminated_length": 3717.4,
"completions/mean_length": 1087.55966796875,
"completions/mean_terminated_length": 1091.1742919921876,
"completions/min_length": 31.2,
"completions/min_terminated_length": 230.2,
"epoch": 2.2596153846153846,
"grad_norm": 0.0005533769144676626,
"learning_rate": 6.009615384615385e-07,
"loss": -0.003,
"num_tokens": 2567122359.0,
"reward": 1.3064433813095093,
"reward_std": 0.08552614152431488,
"rewards/accuracy_reward": 0.7463541626930237,
"rewards/brier_reward": 0.8698168635368347,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967013955116272,
"rewards/mean_confidence_reward": 0.7195225834846497,
"sampling/batch_mean_priority_error": 0.04295157871636351,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.10833333333333332,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012279474921524525,
"sampling/priority_kl": 0.029999899491667748,
"sampling/priority_scale": 0.909530597855337,
"sampling/prob_entropy": 10.278946685791016,
"sampling/prob_max": 6.856414547655732e-05,
"sampling/prob_min": 2.041301559074782e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.251200008392334,
"sampling/prompt_draws_total": 67536.0,
"sampling/seen_fraction": 0.9625,
"sampling/unseen_fraction": 0.0375,
"signal/accuracy_reward/centered_abs_mean": 0.08770616352558136,
"signal/accuracy_reward/group_std_mean": 0.1242572084069252,
"signal/accuracy_reward/group_zero_std_frac": 0.6111111044883728,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04385308176279068,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04385308176279068,
"signal/advantage_abs_mean": 0.0605383463203907,
"signal/advantage_pre_scale_abs_mean": 0.0605383463203907,
"signal/advantage_pre_scale_std": 0.13916278183460234,
"signal/advantage_std": 0.13916278183460234,
"signal/brier_reward/centered_abs_mean": 0.05629269480705261,
"signal/brier_reward/group_std_mean": 0.07803382724523544,
"signal/brier_reward/group_zero_std_frac": 0.3194444477558136,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028146347403526305,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028146347403526305,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0051649304572492834,
"signal/format_reward/group_std_mean": 0.009740457497537136,
"signal/format_reward/group_zero_std_frac": 0.9583333253860473,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025824652286246417,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0025824652286246417,
"signal/mean_confidence_reward/centered_abs_mean": 0.05093696638941765,
"signal/mean_confidence_reward/group_std_mean": 0.06660259738564492,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33333334028720857,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.093696302083118e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.093696302083118e-07,
"step": 940
},
{
"calibration/aurc": 0.09187284481061218,
"calibration/batch_distribution_entropy": 0.5878270899759516,
"calibration/batch_entropy_100bins": 0.32484270689186995,
"calibration/batch_entropy_10bins": 0.5878270899759516,
"calibration/batch_entropy_50bins": 0.3823995786421467,
"calibration/batch_uniqueness": 0.3199851994166748,
"calibration/confidence_entropy": 0.4509366663675958,
"calibration/coverage@0%": 0.15416666666666667,
"calibration/coverage@1%": 0.2591995614035088,
"calibration/coverage@10%": 0.6476913563281571,
"calibration/coverage@15%": 0.7679689432458431,
"calibration/coverage@20%": 0.9062222532458467,
"calibration/coverage@25%": 0.9214854111405835,
"calibration/coverage@30%": 0.9490716180371352,
"calibration/coverage@5%": 0.4574381756447254,
"calibration/distribution_entropy_10": 0.5878270899759516,
"calibration/distribution_entropy_100": 0.32484270689186995,
"calibration/ece": 0.11252378888032692,
"calibration/mean_confidence": 0.7159136752157004,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002951388888888862,
"completions/max_length": 3930.2,
"completions/max_terminated_length": 3930.2,
"completions/mean_length": 1062.620068359375,
"completions/mean_terminated_length": 1065.7474365234375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 254.8,
"epoch": 2.2716346153846154,
"grad_norm": 0.0005658330046571791,
"learning_rate": 5.709134615384615e-07,
"loss": -0.0015,
"num_tokens": 2582458878.0,
"reward": 1.3272025823593139,
"reward_std": 0.08999262303113938,
"rewards/accuracy_reward": 0.7778645753860474,
"rewards/brier_reward": 0.8796510457992553,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.996875,
"rewards/mean_confidence_reward": 0.7279340028762817,
"sampling/batch_mean_priority_error": 0.03774397128747018,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012311909720301628,
"sampling/priority_kl": 0.029999887570738792,
"sampling/priority_scale": 0.9146604239707813,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 6.900994048919529e-05,
"sampling/prob_min": 2.0405833129188978e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2631999492645263,
"sampling/prompt_draws_total": 67896.0,
"sampling/seen_fraction": 0.9637133479118347,
"sampling/unseen_fraction": 0.03628665208816528,
"signal/accuracy_reward/centered_abs_mean": 0.09571940004825592,
"signal/accuracy_reward/group_std_mean": 0.13414473235607147,
"signal/accuracy_reward/group_zero_std_frac": 0.5888888895511627,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04785970002412796,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04785970002412796,
"signal/advantage_abs_mean": 0.061159352958202365,
"signal/advantage_pre_scale_abs_mean": 0.061159352958202365,
"signal/advantage_pre_scale_std": 0.1391223654150963,
"signal/advantage_std": 0.1391223654150963,
"signal/brier_reward/centered_abs_mean": 0.05636544823646546,
"signal/brier_reward/group_std_mean": 0.07898420989513397,
"signal/brier_reward/group_zero_std_frac": 0.2888888955116272,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02818272411823273,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02818272411823273,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0057291666860692205,
"signal/format_reward/group_std_mean": 0.012319867126643657,
"signal/format_reward/group_zero_std_frac": 0.944444453716278,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0028645833430346102,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0028645833430346102,
"signal/mean_confidence_reward/centered_abs_mean": 0.0484912171959877,
"signal/mean_confidence_reward/group_std_mean": 0.0643491841852665,
"signal/mean_confidence_reward/group_zero_std_frac": 0.325,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.849121637562348e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.849121637562348e-07,
"step": 945
},
{
"calibration/aurc": 0.12706838633891895,
"calibration/batch_distribution_entropy": 0.5651185041529028,
"calibration/batch_entropy_100bins": 0.30760167827267476,
"calibration/batch_entropy_10bins": 0.5651185041529028,
"calibration/batch_entropy_50bins": 0.36210371870913616,
"calibration/batch_uniqueness": 0.25386074097927114,
"calibration/confidence_entropy": 0.44389957738306307,
"calibration/coverage@0%": 0.09894179894179894,
"calibration/coverage@1%": 0.24214016328898835,
"calibration/coverage@10%": 0.4060238932750355,
"calibration/coverage@15%": 0.7178011719991159,
"calibration/coverage@20%": 0.78125,
"calibration/coverage@25%": 0.8,
"calibration/coverage@30%": 0.9322916666666666,
"calibration/coverage@5%": 0.27108182408443504,
"calibration/distribution_entropy_10": 0.5651185041529028,
"calibration/distribution_entropy_100": 0.30760167827267476,
"calibration/ece": 0.14062654551231576,
"calibration/mean_confidence": 0.7194144105640516,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.005208333333333348,
"completions/max_length": 3912.8,
"completions/max_terminated_length": 3912.8,
"completions/mean_length": 1094.8569702148438,
"completions/mean_terminated_length": 1100.9157958984374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 279.8,
"epoch": 2.2836538461538463,
"grad_norm": 0.0005399725632742047,
"learning_rate": 5.408653846153847e-07,
"loss": -0.006,
"num_tokens": 2598173102.0,
"reward": 1.3167068004608153,
"reward_std": 0.09733623713254928,
"rewards/accuracy_reward": 0.7733507037162781,
"rewards/brier_reward": 0.8652569413185119,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9947916507720947,
"rewards/mean_confidence_reward": 0.7140972137451171,
"sampling/batch_mean_priority_error": 0.04186446264847028,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08888888888888888,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012349472939968109,
"sampling/priority_kl": 0.03000023923814297,
"sampling/priority_scale": 0.9195403516059741,
"sampling/prob_entropy": 10.27894859313965,
"sampling/prob_max": 6.944662891328335e-05,
"sampling/prob_min": 2.040190011030063e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2751999855041505,
"sampling/prompt_draws_total": 68256.0,
"sampling/seen_fraction": 0.9648200035095215,
"sampling/unseen_fraction": 0.03517999649047852,
"signal/accuracy_reward/centered_abs_mean": 0.10664604753255844,
"signal/accuracy_reward/group_std_mean": 0.1434038370847702,
"signal/accuracy_reward/group_zero_std_frac": 0.5805555641651153,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05332302376627922,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05332302376627922,
"signal/advantage_abs_mean": 0.06883996874094009,
"signal/advantage_pre_scale_abs_mean": 0.06883996874094009,
"signal/advantage_pre_scale_std": 0.15036008059978484,
"signal/advantage_std": 0.15036008059978484,
"signal/brier_reward/centered_abs_mean": 0.06128205507993698,
"signal/brier_reward/group_std_mean": 0.08430658876895905,
"signal/brier_reward/group_zero_std_frac": 0.30555556118488314,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.03064102753996849,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.03064102753996849,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.009016927320044488,
"signal/format_reward/group_std_mean": 0.017384384758770466,
"signal/format_reward/group_zero_std_frac": 0.9277777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.004508463660022244,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.004508463660022244,
"signal/mean_confidence_reward/centered_abs_mean": 0.049524746090173724,
"signal/mean_confidence_reward/group_std_mean": 0.06665968000888825,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3166666716337204,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.952474569108745e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.952474569108745e-07,
"step": 950
},
{
"epoch": 2.2836538461538463,
"eval_calibration/aurc": 0.11860887443162953,
"eval_calibration/batch_distribution_entropy": 0.617508068522075,
"eval_calibration/batch_entropy_100bins": 0.3388462605907164,
"eval_calibration/batch_entropy_10bins": 0.617508068522075,
"eval_calibration/batch_entropy_50bins": 0.39888433548082824,
"eval_calibration/batch_uniqueness": 0.32768699389333367,
"eval_calibration/confidence_entropy": 0.4445777268167821,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.5522648083623694,
"eval_calibration/coverage@15%": 0.725609756097561,
"eval_calibration/coverage@20%": 0.8005226480836237,
"eval_calibration/coverage@25%": 0.9259581881533101,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.617508068522075,
"eval_calibration/distribution_entropy_100": 0.3388462605907164,
"eval_calibration/ece": 0.0331881533101044,
"eval_calibration/mean_confidence": 0.7220383275261325,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.002604166666666685,
"eval_completions/max_length": 2849.6666666666665,
"eval_completions/max_terminated_length": 2849.6666666666665,
"eval_completions/mean_length": 1075.599833170573,
"eval_completions/mean_terminated_length": 1078.4251708984375,
"eval_completions/min_length": 173.16666666666666,
"eval_completions/min_terminated_length": 319.5,
"eval_loss": 0.0,
"eval_num_tokens": 2598173102.0,
"eval_reward": 1.292337954044342,
"eval_reward_std": 0.2966277003288269,
"eval_rewards/accuracy_reward": 0.7265625,
"eval_rewards/brier_reward": 0.8615711629390717,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9965277910232544,
"eval_rewards/mean_confidence_reward": 0.719531257947286,
"eval_runtime": 183.1176,
"eval_samples_per_second": 5.461,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3846028645833333,
"eval_signal/accuracy_reward/group_std_mean": 0.4437861939271291,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19230143229166666,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19230143229166666,
"eval_signal/advantage_abs_mean": 0.24802343298991522,
"eval_signal/advantage_pre_scale_abs_mean": 0.24802343298991522,
"eval_signal/advantage_pre_scale_std": 0.29531795779863995,
"eval_signal/advantage_std": 0.29531795779863995,
"eval_signal/brier_reward/centered_abs_mean": 0.1587554377814134,
"eval_signal/brier_reward/group_std_mean": 0.2169876346985499,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0793777188907067,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.0793777188907067,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"eval_signal/format_reward/group_std_mean": 0.019641855110724766,
"eval_signal/format_reward/group_zero_std_frac": 0.8888889153798422,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.2088215947151184,
"eval_signal/mean_confidence_reward/group_std_mean": 0.24693219860394797,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.0882159598537933e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.0882159598537933e-06,
"eval_steps_per_second": 0.033,
"step": 950
},
{
"epoch": 2.2836538461538463,
"step": 950,
"train_probe_calibration/aurc": 0.09106709452890661,
"train_probe_calibration/batch_distribution_entropy": 0.5885328860111262,
"train_probe_calibration/batch_entropy_100bins": 0.32622672018634385,
"train_probe_calibration/batch_entropy_10bins": 0.5885328860111262,
"train_probe_calibration/batch_entropy_50bins": 0.384028816699254,
"train_probe_calibration/batch_uniqueness": 0.2705508140198376,
"train_probe_calibration/confidence_entropy": 0.43908000945358777,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7621951219512195,
"train_probe_calibration/coverage@15%": 0.8301393728222997,
"train_probe_calibration/coverage@20%": 0.9442508710801394,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5885328860111262,
"train_probe_calibration/distribution_entropy_100": 0.32622672018634385,
"train_probe_calibration/ece": 0.045383275261323915,
"train_probe_calibration/mean_confidence": 0.7391114982578398,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.0034722222222222285,
"train_probe_completions/max_length": 3518.1666666666665,
"train_probe_completions/max_terminated_length": 3518.1666666666665,
"train_probe_completions/mean_length": 1104.3688151041667,
"train_probe_completions/mean_terminated_length": 1108.1290690104167,
"train_probe_completions/min_length": 142.83333333333334,
"train_probe_completions/min_terminated_length": 252.83333333333334,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2598173102.0,
"train_probe_reward": 1.3288832902908325,
"train_probe_reward_std": 0.2760002116362254,
"train_probe_rewards/accuracy_reward": 0.7803819477558136,
"train_probe_rewards/brier_reward": 0.8808420300483704,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.996527781089147,
"train_probe_rewards/mean_confidence_reward": 0.7365450958410898,
"train_probe_runtime": 199.9633,
"train_probe_samples_per_second": 5.001,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.3337131092945735,
"train_probe_signal/accuracy_reward/group_std_mean": 0.411047359307607,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.16685655464728674,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.16685655464728674,
"train_probe_signal/advantage_abs_mean": 0.21701645106077194,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.21701645106077194,
"train_probe_signal/advantage_pre_scale_std": 0.275801161924998,
"train_probe_signal/advantage_std": 0.275801161924998,
"train_probe_signal/brier_reward/centered_abs_mean": 0.13900663703680038,
"train_probe_signal/brier_reward/group_std_mean": 0.19764207303524017,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.06950331851840019,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.06950331851840019,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/group_std_mean": 0.019641855110724766,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888889054457346,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.19654946277538934,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.23597882439692816,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.9654945617730846e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.9654945617730846e-06,
"train_probe_steps_per_second": 0.03
},
{
"calibration/aurc": 0.08837978514640367,
"calibration/batch_distribution_entropy": 0.5270829710202529,
"calibration/batch_entropy_100bins": 0.2865713633885314,
"calibration/batch_entropy_10bins": 0.5270829710202529,
"calibration/batch_entropy_50bins": 0.33734717229516653,
"calibration/batch_uniqueness": 0.15639742451592784,
"calibration/confidence_entropy": 0.4325990407548216,
"calibration/coverage@0%": 0.28445860046751326,
"calibration/coverage@1%": 0.4162294338008466,
"calibration/coverage@10%": 0.6884587140992167,
"calibration/coverage@15%": 0.7218450826805918,
"calibration/coverage@20%": 0.8929811507936508,
"calibration/coverage@25%": 0.9165509259259259,
"calibration/coverage@30%": 0.9470899470899472,
"calibration/coverage@5%": 0.5087330378339463,
"calibration/distribution_entropy_10": 0.5270829710202529,
"calibration/distribution_entropy_100": 0.2865713633885314,
"calibration/ece": 0.12312973907309355,
"calibration/mean_confidence": 0.7541945820958026,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.004427083333333348,
"completions/max_length": 4034.6,
"completions/max_terminated_length": 4034.6,
"completions/mean_length": 1096.816162109375,
"completions/mean_terminated_length": 1101.7827880859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 286.6,
"epoch": 2.295673076923077,
"grad_norm": 0.0005315289017744362,
"learning_rate": 5.108173076923077e-07,
"loss": -0.0047,
"num_tokens": 2613888360.0,
"reward": 1.3091769218444824,
"reward_std": 0.10529560595750809,
"rewards/accuracy_reward": 0.7511284708976745,
"rewards/brier_reward": 0.8719852566719055,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.995225703716278,
"rewards/mean_confidence_reward": 0.72049480676651,
"sampling/batch_mean_priority_error": 0.03887233270399586,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0611111111111111,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012381246872246265,
"sampling/priority_kl": 0.02999982424080372,
"sampling/priority_scale": 0.9230521737830714,
"sampling/prob_entropy": 10.27895393371582,
"sampling/prob_max": 6.980416073929519e-05,
"sampling/prob_min": 2.0411239529494195e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2871999740600586,
"sampling/prompt_draws_total": 68616.0,
"sampling/seen_fraction": 0.9656333327293396,
"sampling/unseen_fraction": 0.0343666672706604,
"signal/accuracy_reward/centered_abs_mean": 0.11203884780406952,
"signal/accuracy_reward/group_std_mean": 0.1537027046084404,
"signal/accuracy_reward/group_zero_std_frac": 0.5361111104488373,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05601942390203476,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05601942390203476,
"signal/advantage_abs_mean": 0.07400249615311623,
"signal/advantage_pre_scale_abs_mean": 0.07400249615311623,
"signal/advantage_pre_scale_std": 0.1581919014453888,
"signal/advantage_std": 0.1581919014453888,
"signal/brier_reward/centered_abs_mean": 0.06311105191707611,
"signal/brier_reward/group_std_mean": 0.08841222822666168,
"signal/brier_reward/group_zero_std_frac": 0.3166666716337204,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.031555525958538055,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.031555525958538055,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007861328218132258,
"signal/format_reward/group_std_mean": 0.015163037367165089,
"signal/format_reward/group_zero_std_frac": 0.9361111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003930664109066129,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003930664109066129,
"signal/mean_confidence_reward/centered_abs_mean": 0.056025400012731555,
"signal/mean_confidence_reward/group_std_mean": 0.0728580117225647,
"signal/mean_confidence_reward/group_zero_std_frac": 0.347222226858139,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.602539886240266e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.602539886240266e-07,
"step": 955
},
{
"calibration/aurc": 0.13481636294467678,
"calibration/batch_distribution_entropy": 0.5443487294614753,
"calibration/batch_entropy_100bins": 0.3051426541946902,
"calibration/batch_entropy_10bins": 0.5443487294614753,
"calibration/batch_entropy_50bins": 0.3592089953512089,
"calibration/batch_uniqueness": 0.24514326070790657,
"calibration/confidence_entropy": 0.45087926568246744,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.4681907894736842,
"calibration/coverage@15%": 0.7244150949338098,
"calibration/coverage@20%": 0.7859006258302415,
"calibration/coverage@25%": 0.8727915578764144,
"calibration/coverage@30%": 0.918699684508268,
"calibration/coverage@5%": 0.2524177631578947,
"calibration/distribution_entropy_10": 0.5443487294614753,
"calibration/distribution_entropy_100": 0.3051426541946902,
"calibration/ece": 0.10919581815995608,
"calibration/mean_confidence": 0.7050144390431039,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.003472222222222254,
"completions/max_length": 3896.4,
"completions/max_terminated_length": 3896.4,
"completions/mean_length": 1087.0146728515624,
"completions/mean_terminated_length": 1090.7575927734374,
"completions/min_length": 0.0,
"completions/min_terminated_length": 210.4,
"epoch": 2.3076923076923075,
"grad_norm": 0.0005695180734619498,
"learning_rate": 4.807692307692308e-07,
"loss": -0.0024,
"num_tokens": 2629494833.0,
"reward": 1.3127452373504638,
"reward_std": 0.09188116788864135,
"rewards/accuracy_reward": 0.7652777791023254,
"rewards/brier_reward": 0.8637569308280945,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9964409708976746,
"rewards/mean_confidence_reward": 0.7389930605888366,
"sampling/batch_mean_priority_error": 0.04523240163427438,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06666666666666667,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012419133633375167,
"sampling/priority_kl": 0.030000034347176553,
"sampling/priority_scale": 0.9262719094520435,
"sampling/prob_entropy": 10.278954315185548,
"sampling/prob_max": 7.014243310550227e-05,
"sampling/prob_min": 2.0422369925654493e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.2991999626159667,
"sampling/prompt_draws_total": 68976.0,
"sampling/seen_fraction": 0.9663866639137269,
"sampling/unseen_fraction": 0.03361333608627319,
"signal/accuracy_reward/centered_abs_mean": 0.0953342005610466,
"signal/accuracy_reward/group_std_mean": 0.13181808739900588,
"signal/accuracy_reward/group_zero_std_frac": 0.6,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0476671002805233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0476671002805233,
"signal/advantage_abs_mean": 0.06448932960629464,
"signal/advantage_pre_scale_abs_mean": 0.06448932960629464,
"signal/advantage_pre_scale_std": 0.14418236911296844,
"signal/advantage_std": 0.14418236911296844,
"signal/brier_reward/centered_abs_mean": 0.05803385302424431,
"signal/brier_reward/group_std_mean": 0.08039816617965698,
"signal/brier_reward/group_zero_std_frac": 0.3277777791023254,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029016926512122156,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029016926512122156,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006287977413740009,
"signal/format_reward/group_std_mean": 0.012360461801290513,
"signal/format_reward/group_zero_std_frac": 0.9472222328186035,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0031439887068700047,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0031439887068700047,
"signal/mean_confidence_reward/centered_abs_mean": 0.05114692747592926,
"signal/mean_confidence_reward/group_std_mean": 0.06721761301159859,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3666666686534882,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.11469272623799e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.11469272623799e-07,
"step": 960
},
{
"calibration/aurc": 0.09925959375408042,
"calibration/batch_distribution_entropy": 0.5466540356910407,
"calibration/batch_entropy_100bins": 0.29627822753163874,
"calibration/batch_entropy_10bins": 0.5466540356910407,
"calibration/batch_entropy_50bins": 0.3487739357086865,
"calibration/batch_uniqueness": 0.21399426906611022,
"calibration/confidence_entropy": 0.45134320648732534,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.07604166666666666,
"calibration/coverage@10%": 0.47854493891797556,
"calibration/coverage@15%": 0.8868116880601843,
"calibration/coverage@20%": 0.8993443251359023,
"calibration/coverage@25%": 0.9755208333333332,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.4654559336823734,
"calibration/distribution_entropy_10": 0.5466540356910407,
"calibration/distribution_entropy_100": 0.29627822753163874,
"calibration/ece": 0.11913866100843437,
"calibration/mean_confidence": 0.7471377391392016,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.002951388888888884,
"completions/max_length": 3694.4,
"completions/max_terminated_length": 3694.4,
"completions/mean_length": 1065.77197265625,
"completions/mean_terminated_length": 1068.911181640625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 284.8,
"epoch": 2.3197115384615383,
"grad_norm": 0.0005898535600863397,
"learning_rate": 4.507211538461539e-07,
"loss": -0.0012,
"num_tokens": 2644892078.0,
"reward": 1.3143615007400513,
"reward_std": 0.09434518218040466,
"rewards/accuracy_reward": 0.7613715410232544,
"rewards/brier_reward": 0.8702881693840027,
"rewards/confidence_one_or_zero": 0.0001736111124046147,
"rewards/format_reward": 0.9970486044883728,
"rewards/mean_confidence_reward": 0.7314235925674438,
"sampling/batch_mean_priority_error": 0.044258274395874334,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.058333333333333334,
"sampling/error_ema_max": 0.25845763087272644,
"sampling/error_ema_mean": 0.012453344091773033,
"sampling/priority_kl": 0.03000020980834961,
"sampling/priority_scale": 0.9294657290214673,
"sampling/prob_entropy": 10.278950691223145,
"sampling/prob_max": 7.047422259347514e-05,
"sampling/prob_min": 2.0431867233128288e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.311199998855591,
"sampling/prompt_draws_total": 69336.0,
"sampling/seen_fraction": 0.9671800017356873,
"sampling/unseen_fraction": 0.03281999826431274,
"signal/accuracy_reward/centered_abs_mean": 0.09511176347732545,
"signal/accuracy_reward/group_std_mean": 0.13041624277830124,
"signal/accuracy_reward/group_zero_std_frac": 0.6111111164093017,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04755588173866272,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04755588173866272,
"signal/advantage_abs_mean": 0.06701920777559281,
"signal/advantage_pre_scale_abs_mean": 0.06701920777559281,
"signal/advantage_pre_scale_std": 0.14776241779327393,
"signal/advantage_std": 0.14776241779327393,
"signal/brier_reward/centered_abs_mean": 0.059418071806430814,
"signal/brier_reward/group_std_mean": 0.08131268173456192,
"signal/brier_reward/group_zero_std_frac": 0.34166666865348816,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029709035903215407,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029709035903215407,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0003255208255723119,
"signal/confidence_one_or_zero/group_std_mean": 0.0006831518840044737,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 3.255208369523643e-09,
"signal/format_reward/centered_abs_mean": 0.005414496501907706,
"signal/format_reward/group_std_mean": 0.011457231175154448,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002707248250953853,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002707248250953853,
"signal/mean_confidence_reward/centered_abs_mean": 0.050473643094301225,
"signal/mean_confidence_reward/group_std_mean": 0.06626100763678551,
"signal/mean_confidence_reward/group_zero_std_frac": 0.37222222685813905,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.047364084020956e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.047364084020956e-07,
"step": 965
},
{
"calibration/aurc": 0.12528094703373654,
"calibration/batch_distribution_entropy": 0.4389142890592641,
"calibration/batch_entropy_100bins": 0.23896849940343157,
"calibration/batch_entropy_10bins": 0.4389142890592641,
"calibration/batch_entropy_50bins": 0.28130985101979333,
"calibration/batch_uniqueness": -0.06004193147009436,
"calibration/confidence_entropy": 0.40077594475257444,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.17447916666666669,
"calibration/coverage@10%": 0.33942558746736295,
"calibration/coverage@15%": 0.512368091818973,
"calibration/coverage@20%": 0.854506636205396,
"calibration/coverage@25%": 0.9072127937336815,
"calibration/coverage@30%": 0.9619614882506529,
"calibration/coverage@5%": 0.2830545583115753,
"calibration/distribution_entropy_10": 0.4389142890592641,
"calibration/distribution_entropy_100": 0.23896849940343157,
"calibration/ece": 0.11179626305483026,
"calibration/mean_confidence": 0.7800371246736294,
"calibration/unique_confidence_per_question": 0.020833333333333332,
"calibration/unique_confidences": 8.0,
"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": 0.003038194444444464,
"completions/max_length": 3829.2,
"completions/max_terminated_length": 3829.2,
"completions/mean_length": 1037.8958618164063,
"completions/mean_terminated_length": 1041.048876953125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 227.6,
"epoch": 2.331730769230769,
"grad_norm": 0.0005680083413608372,
"learning_rate": 4.20673076923077e-07,
"loss": -0.0028,
"num_tokens": 2659941150.0,
"reward": 1.2989899873733521,
"reward_std": 0.09422019422054291,
"rewards/accuracy_reward": 0.7292534708976746,
"rewards/brier_reward": 0.8719236373901367,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9967882037162781,
"rewards/mean_confidence_reward": 0.7369791746139527,
"sampling/batch_mean_priority_error": 0.0409421805639239,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.075,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012488853186368942,
"sampling/priority_kl": 0.02999998964369297,
"sampling/priority_scale": 0.9323509871726856,
"sampling/prob_entropy": 10.278954696655273,
"sampling/prob_max": 7.07887826138176e-05,
"sampling/prob_min": 2.0444312394829466e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.323199987411499,
"sampling/prompt_draws_total": 69696.0,
"sampling/seen_fraction": 0.9679066658020019,
"sampling/unseen_fraction": 0.03209333419799805,
"signal/accuracy_reward/centered_abs_mean": 0.08981662392616271,
"signal/accuracy_reward/group_std_mean": 0.1299781858921051,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04490831196308136,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04490831196308136,
"signal/advantage_abs_mean": 0.06450606808066368,
"signal/advantage_pre_scale_abs_mean": 0.06450606808066368,
"signal/advantage_pre_scale_std": 0.145282381772995,
"signal/advantage_std": 0.145282381772995,
"signal/brier_reward/centered_abs_mean": 0.05912662670016289,
"signal/brier_reward/group_std_mean": 0.08251220434904098,
"signal/brier_reward/group_zero_std_frac": 0.3583333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029563313350081444,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029563313350081444,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.005365668376907706,
"signal/format_reward/group_std_mean": 0.011186510510742664,
"signal/format_reward/group_zero_std_frac": 0.9472222208976746,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002682834188453853,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002682834188453853,
"signal/mean_confidence_reward/centered_abs_mean": 0.04840550422668457,
"signal/mean_confidence_reward/group_std_mean": 0.06418332159519195,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4000000059604645,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.840550218432327e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.840550218432327e-07,
"step": 970
},
{
"calibration/aurc": 0.06388608571442249,
"calibration/batch_distribution_entropy": 0.45754749297228736,
"calibration/batch_entropy_100bins": 0.24964356929574544,
"calibration/batch_entropy_10bins": 0.45754749297228736,
"calibration/batch_entropy_50bins": 0.29387637057584143,
"calibration/batch_uniqueness": 0.01120460609993052,
"calibration/confidence_entropy": 0.4152244460367278,
"calibration/coverage@0%": 0.2739583333333333,
"calibration/coverage@1%": 0.48273906657963445,
"calibration/coverage@10%": 0.6679816688424717,
"calibration/coverage@15%": 0.8481383268059182,
"calibration/coverage@20%": 0.9065532528285465,
"calibration/coverage@25%": 0.9702349869451699,
"calibration/coverage@30%": 0.993733681462141,
"calibration/coverage@5%": 0.6429395126196693,
"calibration/distribution_entropy_10": 0.45754749297228736,
"calibration/distribution_entropy_100": 0.24964356929574544,
"calibration/ece": 0.10643399151436021,
"calibration/mean_confidence": 0.786690192558747,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.0035590277777777677,
"completions/max_length": 3832.0,
"completions/max_terminated_length": 3832.0,
"completions/mean_length": 1066.65029296875,
"completions/mean_terminated_length": 1070.4842529296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 256.8,
"epoch": 2.34375,
"grad_norm": 0.0005234786658547819,
"learning_rate": 3.90625e-07,
"loss": -0.0023,
"num_tokens": 2675320769.0,
"reward": 1.3103954315185546,
"reward_std": 0.0930139496922493,
"rewards/accuracy_reward": 0.7488715291023255,
"rewards/brier_reward": 0.8754635453224182,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9964409708976746,
"rewards/mean_confidence_reward": 0.7411111235618592,
"sampling/batch_mean_priority_error": 0.041501399107933726,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.041666666666666664,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.01252305880188942,
"sampling/priority_kl": 0.02999994046986103,
"sampling/priority_scale": 0.9357503473991529,
"sampling/prob_entropy": 10.278956985473632,
"sampling/prob_max": 7.114256877684965e-05,
"sampling/prob_min": 2.045368164544925e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.335199975967407,
"sampling/prompt_draws_total": 70056.0,
"sampling/seen_fraction": 0.9686466693878174,
"sampling/unseen_fraction": 0.031353330612182616,
"signal/accuracy_reward/centered_abs_mean": 0.0970431849360466,
"signal/accuracy_reward/group_std_mean": 0.13124381005764008,
"signal/accuracy_reward/group_zero_std_frac": 0.6083333373069764,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0485215924680233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0485215924680233,
"signal/advantage_abs_mean": 0.06658430472016334,
"signal/advantage_pre_scale_abs_mean": 0.06658430472016334,
"signal/advantage_pre_scale_std": 0.1472132295370102,
"signal/advantage_std": 0.1472132295370102,
"signal/brier_reward/centered_abs_mean": 0.058478569984436034,
"signal/brier_reward/group_std_mean": 0.07980355173349381,
"signal/brier_reward/group_zero_std_frac": 0.3666666686534882,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029239284992218017,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029239284992218017,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.006190321082249284,
"signal/format_reward/group_std_mean": 0.01252242513000965,
"signal/format_reward/group_zero_std_frac": 0.9444444656372071,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003095160541124642,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003095160541124642,
"signal/mean_confidence_reward/centered_abs_mean": 0.05035320073366165,
"signal/mean_confidence_reward/group_std_mean": 0.06616755574941635,
"signal/mean_confidence_reward/group_zero_std_frac": 0.38888888955116274,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.035319873059052e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.035319873059052e-07,
"step": 975
},
{
"calibration/aurc": 0.10544449408187254,
"calibration/batch_distribution_entropy": 0.5222339767825488,
"calibration/batch_entropy_100bins": 0.2837296727876054,
"calibration/batch_entropy_10bins": 0.5222339767825488,
"calibration/batch_entropy_50bins": 0.3340019800979252,
"calibration/batch_uniqueness": 0.12321258877761894,
"calibration/confidence_entropy": 0.42892239360006473,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.5439225439101444,
"calibration/coverage@15%": 0.8060711492533734,
"calibration/coverage@20%": 0.9552083333333334,
"calibration/coverage@25%": 0.9838541666666666,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.2231053149606299,
"calibration/distribution_entropy_10": 0.5222339767825488,
"calibration/distribution_entropy_100": 0.2837296727876054,
"calibration/ece": 0.10365094626104172,
"calibration/mean_confidence": 0.763162794975775,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.003819444444444442,
"completions/max_length": 3884.2,
"completions/max_terminated_length": 3884.2,
"completions/mean_length": 1080.369384765625,
"completions/mean_terminated_length": 1084.5101318359375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 268.0,
"epoch": 2.355769230769231,
"grad_norm": 0.0006834192317910492,
"learning_rate": 3.6057692307692306e-07,
"loss": -0.0041,
"num_tokens": 2690891136.0,
"reward": 1.3060776233673095,
"reward_std": 0.1066261425614357,
"rewards/accuracy_reward": 0.7427083373069763,
"rewards/brier_reward": 0.8734253406524658,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9960069417953491,
"rewards/mean_confidence_reward": 0.7334027647972107,
"sampling/batch_mean_priority_error": 0.03689815305670817,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0638888888888889,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012555352225899697,
"sampling/priority_kl": 0.030000166967511177,
"sampling/priority_scale": 0.9393207729561255,
"sampling/prob_entropy": 10.278953552246094,
"sampling/prob_max": 7.151352911023423e-05,
"sampling/prob_min": 2.0462764950934798e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3472000122070313,
"sampling/prompt_draws_total": 70416.0,
"sampling/seen_fraction": 0.9693666696548462,
"sampling/unseen_fraction": 0.03063333034515381,
"signal/accuracy_reward/centered_abs_mean": 0.10970052033662796,
"signal/accuracy_reward/group_std_mean": 0.14751950204372405,
"signal/accuracy_reward/group_zero_std_frac": 0.5722222447395324,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05485026016831398,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05485026016831398,
"signal/advantage_abs_mean": 0.07652829140424729,
"signal/advantage_pre_scale_abs_mean": 0.07652829140424729,
"signal/advantage_pre_scale_std": 0.1620244115591049,
"signal/advantage_std": 0.1620244115591049,
"signal/brier_reward/centered_abs_mean": 0.06936425715684891,
"signal/brier_reward/group_std_mean": 0.09475861638784408,
"signal/brier_reward/group_zero_std_frac": 0.32222222685813906,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.034682128578424454,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.034682128578424454,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007085503358393907,
"signal/format_reward/group_std_mean": 0.014527601003646851,
"signal/format_reward/group_zero_std_frac": 0.9361111283302307,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0035427516791969537,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0035427516791969537,
"signal/mean_confidence_reward/centered_abs_mean": 0.05302572175860405,
"signal/mean_confidence_reward/group_std_mean": 0.07050289511680603,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3388888955116272,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.302572049004084e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.302572049004084e-07,
"step": 980
},
{
"calibration/aurc": 0.0821613033038492,
"calibration/batch_distribution_entropy": 0.5489296276167633,
"calibration/batch_entropy_100bins": 0.3007058894796451,
"calibration/batch_entropy_10bins": 0.5489296276167633,
"calibration/batch_entropy_50bins": 0.35398610771490946,
"calibration/batch_uniqueness": 0.1897931844310782,
"calibration/confidence_entropy": 0.43470675216095145,
"calibration/coverage@0%": 0.1484375,
"calibration/coverage@1%": 0.2805897309711286,
"calibration/coverage@10%": 0.725868182620286,
"calibration/coverage@15%": 0.8086414226618075,
"calibration/coverage@20%": 0.8542933169151574,
"calibration/coverage@25%": 0.8920200632128624,
"calibration/coverage@30%": 0.9434547203517933,
"calibration/coverage@5%": 0.4640173884514436,
"calibration/distribution_entropy_10": 0.5489296276167633,
"calibration/distribution_entropy_100": 0.3007058894796451,
"calibration/ece": 0.09899339361837176,
"calibration/mean_confidence": 0.737277937115898,
"calibration/unique_confidence_per_question": 0.022395833333333334,
"calibration/unique_confidences": 8.6,
"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": 0.002517361111111116,
"completions/max_length": 3858.4,
"completions/max_terminated_length": 3858.4,
"completions/mean_length": 1063.3915771484376,
"completions/mean_terminated_length": 1066.071923828125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 280.8,
"epoch": 2.3677884615384617,
"grad_norm": 0.0005046343430876732,
"learning_rate": 3.305288461538462e-07,
"loss": -0.0019,
"num_tokens": 2706243807.0,
"reward": 1.3196176290512085,
"reward_std": 0.09575788974761963,
"rewards/accuracy_reward": 0.7658854126930237,
"rewards/brier_reward": 0.8758524537086487,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9974826455116272,
"rewards/mean_confidence_reward": 0.7436111211776734,
"sampling/batch_mean_priority_error": 0.04085982891623012,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06944444444444445,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.01258925199508667,
"sampling/priority_kl": 0.02999987043440342,
"sampling/priority_scale": 0.9431747139198705,
"sampling/prob_entropy": 10.278960609436036,
"sampling/prob_max": 7.18996481737122e-05,
"sampling/prob_min": 2.0467897411435842e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3592000007629395,
"sampling/prompt_draws_total": 70776.0,
"sampling/seen_fraction": 0.9701733231544495,
"sampling/unseen_fraction": 0.029826676845550536,
"signal/accuracy_reward/centered_abs_mean": 0.10296766757965088,
"signal/accuracy_reward/group_std_mean": 0.13744378834962845,
"signal/accuracy_reward/group_zero_std_frac": 0.5972222328186035,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05148383378982544,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05148383378982544,
"signal/advantage_abs_mean": 0.06796487495303154,
"signal/advantage_pre_scale_abs_mean": 0.06796487495303154,
"signal/advantage_pre_scale_std": 0.1487546980381012,
"signal/advantage_std": 0.1487546980381012,
"signal/brier_reward/centered_abs_mean": 0.05547922253608704,
"signal/brier_reward/group_std_mean": 0.07781999558210373,
"signal/brier_reward/group_zero_std_frac": 0.3777777850627899,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02773961126804352,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02773961126804352,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004616970417555421,
"signal/format_reward/group_std_mean": 0.010467971116304398,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0023084852087777107,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0023084852087777107,
"signal/mean_confidence_reward/centered_abs_mean": 0.048927963525056836,
"signal/mean_confidence_reward/group_std_mean": 0.06495207697153091,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.892796198419091e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.892796198419091e-07,
"step": 985
},
{
"calibration/aurc": 0.11709884279040068,
"calibration/batch_distribution_entropy": 0.6036341462655852,
"calibration/batch_entropy_100bins": 0.3352479102427931,
"calibration/batch_entropy_10bins": 0.6036341462655852,
"calibration/batch_entropy_50bins": 0.3946484156720738,
"calibration/batch_uniqueness": 0.34513761430670165,
"calibration/confidence_entropy": 0.45830963659709384,
"calibration/coverage@0%": 0.1390625,
"calibration/coverage@1%": 0.165625,
"calibration/coverage@10%": 0.5966144878662709,
"calibration/coverage@15%": 0.6706747897035412,
"calibration/coverage@20%": 0.7909831762832924,
"calibration/coverage@25%": 0.8176498429499592,
"calibration/coverage@30%": 0.8764769157953618,
"calibration/coverage@5%": 0.46054532776128393,
"calibration/distribution_entropy_10": 0.6036341462655852,
"calibration/distribution_entropy_100": 0.3352479102427931,
"calibration/ece": 0.13048479223140128,
"calibration/mean_confidence": 0.6936304357062951,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.006336805555555558,
"completions/max_length": 4021.2,
"completions/max_terminated_length": 4021.2,
"completions/mean_length": 1059.2802978515624,
"completions/mean_terminated_length": 1066.165478515625,
"completions/min_length": 0.0,
"completions/min_terminated_length": 281.8,
"epoch": 2.3798076923076925,
"grad_norm": 0.00045987789053469896,
"learning_rate": 3.0048076923076924e-07,
"loss": -0.008,
"num_tokens": 2721561916.0,
"reward": 1.317639660835266,
"reward_std": 0.08361054807901383,
"rewards/accuracy_reward": 0.760243046283722,
"rewards/brier_reward": 0.8814453482627869,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9935763835906982,
"rewards/mean_confidence_reward": 0.7294010400772095,
"sampling/batch_mean_priority_error": 0.03380556619838228,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06944444444444445,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012619511969387532,
"sampling/priority_kl": 0.029999939352273942,
"sampling/priority_scale": 0.9469346225028857,
"sampling/prob_entropy": 10.278962898254395,
"sampling/prob_max": 7.228225003927947e-05,
"sampling/prob_min": 2.0474002667469904e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3711999893188476,
"sampling/prompt_draws_total": 71136.0,
"sampling/seen_fraction": 0.9709333300590515,
"sampling/unseen_fraction": 0.029066669940948486,
"signal/accuracy_reward/centered_abs_mean": 0.0781467005610466,
"signal/accuracy_reward/group_std_mean": 0.11067922562360763,
"signal/accuracy_reward/group_zero_std_frac": 0.6583333492279053,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0390733502805233,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0390733502805233,
"signal/advantage_abs_mean": 0.056924599409103396,
"signal/advantage_pre_scale_abs_mean": 0.056924599409103396,
"signal/advantage_pre_scale_std": 0.13805509656667708,
"signal/advantage_std": 0.13805509656667708,
"signal/brier_reward/centered_abs_mean": 0.05322760194540024,
"signal/brier_reward/group_std_mean": 0.07627559676766396,
"signal/brier_reward/group_zero_std_frac": 0.39166666865348815,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02661380097270012,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02661380097270012,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.010047743003815413,
"signal/format_reward/group_std_mean": 0.01786075867712498,
"signal/format_reward/group_zero_std_frac": 0.9277777910232544,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0050238715019077064,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0050238715019077064,
"signal/mean_confidence_reward/centered_abs_mean": 0.04356338232755661,
"signal/mean_confidence_reward/group_std_mean": 0.059166472405195236,
"signal/mean_confidence_reward/group_zero_std_frac": 0.41666666865348817,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.356338081379363e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.356338081379363e-07,
"step": 990
},
{
"calibration/aurc": 0.10716608850361349,
"calibration/batch_distribution_entropy": 0.531350096709369,
"calibration/batch_entropy_100bins": 0.2954811392641967,
"calibration/batch_entropy_10bins": 0.531350096709369,
"calibration/batch_entropy_50bins": 0.34783561629703386,
"calibration/batch_uniqueness": 0.19805192531066296,
"calibration/confidence_entropy": 0.4487496813944191,
"calibration/coverage@0%": 0.2848886966927763,
"calibration/coverage@1%": 0.307805363359443,
"calibration/coverage@10%": 0.7164352480417755,
"calibration/coverage@15%": 0.799288544386423,
"calibration/coverage@20%": 0.8384,
"calibration/coverage@25%": 0.8565333333333334,
"calibration/coverage@30%": 0.8565333333333334,
"calibration/coverage@5%": 0.5005871301131419,
"calibration/distribution_entropy_10": 0.531350096709369,
"calibration/distribution_entropy_100": 0.2954811392641967,
"calibration/ece": 0.1027855439512618,
"calibration/mean_confidence": 0.7100120691906006,
"calibration/unique_confidence_per_question": 0.021354166666666664,
"calibration/unique_confidences": 8.2,
"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": 0.00234375,
"completions/max_length": 3947.0,
"completions/max_terminated_length": 3947.0,
"completions/mean_length": 1068.986669921875,
"completions/mean_terminated_length": 1071.5660278320313,
"completions/min_length": 0.0,
"completions/min_terminated_length": 269.8,
"epoch": 2.391826923076923,
"grad_norm": 0.0006047965725883842,
"learning_rate": 2.7043269230769233e-07,
"loss": -0.0004,
"num_tokens": 2736978338.0,
"reward": 1.3273440599441528,
"reward_std": 0.08698437064886093,
"rewards/accuracy_reward": 0.7737847208976746,
"rewards/brier_reward": 0.8832326650619506,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.99765625,
"rewards/mean_confidence_reward": 0.7293229222297668,
"sampling/batch_mean_priority_error": 0.03242008661887693,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.0638888888888889,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012646873854100705,
"sampling/priority_kl": 0.02999994680285454,
"sampling/priority_scale": 0.9515143097145483,
"sampling/prob_entropy": 10.278952026367188,
"sampling/prob_max": 7.272460497915745e-05,
"sampling/prob_min": 2.047401649178937e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.3832000255584718,
"sampling/prompt_draws_total": 71496.0,
"sampling/seen_fraction": 0.971780002117157,
"sampling/unseen_fraction": 0.028219997882843018,
"signal/accuracy_reward/centered_abs_mean": 0.09169921875,
"signal/accuracy_reward/group_std_mean": 0.12476410120725631,
"signal/accuracy_reward/group_zero_std_frac": 0.6305555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.045849609375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.045849609375,
"signal/advantage_abs_mean": 0.06181447505950928,
"signal/advantage_pre_scale_abs_mean": 0.06181447505950928,
"signal/advantage_pre_scale_std": 0.13796453922986984,
"signal/advantage_std": 0.13796453922986984,
"signal/brier_reward/centered_abs_mean": 0.05219108313322067,
"signal/brier_reward/group_std_mean": 0.07250193729996682,
"signal/brier_reward/group_zero_std_frac": 0.37777777910232546,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026095541566610335,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026095541566610335,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004215494787786156,
"signal/format_reward/group_std_mean": 0.009954058565199375,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002107747393893078,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002107747393893078,
"signal/mean_confidence_reward/centered_abs_mean": 0.04885038137435913,
"signal/mean_confidence_reward/group_std_mean": 0.06549542620778084,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3861111104488373,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.885038038082712e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.885038038082712e-07,
"step": 995
},
{
"calibration/aurc": 0.12844745222568546,
"calibration/batch_distribution_entropy": 0.5532794175765451,
"calibration/batch_entropy_100bins": 0.2996957400473447,
"calibration/batch_entropy_10bins": 0.5532794175765451,
"calibration/batch_entropy_50bins": 0.35279697614728633,
"calibration/batch_uniqueness": 0.19261521024901582,
"calibration/confidence_entropy": 0.4318324728382888,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.0,
"calibration/coverage@10%": 0.46080953617760045,
"calibration/coverage@15%": 0.641783922338569,
"calibration/coverage@20%": 0.8033785994764397,
"calibration/coverage@25%": 0.9203125,
"calibration/coverage@30%": 0.9375,
"calibration/coverage@5%": 0.2,
"calibration/distribution_entropy_10": 0.5532794175765451,
"calibration/distribution_entropy_100": 0.2996957400473447,
"calibration/ece": 0.0948140619375373,
"calibration/mean_confidence": 0.7466430994798574,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.0026041666666666518,
"completions/max_length": 3874.0,
"completions/max_terminated_length": 3874.0,
"completions/mean_length": 1074.19775390625,
"completions/mean_terminated_length": 1077.0325561523437,
"completions/min_length": 0.0,
"completions/min_terminated_length": 293.0,
"epoch": 2.4038461538461537,
"grad_norm": 0.0006300617242231965,
"learning_rate": 2.403846153846154e-07,
"loss": -0.0012,
"num_tokens": 2752435048.0,
"reward": 1.3379584074020385,
"reward_std": 0.09457544982433319,
"rewards/accuracy_reward": 0.7941840291023254,
"rewards/brier_reward": 0.8844088673591614,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9973090171813965,
"rewards/mean_confidence_reward": 0.7415191173553467,
"sampling/batch_mean_priority_error": 0.03338872647389661,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05277777777777778,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.01267246250063181,
"sampling/priority_kl": 0.030000022053718566,
"sampling/priority_scale": 0.954604584001936,
"sampling/prob_entropy": 10.278950500488282,
"sampling/prob_max": 7.306546904146671e-05,
"sampling/prob_min": 2.0485483037191444e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.39520001411438,
"sampling/prompt_draws_total": 71856.0,
"sampling/seen_fraction": 0.9724133372306824,
"sampling/unseen_fraction": 0.02758666276931763,
"signal/accuracy_reward/centered_abs_mean": 0.10282660573720932,
"signal/accuracy_reward/group_std_mean": 0.14097290933132173,
"signal/accuracy_reward/group_zero_std_frac": 0.5833333373069763,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05141330286860466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05141330286860466,
"signal/advantage_abs_mean": 0.06629122346639633,
"signal/advantage_pre_scale_abs_mean": 0.06629122346639633,
"signal/advantage_pre_scale_std": 0.1459716185927391,
"signal/advantage_std": 0.1459716185927391,
"signal/brier_reward/centered_abs_mean": 0.057567541301250455,
"signal/brier_reward/group_std_mean": 0.0796251080930233,
"signal/brier_reward/group_zero_std_frac": 0.34999999701976775,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028783770650625227,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028783770650625227,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004985894146375358,
"signal/format_reward/group_std_mean": 0.010716632567346096,
"signal/format_reward/group_zero_std_frac": 0.9527777791023254,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.002492947073187679,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.002492947073187679,
"signal/mean_confidence_reward/centered_abs_mean": 0.051600489765405655,
"signal/mean_confidence_reward/group_std_mean": 0.06754546910524369,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3750000029802322,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.160048885954893e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.160048885954893e-07,
"step": 1000
},
{
"epoch": 2.4038461538461537,
"eval_calibration/aurc": 0.1184163209353608,
"eval_calibration/batch_distribution_entropy": 0.594126330909533,
"eval_calibration/batch_entropy_100bins": 0.3282082806303328,
"eval_calibration/batch_entropy_10bins": 0.594126330909533,
"eval_calibration/batch_entropy_50bins": 0.38636147759253836,
"eval_calibration/batch_uniqueness": 0.2892348036578806,
"eval_calibration/confidence_entropy": 0.4452448424888858,
"eval_calibration/coverage@0%": 0.0,
"eval_calibration/coverage@1%": 0.0,
"eval_calibration/coverage@10%": 0.6555944055944056,
"eval_calibration/coverage@15%": 0.7342657342657343,
"eval_calibration/coverage@20%": 0.8033216783216783,
"eval_calibration/coverage@25%": 0.9431818181818182,
"eval_calibration/coverage@30%": 1.0,
"eval_calibration/coverage@5%": 0.0,
"eval_calibration/distribution_entropy_10": 0.594126330909533,
"eval_calibration/distribution_entropy_100": 0.3282082806303328,
"eval_calibration/ece": 0.03129370629370616,
"eval_calibration/mean_confidence": 0.7288461538461539,
"eval_calibration/unique_confidence_per_question": 0.0078125,
"eval_calibration/unique_confidences": 9,
"eval_completions/clipped_ratio": 0.006076388888888895,
"eval_completions/max_length": 3072.1666666666665,
"eval_completions/max_terminated_length": 3072.1666666666665,
"eval_completions/mean_length": 1049.0632222493489,
"eval_completions/mean_terminated_length": 1055.494852701823,
"eval_completions/min_length": 114.0,
"eval_completions/min_terminated_length": 307.8333333333333,
"eval_loss": 0.0,
"eval_num_tokens": 2752435048.0,
"eval_reward": 1.2898423075675964,
"eval_reward_std": 0.3063166191180547,
"eval_rewards/accuracy_reward": 0.7282986044883728,
"eval_rewards/brier_reward": 0.8583159546057383,
"eval_rewards/confidence_one_or_zero": 0.0,
"eval_rewards/format_reward": 0.9930555621782938,
"eval_rewards/mean_confidence_reward": 0.7237847149372101,
"eval_runtime": 200.46,
"eval_samples_per_second": 4.989,
"eval_signal/accuracy_reward/centered_abs_mean": 0.3825412293275197,
"eval_signal/accuracy_reward/group_std_mean": 0.4418593893448512,
"eval_signal/accuracy_reward/group_zero_std_frac": 0.0,
"eval_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.19127061466375986,
"eval_signal/accuracy_reward/weight": 0.5,
"eval_signal/accuracy_reward/weighted_centered_abs_mean": 0.19127061466375986,
"eval_signal/advantage_abs_mean": 0.252433938284715,
"eval_signal/advantage_pre_scale_abs_mean": 0.252433938284715,
"eval_signal/advantage_pre_scale_std": 0.3051137725512187,
"eval_signal/advantage_std": 0.3051137725512187,
"eval_signal/brier_reward/centered_abs_mean": 0.15950087954600653,
"eval_signal/brier_reward/group_std_mean": 0.21895685543616614,
"eval_signal/brier_reward/group_zero_std_frac": 0.0,
"eval_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.07975043977300326,
"eval_signal/brier_reward/weight": 0.5,
"eval_signal/brier_reward/weighted_centered_abs_mean": 0.07975043977300326,
"eval_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_std_mean": 0.0,
"eval_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"eval_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"eval_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"eval_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"eval_signal/format_reward/centered_abs_mean": 0.013346354011446238,
"eval_signal/format_reward/group_std_mean": 0.0362943010404706,
"eval_signal/format_reward/group_zero_std_frac": 0.8055555820465088,
"eval_signal/format_reward/scaled_weighted_centered_abs_mean": 0.006673177005723119,
"eval_signal/format_reward/weight": 0.5,
"eval_signal/format_reward/weighted_centered_abs_mean": 0.006673177005723119,
"eval_signal/mean_confidence_reward/centered_abs_mean": 0.2063693404197693,
"eval_signal/mean_confidence_reward/group_std_mean": 0.24397027492523193,
"eval_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"eval_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 2.0636933678967275e-06,
"eval_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"eval_signal/mean_confidence_reward/weighted_centered_abs_mean": 2.0636933678967275e-06,
"eval_steps_per_second": 0.03,
"step": 1000
},
{
"epoch": 2.4038461538461537,
"step": 1000,
"train_probe_calibration/aurc": 0.0809334296456708,
"train_probe_calibration/batch_distribution_entropy": 0.5681220366191082,
"train_probe_calibration/batch_entropy_100bins": 0.3118720466633525,
"train_probe_calibration/batch_entropy_10bins": 0.5681220366191082,
"train_probe_calibration/batch_entropy_50bins": 0.3671307272846603,
"train_probe_calibration/batch_uniqueness": 0.2208597894839078,
"train_probe_calibration/confidence_entropy": 0.43153479627568575,
"train_probe_calibration/coverage@0%": 0.0,
"train_probe_calibration/coverage@1%": 0.0,
"train_probe_calibration/coverage@10%": 0.7735191637630662,
"train_probe_calibration/coverage@15%": 0.8327526132404182,
"train_probe_calibration/coverage@20%": 0.9416376306620209,
"train_probe_calibration/coverage@25%": 1.0,
"train_probe_calibration/coverage@30%": 1.0,
"train_probe_calibration/coverage@5%": 0.0,
"train_probe_calibration/distribution_entropy_10": 0.5681220366191082,
"train_probe_calibration/distribution_entropy_100": 0.3118720466633525,
"train_probe_calibration/ece": 0.042944250871079906,
"train_probe_calibration/mean_confidence": 0.7481707317073172,
"train_probe_calibration/unique_confidence_per_question": 0.0078125,
"train_probe_calibration/unique_confidences": 9,
"train_probe_completions/clipped_ratio": 0.006770833333333337,
"train_probe_completions/max_length": 2960.6666666666665,
"train_probe_completions/max_terminated_length": 2960.6666666666665,
"train_probe_completions/mean_length": 1050.498291015625,
"train_probe_completions/mean_terminated_length": 1057.6338500976562,
"train_probe_completions/min_length": 132.16666666666666,
"train_probe_completions/min_terminated_length": 261.5,
"train_probe_loss": 0.0,
"train_probe_num_tokens": 2752435048.0,
"train_probe_reward": 1.3347339828809102,
"train_probe_reward_std": 0.2743230958779653,
"train_probe_rewards/accuracy_reward": 0.7873263855775198,
"train_probe_rewards/brier_reward": 0.8855989575386047,
"train_probe_rewards/confidence_one_or_zero": 0.0,
"train_probe_rewards/format_reward": 0.996527781089147,
"train_probe_rewards/mean_confidence_reward": 0.7455729047457377,
"train_probe_runtime": 182.224,
"train_probe_samples_per_second": 5.488,
"train_probe_signal/accuracy_reward/centered_abs_mean": 0.32666015625,
"train_probe_signal/accuracy_reward/group_std_mean": 0.40729354818662006,
"train_probe_signal/accuracy_reward/group_zero_std_frac": 0.0,
"train_probe_signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.163330078125,
"train_probe_signal/accuracy_reward/weight": 0.5,
"train_probe_signal/accuracy_reward/weighted_centered_abs_mean": 0.163330078125,
"train_probe_signal/advantage_abs_mean": 0.21310583502054214,
"train_probe_signal/advantage_pre_scale_abs_mean": 0.21310583502054214,
"train_probe_signal/advantage_pre_scale_std": 0.2730792686343193,
"train_probe_signal/advantage_std": 0.2730792686343193,
"train_probe_signal/brier_reward/centered_abs_mean": 0.1362272302309672,
"train_probe_signal/brier_reward/group_std_mean": 0.19567894438902536,
"train_probe_signal/brier_reward/group_zero_std_frac": 0.0,
"train_probe_signal/brier_reward/scaled_weighted_centered_abs_mean": 0.0681136151154836,
"train_probe_signal/brier_reward/weight": 0.5,
"train_probe_signal/brier_reward/weighted_centered_abs_mean": 0.0681136151154836,
"train_probe_signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_std_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"train_probe_signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"train_probe_signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"train_probe_signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"train_probe_signal/format_reward/centered_abs_mean": 0.006727430348594983,
"train_probe_signal/format_reward/group_std_mean": 0.019641855110724766,
"train_probe_signal/format_reward/group_zero_std_frac": 0.8888889054457346,
"train_probe_signal/format_reward/scaled_weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/format_reward/weight": 0.5,
"train_probe_signal/format_reward/weighted_centered_abs_mean": 0.0033637151742974916,
"train_probe_signal/mean_confidence_reward/centered_abs_mean": 0.18942597756783167,
"train_probe_signal/mean_confidence_reward/group_std_mean": 0.2316800927122434,
"train_probe_signal/mean_confidence_reward/group_zero_std_frac": 0.0,
"train_probe_signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 1.8942597534987726e-06,
"train_probe_signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"train_probe_signal/mean_confidence_reward/weighted_centered_abs_mean": 1.8942597534987726e-06,
"train_probe_steps_per_second": 0.033
},
{
"calibration/aurc": 0.07527226356569491,
"calibration/batch_distribution_entropy": 0.5891084402195206,
"calibration/batch_entropy_100bins": 0.31662396913589846,
"calibration/batch_entropy_10bins": 0.5891084402195206,
"calibration/batch_entropy_50bins": 0.37272461353387987,
"calibration/batch_uniqueness": 0.3001689436576802,
"calibration/confidence_entropy": 0.44116588745456714,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.1296875,
"calibration/coverage@10%": 0.7139041099812193,
"calibration/coverage@15%": 0.7781115300490129,
"calibration/coverage@20%": 0.8766940789473685,
"calibration/coverage@25%": 0.9136732456140351,
"calibration/coverage@30%": 0.9921875,
"calibration/coverage@5%": 0.6627325677935046,
"calibration/distribution_entropy_10": 0.5891084402195206,
"calibration/distribution_entropy_100": 0.31662396913589846,
"calibration/ece": 0.1297788737918555,
"calibration/mean_confidence": 0.7363195441688426,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.002083333333333348,
"completions/max_length": 3804.0,
"completions/max_terminated_length": 3804.0,
"completions/mean_length": 1070.0533203125,
"completions/mean_terminated_length": 1072.345361328125,
"completions/min_length": 0.0,
"completions/min_terminated_length": 279.4,
"epoch": 2.4158653846153846,
"grad_norm": 0.0004999233642593026,
"learning_rate": 2.103365384615385e-07,
"loss": -0.001,
"num_tokens": 2767829198.0,
"reward": 1.324449324607849,
"reward_std": 0.08935754299163819,
"rewards/accuracy_reward": 0.764062511920929,
"rewards/brier_reward": 0.886904525756836,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166626930237,
"rewards/mean_confidence_reward": 0.7524132132530212,
"sampling/batch_mean_priority_error": 0.034781534330840844,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.061111111111111116,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.01270026620477438,
"sampling/priority_kl": 0.030000005289912222,
"sampling/priority_scale": 0.9584945619804784,
"sampling/prob_entropy": 10.278950881958007,
"sampling/prob_max": 7.34629575163126e-05,
"sampling/prob_min": 2.049046161118895e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.407200002670288,
"sampling/prompt_draws_total": 72216.0,
"sampling/seen_fraction": 0.973146665096283,
"sampling/unseen_fraction": 0.02685333490371704,
"signal/accuracy_reward/centered_abs_mean": 0.09141709953546524,
"signal/accuracy_reward/group_std_mean": 0.12482106983661652,
"signal/accuracy_reward/group_zero_std_frac": 0.6277778029441834,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04570854976773262,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04570854976773262,
"signal/advantage_abs_mean": 0.0623251885175705,
"signal/advantage_pre_scale_abs_mean": 0.0623251885175705,
"signal/advantage_pre_scale_std": 0.13974119275808333,
"signal/advantage_std": 0.13974119275808333,
"signal/brier_reward/centered_abs_mean": 0.053202150762081145,
"signal/brier_reward/group_std_mean": 0.07416063845157624,
"signal/brier_reward/group_zero_std_frac": 0.36111111044883726,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026601075381040572,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026601075381040572,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.003982204757630825,
"signal/format_reward/group_std_mean": 0.010536715574562549,
"signal/format_reward/group_zero_std_frac": 0.9444444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019911023788154124,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019911023788154124,
"signal/mean_confidence_reward/centered_abs_mean": 0.04939617142081261,
"signal/mean_confidence_reward/group_std_mean": 0.06459873989224434,
"signal/mean_confidence_reward/group_zero_std_frac": 0.39166667461395266,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.93961704250978e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.93961704250978e-07,
"step": 1005
},
{
"calibration/aurc": 0.19523541298453273,
"calibration/batch_distribution_entropy": 0.5816421537354308,
"calibration/batch_entropy_100bins": 0.3243447138501555,
"calibration/batch_entropy_10bins": 0.5816421537354308,
"calibration/batch_entropy_50bins": 0.3818133492909004,
"calibration/batch_uniqueness": 0.3151588996974953,
"calibration/confidence_entropy": 0.44468967957879874,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.11811023622047243,
"calibration/coverage@10%": 0.33394438976377955,
"calibration/coverage@15%": 0.5759404039287843,
"calibration/coverage@20%": 0.6120143628831644,
"calibration/coverage@25%": 0.627717282830671,
"calibration/coverage@30%": 0.6991380835269285,
"calibration/coverage@5%": 0.13175853018372702,
"calibration/distribution_entropy_10": 0.5816421537354308,
"calibration/distribution_entropy_100": 0.3243447138501555,
"calibration/ece": 0.10923380395825207,
"calibration/mean_confidence": 0.7094446064876683,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.00269097222222221,
"completions/max_length": 3780.2,
"completions/max_terminated_length": 3780.2,
"completions/mean_length": 1046.7429077148438,
"completions/mean_terminated_length": 1049.6477905273437,
"completions/min_length": 40.8,
"completions/min_terminated_length": 246.4,
"epoch": 2.4278846153846154,
"grad_norm": 0.0004883616347797215,
"learning_rate": 1.8028846153846153e-07,
"loss": -0.0024,
"num_tokens": 2782993468.0,
"reward": 1.3204421997070312,
"reward_std": 0.08243466466665268,
"rewards/accuracy_reward": 0.7594618082046509,
"rewards/brier_reward": 0.8840989708900452,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9973090291023254,
"rewards/mean_confidence_reward": 0.7330034852027894,
"sampling/batch_mean_priority_error": 0.035588923942414356,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.08333333333333334,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012730823084712028,
"sampling/priority_kl": 0.0299996517598629,
"sampling/priority_scale": 0.9637907325522974,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 7.395966531476006e-05,
"sampling/prob_min": 2.0483955813688227e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.419200038909912,
"sampling/prompt_draws_total": 72576.0,
"sampling/seen_fraction": 0.9740266680717469,
"sampling/unseen_fraction": 0.025973331928253175,
"signal/accuracy_reward/centered_abs_mean": 0.07844509705901145,
"signal/accuracy_reward/group_std_mean": 0.11526311635971069,
"signal/accuracy_reward/group_zero_std_frac": 0.6250000119209289,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.03922254852950573,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.03922254852950573,
"signal/advantage_abs_mean": 0.05439499020576477,
"signal/advantage_pre_scale_abs_mean": 0.05439499020576477,
"signal/advantage_pre_scale_std": 0.13072411566972733,
"signal/advantage_std": 0.13072411566972733,
"signal/brier_reward/centered_abs_mean": 0.051925729215145114,
"signal/brier_reward/group_std_mean": 0.07489459365606307,
"signal/brier_reward/group_zero_std_frac": 0.3111111164093018,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.025962864607572557,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.025962864607572557,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.004909939365461469,
"signal/format_reward/group_std_mean": 0.010254300013184548,
"signal/format_reward/group_zero_std_frac": 0.955555546283722,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0024549696827307343,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0024549696827307343,
"signal/mean_confidence_reward/centered_abs_mean": 0.04594618603587151,
"signal/mean_confidence_reward/group_std_mean": 0.0627430409193039,
"signal/mean_confidence_reward/group_zero_std_frac": 0.33611111640930175,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.594618360442837e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.594618360442837e-07,
"step": 1010
},
{
"calibration/aurc": 0.1210457860940937,
"calibration/batch_distribution_entropy": 0.5332891838068747,
"calibration/batch_entropy_100bins": 0.29667083825166946,
"calibration/batch_entropy_10bins": 0.5332891838068747,
"calibration/batch_entropy_50bins": 0.3492361106959184,
"calibration/batch_uniqueness": 0.18285195780705218,
"calibration/confidence_entropy": 0.4390206368622095,
"calibration/coverage@0%": 0.2857715948651001,
"calibration/coverage@1%": 0.3003930047867711,
"calibration/coverage@10%": 0.5895207789382071,
"calibration/coverage@15%": 0.6858749456048738,
"calibration/coverage@20%": 0.7202839425587467,
"calibration/coverage@25%": 0.8505208333333332,
"calibration/coverage@30%": 0.8833333333333334,
"calibration/coverage@5%": 0.31761450174064404,
"calibration/distribution_entropy_10": 0.5332891838068747,
"calibration/distribution_entropy_100": 0.29667083825166946,
"calibration/ece": 0.12505969865100078,
"calibration/mean_confidence": 0.7221615263272413,
"calibration/unique_confidence_per_question": 0.021354166666666667,
"calibration/unique_confidences": 8.2,
"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": 0.004513888888888906,
"completions/max_length": 3834.4,
"completions/max_terminated_length": 3834.4,
"completions/mean_length": 1057.0390014648438,
"completions/mean_terminated_length": 1061.9672607421876,
"completions/min_length": 0.0,
"completions/min_terminated_length": 265.4,
"epoch": 2.4399038461538463,
"grad_norm": 0.0005890694446861744,
"learning_rate": 1.5024038461538462e-07,
"loss": -0.0026,
"num_tokens": 2798279613.0,
"reward": 1.3154569149017334,
"reward_std": 0.09527151137590409,
"rewards/accuracy_reward": 0.7602430701255798,
"rewards/brier_reward": 0.8751701235771179,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9954861044883728,
"rewards/mean_confidence_reward": 0.7243923425674439,
"sampling/batch_mean_priority_error": 0.03677215232250921,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.061111111111111116,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012759432010352612,
"sampling/priority_kl": 0.030000021681189537,
"sampling/priority_scale": 0.9682258547050878,
"sampling/prob_entropy": 10.278954124450683,
"sampling/prob_max": 7.439958862960338e-05,
"sampling/prob_min": 2.0484432025114074e-05,
"sampling/prompt_draws_max": 8.0,
"sampling/prompt_draws_mean": 2.4312000274658203,
"sampling/prompt_draws_total": 72936.0,
"sampling/seen_fraction": 0.9747933387756348,
"sampling/unseen_fraction": 0.025206661224365233,
"signal/accuracy_reward/centered_abs_mean": 0.1037109375,
"signal/accuracy_reward/group_std_mean": 0.14046219140291213,
"signal/accuracy_reward/group_zero_std_frac": 0.5888888955116272,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.05185546875,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.05185546875,
"signal/advantage_abs_mean": 0.0689443327486515,
"signal/advantage_pre_scale_abs_mean": 0.0689443327486515,
"signal/advantage_pre_scale_std": 0.14764165878295898,
"signal/advantage_std": 0.14764165878295898,
"signal/brier_reward/centered_abs_mean": 0.05841308310627937,
"signal/brier_reward/group_std_mean": 0.08001031428575515,
"signal/brier_reward/group_zero_std_frac": 0.3361111104488373,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.029206541553139686,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.029206541553139686,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.007063802098855376,
"signal/format_reward/group_std_mean": 0.013031124509871006,
"signal/format_reward/group_zero_std_frac": 0.9444444417953491,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.003531901049427688,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.003531901049427688,
"signal/mean_confidence_reward/centered_abs_mean": 0.04989747107028961,
"signal/mean_confidence_reward/group_std_mean": 0.06642280519008636,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3444444477558136,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.989746742012358e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.989746742012358e-07,
"step": 1015
},
{
"calibration/aurc": 0.08441513061092562,
"calibration/batch_distribution_entropy": 0.5946078369083214,
"calibration/batch_entropy_100bins": 0.331938317642435,
"calibration/batch_entropy_10bins": 0.5946078369083214,
"calibration/batch_entropy_50bins": 0.3907524168116919,
"calibration/batch_uniqueness": 0.33384874131944436,
"calibration/confidence_entropy": 0.45210384409773424,
"calibration/coverage@0%": 0.31875,
"calibration/coverage@1%": 0.31875,
"calibration/coverage@10%": 0.7119791666666667,
"calibration/coverage@15%": 0.7380208333333333,
"calibration/coverage@20%": 0.8348958333333334,
"calibration/coverage@25%": 0.8619791666666666,
"calibration/coverage@30%": 0.9552083333333332,
"calibration/coverage@5%": 0.6125,
"calibration/distribution_entropy_10": 0.5946078369083214,
"calibration/distribution_entropy_100": 0.331938317642435,
"calibration/ece": 0.13171874999999988,
"calibration/mean_confidence": 0.6949479166666668,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002083333333333326,
"completions/max_length": 3989.0,
"completions/max_terminated_length": 3989.0,
"completions/mean_length": 1093.742724609375,
"completions/mean_terminated_length": 1096.0311279296875,
"completions/min_length": 0.0,
"completions/min_terminated_length": 257.6,
"epoch": 2.451923076923077,
"grad_norm": 0.0004692086367867887,
"learning_rate": 1.201923076923077e-07,
"loss": -0.0003,
"num_tokens": 2814016169.0,
"reward": 1.3153839826583862,
"reward_std": 0.08586094528436661,
"rewards/accuracy_reward": 0.7405381917953491,
"rewards/brier_reward": 0.8923853874206543,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9978298664093017,
"rewards/mean_confidence_reward": 0.7217013955116272,
"sampling/batch_mean_priority_error": 0.0273770647804218,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012783795781433582,
"sampling/priority_kl": 0.02999986857175827,
"sampling/priority_scale": 0.97209588882979,
"sampling/prob_entropy": 10.278952407836915,
"sampling/prob_max": 7.480340282199905e-05,
"sampling/prob_min": 2.0359152040327898e-05,
"sampling/prompt_draws_max": 8.4,
"sampling/prompt_draws_mean": 2.4432000160217284,
"sampling/prompt_draws_total": 73296.0,
"sampling/seen_fraction": 0.9754666566848755,
"sampling/unseen_fraction": 0.024533343315124512,
"signal/accuracy_reward/centered_abs_mean": 0.084326171875,
"signal/accuracy_reward/group_std_mean": 0.12386898547410966,
"signal/accuracy_reward/group_zero_std_frac": 0.600000011920929,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.0421630859375,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.0421630859375,
"signal/advantage_abs_mean": 0.057266923040151595,
"signal/advantage_pre_scale_abs_mean": 0.057266923040151595,
"signal/advantage_pre_scale_std": 0.13469514846801758,
"signal/advantage_std": 0.13469514846801758,
"signal/brier_reward/centered_abs_mean": 0.05296983122825623,
"signal/brier_reward/group_std_mean": 0.07545235157012939,
"signal/brier_reward/group_zero_std_frac": 0.2972222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.026484915614128114,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.026484915614128114,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.003965928766410798,
"signal/format_reward/group_std_mean": 0.00823423620313406,
"signal/format_reward/group_zero_std_frac": 0.9638888835906982,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.001982964383205399,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.001982964383205399,
"signal/mean_confidence_reward/centered_abs_mean": 0.050093865394592284,
"signal/mean_confidence_reward/group_std_mean": 0.0671265460550785,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3194444507360458,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.009386313759023e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.009386313759023e-07,
"step": 1020
},
{
"calibration/aurc": 0.09942333186485555,
"calibration/batch_distribution_entropy": 0.553478276317741,
"calibration/batch_entropy_100bins": 0.30524244330208805,
"calibration/batch_entropy_10bins": 0.553478276317741,
"calibration/batch_entropy_50bins": 0.3593264654738634,
"calibration/batch_uniqueness": 0.23467320477290174,
"calibration/confidence_entropy": 0.4469483642342733,
"calibration/coverage@0%": 0.0,
"calibration/coverage@1%": 0.22552083333333334,
"calibration/coverage@10%": 0.590405701754386,
"calibration/coverage@15%": 0.7170581886653806,
"calibration/coverage@20%": 0.8068247393680537,
"calibration/coverage@25%": 0.9029934784605494,
"calibration/coverage@30%": 0.9449287280701754,
"calibration/coverage@5%": 0.5260197368421052,
"calibration/distribution_entropy_10": 0.553478276317741,
"calibration/distribution_entropy_100": 0.30524244330208805,
"calibration/ece": 0.12965119236245054,
"calibration/mean_confidence": 0.7391239178607514,
"calibration/unique_confidence_per_question": 0.021875,
"calibration/unique_confidences": 8.4,
"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": 0.0028645833333333258,
"completions/max_length": 3859.4,
"completions/max_terminated_length": 3859.4,
"completions/mean_length": 1073.966015625,
"completions/mean_terminated_length": 1077.1373291015625,
"completions/min_length": 54.4,
"completions/min_terminated_length": 296.8,
"epoch": 2.4639423076923075,
"grad_norm": 0.0006311958422884345,
"learning_rate": 9.014423076923076e-08,
"loss": -0.0027,
"num_tokens": 2829481569.0,
"reward": 1.3142542600631715,
"reward_std": 0.0905083417892456,
"rewards/accuracy_reward": 0.7552083373069763,
"rewards/brier_reward": 0.8762369632720948,
"rewards/confidence_one_or_zero": 8.680555620230735e-05,
"rewards/format_reward": 0.9970486164093018,
"rewards/mean_confidence_reward": 0.7326996445655822,
"sampling/batch_mean_priority_error": 0.036499032662617364,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.04722222222222223,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012810037657618522,
"sampling/priority_kl": 0.030000165849924088,
"sampling/priority_scale": 0.9763287365669384,
"sampling/prob_entropy": 10.278953170776367,
"sampling/prob_max": 7.523820822825655e-05,
"sampling/prob_min": 2.0165404202998614e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.4552000522613526,
"sampling/prompt_draws_total": 73656.0,
"sampling/seen_fraction": 0.9761400103569031,
"sampling/unseen_fraction": 0.023859989643096925,
"signal/accuracy_reward/centered_abs_mean": 0.09184027761220932,
"signal/accuracy_reward/group_std_mean": 0.12902664840221406,
"signal/accuracy_reward/group_zero_std_frac": 0.6055555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04592013880610466,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04592013880610466,
"signal/advantage_abs_mean": 0.06400088593363762,
"signal/advantage_pre_scale_abs_mean": 0.06400088593363762,
"signal/advantage_pre_scale_std": 0.14083140045404435,
"signal/advantage_std": 0.14083140045404435,
"signal/brier_reward/centered_abs_mean": 0.05662917345762253,
"signal/brier_reward/group_std_mean": 0.07765991985797882,
"signal/brier_reward/group_zero_std_frac": 0.3083333373069763,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028314586728811264,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028314586728811264,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0001681857625953853,
"signal/confidence_one_or_zero/group_std_mean": 0.0004910463467240334,
"signal/confidence_one_or_zero/group_zero_std_frac": 0.9972222208976745,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 1.6818574977151001e-09,
"signal/format_reward/centered_abs_mean": 0.0051757812732830645,
"signal/format_reward/group_std_mean": 0.009404184995219111,
"signal/format_reward/group_zero_std_frac": 0.9638888955116272,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0025878906366415323,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0025878906366415323,
"signal/mean_confidence_reward/centered_abs_mean": 0.05497342348098755,
"signal/mean_confidence_reward/group_std_mean": 0.07288424968719483,
"signal/mean_confidence_reward/group_zero_std_frac": 0.31666666865348814,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.497342044691323e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.497342044691323e-07,
"step": 1025
},
{
"calibration/aurc": 0.19010450098037168,
"calibration/batch_distribution_entropy": 0.5954675679330762,
"calibration/batch_entropy_100bins": 0.33528836429101044,
"calibration/batch_entropy_10bins": 0.5954675679330762,
"calibration/batch_entropy_50bins": 0.3946960375230943,
"calibration/batch_uniqueness": 0.3225400975188956,
"calibration/confidence_entropy": 0.44998980769433705,
"calibration/coverage@0%": 0.09166666666666666,
"calibration/coverage@1%": 0.09166666666666666,
"calibration/coverage@10%": 0.20758850043975374,
"calibration/coverage@15%": 0.5856291879941489,
"calibration/coverage@20%": 0.6444802186885005,
"calibration/coverage@25%": 0.7455177626638744,
"calibration/coverage@30%": 0.7857519788918206,
"calibration/coverage@5%": 0.19092183377308708,
"calibration/distribution_entropy_10": 0.5954675679330762,
"calibration/distribution_entropy_100": 0.33528836429101044,
"calibration/ece": 0.11791738083214551,
"calibration/mean_confidence": 0.6998125620962554,
"calibration/unique_confidence_per_question": 0.022916666666666665,
"calibration/unique_confidences": 8.8,
"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": 0.002083333333333348,
"completions/max_length": 3790.8,
"completions/max_terminated_length": 3790.8,
"completions/mean_length": 1066.6471557617188,
"completions/mean_terminated_length": 1068.9392211914062,
"completions/min_length": 63.0,
"completions/min_terminated_length": 250.0,
"epoch": 2.4759615384615383,
"grad_norm": 0.0005696768057532609,
"learning_rate": 6.009615384615386e-08,
"loss": -0.0005,
"num_tokens": 2844893344.0,
"reward": 1.3270651817321777,
"reward_std": 0.08265040963888168,
"rewards/accuracy_reward": 0.7684027791023255,
"rewards/brier_reward": 0.8877960085868836,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9979166626930237,
"rewards/mean_confidence_reward": 0.7462586641311646,
"sampling/batch_mean_priority_error": 0.03507078950753996,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.03888888888888889,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.01284012570977211,
"sampling/priority_kl": 0.030000190064311028,
"sampling/priority_scale": 0.9783201873069629,
"sampling/prob_entropy": 10.278951263427734,
"sampling/prob_max": 7.551889138994739e-05,
"sampling/prob_min": 2.01863469555974e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.467199945449829,
"sampling/prompt_draws_total": 74016.0,
"sampling/seen_fraction": 0.9765199780464172,
"sampling/unseen_fraction": 0.023480021953582765,
"signal/accuracy_reward/centered_abs_mean": 0.08758680522441864,
"signal/accuracy_reward/group_std_mean": 0.12149299085140228,
"signal/accuracy_reward/group_zero_std_frac": 0.6305555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04379340261220932,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04379340261220932,
"signal/advantage_abs_mean": 0.057532144337892534,
"signal/advantage_pre_scale_abs_mean": 0.057532144337892534,
"signal/advantage_pre_scale_std": 0.13342130929231644,
"signal/advantage_std": 0.13342130929231644,
"signal/brier_reward/centered_abs_mean": 0.05133935660123825,
"signal/brier_reward/group_std_mean": 0.07152377292513848,
"signal/brier_reward/group_zero_std_frac": 0.39166667461395266,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.025669678300619125,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.025669678300619125,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0038519965251907705,
"signal/format_reward/group_std_mean": 0.00847507119178772,
"signal/format_reward/group_zero_std_frac": 0.9611111044883728,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0019259982625953853,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0019259982625953853,
"signal/mean_confidence_reward/centered_abs_mean": 0.04585125148296356,
"signal/mean_confidence_reward/group_std_mean": 0.06037860438227653,
"signal/mean_confidence_reward/group_zero_std_frac": 0.4138888895511627,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.585124941058893e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.585124941058893e-07,
"step": 1030
},
{
"calibration/aurc": 0.07878366204318273,
"calibration/batch_distribution_entropy": 0.5279258673150643,
"calibration/batch_entropy_100bins": 0.2894325308041625,
"calibration/batch_entropy_10bins": 0.5279258673150643,
"calibration/batch_entropy_50bins": 0.34071529228354647,
"calibration/batch_uniqueness": 0.16018399170898742,
"calibration/confidence_entropy": 0.4365498270125993,
"calibration/coverage@0%": 0.10364583333333333,
"calibration/coverage@1%": 0.2505208333333333,
"calibration/coverage@10%": 0.6958333333333334,
"calibration/coverage@15%": 0.8461184210526316,
"calibration/coverage@20%": 0.8843311403508773,
"calibration/coverage@25%": 0.8968311403508773,
"calibration/coverage@30%": 0.9382401315789475,
"calibration/coverage@5%": 0.5854166666666667,
"calibration/distribution_entropy_10": 0.5279258673150643,
"calibration/distribution_entropy_100": 0.2894325308041625,
"calibration/ece": 0.123923793859649,
"calibration/mean_confidence": 0.7415060307017545,
"calibration/unique_confidence_per_question": 0.02239583333333333,
"calibration/unique_confidences": 8.6,
"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": 0.002517361111111094,
"completions/max_length": 3987.2,
"completions/max_terminated_length": 3987.2,
"completions/mean_length": 1090.0648681640625,
"completions/mean_terminated_length": 1092.80859375,
"completions/min_length": 0.0,
"completions/min_terminated_length": 243.4,
"epoch": 2.487980769230769,
"grad_norm": 0.0004971089074388146,
"learning_rate": 3.004807692307693e-08,
"loss": -0.0015,
"num_tokens": 2860511307.0,
"reward": 1.3260030269622802,
"reward_std": 0.09149017930030823,
"rewards/accuracy_reward": 0.7737847208976746,
"rewards/brier_reward": 0.8809843420982361,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9972222208976745,
"rewards/mean_confidence_reward": 0.7359982609748841,
"sampling/batch_mean_priority_error": 0.03327463871940677,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.06388888888888888,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012865498848259448,
"sampling/priority_kl": 0.030000067502260207,
"sampling/priority_scale": 0.9826615870231763,
"sampling/prob_entropy": 10.278948020935058,
"sampling/prob_max": 7.595972710987553e-05,
"sampling/prob_min": 2.0185628454783e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.479199981689453,
"sampling/prompt_draws_total": 74376.0,
"sampling/seen_fraction": 0.9772266626358033,
"sampling/unseen_fraction": 0.022773337364196778,
"signal/accuracy_reward/centered_abs_mean": 0.08817274421453476,
"signal/accuracy_reward/group_std_mean": 0.12668398320674895,
"signal/accuracy_reward/group_zero_std_frac": 0.5944444477558136,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04408637210726738,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04408637210726738,
"signal/advantage_abs_mean": 0.06206140071153641,
"signal/advantage_pre_scale_abs_mean": 0.06206140071153641,
"signal/advantage_pre_scale_std": 0.14209077060222625,
"signal/advantage_std": 0.14209077060222625,
"signal/brier_reward/centered_abs_mean": 0.057836859673261645,
"signal/brier_reward/group_std_mean": 0.0815641537308693,
"signal/brier_reward/group_zero_std_frac": 0.32777778804302216,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.028918429836630823,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.028918429836630823,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.00514322918606922,
"signal/format_reward/group_std_mean": 0.011155045032501221,
"signal/format_reward/group_zero_std_frac": 0.95,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.00257161459303461,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.00257161459303461,
"signal/mean_confidence_reward/centered_abs_mean": 0.05260037705302238,
"signal/mean_confidence_reward/group_std_mean": 0.07008446007966995,
"signal/mean_confidence_reward/group_zero_std_frac": 0.3527777850627899,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 5.260037369225756e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 5.260037369225756e-07,
"step": 1035
},
{
"calibration/aurc": 0.1045495803451458,
"calibration/batch_distribution_entropy": 0.5107908600789777,
"calibration/batch_entropy_100bins": 0.27355090882409727,
"calibration/batch_entropy_10bins": 0.5107908600789777,
"calibration/batch_entropy_50bins": 0.3220197038511045,
"calibration/batch_uniqueness": 0.11296319207501355,
"calibration/confidence_entropy": 0.42644262978301706,
"calibration/coverage@0%": 0.09347258485639687,
"calibration/coverage@1%": 0.25805591818973017,
"calibration/coverage@10%": 0.5873449738903395,
"calibration/coverage@15%": 0.7735653285465622,
"calibration/coverage@20%": 0.8299037206266318,
"calibration/coverage@25%": 0.932186956048738,
"calibration/coverage@30%": 1.0,
"calibration/coverage@5%": 0.29720001087902526,
"calibration/distribution_entropy_10": 0.5107908600789777,
"calibration/distribution_entropy_100": 0.27355090882409727,
"calibration/ece": 0.08730907310704951,
"calibration/mean_confidence": 0.7545944843342038,
"calibration/unique_confidence_per_question": 0.020833333333333336,
"calibration/unique_confidences": 8.0,
"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": 0.0019097222222222098,
"completions/max_length": 3708.6,
"completions/max_terminated_length": 3708.6,
"completions/mean_length": 1063.2446166992188,
"completions/mean_terminated_length": 1065.3385620117188,
"completions/min_length": 61.0,
"completions/min_terminated_length": 245.8,
"epoch": 2.5,
"grad_norm": 0.0005916485097259283,
"learning_rate": 0.0,
"loss": 0.0,
"num_tokens": 2875820461.0,
"reward": 1.3269983053207397,
"reward_std": 0.08716974556446075,
"rewards/accuracy_reward": 0.7735243082046509,
"rewards/brier_reward": 0.8823671817779541,
"rewards/confidence_one_or_zero": 0.0,
"rewards/format_reward": 0.9980902791023254,
"rewards/mean_confidence_reward": 0.7451302289962769,
"sampling/batch_mean_priority_error": 0.03641648332159923,
"sampling/batch_prompt_draws": 72.0,
"sampling/batch_unique_prompts": 72.0,
"sampling/batch_unseen_draw_fraction": 0.05,
"sampling/error_ema_max": 0.27569612860679626,
"sampling/error_ema_mean": 0.012894181534647942,
"sampling/priority_kl": 0.030000076070427893,
"sampling/priority_scale": 0.9869442045921459,
"sampling/prob_entropy": 10.27895736694336,
"sampling/prob_max": 7.640240946784615e-05,
"sampling/prob_min": 2.0186353867757135e-05,
"sampling/prompt_draws_max": 9.0,
"sampling/prompt_draws_mean": 2.4911999702453613,
"sampling/prompt_draws_total": 74736.0,
"sampling/seen_fraction": 0.9778599977493286,
"sampling/unseen_fraction": 0.022140002250671385,
"signal/accuracy_reward/centered_abs_mean": 0.08924153745174408,
"signal/accuracy_reward/group_std_mean": 0.12243999242782592,
"signal/accuracy_reward/group_zero_std_frac": 0.6305555582046509,
"signal/accuracy_reward/scaled_weighted_centered_abs_mean": 0.04462076872587204,
"signal/accuracy_reward/weight": 0.5,
"signal/accuracy_reward/weighted_centered_abs_mean": 0.04462076872587204,
"signal/advantage_abs_mean": 0.061470860987901686,
"signal/advantage_pre_scale_abs_mean": 0.061470860987901686,
"signal/advantage_pre_scale_std": 0.1358793631196022,
"signal/advantage_std": 0.1358793631196022,
"signal/brier_reward/centered_abs_mean": 0.05750547796487808,
"signal/brier_reward/group_std_mean": 0.07936549633741379,
"signal/brier_reward/group_zero_std_frac": 0.3472222238779068,
"signal/brier_reward/scaled_weighted_centered_abs_mean": 0.02875273898243904,
"signal/brier_reward/weight": 0.5,
"signal/brier_reward/weighted_centered_abs_mean": 0.02875273898243904,
"signal/confidence_one_or_zero/centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/group_std_mean": 0.0,
"signal/confidence_one_or_zero/group_zero_std_frac": 1.0,
"signal/confidence_one_or_zero/scaled_weighted_centered_abs_mean": 0.0,
"signal/confidence_one_or_zero/weight": 9.999999747378752e-06,
"signal/confidence_one_or_zero/weighted_centered_abs_mean": 0.0,
"signal/format_reward/centered_abs_mean": 0.0035590278683230283,
"signal/format_reward/group_std_mean": 0.008442434668540954,
"signal/format_reward/group_zero_std_frac": 0.9583333492279053,
"signal/format_reward/scaled_weighted_centered_abs_mean": 0.0017795139341615141,
"signal/format_reward/weight": 0.5,
"signal/format_reward/weighted_centered_abs_mean": 0.0017795139341615141,
"signal/mean_confidence_reward/centered_abs_mean": 0.049709752947092053,
"signal/mean_confidence_reward/group_std_mean": 0.06713559478521347,
"signal/mean_confidence_reward/group_zero_std_frac": 0.35277778208255767,
"signal/mean_confidence_reward/scaled_weighted_centered_abs_mean": 4.970974998741439e-07,
"signal/mean_confidence_reward/weight": 9.999999747378752e-06,
"signal/mean_confidence_reward/weighted_centered_abs_mean": 4.970974998741439e-07,
"step": 1040
},
{
"epoch": 2.5,
"step": 1040,
"total_flos": 0.0,
"train_loss": -0.0043364950483123724,
"train_runtime": 239563.7885,
"train_samples_per_second": 0.313,
"train_steps_per_second": 0.004
}
],
"logging_steps": 5,
"max_steps": 1040,
"num_input_tokens_seen": 2875820461,
"num_train_epochs": 3,
"save_steps": 60,
"stateful_callbacks": {
"TrainerControl": {
"args": {
"should_epoch_stop": false,
"should_evaluate": false,
"should_log": false,
"should_save": true,
"should_training_stop": true
},
"attributes": {}
}
},
"total_flos": 0.0,
"train_batch_size": 6,
"trial_name": null,
"trial_params": null
}